Content uploaded by Aidan Meyler
Author content
All content in this area was uploaded by Aidan Meyler on Nov 29, 2021
Content may be subject to copyright.
Occasional Paper Series
Inflation expectations and their role
in Eurosystem forecasting
Work stream on inflation expectations
No 264 / September 2021
Disclaimer: This paper constitutes staff input into the Governing Council’s deliberation in the context of the ECB’s
monetary policy strategy review. This paper should not be reported as representing the views of the Eurosystem.
The views expressed are those of the authors and do not necessarily reflect those of the Eurosystem.
Acknowledgements
This report has been jointly produced by the Eurosystem work stream on inflation expectations, comprising staff from the European
Central Bank (ECB) and the national central banks (NCBs) of the EU Member States. This work stream drew on a pre-existing network of
experts on inflation expectations that preceded the strategy review. The objectives of that expert group were (i) to review the nature and
behaviour of inflation expectations, with a focus on the degree of anchoring, and (ii) to explore the role that measures of expectations can
play in forecasting inflation. The report fed into the Governing Council’s deliberations on the monetary policy strategy review 2020-21.
Group co-chairs
Ursel Baumann
European Central Bank
Matthieu Darracq Paries
European Central Bank
Thomas Westermann
European Central Bank
Marianna Riggi
Banca d’Italia
Report coordinators
Elena Bobeica
European Central Bank
Aidan Meyler
European Central Bank
Benjamin Böninghausen
European Central Bank
Additional contributing authors
Friedrich Fritzer
Oesterreichische Nationalbank
Riccardo Trezzi
European Central Bank
Jana Jonckheere
Nationale Bank van België/Banque Nationale de Belgique
Dmitry Kulikov
Eesti Pank
Dilyana Popova
Българска народна банка (Bulgarian National Bank)
Sulev Pert
Eesti Pank
Nektarios Michail
Central Bank of Cyprus
Maritta Paloviita
Suomen Pankki – Finlands Bank
František Brázdik
Česká národní banka
Harri Pönkä
Suomen Pankki – Finlands Bank
Mikkel Bess
Danmarks Nationalbank
Lauri Vilmi
Suomen Pankki – Finlands Bank
Casper Jørgensen
Danmarks Nationalbank
Pierre-Antoine Robert
Banque de France
Alexander Al-Haschimi
European Central Bank
Philipp Gmehling
Deutsche Bundesbank
Marta Bańbura
European Central Bank
Matthias Hartmann
Deutsche Bundesbank
Evangelos Charalampakis
European Central Bank
Jan-Oliver Menz
Deutsche Bundesbank
Benny Hartwig
European Central Bank
Fabian Schupp
Deutsche Bundesbank
John Hutchinson
European Central Bank
Christian Speck
Deutsche Bundesbank
Joan Paredes
European Central Bank
Ute Volz
Deutsche Bundesbank
Lovisa Reiche
European Central Bank
Zacharias Bragoudakis
Bank of Greece
Marcel Tirpák
European Central Bank
Evangelia Kasimati
Bank of Greece
Veronika Tengely
Magyar Nemzeti Bank
Tomasz Łyziak
Narodowy Bank Polski
Alex Tagliabracci
Banca d’Italia
Ewa Stanisławska
Narodowy Bank Polski
Andrejs Bessonovs
Latvijas Banka
Nikolay Iskrev
Banco de Portugal
Olegs Krasnopjorovs
Latvijas Banka
Miroslav Gavura
Národná banka Slovenska
Tomas Reichenbachas
Lietuvos bankas
Milan Damjanović
Banka Slovenije
Roberta Colavecchio
Banque centrale du Luxembourg
Matjaz Maletic
Banka Slovenije
Gabriele Galati
De Nederlandsche Bank
Danilo Leiva-León
Banco de España
Ide Kearney
De Nederlandsche Bank
Pär Stockhammar
Sveriges Riksbank
This report is part of a set of papers within the ECB’s Occasional Paper Series, related to the ECB’s Strategy review 2020-21. This set
includes the following papers:
Set of Occasional Papers related to the ECB’s Strategy review 2020-21
No 263, “The implications of globalisation for the ECB monetary policy strategy”.
No 264, “Inflation expectations and their role in Eurosystem forecasting”.
No 265, “Inflation measurement and its assessment in the ECB’s monetary policy strategy review”.
No 266, “Digitalisation: channels, impacts and implications for monetary policy in the euro area”.
No 267, “Review of macroeconomic modelling in the Eurosystem: current practices and scope for improvement”.
No 268, “Key factors behind productivity trends in EU countries”.
No 269, “The ECB’s price stability framework: past experience, and current and future challenges”.
No 270, “Non-bank financial intermediation in the euro area: implications for monetary policy transmission and key vulnerabilities”.
No 271, “Climate change and monetary policy in the euro area”.
No 272, “The role of financial stability considerations in monetary policy and the interaction with macroprudential policy in the euro area”.
No 273, “Monetary-fiscal policy interactions in the euro area”.
No 274, “Clear, consistent and engaging: ECB monetary policy communication in a changing world”.
No 275, “Employment and the conduct of monetary policy in the euro area”.
No 276, “The mandate of the ECB: Legal considerations in the ECB’s monetary policy strategy review”.
No 277, “Evolution of the ECB’s analytical framework”.
No 278, “Assessing the efficacy, efficiency and potential side effects of the ECB’s monetary policy instruments since 2014”.
No 279, “The need for an inflation buffer in the ECB’s price stability objective – the role of nominal rigidities and inflation differentials”.
No 280, “Understanding low inflation in the euro area from 2013 to 2019: cyclical and structural drivers”.
ECB Occasional Paper Series No 264 / September 2021
1
Contents
Abstract 3
Executive summary 4
1 Introduction 10
1.1 Motivation and mandate 10
1.2 The conceptual framework for analysing inflation expectations 11
2 The nature and use of measures of inflation expectations 14
2.1 Availability and use of measures of expectations for the euro area
and individual Member States 14
2.2 The nature and scope of information in survey and market-based
measures 17
2.3 The relationship between market and survey-based expectations 28
Box 1 Inflation perceptions and expectations: evidence from the ECB
Consumer Expectations Survey 31
Box 2 Technical factors as drivers of market-based measures of inflation
compensation 36
3 Drivers and (un)anchoring of inflation expectations 41
3.1 The drivers of inflation expectations 41
Box 3 The link between inflation expectations and oil prices. Does the
nature of the shock to the global oil market matter? 53
3.2 Defining and measuring risks to anchoring 56
Box 4 What levels of inflation are consistent with the ECB’s definition of
price stability, according to the SPF? 58
Box 5 Inflation expectations in advanced economies: are there signs of
unanchoring? 70
4 Inflation expectations in macroeconomic forecasting 76
4.1 Indicators of inflation expectations as standalone forecasts 76
4.2 Predictive power of indicators of inflation expectations 84
4.3 The use of inflation expectations in E(S)CB projection models 91
Box 6 Conditional forecasts of inflation expectations 98
ECB Occasional Paper Series No 264 / September 2021
2
Box 7 Econometric models for inflation forecasting 99
5 Conclusions and implications 103
5.1 Conclusions 103
5.2 Implications 106
Bibliography 108
Annex A – Households’ and firms’ inflation expectations 118
A.1 What do we understand of firms’ observed inflation expectations?
118
A.2 What do we understand of households’ observed inflation
expectations? 125
A.3 The role of inflation expectations for households’ and firms’
choices 130
Annex B – Constructing a heat map of inflation expectation anchoring 133
B.1 Introduction 133
B.2 How the heat maps are constructed 133
ECB Occasional Paper Series No 264 / September 2021
3
Abstract
This paper summarises the findings of the Eurosystem’s Expert Group on Inflation
Expectations (EGIE), which was one of the 13 work streams conducting analysis that
fed into the ECB’s monetary policy strategy review. The EGIE was tasked with
(i) reviewing the nature and behaviour of inflation expectations, with a focus on the
degree of anchoring, and (ii) exploring the role that measures of expectations can play
in forecasting inflation. While it is households’ and firms’ inflation expectations that
ultimately matter in the expectations channel, data limitations have meant that in
practice the focus of analysis has been on surveys of professional forecasters and on
market-based indicators. Regarding the anchoring of inflation expectations, this paper
considers a number of metrics: the level of inflation expectations, the responsiveness
of longer-term inflation expectations to shorter-term developments, and the degree of
uncertainty. Different metrics can provide conflicting signals about the scale and
timing of potential unanchoring, which underscores the importance of considering all
of them. Overall, however, these metrics suggest that in the period since the global
financial and European debt crises, longer-term inflation expectations in the euro area
have become less well anchored. Regarding the role measures of inflation
expectations can play in forecasting inflation, this paper finds that they are indicative
for future inflationary developments. When it comes to their predictive power, both
market-based and survey-based measures are found to be more accurate than
statistical benchmarks, but do not systematically outperform each other. Beyond their
role as standalone forecasts, inflation expectations bring forecast gains when included
in forecasting models and can also inform scenario and risk analysis in projection
exercises performed using structural models. In terms of the implications for the ECB’s
economic and monetary analysis going forward, the work of the EGIE essentially
highlights the need for (i) more data on households’ and firms’ inflation expectations,
(ii) a comprehensive framework for assessing (un)anchoring and (iii) further
considerations regarding the use of observed expectation measures in forecasting
models.
JEL codes: D84, E31, E37, E52.
Keywords: Inflation expectations, anchoring, forecasting, macroeconomics,
monetary policy.
ECB Occasional Paper Series No 264 / September 2021
4
Executive summary
This paper summarises the findings of the Eurosystem’s Expert Group on
Inflation Expectations (EGIE), which was one of the 13 work streams
conducting analysis that fed into the ECB’s monetary policy strategy review.
The EGIE was tasked with (i) reviewing the nature and behaviour of inflation
expectations, with a focus on the degree of anchoring, and (ii) exploring the role that
measures of expectations can play in forecasting inflation.
Inflation expectations play a key role in monetary policy transmission through
the “expectations channel”. In this context, they influence price and wage setting
and – via expected real interest rates – consumption, investment, borrowing and
saving. The effectiveness of the expectations channel depends on the credibility of the
central bank in pursuing its price stability objective and is thus intimately related to the
anchoring of longer-term inflation expectations. When these expectations are firmly
anchored, they should not be responsive to macroeconomic news. However, if they
become unanchored, the ability of monetary policy to re-anchor expectations could
hinge on the responsiveness of expectations to monetary policy actions.
Looking at available empirical measures, it would seem that inflation
expectations data for euro area households and firms – the agents most
relevant for monetary transmission – are incomplete. This is particularly true of
longer-term horizons that would allow analysis of the term structure of expectations
and the degree of anchoring. Other major central banks appear to benefit from better
availability of measures of households’ and firms’ longer-term expectations for their
jurisdictions. The ECB’s Consumer Expectations Survey (CES) may close some of
this area-wide data gap going forward.
Available measures of households’ inflation expectations point to an apparent
upward bias relative to actual inflation but broad co-movement with inflation.
Some of this bias seems to be related to the fact that agents who are more uncertain
typically report their inflation expectations using round figures (in multiples of five) and
these are higher than actual inflation on average. Analysis based on consumers’
inflation expectations as reported in the European Commission Consumer Survey
(ECCS) shows that consumers’ perception of inflation and uncertainty varies
according to socio-economic factors. Consumers are more likely to be uncertain about
their inflation perceptions and expectations and to have higher inflation perceptions
and expectations if they are younger, female, have lower levels of formal education
and belong to lower-income groups. However, consumers appear to be
knowledgeable about broad inflation regimes, given that changes in their expectations
are correlated with changes in actual inflation. A key question for monetary
policymakers is to what extent households’ and firms’ inflation expectations influence
their actual decisions. One recent strand of the literature promotes a supply-side
narrative, whereby expectations of higher inflation are accompanied by a more
pessimistic general economic outlook (“inflation is bad for the economy”), whereas the
ECB Occasional Paper Series No 264 / September 2021
5
uncertainty framework referred to above suggests that this correlation could reflect
increased uncertainty – which, in turn, increases inflation expectations.
Reflecting the availability of data, the ECB’s economic and monetary analysis
has focused, in practice, on inflation expectations reported in surveys of
professional forecasters and those implied by financial market prices. This
focus is common across all major central banks. It is supported by the fact that the
expectations of financial market participants matter for the setting of financial prices
and thus the financing conditions of the non-financial sector. Moreover, while
professional forecasters do not represent a specific group of economic agents, their
expectations may matter because they are an information source for agents who are
relevant for the monetary transmission mechanism. There is some evidence, for
individual euro area countries, showing that firms and trade unions use information
from professional forecasters as an input when forming their own expectations.
Given the prominence of professional forecasters’ expectations and
market-based indicators, a recurring issue in economic and monetary analysis
is their potentially differing signals. When using them for policy inference, it is
therefore important to understand their partly different nature. Professional forecasters
provide figures for the expected level of inflation and the physical probabilities
surrounding it. Market-based measures, meanwhile, are inferred from the prices that
market participants pay in hedging against inflation (or deflation) risks, which implies
that they also contain a risk premium. Differences between the levels of survey and
market-based measures of inflation expectations become considerably smaller when
market-based measures are adjusted for such risk premia. While there is thus value in
distinguishing between estimates of risk premia (which can move sharply in times of
market stress) and estimates of the “genuine” expectations component of
market-based indicators, the inflation risk premium is also useful for policy analysis, as
it reflects investors’ perceived uncertainty. Analysis of market-based measures of
inflation expectations also needs to consider market distortions (or technical factors)
that may blur their information content, particularly in times of stress.
Uncertainty in inflation expectations can be assessed by looking at higher
moments of distributions. The standard deviation of the aggregated Survey of
Professional Forecasters (SPF) probability distribution (second moment) for long-term
expectations exhibited an upward shift in the period following the global financial crisis
(GFC) and a further one as of the end of 2018. In addition, indicators such as the
balance of risks suggest that the distribution became more skewed to the downside
(third moment) in these periods. The distribution of market participants’ inflation
expectations can be gauged from the prices of euro area inflation options. However,
directly comparing these “risk-neutral” probabilities with the “physical” probabilities in
surveys is misleading owing to the presence of the aforementioned premia. In the
context of the distribution of inflation expectations, risk-neutral probabilities tend to
overstate the corresponding physical probabilities for tail events, as risk-averse
investors value the pay-off from inflation options more highly in deflation and
high-inflation regimes. This suggests that changes over time may be more telling than
the levels of uncertainty.
ECB Occasional Paper Series No 264 / September 2021
6
Overall (as is done in practice), policymakers need to look at both survey and
market-based measures, and at all of their different moments. This is because
they all contain independent information that cannot be inferred from robust
relationships between sources or between moments. Standard causality analysis and
dedicated model-based analysis testing the responsiveness of SPF data to
inflation-linked swap (ILS) data has not pointed to systematic lead/lag relationships or
significant response coefficients. Similarly, causality analysis has not generated
robust results for changes in higher moments (such as skewness) preceding changes
in central tendencies. Although one benefit of market-based measures is their high
levels of frequency and timeliness, for instance when it comes to assessing the impact
of policy measures, these data require careful interpretation, as they always need to
be decomposed into estimates of “true” expectations and the various premia.
This paper delves into the drivers of inflation expectations and analyses the
role of oil prices and monetary policy, which tend to feature most prominently
in the policy debate. Notably, oil prices are an acknowledged driver of headline
inflation in the short to medium term via direct and indirect effects, and potentially also
in the longer term if they give rise to second-round effects. Model-based estimates
confirm the expected strong impact on short-term expectations and suggest that it is
stronger for market than for survey-based data. The impact on longer-term
expectations is very muted, which also holds for market data when these are corrected
for inflation risk premia. Crucially, the origin – or nature – of the oil shock appears to
matter, with the impact on expectations being strongest when the shock reflects
underlying developments in global economic activity.
The role of monetary policy shocks in driving inflation expectations is relevant
in a situation where longer-term inflation expectations have declined and the
task is to re-anchor long-term expectations to the inflation objective. A
model-based analysis using market-based expectations distinguishes between the
impact of a pure policy shock (e.g. an unexpected interest rate hike/cut) and that of an
information shock (e.g. an unexpected change in the central bank’s assessment of the
macroeconomic outlook). While there is no evidence of information shocks having a
significant impact, pure policy shocks have led to statistically significant increases in
spot ILS rates across short to medium-term maturities – but only in the period since
the establishment of the public sector purchase programme (PSPP). Monetary policy
can also drive inflation expectations by changing the inflation objective or the strategy
more generally. It appears that such adjustments have led to changes in longer-term
survey-based inflation expectations for the euro area, the United States and Japan,
but these have mainly been visible in the modes and the widths of cross-sectional
distributions, rather than the average level of inflation expectations. For the euro area,
there has been a fairly strong empirical association between longer-term inflation
expectations and trends in actual inflation, as one would expect if inflation trends are
seen as a track record of central bank credibility in pursuing inflation objectives.
However, this association broke down in 2015, and it was not until 2019 that
longer-term inflation expectations from the SPF started to fall towards the by then
already lower trend measures. This highlights the challenges in anchoring inflation
expectations if inflation trends are persistently below target and suggests that
ECB Occasional Paper Series No 264 / September 2021
7
re-anchoring efforts may have limited effectiveness if signs of successful changes to
inflation trends do not become evident.
This paper puts forward a range of metrics aimed at assessing the degree of
anchoring, highlighting the fact that this is a complex and multi-faceted
concept and policymakers need to look at and distil possibly mixed signals at
each point in time. Mostly, anchoring is assessed in terms of the level of long-term
expectations and their responsiveness to short-term developments. Different metrics
can provide conflicting signals about the scale and timing of potential unanchoring,
which underscores the importance of considering all of them. In practice, each metric
has its strengths and weaknesses, and metrics could also be interlinked if, for
instance, responsiveness ceases once the level has adjusted to a new “steady state”.
This paper also presents heat maps, providing a visual synthetic overview of the
different dimensions of the anchoring of inflation expectations.
Overall, the evidence suggests that longer-term inflation expectations in the
euro area have become less well anchored in the post-GFC period. Regarding
their level, during 2019, average longer-term SPF inflation expectations moved to the
bottom of – or slightly outside of – the 1.7-2.0% range, and formal break-point tests
suggest a downward shift. The responsiveness-based concept points to positive – but
only sporadically significant – correlations with different shorter-term developments.
While there is some evidence of a positive and statistically significant pass-through
from short-term to longer-term ILS rates throughout the low-inflation period, this might
reflect co-movement of premia and technical factors across horizons. Other results
using single-equation models with stochastic volatility also show a positive – but only
temporarily significant – pass-through from short-term to longer-term SPF
expectations during the GFC and at a time that coincided with the start of quantitative
easing. The different timings highlight the problem that the responsiveness metric can,
in principle, capture the effect of very persistent shocks, unanchoring and/or
re-anchoring. Finally, models testing explicitly for the responsiveness of survey-based
measures of longer-term expectations to surprises in euro area inflation releases find
that only negative surprises (e.g. post-2013) have had a significant impact on
longer-term inflation expectations.
There is some evidence showing that, in recent years, it has taken longer after a
shock for inflation expectations to reach their new “steady-state” level.
Modelling the term structure of the inflation expectations curve using Consensus
Economics data up to ten years ahead suggests that, in the post-GFC period, the
horizon over which shocks are expected to fade out has lengthened, at the same time
as data-implied steady-state inflation expectations have shifted downwards. The
evidence for the euro area is also supported by analysis of longer-term inflation
expectations in other advanced economies, suggesting that there is some evidence of
a common global factor.
Market and survey-based expectations both outperform statistical forecast
benchmarks and are broadly similar, on average, to the forecasts produced by
the Eurosystem’s (Broad) Macroeconomic Projection Exercises ((B)MPEs).
Forecast performance tests for market and survey-based measures suggest that both
are credible benchmarks and have, on average, a level of forecast accuracy for actual
ECB Occasional Paper Series No 264 / September 2021
8
inflation which is similar to that of Eurosystem projections. This relative forecast
performance differs slightly across horizons, with those measures tending to be more
favourable for the Eurosystem projections at the two-year-ahead horizon than at the
one-year-ahead horizon. While survey-based inflation expectations have a relatively
high degree of accuracy as regards the baseline, they are less precise when it comes
to the surrounding probabilities and densities. Comparing ex ante probabilities and ex
post outcomes, the “probability integral transform” (PIT) suggests that outcomes are
too often in the tails of the reported probability distribution. This under-estimation of
the true degree of uncertainty has remained the case even after the GFC led to an
increase in ex ante uncertainty, as reflected in the step increase in the standard
deviation of the aggregated SPF probability.
Including observed inflation expectations in models that can be used in the
Eurosystem projection process marginally improves their inflation forecast
performance. An extensive real-time forecast evaluation for the euro area and a
number of Member States covering a diverse set of models suggests that indicators of
inflation expectations bring some gains to the accuracy of inflation forecasts compared
with models that do not include such expectations, but those gains are typically
modest. The predictive gains are somewhat higher for the two-year-ahead horizon,
and they are somewhat higher for HICP inflation excluding energy and food than for
headline inflation. Gains can come from using long-term expectations to inform the
inflation trend and from using short-term inflation expectations to drive current inflation
dynamics. For reasons of robustness, it is hence advisable to follow a comprehensive
approach, looking at different models and specifications. A “thick” Phillips curve
framework is regularly applied in the Eurosystem projection exercises as a
cross-checking device for underlying inflation outcomes. A real-time out-of-sample
forecast exercise suggests that including observed expectations brings modest
forecast gains relative to specifications that are based on lagged inflation only. For
more recent years, including market-based measures would have improved the
forecast performance, but time variation in the forecast performance is such that it is
not clear whether this result would be robust if it were tested over a longer period.
Finally, observed inflation expectations can inform scenario and risk analysis
in projection exercises performed using structural models. The ECB’s main
(semi-)structural projection models include a behavioural equation for agents’
long-term expectations that allows for interaction with both the inflation objective and
actual inflation. These behavioural equations are calibrated on the historical behaviour
of SPF and Consensus Economics longer-term inflation expectations. Scenarios can
use observed measures of inflation expectations, for instance by assuming a shock
that would shift the central tendency to specific (lower) percentiles of the aggregate
probability distribution or the cross-sectional distribution. Longer-term inflation
expectations can also be made an endogenous variable in a dedicated satellite model
that uses the main macroeconomic variables featuring in the workhorse forecasting
models. The satellite model can be used to produce conditional forecasts of
longer-term inflation expectations, which can then be fed into the main forecasting
models as a scenario path.
ECB Occasional Paper Series No 264 / September 2021
9
There is some evidence that the official Eurosystem projections influence the
inflation expectations of professional forecasters. Central bank projections are an
important element of communication and, especially as regards forward guidance and
re-anchoring, can be expected to have an impact on private sector expectations.
Regression analysis suggests that Eurosystem projections do have some influence on
subsequent Consensus Economics expectations even after controlling for
macroeconomic news (on oil prices, inflation surprises and monetary policy
measures) emerging between the two forecasts.
In terms of the implications for the ECB’s economic and monetary analysis
going forward, the work of the EGIE essentially highlights the need for (i) more
data on households’ and firms’ inflation expectations, (ii) a comprehensive
framework for assessing (un)anchoring and (iii) the use of observed measures
in forecasting models. At the same time, a set of additional issues has emerged that
should be part of future analytical agendas. For instance, given the strong role played
by longer-term expectations as benchmarks for monetary policy credibility,
robustification is needed with regard to the finding that the horizon over which agents
typically expect (unavoidable) shocks to fade out and credibility to kick in has
lengthened, and what this implies for unanchoring if there is both lengthening of that
horizon and a lower level of longer-term expectations. Also, further assessment of the
role of expectations in monetary transmission would be greatly enhanced if it could
rely more on data for households’ and firms’ expectations. This would allow the testing
of some recent hypotheses regarding the formation of those agents’ expectations and
would be a first step in investigating whether and how those expectations shape real
economic decisions. In this respect, better data on households’ and firms’
expectations may confirm the substantial heterogeneity across agents within the same
institutional sector and put a premium on models that can deal with such
heterogeneous agents.
ECB Occasional Paper Series No 264 / September 2021
10
1 Introduction
1.1 Motivation and mandate
This paper reviews the nature and behaviour of inflation expectations and their
role in forecasting inflation. Information on inflation expectations plays an important
role in the regular assessment of the economic and monetary situation, but it can
occasionally be difficult to interpret and square across its different sources. It also
plays an important role as a regular cross-check of the ESCB’s own projections for the
inflation outlook, but the potential for it to formally feed in to those projections via its
inclusion in models has not been explored to the same extent. Against this
background, the EGIE was set up and given the following tasks:
• Review the measurement and drivers of inflation expectations. Reflecting the
availability of data, the focus here is on expectations reported in surveys of
professional forecasters and indicators derived from financial market data.
Moreover, given the central question of the anchoring of expectations in a
low-inflation period, emphasis is placed on longer-term inflation expectations.
• Evaluate the role of inflation expectations in forecasting. This role can be viewed
both (i) from the perspective of the forecast performance of different measures of
inflation expectations and (ii) from the perspective of whether empirical
measures of inflation expectations should be incorporated in macroeconomic
models in order to improve forecast performance or exploit the correlation
between inflation expectations and “trend inflation”.
The work of the EGIE provided insights that fed into the ECB’s review of its
monetary policy strategy. These insights were used, for instance, in the seminar on
inflation measurement and trends by complementing the inflation narrative with an
assessment of the anchoring of inflation expectations. A change in the extent to which
inflation expectations are anchored or there is uncertainty surrounding such anchoring
could be a factor that helps to explain a protracted period of low actual inflation. The
information generated by the EGIE also fed into the seminar on the ECB’s economic
and monetary analysis, reflecting the prominent role that inflation expectations play in
these analytical paradigms. One example in this regard is the role of inflation
expectations for second-round effects possibly triggered by temporary cost shocks to
inflation. Another is the role of expectations in forecasting, which is a key element of
economic analysis and the analysis of short to medium-term risks to price stability.
In the light of that mandate, this paper is organised as follows. The remainder of
this section describes the main elements of the conceptual framework for analysing
inflation expectations. Section 2 takes stock of the available empirical measures of
inflation expectations, how they are used in the Eurosystem and beyond, and what
their relative characteristics are. The focus here is on measures that are (i) available
for the euro area as a whole, (ii) cover longer horizons and (iii) are regularly discussed
in policy deliberations. Section 3 reviews determinants of longer-term inflation
expectations, focusing on some frequently referenced potential influences, such as oil
ECB Occasional Paper Series No 264 / September 2021
11
prices or monetary policy actions. This section also discusses different ways of
measuring unanchoring and offers a synthetic approach to looking at them in the form
of a heat map. Section 4 discusses selected issues relating to the use of inflation
expectations in forecasting. On the one hand, it looks at the information content of
expectations as a cross-check for the Eurosystem’s own forecasts and projections.
And on the other hand, it examines whether including empirical measures of inflation
expectations as forward-looking variables in models can enhance forecast accuracy
and narratives. Finally, Section 5 concludes.
1.2 The conceptual framework for analysing inflation
expectations
Central bankers look closely at inflation expectations, as they play an important
role in the monetary transmission mechanism. This role is typically captured in the
“expectations channel” and operates through the impact that agents’ inflation
expectations have on price/wage setting and – via expected real interest rates –
consumption, investment, borrowing and saving. The influence on expected real
interest rates is especially relevant if nominal interest rates are constrained by a lower
bound. The effectiveness of the expectations channel depends on the credibility of the
central bank in pursuing its price stability objective and is thus intimately related to the
role of longer-term inflation expectations. If economic agents believe in the central
bank’s ability to maintain price stability and its commitment to doing so, longer-term
inflation expectations will remain firmly anchored and monetary policy can influence
agents’ wage and price-setting behaviour, which might otherwise contain an unstable
element resulting from inflation/deflation fears. Thus, well-anchored longer-term
inflation expectations act as automatic stabilisers and enhance the effectiveness of
monetary policy in the transmission mechanism. This makes them a desirable feature
in any monetary policy framework.
However, longer-term inflation expectations hardly ever feature explicitly in the
modelling of the monetary transmission mechanism. This is to do with the notion
that well-anchored expectations should not react to macroeconomic news, including
monetary policy shocks. Thus, they tend to be used in empirical models as exogenous
assumptions, rather than endogenous variables: their response to co-variables in the
model is restricted to zero. If longer-term inflation expectations represent a nominal
anchor and act as gravitation points for actual inflation, a failure to recognise the
unanchoring of longer-term expectations could imply a distorted view of the model’s
structural properties, its adjustment dynamics and the effectiveness of monetary
policy. Hence, developing an understanding of the degree of anchoring of longer-term
inflation expectations and what drives these expectations is of crucial importance for
central banks.
In periods of persistently low inflation, the expectations channel of monetary
transmission is usefully augmented by a re-anchoring channel. The role of
longer-term inflation expectations and risks of unanchoring in explaining the
low-inflation period seen since 2014 was analysed by a Eurosystem expert group
(Ciccarelli & Osbat, 2017) and has been revisited in the context of the ongoing
ECB Occasional Paper Series No 264 / September 2021
12
monetary policy strategy review. If longer-term inflation expectations have shifted
away from the central bank’s inflation objective, any assumption of zero response to
macroeconomic news will then restrict their capacity to contribute to re-anchoring in
monetary transmission models. Once there is reason to believe that expectations are
unanchored, the responsiveness of long-term inflation expectations to monetary policy
becomes desirable, rather than undesirable (Diegel & Nautz, 2020). In addition, the
models then also need to capture the impact that changes in longer-term inflation
expectations have on actual inflation – i.e. reverse the causality that is often explored
in the analysis of longer-term expectations. If there is such a relationship, then
monetary policy can influence inflation outcomes both indirectly through the regular
monetary transmission mechanism and directly by re-anchoring longer-term
expectations. Developing an understanding of the role that longer-term expectations
play in actual inflation dynamics has thus become more important for central banks.
In practice, inflation expectations cannot be captured in a single measure and
number. This holds for both shorter and longer-term expectations. The effectiveness
of monetary transmission may be impacted if inflation expectations diverge
substantially across the different sectors of the economy, such as households,
non-financial corporations or the financial sector (Darracq Paries & Zimic, 2021).
Notably, households’ and firms’ expectations matter directly for the expectations
channel via the impact that real interest rates have on spending decisions and via
price/wage setting. Those of financial market participants matter directly for the setting
of financial prices and thus the financing conditions of the non-financial sector (see
Figure 1). And finally, the expectations of professional forecasters may matter
because they are an information source for other agents. This issue of multiple
expectations is magnified if there is substantial heterogeneity of inflation expectations
among agents within a specific institutional sector. And it is further magnified if the
individual agents have very different perceptions of the uncertainty surrounding their
central expectations. Central banks should therefore consider broad sets of
information on expectations and the different moments of their distributions.
Figure 1
Stylised overview of the transmission mechanism relating to inflation expectations
Source: Eurosystem staff.
Against that background, the EGIE’s work on inflation expectations links up
with other work streams of the monetary policy strategy review. This is
Inflation
expectations
Households
Consumption
(real interest
rate) Wage
negotiations
Firms
Investment (real
interest rate) Price setting
Financial
markets
Financial prices
and conditions
Professional
Forecasters
ECB Occasional Paper Series No 264 / September 2021
13
particularly true of the work stream on Eurosystem modelling as regards the empirical
modelling of expectations in workhorse models. However, the EGIE’s work also has
interfaces with the work stream on the price stability objective – for instance, on the
question of whether the nature of the inflation objective has a bearing on the anchoring
of longer-term inflation expectations. And it has implications for analysis of the
effectiveness of monetary policy instruments, as the latter depends on the behaviour
of expectations. The issues raised and the analysis carried out are also relevant for
the work stream on monetary policy communication.
ECB Occasional Paper Series No 264 / September 2021
14
2 The nature and use of measures of
inflation expectations
Empirical measures of inflation expectations have assumed an important role
in central banks’ regular analysis and communication. This originates, for
instance, from the role that levels of longer-term expectations play as a sounding
board for how agents perceive monetary policy’s efforts to achieve quantitative
inflation objectives. At the same time, changes in short to medium-term expectations
provide important information on the nature and duration of shocks that agents see
influencing the inflation outlook at different points in time. This is an important
cross-check for central banks’ own forecasts and their revisions. For instance, the
ECB Survey of Professional Forecasters, with its questions on inflation expectations at
both shorter and longer horizons and on key correlates of inflation (oil prices, GDP and
unemployment), was set up with such information needs in mind. This section
discusses the availability and nature of inflation expectations used in the Eurosystem
and beyond.
2.1 Availability and use of measures of expectations for the
euro area and individual Member States
The taxonomy of empirical measures of inflation expectations for the euro area
spans different agents and methodologies. For the euro area as an entity,
expectations have traditionally been available from surveys addressed to professional
forecasters (ECB Survey of Professional Forecasters, Consensus Economics (CE),
Euro Zone Barometer (EZB) and, more recently, the ECB Survey of Monetary
Analysts (SMA)) and households and firms (both European Commission (EC)).
However, while the surveys of professional forecasters generate quantitative
expectations for horizons up to many years ahead, those for households and firms are
more limited. They are mostly qualitative in nature (although quantitative expectations
for households are available via the joint harmonised EU programme of consumer
surveys), cover short horizons (generally one year or less) and, in the case of firms,
relate to producer rather than consumer prices. Moreover, surveys of professionals
directly elicit expectations for the euro area as a whole (top down), while surveys of
households and firms involve the aggregation of expectations for the respective
countries (bottom up). A second source of empirical measures of inflation
expectations is the information derived from prices of financial market products linked
to euro area inflation. These expectations are available for both shorter and longer
horizons. In practice, the ECB’s economic and monetary analysis has focused on
expectations derived from the SPF and on market-based indicators (as discussed in
more detail in Section 2.2), partly because, in addition to mean expectations, they also
provide explicit information on risks and uncertainties that can only be gauged from
the cross-sectional dimension in the case of other surveys. Table 1 provides a
ECB Occasional Paper Series No 264 / September 2021
15
synoptic overview of the key features of different survey-based sources for the euro
area.
Table 1
Key features of different empirical measures of inflation expectations for the euro area
Name Agent Geog. Horizon Target Frequency Sample Size Comments
ECCS HH CC 1ya Consumer
prices Monthly 1972-2004 ~25,000
CES HH CC 1ya, 3ya Prices Monthly 2020 ~8,000 Also probabilities
ECBS NFC CC 3ma Producer
prices Monthly 1962 ~70,000 Only qualitative
SPF Prof EA Multiple HICP (HICPX) Quarterly 1999 ~75 Also probabilities
SMA Prof EA Multiple HICP (HICPX) 6-weekly 2019 ~30 Also probabilities
CE Prof CC/EA Multiple HICP (HICPX) Monthly/
quarterly 1990 ~30
EZB Prof CC/EA Multiple HICP Monthly 2002 ~30
Swaps/bonds Market EA (CC) Multiple HICP excl.
Tobacco Daily 2005 Also (risk-neutral)
probabilities from options
Notes: ECCS = European Commission Consumer Survey; CES = ECB Consumer Expectations Survey; ECBS = European Commission
Business Survey; SPF = ECB Survey of Professional Forecasters; SMA = ECB Survey of Monetary Analysts; CE = Consensus
Economics; EZB = Euro Zone Barometer; HH = households; NFC = non-financial corporations; Prof = professionals; CC = individual
countries; EA = euro area; “1ya” = one year ahead; “3ya” = three years ahead; “3ma” = three months ahead.
The situation as regards the availability and use of measures of inflation
expectations at Eurosystem national central banks (NCBs) largely mirrors the
picture for the euro area as a whole. Most NCBs mainly have access to the
“standard” measures of inflation expectations discussed above for the euro area. At
some NCBs, additional “non-standard” measures of expectations are available for
their country. In most cases, those additional measures are derived from surveys of
professional forecasters, but in some cases they involve surveys of firms and
households. The set of empirical measures of inflation expectations that is available
for the euro area and its Member States appears to broadly correspond to that
available for other jurisdictions. This is particularly true for surveys of professional
forecasters and market-based indicators, which are, for instance, available for the
United States, Japan, the United Kingdom and many other countries. In contrast,
some notable differences exist with regard to households’ inflation expectations. In the
euro area, such expectations are, thus far, only available for a one-year horizon,
whereas in the United States and the United Kingdom they are also available for
longer horizons of around five years.
In practice, the ECB’s regular economic and monetary analysis is mostly based
on expectations derived from surveys of professional forecasters and market
participants. This reflects their widespread availability, their well-understood designs,
and the fact that their quantitative results broadly correspond to actual inflation. In
principle, however, the main focus of monetary policy might be on households’ and
firms’ expectations, as in theoretical macroeconomic monetary transmission models
such expectations influence choices, which directly affect prices and quantities. This
dichotomy between conceptual and practical importance reflects, on the one hand, the
fact that data on the inflation expectations of households and – especially – firms are
scarce and, on the other hand, the fact that quantitative results have often not been in
line with actual inflation data.
ECB Occasional Paper Series No 264 / September 2021
16
Research on households’ and firms’ inflation expectations has recently
gathered pace and provided some valuable findings. Currently available data for
the euro area and other countries allow for some initial insights on the factors
influencing the formation of those agents’ expectations and how that guides their
decisions (see Annex A for an overview). Firms’ observed inflation expectations seem
to be driven by several factors, including awareness of news on current inflation, the
dynamics of wages and input prices, and the monetary policy stance and its
communication. Evidence for Italy suggests that the causal effects which firms’
inflation expectations have on their behaviour vary depending on whether monetary
policy is constrained at the effective lower bound. Observed consumer expectations
tend to co-move with actual price dynamics, but have an upward bias. This bias might,
at least in part, be ascribable to uncertainty that leads consumers to resort to rounding,
thereby increasing aggregate inflation expectations. This uncertainty can have several
sources, such as the economic outlook or, more structurally, socio-economic
characteristics. How households’ inflation expectations affect consumption depends
on the net effect of intertemporal substitution and income effects. Empirically, the sign
of this net effect is ambiguous and appears to depend on whether there is a low or
high-inflation regime. Overall, there is broad agreement that the way in which
households and firms form and adjust their inflation expectations depends heavily on
the way that they interpret the source of the news and its implications for the broader
economic outlook. This also implies that monetary policy communication is highly
important for the effectiveness of monetary policy transmission via the expectations
channel.
At the same time, a number of open questions remain in relation to households’
and firms’ inflation expectations. For example, more dedicated studies are required
in order to establish a broader consensus regarding the way in which
firms/households form their inflation expectations, to what extent these expectations
affect their decisions and thus, ultimately, the inflation process, and whether central
banks can influence firms’ and households’ inflation expectations. Available research
suggests that there is potential for the relevance of households’ and firms’ inflation
expectations for monetary policy to increase, but several data gaps still need to be
closed before open questions can be answered for the euro area. The ECB’s
Consumer Expectations Survey may close some of this area-wide data gap going
forward (see Box 1). Firmer and broader insights could then also help the empirical
modelling of inflation expectations – for instance in wage equations, and thus,
ultimately, in price equations. Responses to an EGIE questionnaire suggest that
NCBs’ modelling of wage setting is predominantly reliant on backward-looking
expectations. In this respect, the Reserve Bank of Australia and Sveriges Riksbank
seem to be the only major central banks that have access to data on the explicit
inflation expectations of unions and/or employer organisations.
ECB Occasional Paper Series No 264 / September 2021
17
2.2 The nature and scope of information in survey and
market-based measures
Assessing survey and market-based measures of inflation expectations in the
policy process requires an understanding of their different natures. Surveys
providing quantitative expectations reflect answers to direct questions on inflation
outcomes at different horizons. They are typically presented as aggregations of
participants’ individual inflation forecasts.1 By contrast, market-based inflation
expectations are measures derived from the prices of financial instruments linked to
future inflation. These instruments include inflation-linked swaps (ILSs),
inflation-linked bonds (ILBs) or inflation options and are traded by informed investors
on a continuous basis.
2.2.1 The assessment of central tendencies
To ensure a proper comparison between survey and market-based measures,
some adjustments are needed. First of all, survey-based expectations typically
relate to overall HICP, while market-based measures relate to HICP excluding
tobacco. In the period since 2004, these measures of inflation have differed by an
average of 0.1 percentage points (p.p.). It is not clear whether market participants take
this difference into account in the formation of their expectations, but if they did, this
would bring the central tendencies of survey and market-based measures somewhat
closer together. Second, the high-frequency information available for market-based
measures implies that comparisons with survey-based measures will need to consider
the timing of their measurement – especially for expectations relating to shorter
horizons, where additional information typically has some impact. This difference
should, in principle, become less relevant when moving towards expectations at
longer-term horizons if these are not responsive to news. Third, a feature that is very
relevant also for comparisons at longer horizons is the fact that market prices, rather
than just reflecting genuine expectations, also reflect risk premia and are potentially
driven by other distorting factors (e.g. liquidity conditions), thus providing a less direct
gauge of market participants’ inflation expectations.
Distilling investors’ unobserved inflation expectations from observed
market-based indicators requires empirical estimations. The two components –
investors’ inflation expectations and the inflation risk premium – can, for example, be
recovered by modelling the inflation-linked swap curve on the basis of “term structure
models”.2 The “genuine” inflation expectations derived using these models are closer
to those signalled by the central tendencies of survey-based measures (see Chart 1).
In particular, market and survey-based inflation expectations differ in times of
economic and financial stress: in such periods, the inflation risk premium is the most
1 For the remainder of this paper, references to survey-based measures of inflation expectations should be
understood as referring to the ECB Survey of Professional Forecasters.
2 See, for example, Camba-Méndez & Werner (2017). Note that any such estimate is subject to statistical
and conceptual uncertainty, such that the precise level of the model-implied expectation and risk
components should be considered with some caution.
ECB Occasional Paper Series No 264 / September 2021
18
important source of variation in market-based measures, although market-based
measures still retain a higher degree of volatility after adjustment for inflation risk
premia than their survey-based counterparts.
Chart 1
SPF inflation expectations and risk premium-adjusted 1y4y ILS rate
(percentages per annum)
Sources: Bloomberg, Refinitiv and Eurosystem staff calculations.
Notes: The “1y4y ILS” is the one-year ILS rate four years ahead. The risk adjustment is based on an affine term structure model and fitted
to the euro area zero-coupon ILS curve. The estimation method follows Joslin et al. (2011). For details, see Camba-Méndez & Werner
(2017). Latest observations: Q1 2021 (SPF); March 2021 (market data).
The inflation risk premium also provides useful information with regard to
expectations. This risk premium compensates investors for the uncertainty
surrounding their inflation expectations. However, at least from a theoretical point of
view, a negative (positive) inflation risk premium cannot necessarily be equated with a
deflationary (inflationary) scenario. For example, following consumption-based asset
pricing theory, a negative (positive) risk premium reflects the fact that investors expect
future inflation to be positively (negatively) correlated with future growth.3 For
example, the shift from a positive to a negative risk premium after 2014 suggests that
medium-term expectations were – with some degree of variation – predominantly for a
low-inflation/low-growth regime (see Chart 2). This change is also confirmed by
developments in the BoRI – a measure of the balance of risks derived from SPF data.
While the level of the inflation risk premium – in absolute terms – suggests comparable
levels of uncertainty in the pre-2014 and post-2014 regimes, it declined overall. This
may, among other things, reflect the fact that investors in inflation-linked products
were continually surprised on the downside by inflation outcomes in the post-2014
period, similar to what was observed for forecasters participating in surveys. Paying
more for these inflation protection products than they had anticipated may have
reduced investors’ appetite for the products in question. This may have led to a
3 See Camba-Méndez & Werner (2017) and the discussion therein. It holds that an asset whose pay-out is
eroded by high inflation in a state of the world in which this pay-out is valued more should yield a positive
inflation risk premium, and vice versa. The value of an asset’s pay-out to the investor is higher when the
marginal utility of consumption is high – i.e. when overall consumption is low. A typical scenario in which
this would be the case would be a recession driven by weak aggregate demand, which would lead to
weak economic activity in tandem with low inflation (Di Iorio & Fanari, 2020).
0.5
1.0
1.5
2.0
2.5
3.0
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Five-year ahead SPF poi nt forecast
Five-year ahead SPF mean 1y4y ILS
Risk-adjusted 1y4y ILS
ECB Occasional Paper Series No 264 / September 2021
19
broader downward (re)pricing of inflation risks and a corresponding decline in the
perceived value of inflation protection – i.e. a smaller risk premium.
Chart 2
Estimated inflation risk premium (IRP) for 1y4y ILS and SPF Balance of Risk Indicator
(BoRI)
(percentage points)
Source: Eurosystem staff calculations.
Notes: The BoRI is calculated as the difference between the mean of the aggregate probability distribution and the average point forecast
in the SPF. The estimated inflation risk premium is calculated as the difference between the raw and risk-adjusted 1y4y ILS.
The cross-sectional nature of survey-based measures suggests a need for
robustness checks with regard to the sample. First of all, in order to allow for the
possibility that outliers are unduly influencing the average across panellists, the
headline figures in the SPF reports, couched in terms of averages as measured by the
mean response, are regularly checked against corresponding medians. Second, given
that the composition of the panel normally changes somewhat from survey round to
survey round, it is useful to check the average for the panel responses against the
average for a constant or balanced panel (i.e. those who responded in both rounds).
For the available data on long-term inflation expectations since 2001, the differences
from one round to the next between unbalanced and balanced panels have, overall,
been marginal, with a standard deviation of 2 basis points. Thus, looking at an
unbalanced panel would not normally have an impact on average long-term
expectations reported to one decimal place (see Chart 3).
-1.0
-0.5
0.0
0.5
1.0
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
BoRI (left-hand scale)
Estimated IRP 1y4y (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
20
Chart 3
Changes in SPF longer-term inflation expectations
(percentage points)
Sources: ECB (SPF) and Eurosystem staff calculations.
Note: The balanced panel is always between two consecutive survey rounds.
In the case of both survey and market-based indicators, the reliability of
measures of expectations can be affected by “thinness”. For survey measures,
this relates to a potentially small number of panellists, particularly for longer-term
expectations. A Monte Carlo-type exercise randomly drawing samples of different
sizes from SPF long-term inflation expectations data suggests that variation in the
sample average increases as the number of panellists decreases (see Chart 4). For a
panel comprising around ten panellists (approximately the number reporting
longer-term expectations for Consensus Economics and the Euro Zone Barometer)
the variation is above 0.2 p.p. For a sample of around 20 panellists (approximately the
number in the Survey of Monetary Analysts) the variation is around 0.15 p.p., whereas
once the sample size exceeds 40 (e.g. in the case of the SPF) it starts to level off at
around 0.1 p.p. Thus, for samples of 20 panellists or less, “outliers” can lead to
changes in average outcomes for longer-term expectations, which need to be
interpreted with caution as regards indications of the degree of anchoring.4
4 Over the period (Q3 2014 to Q1 2021) when quarterly longer-term expectations have been available for
the SPF, Consensus Economics and the Euro Zone Barometer, the variability (standard deviation of
quarter-on-quarter changes) of the last two has been substantially higher (SPF: 0.034 p.p.;
CE: 0.088 p.p.; EZB: 0.065 p.p.).
-0.10
-0.08
-0.06
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021
Revision
Revision with balanced panel
ECB Occasional Paper Series No 264 / September 2021
21
Chart 4
The sample average variation of longer-term inflation expectations as a function of
sample size
(percentage points)
Sources: Eurosystem staff calculations based on SPF inflation expectations five years ahead.
Notes: Monte Carlo simulation based on 1,000 iterations. The red horizontal lines denote 0.10, 0.15 and 0.20 percentage points.
For market-based indicators of inflation expectations, “thinness” can occur in
the presence of market imperfections. This implies that prices of inflation-linked
products reflect not only economic fundamentals, but also technical features of the
market at the relevant point in time. This is particularly true of the bond-derived
break-even inflation rate (BEIR), which can be subject to significant liquidity effects if
illiquid bonds cannot easily be traded at a fair market value and thus need to pay an
additional return (premium). Such liquidity effects introduce a bias that is
time-varying – i.e. more likely to arise in periods of financial turmoil. As they are not
easily separable from other risk premia, their presence is an argument in favour of
focusing on prices in inflation-linked swap markets, which are less prone to liquidity
distortions.5 The smaller and more concentrated markets for inflation-linked assets
are, the more prone they are to being influenced by individual investment strategies
(such as those of large buy-and-hold investors), which may again blur their link to the
inflation outlook as perceived by market participants. Overall, while technical factors
can be associated with a certain bias in market-based indicators of inflation
compensation during periods of stress, there seems to be little reason to conclude that
they have been the key driver of the decline in market-based measures of inflation
compensation in the euro area over the last decade (see Box 2 for more details).
However, the uncertainty around such distortions remains considerable, and a more
comprehensive quantification of trading costs and demand-supply imbalances for safe
assets in inflation-linked markets is needed.
5 This is consistent with feedback from market participants. See also Garcia & Werner (2018).
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
010 20 30 40 50 60
95th to 5th percentile range (mean)
95th to 5th percentile range (median)
ECB Occasional Paper Series No 264 / September 2021
22
2.2.2 The assessment of distributions
Data on the distributions of inflation expectations can provide information on
the risks and uncertainties surrounding central tendencies. This is important for
monetary policy if central bank credibility depends not only on anchoring mean
inflation expectations but also on minimising perceived inflation uncertainty and risks,
or if policy options are dependent on specific features of distributions (such as their
variances, skewness or kurtosis). The assessment of distributions of inflation
expectations is a regular feature of economic and monetary analysis.
SPF data provide information on uncertainty and risks via their cross-sectional
distributions and their probability distributions. The cross-sectional distribution
captures the heterogeneity of survey respondents’ expectations and can be quantified
in standard deviations or coefficients of variation. The probability distributions that
professional forecasters provide for expected outcomes at different horizons can be
summarised both in terms of the average standard deviation of the individual
distributions (uncertainty) and in terms of distributional asymmetries (balance of risk).
The distributional asymmetry, among other measures, makes use of the skewness in
distributions. For expectations five years ahead, Chart 5 points to higher uncertainty
and persistent downward risks to central expectations since the financial crisis – albeit
with some variation.
Chart 5
Perceived balance of risks and uncertainty around longer-term inflation expectations
(variance-scaled percentage points; percentage points)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: Uncertainty is measured as the average individual standard deviation; the balance of risks is measured as the average individual
distribution asymmetry (calculated as the average of 12 alternative measures). Values less than zero signify downside risk, while those
above zero signify upside risk.
The distribution of market participants’ inflation expectations can be gauged
from prices of euro area inflation options. These options are instruments with
“non-linear” pay-offs, either (i) paying out if inflation as measured by euro area HICP
excluding tobacco exceeds a certain threshold, and zero otherwise (inflation caps) or
(ii) paying out if inflation falls short of a certain threshold, and zero otherwise (inflation
floors). By comparing the prices of options that insure against different outcomes, it is
possible to infer the probabilities that investors assign to those different outcomes.
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
-3.0
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Balance of risks (di stributional asymmetry) (left-hand scale)
Uncertainty (distributional standard deviation) (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
23
However, such option-implied probabilities must not be interpreted as reflecting
underlying “physical” probabilities. Akin to the information derived from ILB or ILS
rates, information obtained from options represents “risk-neutral” expectations and, as
such, includes a premium component. For example, an option-implied probability of
25% for deflation does not imply that investors believe there is a one-in-four chance
that deflation will actually emerge. Risk-neutral probabilities tend to overstate the
corresponding physical probabilities for tail events (such as outright deflation or very
high inflation), and vice versa for non-tail events.
The impact that risk premia have on the level of option-implied risk-neutral
probabilities can be illustrated by comparing them with SPF results. The
probabilities reported in the SPF can be interpreted as physical probabilities and
therefore provide a natural reference point for assessing the degree to which market
participants price risk premia into inflation options. As an example, Chart 6 compares
option-implied probabilities with SPF-based probabilities for euro area inflation one
year ahead, which is the horizon that allows for the closest matching given the
availability of euro area inflation options and the horizons considered in the SPF. The
former exceed the latter for outcomes where inflation is either negative or above 3%.
In each of those two cases, risk-neutral probabilities are roughly three times higher
than physical probabilities (see the red markers). In contrast, physical probabilities
tend to be higher than their risk-neutral counterparts for the low, but positive, inflation
outcomes in between the two tail events. These observations are consistent with
risk-averse investors valuing the pay-off from inflation options more highly in
deflationary and high-inflation regimes, resulting in a larger wedge between the
associated risk-neutral and physical probabilities. However, this wedge can also
reflect the fact that professional forecasters tend to be overly confident in the accuracy
of their central scenarios and assign too small a probability to tail events.6
6 In this respect, it is important to note that probabilities reported in surveys rely on subjective distributions,
and that these – according to the results of special SPF questionnaires (link) – are often based on
judgements rather than models. Evidence for overly narrow SPF distributions is presented in Section 4.1.
ECB Occasional Paper Series No 264 / September 2021
24
Chart 6
Option-implied risk-neutral probabilities versus physical probabilities derived from the
SPF
(left-hand scale: percentages; right-hand scale: percentage points)
Sources: Bloomberg, Refinitiv and Eurosystem staff calculations.
Notes: “Option-implied” refers to the risk-neutral probability of a given inflation outcome, as extracted from the prices of one-year
zero-coupon options based on (three-month-lagged) euro area HICP inflation excluding tobacco (HICPxT). “SPF” refers to physical
probabilities for euro area HICP inflation over the next year, as implied by the responses of professional forecasters surveyed by the ECB
(based on results for the first quarter of 2018). For ease of comparison, risk-neutral probabilities are evaluated on the date of the deadline
for SPF participants to respond (11 January 2018). The ratio of risk-neutral probabilities to physical probabilities is calculated by dividing
the option-implied probabilities by the SPF-implied probabilities. This metric serves to illustrate the stylised fact that risk-neutral
probabilities tend to exceed their physical counterparts in the tails of the distributions, without aiming for an exact quantification of this
stylised fact. The illustration is adapted from Böninghausen et al. (2018).
Option-implied probabilities need to be interpreted with caution as regards their
levels, but their changes over time still convey important information. This is
because changes in risk-neutral probabilities and their physical counterparts will
broadly be in line with each other, unless there is a negative correlation between the
true, physical probabilities and risk premia. However, this would implausibly imply that,
in the case of deflation, for example, investors systematically revise downwards the
risk premium for deflation whenever the odds of this event materialising are seen as
increasing. If, instead, physical probabilities and risk premia can be expected to move
together in terms of direction over time, the evolution of option-implied probabilities
provides useful and timely signals regarding shifts in investors’ underlying inflation
outlook.7
Option-implied risk-neutral probabilities and physical probabilities both point
to changes in the distribution of inflation expectations in recent years. The
risk-neutral probabilities in Chart 7 are those implied by zero-coupon options whose
pay-offs depend on average euro area inflation over a five-year period starting today
(spot inflation expectations). The physical probabilities in Chart 8 are those related to
SPF inflation expectations in five years. Both distributions suggest that at the point in
time when the ECB’s asset purchase programme (APP) was introduced, the balance
of probabilities was tilting towards low, but positive, outcomes (between 0% and
1.5%), and – in the case of option-implied probabilities – also towards deflation (below
7 While there is no information available on the physical distribution underlying inflation options, this type of
co-movement with risk premia can also be seen in the fact that the sub-components of ILS rates
(i.e. estimated “genuine” expectations and premia) tend to be positively correlated.
-18
-12
-6
0
6
0
10
20
30
40
<-1.0% -1.0% to
-0.6% -0.5% to
-0.1% 0.0% to
0.4% 0.5% to
0.9% 1.0% to
1.4% 1.5% to
1.9% 2.0% to
2.4% 2.5% to
2.9% 3.0% to
3.4% 3.5% to
3.9% ≥4.0%
Option-implied (left-hand scale)
SPF (left-hand scale)
Percentage point difference between risk-neutral probabil ities and physical probabilities (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
25
0%). These were clear shifts relative to the situation two to three years earlier, when
both market participants and professional forecasters had assigned greater probability
to high-inflation outcomes (i.e. above 2.5%). The introduction of the APP broadly
coincided with the end of this tilting and, with the general inflation outlook improving
towards the end of 2016, the spectre of low inflation in the euro area gradually receded
and that of deflation in market-based measures vanished. This situation prevailed until
2019 in the case of physical probabilities and until the outbreak of coronavirus
(COVID-19) in the case of market-based probabilities, when probabilities for low
inflation and deflation returned to levels similar to those observed in 2015 and 2016.
Chart 7
Option-implied risk-neutral distribution of average euro area inflation over the next five
years
(percentages)
Sources: Bloomberg, Refinitiv and Eurosystem staff calculations.
Notes: Probabilities implied by five-year zero-coupon inflation options, smoothed over five business days. Risk-neutral probabilities may
differ significantly from physical probabilities. The latest observations are for 26 March 2021.
Chart 8
SPF physical probabilities for average euro area inflation five years ahead
(percentages)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: Based on SPF inflation expectations five years ahead. The latest observations are for Q1 2021.
0
10
20
30
40
50
60
70
80
90
100
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Below 0%
Between 0% and 1.5%
Between 1.5% and 2.0%
Between 2.0% and 2.5%
Above 2.5%
0
10
20
30
40
50
60
70
80
90
100
2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021
Below 0%
Between 0% and 1.5%
Between 1.5% and 2.0%
Between 1.5% and 2.0%
Between 2.0% and 2.5%
ECB Occasional Paper Series No 264 / September 2021
26
Changes in option-implied probabilities also suggest that investors have
started to price in gradually decreasing levels of inflation uncertainty. This is
evident from Chart 7 in the decline in the option-implied risk of deflation up until around
the time of the resumption of net asset purchases under the APP in 2019. It is also
evident from Chart 9, in so far as the downward movement in option-implied
volatilities – a gauge of the spread of the distribution – over the period from 2012 to
2014 was not subsequently reversed when the swap rate – a gauge of the central
tendency of the distribution – increased again. This supports the interpretation that
investors’ uncertainty regarding euro area inflation and the risk premia they are
demanding continue to be relatively low (see also Chart 2).
Chart 9
Inflation uncertainty as implied by euro area inflation options
(left-hand scale: basis points; right-hand scale: percentages)
Sources: Bloomberg, Refinitiv and Eurosystem staff calculations.
Notes: “Implied volatility” refers to the average implied volatility across five-year zero-coupon inflation options with different strike rates
(both “cap” and “floor” options). “Swap rate” refers to five-year euro area HICPxT-linked swaps. The latest observations are for 26 March
2021.
2.2.3 Assessing the relationship between moments
In the first instance, different moments of an inflation expectations distribution
provide independent pieces of information. In forecast terminology, they can be
seen as providing information on the baseline, risks and uncertainty. At the same time,
there is the question of whether movements in the shape of the distribution (e.g. as
measured by skewness) have indicator properties for movements in central
tendencies (e.g. as measured by means or medians). A negatively skewed distribution
signals downside risks to mean expected values, and if events such as drops in
inflation are realised more often than was predicted, then one could expect to see a
subsequent drop in the central tendency. Such indicator properties of higher moments
for central tendencies can theoretically be rationalised for both market-based and
survey-based measures, yet no clear evidence has been found empirically – for either
measure.
In the case of market-based measures, there is no clear evidence for skewness
or kurtosis having an indicator property for the mean. This is consistent with
0.0
0.5
1.0
1.5
2.0
2.5
0
100
200
300
400
500
2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Implied volatility (left-hand scale)
Swap rate (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
27
theoretical considerations indicating that skewness and kurtosis tend to change
simultaneously with changes in inflation expectations, given that investors’ risk
preferences (i.e. risk aversion) are state-dependent. An indicator property would rest
on the assumption that a shock to investors’ preferences translates directly to a
change in the higher-order moments, but takes some time to manifest itself in the
mean. To explore this hypothesis empirically, risk-neutral probability distributions for
maturities of one to ten years, generated from option prices, are used to compute
weekly values for the mean, variance, three different measures of skewness and
kurtosis. A bivariate VAR model is then fitted to the weekly series and the optimal lag
length is identified, and a test for Granger causality is then conducted for that lag
length. On the basis of the results, the hypothesis of higher-order moments being
informative for the future mean of inflation expectations is not supported. This is
consistent with the theoretical considerations expressed above.
Chart 10
Bowley skewness and inflation expectations for five-year inflation options
(left-hand scale: percentages; right-hand scale: percentage points)
Source: Eurosystem staff calculations.
Notes: “Bowley skewness” is the weekly change in that skewness measure for inflation option densities with a five-year maturity. The
latest observations are for 11 November 2020.
In the case of survey measures, participants may initially be hesitant about
expressing changes in their views directly in the form of changes in central
tendencies. This may be particularly true of longer-term inflation expectations, given
their association with the inflation objective and the central bank’s credibility in terms
of achieving this objective. In the event of doubts, forecasters may, instead, first
change their assessment of the shape of the inflation expectations distribution,
increasing skewness, and only change its mean once perceived risks have
materialised. However, Granger causality tests for the bivariate vector autoregressive
relationships between the average point forecast, the mean of the probability
distribution and different measures of risk and skewness at the two-year and five-year
horizons do not point to robust relationships (see Table 2).8 Using rolling window
8 The simple Balance of Risks Indicator is the difference between the mean of the probability distribution
and the average point forecast. The synthetic BoRI is essentially the average over different skewness
measures (including the simple BoRI, skewness itself and quantile skewness – with three distinct
interpolation methods applied to smooth over the probabilities in the bins of the survey.
-0.10
-0.05
0.00
0.05
0.10
0.15
0.0
0.5
1.0
1.5
2.0
2.5
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Expected inflation rate (left-hand scale)
Bowley skewness (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
28
regressions, there is some limited evidence of causality running from the BoRI to
average point forecasts and the mean of the probability distribution, but this is less
clear when estimated using the full sample. Thus, while higher moments of inflation
expectations distributions may provide information on risks and uncertainties, there is
no clear evidence that they anticipate movements in central tendencies.
Table 2
Granger causality analyses for relationships between different moments of medium
and longer-term SPF inflation expectations
Cause
Full sample regression Rolling window regression
Mean of probability
distribution Average point
forecast Mean of probability
distribution Average point
forecast
2y 5y 2y 5y 2y 5y 2y 5y
Standard
deviation 2y No No No No No No No Yes
5y No No No No No No No No
Skew 2y No No No No No No No No
5y No No No No No No No No
Simple BoRI 2y No No No No No No No No
5y No No No No No Yes No Yes
BoRI 2y No No No No No No No No
5y No VAR(2) No No No No No Yes
Source: Eurosystem staff calculations.
Notes: “VAR(X)” denotes a VAR with X lags. The estimation period is 2010-19. The assessment for the rolling regressions is based on the
median sum of coefficients. The rolling regressions cover an eight-year window and are based on bivariate regressions with up to four
lags.
2.3 The relationship between market and survey-based
expectations
The existence of some strong co-movement between market and survey-based
inflation expectations suggests that they may not contain fully independent
information. When interpreting the two measures of expectations, it is important to
know whether the signals they provide contain new information about true underlying
expectations, or whether one measure is simply following the other (i.e. survey
expectations are following market expectations, or vice versa). For instance, in a
special SPF survey in 2018, around 50% of respondents indicated that they used
market-based measures as one of several inputs when forming their longer-term
inflation expectations. This percentage had increased since the previous special
survey in 2013. From that perspective, co-movement and possibly even leading
indicator properties between market-based and survey-based inflation expectations
would have a natural explanation. That issue is explored here using two different
approaches.
The first approach uses a regression-based mixed data sampling model. This
model accounts for the frequency mismatch between the quarterly SPF responses
and the daily ILS rates by letting the daily information on ILS rates decay over the
ECB Occasional Paper Series No 264 / September 2021
29
quarter and thus allowing more recent observations to have greater importance.9 It
estimates the interlinkages at the one-year, two-year and five-year horizons and
allows for the presence of possible common elements in the inflation outlook by both
market and survey participants by the inclusion of different sets of control variables,
ranging from no controls beyond an autoregressive component (Model 1) to HICP
projections for the same horizon as the expectations10 (Model 2) to macroeconomic
factors that relate to the current state of the economy, namely euro area HICP and
manufacturing and services PMIs11 (Model 3). The results point to daily ILS rates
having statistically significant information content for SPF responses at all horizons,
irrespective of the choice of controls (see Chart 11). However, the regression
coefficient for ILS rates decreases when adding controls in Models 2 and 3,
suggesting that common factors explain a fair amount of the correlation between
market and survey-based measures at the short end. At longer horizons, the
interlinkage is economically negligible, with a very small regression coefficient for ILS
rates implying that a 1 p.p. increase in the five-year ILS rate over the quarter
corresponds to a rise of just 0.04 p.p. in the SPF expectation for that same horizon.
Chart 11
Estimated impact of ILS rates on SPF responses
(percentage points per 1 percentage point change in ILS rates)
Sources: Refinitiv, ECB and Eurosystem staff calculations.
Notes: This chart depicts the coefficients for the ILS term and their respective 95% confidence bands for three models. Model 1
represents the baseline model, which estimates the impact of the weighted average of ILS rates over the 65 business days before the
mandatory reply date for survey respondents, accounting for an autoregressive component. Model 2 adds other inflation projections
released just before the SPF response date – namely, (B)MPE projections for the one-year and two-year horizons, as well as Consensus
Economics projections for the five-year horizon. Model 3 expands on Model 2 by adding average actual HICP and manufacturing and
business PMIs over the quarter. All regressions employ a decaying weighting structure estimated using a restricted beta weighting
model. Different weighting structures produce very similar results.
9 The model follows Hanoma & Nautz (2019) and estimates the effect of the weighted average of ILS rates
over the quarter before the mandatory reply date for survey respondents. Specifically, as there are
around 65 business days on average between the reply deadlines for SPF rounds, the estimated model
takes the following form: = ++(;)++
.
10 HICP forecasts from the Eurosystem’s Broad Macroeconomic Projection Exercises are used for the one
and two-year horizons, while a Consensus Economics forecast is used for the five-year horizon.
11 Market-based measures of inflation expectations may themselves be informed by previous SPF survey
results. Controlling for past SPF survey responses via the lagged dependent variable helps to capture
such two-way feedback effects.
0.0
0.1
0.2
0.3
0.4
Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 Model 1 Model 2 Model 3
1y 2y 5y
ECB Occasional Paper Series No 264 / September 2021
30
The second approach applies the concept of Granger causality analysis using a
VAR model. The model has quarterly frequency and alternative specifications with
one or four lags, and with levels or changes respectively. It tests for causality between
longer-term expectations, using the average point forecast and the mean of the
probability distribution five years ahead on the SPF side, and the 1y4y and 5y5y ILS
rates on the market side. The results point to some partial causality running from
survey to market data (and, albeit to a lesser extent, in the opposite direction), but also
some variation in the significance of this causality over time (see Chart 12).
Chart 12
Granger causality of survey and market-based longer-term inflation expectations
(p-values)
Source: Eurosystem staff calculations.
Notes: These results are for the model specification in levels with one lag. The x-axis denotes the rolling windows over which the Granger
causality tests are carried out (e.g. “Q1 11 Q4 20” denotes the period from Q1 2011 to Q4 2020). Values below the red line denote
statistical significance at the 10% level.
Overall, empirical evidence on the nature of interlinkages between survey and
market-based longer-term expectations remains inconclusive. The sometimes
high degree of correlation is not unambiguously grounded in causal relationships
between information sources. Whether such relationships are evident appears to
depend on the set-up of the models and tests in question. This suggests that market
and survey-based measures contain complementary information that may be useful to
policymakers, confirming that both types of measure should be monitored in parallel.12
At the same time, policymakers might be interested in the signal that emerges across
the two sets of measures. The Federal Reserve System and the Bank of Japan have
developed synthetic indicators which distil the common components across consumer
price expectations in market prices and different surveys and across both shorter and
longer horizons.13 When applying dynamic factor model analysis, the set of
expectations that is considered can greatly influence the results. For instance, when
12 This finding is consistent with information in surveys that are absent from (i.e. not spanned by) the
inflation swap curve that helps to predict future returns for inflation-linked swaps in Speck (2019).
Meanwhile, Reis (2020) develops a model where people, markets and traders form their inflation
expectations differently. In his framework, the role of financial shocks relative to natural rate shocks
determines whether the inflation expectations of markets or people have better forecasting performance.
13 For the Federal Reserve, see Ahn & Fulton (2021); for the Bank of Japan, see Nishino et al. (2016).
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Q1
05 Q2
05 Q3
05 Q4
05 Q1
06 Q2
06 Q3
06 Q4
06 Q1
07 Q2
07 Q3
07 Q4
07 Q1
08 Q1
08 Q2
08 Q3
08 Q4
08 Q1
09 Q1
09 Q2
09 Q3
09 Q4
09 Q1
10 Q1
10 Q2
10 Q3
10 Q4
10 Q1
11 Q1
11 Q2
11 Q3
11 Q4
11 Q1
11
Q4
12 Q1
13 Q2
13 Q3
13 Q4
13 Q1
14 Q2
14 Q3
14 Q4
14 Q1
15 Q2
15 Q3
15 Q4
15 Q1
16 Q2
16 Q3
16 Q4
16 Q1
17 Q2
17 Q3
17 Q4
17 Q1
18 Q2
18 Q3
18 Q4
18 Q1
19 Q2
19 Q3
19 Q4
19 Q1
20 Q2
20 Q3
20 Q4
20
SPF_1y4y_pnt Granger causes Mkt_1y4y
Mkt_1y4y Granger causes SPF_1y4y_pnt SPF_1y4y_prb Granger causes Mkt_1y4y
Mkt_1y4y Granger causes SPF_1y4y_prb
ECB Occasional Paper Series No 264 / September 2021
31
examining euro area data, including all the different series of short-term expectations
for households and firms and combining them with the few longer-term expectations
derived from surveys and markets results in the dynamic factor model picking up
movements in actual inflation. Focusing only on longer-term market and survey-based
expectations yields a different result (see Chart 13).14 Hence, while a synthetic
inflation expectation indicator might be useful in principle for communication purposes,
its actual construction is, in practice, subject to uncertainties that could defeat the
original purpose.
Chart 13
Euro area inflation and dynamic factor model analysis of different measures of
expectations
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: Expectations comprise series derived from the SPF, Consensus Economics, market prices, EC household surveys and EC
business surveys. “S14” refers to a model based on 14 series (three SPF, three CE, four ILS, one ECCS and three ECBS). “S3” refers to
a model based on three series (one SPF, one CE and one ILS). The estimates were made using the method employed by Ahn & Fulton
(2021).
Box 1
Inflation perceptions and expectations: evidence from the ECB Consumer Expectations
Survey
Monitoring and managing consumers’ inflation expectations are major goals for
policymakers. To that end, there is demand for high-quality survey-based measures of inflation
expectations. With that in mind, the ECB’s new Consumer Expectations Survey aims to enrich our
understanding of consumers’ inflation expectations. The CES was launched in January 2020 and
collects, using a panel structure, monthly data on consumers’ price development expectations for
each of the six largest euro area economies (namely, Germany, France, Italy, Spain, the Netherlands
and Belgium). In April 2020, the CES reached its target sample size, surveying 10,000 households in
total. The CES provides a unique cross-country perspective on various aspects of consumers’
14 In this instance, three inflation expectations series were considered: (i) expectations five years ahead
derived from the SPF, (ii) expectations six to ten years ahead derived from Consensus Economics, and
(iii) the five-year ILS rate five years ahead. If the one-year ILS rate four years ahead was also included,
the estimated coefficient moved almost exactly in line with the market-based measures as a result of its
strong co-movement with the five-year rate five years ahead.
-3
-2
-1
0
1
2
31999 2002 2005 2008 2011 2014 2017 2020
S14 (left-hand scale)
HICP_norm (right-hand scale)
-3
-2
-1
0
1
2
31999 2002 2005 2008 2011 2014 2017 2020
S3 (left-hand scale)
HICP_norm (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
32
inflation expectations and behaviour. An important feature of the CES is the panel dimension, which
allows it to track individual inflation expectations and types of consumer behaviour over time.15
The CES provides both qualitative and quantitative measures of consumers’ inflation
perceptions. These measures are backward-looking, as the CES asks respondents about their
current perception of prices in general compared with 12 months ago. In addition, the CES elicits both
qualitative and quantitative forward-looking measures of short and medium-term inflation
expectations, asking about inflation expectations over the next 12 months and between two and three
years ahead respectively.16 The CES elicits a probabilistic measure of inflation expectations within
the spirit of a large and growing body of economic research led by Manski (2004). That measure
provides density forecasts, which enable respondents to express their uncertainty about their own
inflation expectations.17
Currently available survey results suggest that median short and medium-term inflation
expectations are well anchored, as they are aligned with the ECB’s inflation target of
“below, but close to, 2%” (see Chart A). The distributions of inflation perceptions and inflation
expectations are skewed to the right, as the means are higher than the corresponding medians. The
interquartile range (i.e. the difference between the 75th and 25th percentiles) represents a measure
of disagreement among consumers and conveys information about inflation uncertainty. It shows that
there is, on average, slightly more disagreement about short-term inflation expectations than there is
about inflation perceptions and medium-term inflation expectations. Chart B suggests that, with the
exception of Italy, there is relatively little cross-country heterogeneity in short-term inflation
expectations.
15 Georgarakos and Kenny (2021) provide a more detailed description of the CES and ECB (2021) contains
a first evaluation of the survey.
16 In particular, for medium-term inflation expectations , the CES asks respondents by about what
percentage they expect prices in general in the country you currently live in to increase (decrease) over
the 12-month period <between survey month year+2 and survey month year+3>.
17 In particular, the CES asks respondents to indicate the probability that inflation over the next 12 months
will fall within eight different pre-specified categories: “prices will increase by 8% or more”; “prices will
increase by 4% or more, but less than 8%”; “prices will increase by 2% or more, but less than 4%”; “prices
will increase by less than 2%”; “prices will decrease by less than 2%”; “prices will decrease by 2% or
more, but less than 4%”; “prices will decrease by 4% or more, but less than 8%”; and “prices will decrease
by 8% or more”.
ECB Occasional Paper Series No 264 / September 2021
33
Chart A
Quantitative measures of inflation
(annual percentage changes)
Source: ECB CES.
Note: Pooled and weighted data across waves from April to December 2020. Using weighted data. Statistics computed from open-end questions on inflation with
different time horizons (12 months before interview date and 12 months / 3 years ahead of interview date). Question(s) asked: (a) Open-ended (quantitative)
questions on inflation perceptions (past 12 months) and inflation expectations (12 months ahead and 3 years ahead. Latest observation: December 2020.
Chart B
Median inflation expectations and uncertainty over next 12 months across countries
Source: ECB CES.
Note: Pooled data across waves. Using weighted data. Disagreement and inflation expectations are obtained from the open-ended question about individual
expectation of changes in prices in general over the next 12 months. Question asked: How much higher (lower) do you think prices in general will be 12 months
from now in the country you currently live in? Please give your best guess of the change in percentage terms. You can provide a number up to one decimal point.
This chart shows the median and the average interquartile range for inflation expectations over the next 12 months across countries. Latest observation:
December 2020.
Making use of the rich set of individual characteristics in the CES data, we provide useful
insights into the heterogeneity of consumers’ inflation expectations across specific
demographic and socio-economic groups (see Table A). CES results reveal that inflation
expectations for the next 12 months and in three years’ time are higher for female consumers than
male consumers, increase with age and decrease with a high level of education, financial literacy and
income. This is in line with previous studies.18 In addition, recent studies using the EC surveys
18 See, for example, Bryan & Venkatu (2001), Lusardi (2008), Bruine de Bruin et al. (2010), Bruine de Bruin
et al. (2011), Diamond et al. (2020) and D’Acunto et al. (2021).
0
1
2
3
4
5
6
Inflation perceptions Inflation expectations 12 months ahead Inflation expectations 3 years ahead
Median
Mean
Average interquartile range
0
1
2
3
4
5
6
7
8
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
EA BE DE ES FR IT NL
Median inflation expectations 12 months ahead (left-hand scale)
Inflation disagrement (average interquartile range) (ri ght-hand scale)
ECB Occasional Paper Series No 264 / September 2021
34
provide similar evidence on the effects of age, gender, education and income on inflation
expectations.19 In addition, consumers whose household consists of more than five members tend,
on average, to have higher short and medium-term inflation expectations.
Table A
Heterogeneity in inflation expectations and inflation uncertainty
(percentages)
Source: ECB CES.
Notes: Pooled data across waves. Averages using weighted data. Data has been winsorised at the 2nd and 98th percentile. The interquartile range is averaged.
Medians are taken over the full sample. Inflation Uncertainty is derived as the standard deviation from a probabilistic question asking respondents to distribute
100 points in pre-defined intervals. Question(s) asked: Open-ended (quantitative) questions on inflation perceptions (past 12 months) and inflation expectations
(12 months ahead and 3 years). Latest observation: December 2020.
Consumers who have temporary employment and liquidity constraints tend to have higher
inflation expectations 12 months and three years ahead. Also, a low level of trust in the ECB
tends, on average, to be associated with higher short and medium-term inflation expectations,
consistent with Christelis et al. (2020).
19 See Arioli et al. (2016) and Meyler & Reiche (2021).
Inflation expectations
over next 12 months Inflation expectations
three years ahead Inflation uncertainty over next
12 months
Mean IQR Mean IQR Mean IQR
Gender Male 3.2 4.4 3.3 4.2 1.6 1.8
Female 4.4 5.7 4.0 5.2 1.7 2.1
Age 18-34 3.4 4.5 3.00 3.9 1.9 2.3
35-49 4.0 5.3 3.7 4.7 1.7 2.1
50-64 4.2 5.1 4.0 4.8 1.6 1.8
65+ 3.7 4.5 3.9 4.9 1.4 1.4
Education Primary 4.2 5.5 4.1 5.4 1.7 2.1
Secondary 4.2 5.6 4.0 5.2 1.7 2.1
Tertiary 3.5 4.6 3.4 4.4 1.6 1.8
Household size 1 3.6 5.0 3.5 4.7 1.5 1.7
2 3.7 4.6 3.5 4.5 1.5 1.7
3 3.9 5.1 3.8 4.8 1.8 2.2
4 4.0 5.4 3.7 4.7 1.8 2.2
5 or more 4.5 5.8 4.4 5.8 2.0 2.4
Employment type Permanent 3.5 4.6 3.3 4.3 1.6 1.9
Temporary 4.2 5.4 4.2 5.3 1.9 2.4
Financial literacy Below median 4.5 6.3 4.2 5.7 1.8 2.4
Median or above 3.4 4.5 3.4 4.2 1.5 1.8
Trust in the ECB Low level of trust 5.1 5.9 5.0 6.1 1.8 2.2
Neither 3.8 5.1 3.7 5.0 1.7 2.0
High level of trust 3.1 4.2 2.9 3.9 1.5 1.8
Income quartile 1 4.6 6.3 4.4 5.6 1.8 2.4
2 4.0 5.2 3.7 4.8 1.7 2.1
3 3.4 4.6 3.3 4.4 1.5 1.8
4 3.2 4.2 3.2 4.0 1.5 1.7
Liquidity constrained? Yes 4.9 6.6 4.8 6.3 2.0 2.6
No 3.4 4.6 3.2 4.3 1.5 1.8
ECB Occasional Paper Series No 264 / September 2021
35
Turning to the probabilistic measure of inflation expectations over the next 12 months
provided by the CES, Chart C displays the average distribution of the probabilistic responses
for inflation expectations, which are allocated to eight bins. 85% of the responses are in bins
associated with prices increases, being spread fairly evenly between “less than 2%”, “2% or more, but
less than 4%”, “4% or more, but less than 8%” and “more than 8%”.
Chart C
Average histogram for the probabilistic measure of inflation expectations
(average probability allocated by respondents to category)
Source: ECB CES.
Notes: Pooled data across waves from April to December 2020. Using weighted data to compute the shares. Questions asked: Individual-level data on inflation
expectation for prices in general over the next 12 months is derived from a probabilistic question asking respondents to distribute 100 points in pre-defined
intervals. Point-forecasts of 12 months ahead inflation expectations are obtained from asking respondents about the numerical forecast in a range from -100 to
100 percent allowing also for the use of decimals.
Eliciting consumers’ subjective probability distribution for future inflation outcomes allows
us to construct an individual measure of inflation uncertainty, which is the standard deviation
of each individual’s probability distribution. As with inflation expectations, there is heterogeneity
in inflation uncertainty across specific demographic and socio-economic groups (see the last two
columns of Table A). Female, younger, less educated and liquidity constrained consumers all tend, on
average, to have higher uncertainty in their inflation expectations, as do consumers with a temporary
job, a low level of income and a low level of financial literacy. The negative correlation between
financial literacy and inflation uncertainty is also documented in Bruine de Bruin et al. (2011). We also
provide evidence that inflation uncertainty decreases with trust in the ECB, which is in line with
Christelis et al. (2020). Meyler & Reiche (2021), using the data from the EC surveys, also provide
evidence on inflation uncertainty, focusing on the share of respondents that report inflation
expectations using round numbers (specifically, multiples of five). In line with the findings of the CES,
they show that older and male consumers and consumers with high levels of education and income
are more certain about their inflation expectations, as they are less likely to report inflation
expectations using round numbers.
0
5
10
15
20
25
30
8% or more 4% or more, but
less than 8% 2% or more, but
less than 4% Less than 2% Less than 2% 2% or more, but
less than 4% 4% or more, but
less than 8% 8% or more
Price decreases Price increases
ECB Occasional Paper Series No 264 / September 2021
36
Box 2
Technical factors as drivers of market-based measures of inflation compensation
In interpreting the economic signals from market-based measures of inflation compensation,
central banks need to consider not only the presence of risk premia for unexpectedly high or
low future inflation, but also that of market imperfections, or technical factors.
This box takes stock of pertinent features of euro area inflation-linked markets with a view to
assessing the extent to which such imperfections may interfere with the economic
interpretation of information from market-based indicators. In assessing possible distortions to
the signals from market-based measures of inflation compensation, it is useful to distinguish between
the effects on ILS rates and bond-implied BEIRs. ILS rates are widely regarded as the preferred
measure of inflation compensation in the euro area, including by market participants. They are readily
available with fixed maturities, while the calculation of bond-based BEIRs with fixed maturities is
complicated in the euro area by the small number of inflation-linked bonds per country and the
differences in credit risk across countries. However, there is little data available about the
microstructure of inflation swap markets, while somewhat better datasets are available for
inflation-indexed bonds. Given that ILS rates and BEIRs are connected via no-arbitrage
considerations of some kind, two key questions arise from a monetary policy point of view. First, do
any distortions emanating from bond markets also translate into significant distortions of ILS rates?
And second, if they do, has the impact of such distortions changed significantly over time, such that
longer-term trends in ILS rates and BEIRs would need to be re-evaluated in the light of the existing
evidence?
In general, markets for inflation-linked assets are considerably smaller than other markets in
terms of volume and tend to be more concentrated in terms of investor activity. In the
inflation-linked bond market, for example, the outstanding amount of bonds for Germany, France and
Italy amounts to only around 8% of the overall market for their nominal peers (Di Iorio & Fanari, 2020).
Likewise, order book data from MTS, the largest electronic trading platform for European fixed
income assets, reveal that the volume of offers to buy and sell nominal bonds from those three
countries exceeds the volume of inflation-linked bonds. In the inflation-linked swap market, trades are
significantly fewer in number and involve less substantial volumes relative to other derivative markets.
For instance, a ballpark estimate suggests that the outstanding notional amount of EONIA and
EURIBOR-linked interest rate derivatives may exceed that of euro area inflation-linked swaps by a
factor of as much as 50 (see Chart A).
ECB Occasional Paper Series No 264 / September 2021
37
Chart A
Relative depth of different interest rate derivative markets vs euro area HICP-linked swap market
(multiples based on outstanding notional amounts (ratios))
Sources: EMIR database and Eurosystem staff calculations.
Note: The ranges depicted are based on the highest and lowest market depth multiples as of the 16 May 2018, 12 June 2018, and 19 December 2019 reporting
dates.
It is thus important to be mindful of their relatively small size, albeit the signals from
inflation-linked markets are not uninformative per se as long as activity is deemed sufficient.
In fact, many of the above-mentioned non-inflation-linked markets that serve as reference points for
activity in inflation-linked markets are themselves significantly smaller than other bond markets
(e.g. US Treasury bonds) or derivatives markets (e.g. foreign exchange derivatives). In other words, a
relatively low level of activity is not enough to dismiss outright the signals from inflation-linked
markets – or, indeed, any other market. Conceptually, the relevant benchmark for assessing the
usefulness of these signals is whether inflation-linked market activity is sufficient in its own right. In
practice, however, this benchmark is unknown and therefore ultimately subjective. Hence, the fact
that inflation-linked markets have low levels of activity relative to other markets justifies a closer
examination of market distortions.
Conceptually, the possible distortion of signals from inflation-linked markets may be further
broken down into a discussion of levels and dynamics, but those two concepts interact in
practice and empirical evidence is as yet incomplete. The effects of Eurosystem public sector
asset purchases on market-based measures of inflation compensation in the euro area are a case in
point. On the one hand, it has been argued that a key source of bias arising from these purchases has
been a duration extraction channel.20 According to this argument, the Eurosystem might have
reduced the free float of nominal government bonds held by price-sensitive investors by more than
that of inflation-linked bonds, thus compressing nominal yields (on the former) by more than real
yields (on the latter). As a result, BEIRs might have been compressed for technical reasons.
However, in line with its market neutrality principle, the Eurosystem’s purchases actually helped to
reduce the free float of nominal and inflation-linked bonds to a very similar extent. This would suggest
that whatever distortions might have existed in the levels of BEIRs before the start of purchases were
not aggravated further. In other words, the dynamics of market-based measures of inflation
compensation were arguably largely unaffected by these purchases. On the other hand, the free float
20 For information on the duration extraction channel more generally, see Eser et al. (2019).
0
10
20
30
40
50
60
EONIA swaps EURIBOR futures EURIBOR swaps EONIA/EURIBOR swaps and
futures combined
ECB Occasional Paper Series No 264 / September 2021
38
of nominal bonds was already visibly lower than that of inflation-linked bonds before the start of
purchases, so it also stands at a noticeably lower level today following the Eurosystem’s purchases.
Thus, if the yield impact of a given reduction in the free float – say, 1 p.p. of the relevant outstanding
amount – were to increase as the level of the free float approached “low” values, market-neutral
Eurosystem purchases could still have added to the downward trend in BEIRs by compressing
nominal yields more than real yields.21 Such complex interactions notwithstanding, the evidence
presented below focuses on assessing evidence on the impact that technical factors have on the
dynamics of market-based measures of inflation compensation, which appears, at present, to be
somewhat more robust than the impact on the levels of these measures.
For ILS markets, first evidence from swap repository data that have recently become
available does not suggest that the dynamics of ILS rates – in the form of a significant recent
decline – can be attributed to variations in market activity. More specifically, trade repository
data from EMIR help to trace activity in euro area ILS markets. While comprehensive historical
analysis is not possible for data coverage and quality reasons, that part of the data which is deemed
reliable enough covers both the notable decline in market-based indicators of longer-term inflation
expectations between the autumn of 2018 and mid-2019, as well as the more volatile developments
in 2020 during the pandemic. The data indicate that activity remained broadly stable during this
decline, rather than being systematically higher or lower than before. Similarly, there do not appear to
be significant differences in terms of euro area ILS market activity between the steadier decline and
the more tumultuous 2020 (see Chart B). This initial evidence does not, therefore, suggest that the
informational content of euro area ILS rates has changed systematically. Hence, variations in market
activity do not alleviate concerns about the decline in longer-term ILS rates in the euro area.
Chart B
Activity in euro area ILS markets and the euro area 5y5y ILS rate
(left-hand scale: standardised values; right-hand scale: percentages per annum)
Sources: Bloomberg, EMIR database and Eurosystem staff calculations.
Notes: Activity is based on newly logged euro area HICPxT-linked ILS transactions between adjacent de-duplicated EMIR trade state reports (generally weekly
frequency). New transactions are filtered in nine different ways on the basis of their execution and effective dates. Both the number of new transactions and their
volumes are summed over rolling windows of ten business days and standardised in-sample for comparability. The chart shows the minimum, maximum and
median for these standardised values. The latest observations are for 7 October 2020.
21 The aforementioned arguments imply identical sensitivity to changes in the respective free-float ratios on
the part of inflation-indexed and nominal bonds. Allowing for the possibility of a stronger general impact
on nominal yields than on real yields would therefore provide further grounds for exploring a downward
impact on BEIRs.
0.6
0.8
1.0
1.2
1.4
1.6
1.8
-2
-1
0
1
2
3
4
07/18 10/18 01/19 04/19 07/19 10/19 01/20 04/20 07/20 10/20
Median
Min-max range
5y5y (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
39
In inflation-linked bond markets, trading costs may be one source of market distortions that
is particularly relevant in times of stress. These costs can be observed in terms of bid-ask
spreads, for example, as the difference between the prices at which market participants are willing to
sell or buy inflation-linked bonds. As Chart C reveals, bid-ask spreads as a proxy for trading costs for
inflation-linked bonds tend to increase particularly in times of heightened general market stress. This
means that, in such periods, the price signal from inflation-linked bonds may be much less precise
than it is in normal periods. Furthermore, the volume of offers to buy and sell bonds in the order book
declines. Higher costs and lower supply both prevent market participants from trading in times of
crisis, such that BEIRs may deviate considerably from their “economically justified” level. Three
periods stand out in particular: the sovereign debt crisis (2010-13), which was marked by substantial
safe-haven flows into nominal safe assets; the period between 2015 and 2017; and early 2020. In the
second and third of those periods, there was arguably a lack of available safe assets on account of
central bank asset purchases and the turmoil at the start of the pandemic respectively. High costs and
limited supply can both result in market distortions and biases in the prices of inflation-linked assets,
as they can potentially impair arbitrage and price-finding mechanisms.22
Chart C
Relative illiquidity of inflation-linked bonds
(yield bid-ask spreads for German and French bonds (20-day average; median and interquartile range))
Sources: MTS and Speck (2021).
Notes: Bid-ask spreads for inflation-indexed bonds and their maturity-matched nominal equivalents. The median across all ISIN pairs for German and French
bonds is the solid line; the shaded areas indicate the interquartile range. Bid-ask spreads are quoted in MTS for bond prices in euro. The “yield” bid-ask spread
displayed in the chart is calculated as the quoted bid-ask spread divided by the mid-price and the bond’s remaining time to maturity.
An empirical quantification of these effects suggests that their overall contribution accounts
for only part of the long-term decline in market-based measures of long-term inflation
compensation. The relationship between ILS rates – as a stand-in for market-based measures of
inflation compensation more generally, due to their connection to BEIRs via a (loose) arbitrage
relationship – and proxies for market factors can be studied by estimating rolling regressions23 of the
following kind:
=++
22 See Christensen & Gillan (2019), Driessen et al. (2017), Fleckenstein et al. (2014) and Haubrich &
Pennacchi (2012).
23 See Speck (2021). The estimation is conducted with a 250-day rolling window.
0
5
10
15
20
25
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Inflation-linked bonds
Nominal bonds
ECB Occasional Paper Series No 264 / September 2021
40
The explanatory variable contains the median of the average yield bid-ask spread (BAS) as a
proxy for trading costs. As additional variables, it contains the first three principal components of the
KfW-Bund spread curve, a commonly used proxy for general market imperfections (such as asset
scarcity) in the Bund market (Hördahl & Tristani, 2015), and the difference in credit default swap
(CDS) spreads between subordinate and senior bank debt as a proxy for the availability of arbitrage
capital. Most importantly, the exercise confirms that in the specific stress periods mentioned above,
the included measures of market imperfections do indeed have significant explanatory power for
market-based measures of inflation compensation at times. Bond market trading costs (BASs) played
an important role in the COVID-related spikes in early 2020, but the general market imperfections
represented by the KfW-Bund and the banking sector also contributed as well (see Chart D).
Compared with the results seen in the literature (which mostly relate to US data), the magnitude of the
premia for technical factors is fairly low.24 In the case of the five-year forward ILS rate five years
ahead, the above measures of market imperfections only seem to account for around 30 basis points
of the substantial decline seen since 2008 (which has totalled around 100 basis points).
Chart D
Bias of technical factors in five-year forward inflation swap rates five years ahead
(basis points)
Sources: Bloomberg, MTS and calculations of Speck (2021).
Notes: Cumulative contributions since 2007 for the factors β_i * ΔXt. Each data point t is contained in multiple rolling regression windows i. For each t, the
cumulative contributions are evaluated by simply averaging β_i across all samples containing data point t. The latest observations are for 8 January 2021.
Overall, while the presence of technical factors points to a certain bias in market-based
indicators of inflation compensation in periods of stress, there seems to be little ground for
concluding that such distortions were the key driver of the trend decline seen in
market-based measures of inflation compensation in the euro area over the last decade.
However, this conclusion on the dynamics of market-based measures is consistent with a wide range
of possible distortions in the levels of ILS rates and BEIRs. Moreover, the uncertainty around such
possible distortions remains considerable. To this end, research is needed on trading costs and
demand-supply imbalances for safe assets in particular, which may help to arrive at a more
comprehensive quantification of the impact of technical factors in inflation-linked markets.
24 See the literature review in Kupfer (2018).
-160
-140
-120
-100
-80
-60
-40
-20
0
20
40
60
2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 202
KfW1
KfW2
KfW3
BAS
CDS
Combined contribution of technical factors
Change in euro area 5y5y ILS rate
ECB Occasional Paper Series No 264 / September 2021
41
3 Drivers and (un)anchoring of inflation
expectations
3.1 The drivers of inflation expectations
3.1.1 Conceptual framework for the drivers of inflation expectations
Inflation expectations can be driven by myriad factors – differing across
horizons and different types of agent. Figure 2 presents a stylised overview of the
drivers of both short and long-term expectations. The former relate, to a large extent,
to shocks that drive inflation developments at the business cycle frequency, whereas
the latter are linked to the credibility of the central bank in pursuing its inflation aim. If
the delineation between short and long-term expectations is associated with the time
horizon at which the impact of shocks fades out, then the fact that some shocks can be
very persistent complicates the assessment. While this fading impact can be traced
along the inflation expectations curve, information from the SPF suggests that some
respondents consider the impact of some shocks to be more persistent than the
five-year horizon covered in the survey.25
Figure 2
Drivers of expectations at different horizons
Source: Eurosystem staff.
Regression-based analysis suggests that the horizon over which the impact of
shocks fades out along the inflation expectations curve has lengthened since
25 See de Vincent-Humphreys et al. (2019).
Shorter term
Information on outlook
and shocks
Benchmark for central
bank’s projection
Longer term
Information on
perceived central
bank credibility
Benchmark for
effectiveness of
monetary policy
(including forward
guidance)
When do shocks
fade out?
ECB Occasional Paper Series No 264 / September 2021
42
the GFC. Consensus Economics expectations for the euro area for different horizons
are explained as the weighted sum of estimated steady-state inflation expectations
and actual inflation, where the coefficient for actual inflation captures the decaying
impact of new information.26 Regression results suggest that the coefficient for actual
inflation becomes zero (and the coefficient for steady-state inflation expectations turns
unity) after two to three years in the period 1998-2008 and after four to five years in the
period 2009-20 (see Chart 14). This lengthening of the adjustment process comes on
top of some downward movement in estimated steady-state inflation expectations.
This lengthening and downward movement could be interpreted as signs of
unanchoring, but it could also be argued that they are the result of more severe or
persistent shocks. The estimated adjustment for the euro area is longer than the US
equivalent, but shorter than that seen for Japan. And within the euro area, the
lengthening of that horizon in the post-GFC period has been particularly noticeable for
Italy (results not shown here).
Chart 14
Coefficient for estimated “steady-state inflation expectations” in relation to the forecast
horizon (left-hand panel) and estimated steady-state inflation (right-hand panel)
Estimated “steady-state inflation
expectations” in relation to the forecast
horizon
Estimated steady-state inflation
(coefficients) (annual percentage changes)
Source: Eurosystem staff calculations.
Note: Beta denotes the coefficient for steady-state inflation expectations in the regression explained in footnote 26.
The speed at which the impact of shocks fades out in economic models is
influenced by the way in which agents form their expectations. In economic
modelling, the embedded expectation formation process typically falls somewhere
between purely adaptive expectations (based only on past information) and
“model-consistent” full information rational expectations (FIRE). With adaptive
expectations, the inflation expectations curve reflects the impact of shocks on actual
26 This approach is based on Mehrotra & Yetman (2018). Here, it is implemented in the form of the
state-space model =+(1)+, where h = 0, 1, 2, 3, 4, 5 and 6-10 years is the forecast
horizon, while unobserved steady-state inflation expectations are given by =
+.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
β0 β1 β2 β3 β4 β5 β6 β7 β8 β9 β10
1989-1998
1999-2008
2009-2020
1.4
1.6
1.8
2.0
2.2
2.4
2.6
2.8
3.0
3.2
3.4
3.6
1989 1993 1997 2001 2005 2009 2013 2017
1989-2020
1989-1998 1999-2008
2009-2020
ECB Occasional Paper Series No 264 / September 2021
43
inflation, while with rational expectations it is crucially determined by the central bank’s
credibility in terms of achieving its inflation aim. Available findings suggest that
survey-based measures of expectations embed an “intermediate” level of rationality
(that is to say, they are neither fully rational, nor do they follow a simple autoregressive
model) and that this varies with the type of agent. More recent work emphasises
deviation from rational expectations owing to information rigidities (e.g. rational
inattention), model misspecification (e.g. bounded rationality) and learning (as agents
often observe shocks, but do not know the parameters governing dynamics in the
economy and act as an econometrician in each period to estimate a perceived law of
motion). These hybrid formation mechanisms mixing adaptive elements and rational
expectations imply that if an economic shock affects inflation, expectations will also be
affected, but not as much as actual inflation.27
It is probably the case that different agents form expectations in different ways.
This holds both across and within categories of agents. For instance, quantitative data
on consumers’ inflation expectations one year ahead as derived from the European
Commission Consumer Surveys co-move closely with actual inflation, thus pointing to
a more adaptive expectation formation process. There is also considerable
heterogeneity across individual households in their reported inflation expectations. In
contrast, professional forecasters use a mix of models and informed judgement to
form their inflation expectations. The 2018 special SPF survey revealed that
reduced-form models were the main tool used for short and medium-term
expectations, while for longer-term expectations models with economic structure were
more widely used. When combining models and judgement for longer-term
expectations, expert judgement is the dominant factor (see Chart 15) and can be
linked to belief in the monetary authority’s ability to achieve its inflation objective. The
inflation expectations of financial market participants can be expected to be formed in
a similar way to those of professional forecasters, if only because the SPF includes
many participants from the financial sector. However, market-based measures of
inflation expectations also contain risk premia, implying that, in addition to factors
which are relevant for the central tendency of the inflation outlook, they also reflect
factors relating to risk sentiment and uncertainty.
27 For an overview of literature on expectation formation mechanisms, see Coibion et al. (2018). Nakamura
(2021) argues that by “solving forward” the New Keynesian Phillips curve, one can show that long-run
inflation expectations may drive large movements in current inflation.
ECB Occasional Paper Series No 264 / September 2021
44
Chart 15
Formation of inflation expectations in the SPF
(percentages)
Source: ECB (2018 special SPF survey).
In what follows, the discussion of drivers of inflation expectations focuses on a
limited subset. This subset is meant to be representative of factors that operate at
different horizons of the inflation expectations curve and includes oil price shocks,
monetary policy shocks and the link between inflation expectations and inflation trends
and the macroeconomy more generally.28 Owing to data limitations, the focus of the
empirical analysis is on professional surveys and market-based measures of
expectations.
3.1.2 The role of oil prices for inflation expectations
The oil price is a key driver of headline inflation rates, and therefore of
short-term inflation expectations. Oil price changes have direct and indirect effects
on inflation that are typically reflected in corresponding changes in shorter-term
inflation expectations. However, the correlation between oil price changes and
inflation expectations fades as the horizon lengthens, in line with the notion that the
impact of isolated oil price shocks on headline inflation tends to be transitory
(see Chart 16).
28 Many other factors can potentially influence inflation expectations. One such factor that may operate in
parallel with determinants of inflation in standard models is house prices. There is evidence for the euro
area suggesting that prices for residential real estate contribute significantly to consumers’ inflation
perceptions (Döhring & Mordonu, 2007), and this link between perceptions and house prices is confirmed
by analysis conducted in the context of the work stream report on inflation measurement (see Box 4.1 in
the forthcoming paper). In the context of the EGIE, a panel VAR analysis suggests that – possibly through
their impact on perceptions – house prices also have some impact on consumers’ inflation expectations
one year ahead. Another factor not considered here is the role of exchange rate movements. There is
some evidence pointing to an increase in the effect that exchange rate shocks have on observed inflation
in the euro area (e.g. Leiva-León et al. (2021) and Ortega & Osbat (2020)). In this context, one possible
avenue for future work could be to investigate whether exchange rates have also had a greater impact on
inflation expectations in the euro area.
0
10
20
30
40
50
60
70
80
90
100
Short term Medium term Longer term
Essentially model-based
Model-based with judgemental adjustments
Essentially judgement-based
ECB Occasional Paper Series No 264 / September 2021
45
Chart 16
Correlation between oil price changes and survey-based inflation expectations for
various horizons
(correlation coefficients in percentages)
Source: Eurosystem staff calculations.
Notes: “CY” means calendar year, and “x(C)Y” denotes a horizon x (calendar) years ahead. Sample: 2000 to Q3 2020 for SPF measures;
2000 to Q2 2020 for Consensus Economics measures. Correlations are contemporaneous and calculated on the basis of annual growth
rates.
The fading correlation is more pronounced for survey than market-based
measures. This is probably due to the risk premia embedded in market-based
measures, which tend to be correlated across the term structure for a given asset
class. Hence, the generally higher correlations between oil prices and market-based
indicators of inflation expectations may reflect such correlation of risk premia across
time, rather than necessarily implying that market participants genuinely expect oil
price shocks to affect actual inflation in the distant future. This possibility is
corroborated by the fact that the correlation between oil prices and market-based
measures appears to be higher in environments characterised by significant financial
market uncertainty (or adverse risk sentiment) – since, in these environments, the role
played by risk premia in the repricing across asset classes is arguably greater
(see Chart 17). For changes in oil prices to trigger a response in long-term
expectations, one would have to assume either (i) a notable serial correlation between
oil price shocks, such that oil price changes today were expected to keep
systematically occurring in the years ahead, and/or (ii) significant second-round
effects (for instance, via wage setting) that extend the impact which an isolated oil
price shock has on inflation rates well beyond the near term.
-20
-10
0
10
20
30
40
50
1y 2y 5y 1cy 2cy 3cy 4cy 5cy 6-10y
SPF Consensus Economics
ECB Occasional Paper Series No 264 / September 2021
46
Chart 17
Correlation between oil price changes and market-based indicators of inflation
expectations for various horizons
(correlation coefficients)
Sources: Refinitiv and Eurosystem staff calculations.
Notes: This chart shows, for each horizon, the pairwise correlation between daily changes in euro area inflation-linked swap rates and
daily percentage changes in the spot price of oil (Brent crude). The sample period is 1 April 2005 to 18 September 2020. For the full
sample and the low/high-uncertainty environment subsamples, a curve is fitted across the term structure through the pairwise
correlations between euro area inflation-linked swap rates and the price of oil. A low (high)-uncertainty environment is defined as one in
which the VIX is below (above) the 25th (75th) percentile of its distribution, based on the full sample.
The impact that oil price changes have on short to medium-term inflation
expectations depends on the underlying nature of the shock. Using an empirical
model for the global crude oil market based on Kilian & Murphy (2014), changes in oil
prices can be broken down into global activity shocks and oil-specific demand and
supply shocks (see Box 3 for a detailed discussion). Chart 18 and Chart 19show that
measures of short-term inflation expectations react in a statistically significant manner
to the identified global activity shock, that oil-specific demand shocks also tend to have
a significant positive impact on expectations, and that oil-specific supply shocks
trigger more muted responses.29 In general, market-based measures tend to react
more strongly than survey-based ones, but not when one corrects for the impact of
inflation risk premia.30
29 This confirms economic thinking and results in the literature (Aastveit et al., 2020), whereby a common
global demand shock driving the oil price has the potential to trigger a stronger reaction in inflation
expectations compared with a pure oil supply shock or a precautionary oil demand shock. Venditti &
Veronese (2020) suggest that in the case of market-based measures of inflation expectations, a risk
sentiment shock might be even more important than global demand shocks.
30 Even after controlling for inflation risk premia, market-based measures of inflation expectations continue
to react strongly to oil-specific demand shocks, although the reaction becomes somewhat less
pronounced.
0.0
0.1
0.2
0.3
0.4
1y1y 1y2y 1y3y 1y4y 1y5y 1y6y 1y7y 1y8y 1y9y
Full sample
High-uncertainty environment
Low-uncertainty environment
Log (full sample)
Log (high-uncertainty environment)
Log (low-uncertainty environment)
ECB Occasional Paper Series No 264 / September 2021
47
Chart 18
Response of inflation expectations one year ahead to identified oil price shocks
Survey-based measures Market-based measures Market-based measures
adjusted for inflation risk
premia
(impulse response function)
Source: Eurosystem staff calculations.
Notes: Local projections of the impact on measures of expectations of particular oil shocks and lagged variables using an oil market
model with up to 12 lags. The shaded areas depict 68% confidence intervals for the global activity shock. The sample period is January
2001 to December 2019. Circles on lines mean that the response is statistically significant at the 10% confidence level.
As expected, the impact of oil price shocks on long-term expectations is
generally very muted (see Chart 19). The reaction tends to be stronger (and is
statistically significant) where oil price changes are related to global activity shocks,
but its magnitude is very muted overall when compared with the responses of
short-term inflation expectations. Market-based measures post stronger impacts, but
this appears to be related to developments in inflation risk premia.
-0.04
0.00
0.04
0.08
0.12
0.16
1 2 3 4 5 6 7 8 91011 12
Oil supply shock
Global activity shock
Oil-specific demand shock
-0.04
0.00
0.04
0.08
0.12
0.16
123456789101112
Oil supply shock
Global activity shock
Oil-specific demand shock
-0.04
0.00
0.04
0.08
0.12
0.16
1 2 3 4 5 6 7 8 9 10 1112
Oil supply shock
Global activity shock
Oil-specific demand shock
ECB Occasional Paper Series No 264 / September 2021
48
Chart 19
Response of long-term inflation expectations to identified oil price shocks
Survey-based measures Market-based measures Market-based measures
adjusted for inflation risk
premia
(impulse response function)
Source: Eurosystem staff calculations.
Notes: Local projections of the impact on measures of expectations of particular oil shocks and lagged variables using an oil market
model with up to 12 lags. The shaded areas depict 68% confidence intervals for the global activity shock. The sample period is January
2001 to December 2019. Circles on lines mean that the response is statistically significant at the 10% confidence level.
3.1.3 Inflation expectations and monetary policy
Monetary policy is intimately related to the full term structure of the inflation
expectations curve. With a given central bank strategy and anchored inflation
expectations, adjustments to the monetary policy stance can lead to changes in short
to medium-term expectations, but should not, in principle, influence long-term
expectations. By contrast, the re-anchoring channel sees an explicit role for monetary
policy actions in impacting longer-term inflation expectations. For instance, monetary
easing may lead to reassurance regarding the central bank’s willingness to re-anchor
inflation expectations at levels consistent with price stability. Andrade et al. (2016)
found evidence consistent with such a re-anchoring channel in the euro area after the
announcement of the APP.
Market-based measures of inflation compensation can provide a useful testing
ground for the impact of monetary policy actions on inflation expectations
owing to their high frequency. Their timely nature allows, in principle, for a cleaner
identification of the impact by comparison with lower-frequency indicators
(e.g. surveys) by reducing the number of potentially confounding factors. More
specifically, regression analysis can gauge the reaction of ILS rates to two types of
monetary policy shock on a day-to-day basis. The first type of shock is a pure policy
shock – such as an unexpected interest rate hike or cut – that moves bond and equity
prices in the same direction. The second type of shock is an information shock –
triggered by the central bank signalling an unexpected change in its assessment of the
macroeconomic outlook – that moves bond and equity prices in opposite directions.
-0.02
0.00
0.02
0.04
1 2 3 4 5 6 7 8 91011 12
Oil supply shock
Global activity shock
Oil-specific demand shock
-0.02
0.00
0.02
0.04
123456789101112
Oil supply shock
Global activity shock
Oil-specific demand shock
-0.02
0.00
0.02
0.04
12345678 9 10 1112
Oil supply shock
Global activity shock
Oil-specific demand shock
ECB Occasional Paper Series No 264 / September 2021
49
Controlling for macroeconomic data releases and technical factors that may affect
market-based indicators of inflation expectations at varying times and to various
degrees, analysis finds that information shocks are not associated with significant
changes in the term structure of euro area ILS rates. However, since the PSPP has
been in place, pure policy shocks on the day of a monetary policy Governing Council
meeting have started to be followed by statistically significant increases in spot ILS
rates across maturities (see Chart 20). On average, a tightening (loosening) of
monetary policy over and above market participants’ expectations for the policy
meetings during that period have been associated (in line with economic priors) with a
downward (upward) shift in market-based inflation curves. At the same time,
longer-term forward ILS rates, for instance at the 5y5y tenor, have not shown a
significant reaction to monetary policy surprises, possibly as a result of being
anchored by the ECB’s inflation aim.
Chart 20
Monetary policy shocks and ILS rates
Information shock Pure policy shock
(response coefficients) (response coefficients)
Sources: EUREX, Refinitiv, Bloomberg and calculations by Kerßenfischer (2019) and Speck (2020).
Notes: The markers ◊, x and + indicate statistical significance at the 1%, 5% and 10% levels respectively. The parameters for the 5y5y
forward are not statistically significant.
Survey-based measures can be used to test the impact that central banks have
on inflation expectations via their communication of projections. Here, the
approach used by Hattori et al. (2016) for Japan is applied in order to test whether
Consensus Economics inflation expectations for the euro area respond to Eurosystem
projections.31 The analysis controls for developments in key variables influencing the
forecasts at the time of their production, such as oil prices (,
), exchange rates
(,
) and inflation surprises (). The underlying testing equation is specified in
terms of overall HICP and regresses changes in the Consensus Economics forecasts
31 Hattori et al. (2016) use a sample running from 2004 to 2015 with 38 private forecasts. Similar analyses
have been carried out by Romer & Romer (2000), Fujiwara (2005), Hubert (2015), Pedersen (2015) and
Łyziak & Paloviita (2018).
-0.004
-0.002
0.000
0.002
0.004
0.006
0.008
0.010
2y 3y 4y 5y 6y 7y 8y 9y 10y 5y5y
Pre-crisis (before June 2008)
Crisis (June 2008 to June 2014)
APP (since July 2014)
-0.014
-0.012
-0.010
-0.008
-0.006
-0.004
-0.002
0.000
0.002
0.004
0.006
2y 3y 4y 5y 6y 7y 8y 9y 10y 5y5y
Pre-crisis (before Jun 2008)
Crisis (Jun 2008 -Jun 2014)
APP (since Jul 2014)
ECB Occasional Paper Series No 264 / September 2021
50
,
for the next calendar year on the difference between the latest (B)MPE
projection ,
and the Consensus Economics forecast available at that time.32
,
=
+
,
+
+
,
+
,
+
,
,
+
The results point to robust, statistically and economically significant adjustment of
private sector forecasts to Eurosystem projections. If a Consensus Economics
forecast is below the (B)MPE forecast, the model results suggest it will subsequently
be revised upwards – and vice versa – after controlling for data news and changes in
key assumption variables. There is, however, no evidence that this adjustment is
significantly stronger where there is a larger initial difference between the private
sector forecast and the central bank forecast. At the same time, the adjustment
appears to be somewhat stronger where private forecasts are above – rather than
below – central bank forecasts. The relevant coefficients suggest that this adjustment
was somewhat stronger in the pre-GFC period than it is now.33 While this analysis
only explores the impact that Eurosystem projections have on forecasters’ short-term
expectations, any such exogenous impact might also apply to the whole of the term
structure, implying that the communicated endpoint for the Eurosystem projections
also has an impact on private sector medium-term inflation expectations.
Changes to monetary policy targets and strategies should also influence
longer-term inflation expectations.34 Identifying the impact of such changes
empirically is challenging, as changes to targets or strategies are typically rare (with
quantitative inflation targeting only starting in the 1990s), can take various different
forms (such as a change from a range to a point target or a change of point target)35,
and are not normally introduced as a kind of shock at a specific point in time, but rather
phased into the public domain via tailored communication over a longer period.
Bearing this in mind, visual inspection of the evolution of longer-term SPF inflation
expectations for the euro area over time (as market-based indicators were not yet
available at that time) suggests that the main impact of the 2003 clarification of the
ECB’s inflation aim as “below, but close to, 2%” was a gradual increase in the
percentage reporting 1.9% as their expectations five years ahead and a
32 π is the forecast at time t for the next calendar year (ncy), π_t^S is the inflation surprise at time t
(calculated as the difference between Bloomberg expectations and the actual HICP flash release),
∆e_(t,12m)^CE is the forecast change in the Consensus Economics USD:EUR exchange rate over the
next 12 months, ∆
〖
oil
〗
_(t,12m)^CE is the forecast change in the Consensus Economics oil price (USD
per barrel) over the next 12 months. The Eurosystem inflation forecast for the next calendar year is
finalised in month t-1 and published in month t.
33 As an additional robustness check, dummies were included for the periods with the largest outliers
(Q4 2008, Q1 2010, Q1 2013, Q4 2013, Q1 2014 and Q1 2020) in the basic specification (1), but the
results did not change in any meaningful way.
34 Other dimensions not considered here include (i) the question of whether there are regime-dependent
effects (that is to say, whether there is a difference between regimes with low and high inflation
expectations) and (ii) the question of whether there are asymmetric effects (that is to say, whether
positive shocks have similar effects to negative shocks).
35 Grosse-Steffen et al. (2020) find, on the basis of a panel of 29 countries, that moving from a range to a
point target subsequently increases the degree of anchoring for two to ten-year inflation expectations.
ECB Occasional Paper Series No 264 / September 2021
51
corresponding shift to a more unimodal distribution of individual point expectations
(see panel a of Chart 21).
Chart 21
Impact of selected changes to monetary policy
Frequency of specific longer-term point
inflation expectations in the SPF Dispersion of US longer-term inflation
expectations in Consensus Economics data
(percentages) (percentages per annum)
Sources: left-panel: ECB (SPF) and Eurosystem staff calculations; right-panel: Consensus Economics and Eurosystem staff
calculations.
Notes: left-panel: Vertical line denotes the conclusion of the Eurosystem’s 2003 monetary policy strategy review. Right-panel: Vertical
lines denote (a) the US Federal Reserve’s first announcement of an explicit inflation target in January 2012, (b) the clarification of
symmetry in January 2016 and (c) the explicit statement in August 2020 indicating that the target is an average “over time”.
For the United States and Japan, the impact of changes in monetary policy
targets on inflation expectations has differed. The US Federal Reserve’s
announcement of an explicit inflation target of 2% in 2012 was followed by a narrowing
of both the standard deviation of individual longer-term inflation expectations and the
high-low range in the Consensus Economics survey (see panel b of Chart 21). There
was only a small impact on the level of average longer-term inflation expectations, as
this was already largely consistent with the target at the time of the announcement.
The amendment in 2016, defining the inflation objective as symmetrical, had little
discernible impact on the level or dispersion of longer-term expectations. In contrast,
the Federal Reserve’s announcement in 2020 of a new strategy, including a make-up
element, was followed by an increase in both the dispersion of Consensus Economics
expectations and the high-low range.36 In Japan, the explicit adoption of an inflation
objective in 2012 and the increase in the inflation target in 2013 do not appear to have
had a significant impact on the average level or dispersion of individual longer-term
inflation expectations. According to Nakata (2020), the Japanese experience shows
that the announcement of a higher inflation target does not guarantee that inflation will
increase to the new target level, even if the announcement is accompanied by a
historically unprecedented degree of monetary accommodation. However, mapping
36 Naggert et al. (2021) report that in the US SPF there was “an upward shift in the lower end (below
2 percent) of the distribution of inflation expectations and a stronger anchoring of expectations around
the 2 percent inflation objective following the announcement”.
0
5
10
15
20
25
30
35
40
45
50
1999 2000 2001 2002 2003 2004 2005 2006 2007
1.5%
1.8% 1.9%
2.0%
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
2005 2007 2009 2011 2013 2015 2017 2019 2021
US longer-term inflation expectations
High and low values
+/-1 standard deviation
ECB Occasional Paper Series No 264 / September 2021
52
such experiences to other jurisdictions is not straightforward, as Japan also has a
somewhat different structural environment. Overall, with the benefit of hindsight, data
analysis suggests that the impact can be (a) subtle (particularly if long-term
expectations are already relatively close to the stated policy aim) and/or
(b) state-dependent (as changes in a challenging environment for monetary policy
may be more difficult to endure).
3.1.4 The link between long-term inflation expectations and inflation
trends
Longer-term survey expectations and longer-term trends in actual inflation
co-moved closely prior to the recent low-inflation period. In the 2018 special SPF
survey, respondents indicated that, in addition to the ECB’s inflation objective, a key
factor driving their longer-term expectations was trends in actual inflation. Such trends
could be seen as capturing the ECB’s track record in terms of inflation performance,
but they would also imply a strongly backward-looking element in the formation of
longer-term inflation expectations. They could also reflect structural drivers, and
agents’ longer-term inflation expectations would then need to incorporate an explicit
view as to whether, and to what extent, monetary policy can counteract such drivers in
pursuing the inflation aim. Chart 22 illustrates the co-movement between SPF
expectations five years ahead and two proxies for the inflation trend: an expanding
average of headline inflation and an exponentially weighted moving average (EWMA)
of inflation.37
Chart 22
SPF long-term euro area inflation expectations and a measure of inflation trends
(year-on-year growth rates)
Sources: ECB (SPF) and Eurosystem staff calculations (based on the HICP).
Notes: The expanding average and the exponentially weighted moving average (EWMA) are available for periods from 1999 onwards.
The shaded area denotes the period of quantitative easing.
37 The EWMA of past inflation is computed as: = (1)
. This approach assumes a
“statistical” trend, based on a weighted average of past inflation rates with a “smoothing” parameter
= 0.02, which implies a very low “forgetting” factor.
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Inflation trend –expanding average
Inflation trend –EWMA
SPF expectations five years ahead
ECB Occasional Paper Series No 264 / September 2021
53
A decoupling of SPF expectations five years ahead and long-term inflation
trends occurred around 2015. The timing of that development coincides with the
ECB starting its quantitative easing via the APP at a point in time when low inflation
was a broad-based phenomenon across countries. Longer-term expectations have
remained relatively stable following the launch of the APP, while actual inflation has
fallen steadily. It was only in 2019 that SPF expectations five years ahead started to
fall (declining more markedly than the considered measures of inflation trends), raising
concerns over a possible unanchoring of expectations in the euro area. By that time,
inflation trends were persistently low and using the Eurosystem’s inflation projections
at the time to assess future developments would not have made much difference, as
they were pointing to a delayed return to the targeted level of inflation. This might have
been seen as confirmation of downside risks to long-term expectations and led survey
respondents to revise downward their assessment of central tendencies. More
generally, the analysis highlights the challenges in anchoring inflation expectations if
inflation trends are persistently below target and shows that re-anchoring efforts on
the part of monetary policy may only be effective if signs of successfully changing
inflation trends become evident.
Box 3
The link between inflation expectations and oil prices. Does the nature of the shock to the
global oil market matter?
In interpreting the link between long-term inflation expectations and oil price movements, it is
important to observe the nature of the shock underlying oil price changes
On the basis of a simple general co-movement between the price of oil and long-term inflation
expectations (particularly in the case of market-based expectations), one might be tempted to
conclude that there was causality running from the former to the latter (Darvas & Hüttl, 2016;
Elliott et al., 2015). However, in regression analysis, the notion of a causal link loses ground when
more controls are considered in the form of additional relevant variables (Conflitti & Cristadoro, 2018)
or when accounting for the different kinds of demand and supply shock that are driving global oil price
movements (Aastveit et al., 2020).
Given the considerable uncertainty as to which underlying shocks are driving the oil price at a given
point in time, this box identifies those shocks using two structural VAR models that feature prominently
in the literature, namely those employed in Kilian & Murphy (2014) (“KM14”) and Baumeister &
Hamilton (2019) (“BH19”).38 It then assesses their impact on euro area inflation expectations.
First, in order to pin down the transmission of shocks, we employ a local projection (LP) approach
using shocks identified in the two benchmark models for the global oil market.
This is a very flexible approach which allows for the direct computation of impulse response functions
(IRFs). In line with Jorda (2005), we regress the dependent variable at time + (inflation swaps or
survey expectations) on the information set at time as follows:
38 BH19 and KM14 differ in terms of (i) the precise data chosen to characterise the crude oil market, (ii) the
setting-up of the priors of the model and (iii) the number of identified shocks. However, the main
difference lies in their assumptions on the price elasticity of oil supply. This feature determines the
relative importance of supply and demand as drivers of the real price of oil.
ECB Occasional Paper Series No 264 / September 2021
54
=++()+
where is a particular shock to the global oil market = {1,2,3}, is a set of control variables (12
lags of global oil market variables), and is the residual. The parameter is an estimate of the
impulse response function of variable to the shock at horizon = 0,1, … ,12.39
Table A
The identification scheme of the BVAR-KM14 model
Notes: Asterisks mean the sign is left unrestricted. (+) means that the restriction is imposed only for short-term expectations, while long-term expectations are left
unrestricted.
Second, we build a structural VAR model (BVAR-KM14) with block exogeneity which identifies the
transmission of oil demand and supply shocks to both foreign and domestic variables through the
real price of oil (see Table A).
This model incorporates a foreign block (F) pertaining to the global oil market40 as developed in Kilian
& Murphy (2014) and a block of domestic euro area variables (D) comprising expected and realised
inflation (see Chart A).41 The estimation of the model follows a two-step approach, as advocated by
Canova (2005). We rely on the assumption that domestic variables do not affect developments in the
global oil market. This block exogeneity is reflected in the zero restrictions in dynamic matrix A()
and impact matrix , as follows:
=A()0
A()A()
+ 0
and relate to the blocks of foreign and domestic variables respectively. = () is a
vector of structural shocks with covariance matrix ()=. Structural shocks are recovered by
39 A Newey-West correction is applied to account for autocorrelation of the residuals. Also, the dynamic
specification is augmented to incorporate the previous horizon’s residual, as originally postulated by
Jorda (2005). The IRFs are computed up to 12 months ahead, as at longer horizons the LP estimator
becomes increasingly inefficient and more susceptible to model misspecification errors. The data sample
for euro area inflation expectations starts in January 2001 for survey and market-based expectations and
in June 2005 for market-based expectations adjusted for the impact of inflation risk premia.
40 The variables included in the foreign block are: (i) global crude oil production; (ii) an index of global real
activity: the dry cargo shipping rate index developed in Kilian (2009); (iii) the real price of oil, defined as
US refiners’ acquisition cost for imported crude oil deflated by US CPI inflation; and (iv) total US crude oil
inventories, scaled by the ratio of OECD petroleum stocks to US petroleum stocks.
41 Additional restrictions are implemented in order to characterise the global oil market (see Kilian &
Murphy (2014)): (i) bounds on the price elasticity of oil supply: 0 < < 0.0258; and (ii) bounds on
the price elasticity of oil demand: 0.8 < < 0.
Global crude oil market
Oil supply Global demand Oil-specific demand
Oil production - + + * 0 0
Global real activity - + - * 0 0
Real oil price + + + * 0 0
Oil inventories * * + * 0 0
Inflation expectations (+) (+) (+) * * 0
Inflation + + + * * *
ECB Occasional Paper Series No 264 / September 2021
55
imposing a mixture of sign and zero restrictions, in line with Mumtaz & Surico (2009). Finally, the A()
and matrices of coefficients are used to compute impulse responses.42
Chart A
The response of expectations to various shocks driving the oil price
(percentage points)
Source: Eurosystem staff calculations.
Notes: Shaded areas denote 68% confidence intervals for the impulse response functions derived from the block BVAR-KM14 model. “BVAR” = Bayesian vector
autoregression model; “LP” = local projection model.
42 The model uses 12 lags for survey-based measures of expectations and six lags for market-based
measures owing to the smaller sample for the latter. It is estimated using a Bayesian approach.
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.10
0.12
13 5 7 9 11 1 3 5 7 9 11 1357911 1357911 1357911 1 3 5 7 9 11
Oil supply Global activity Oil-specific demand Oil supply Global activity Oil-specific demand
SPF1y SPF5y
BVAR-KM14
LP-KM14
LP-BH19
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
1357911 1 3 5 7 9 11 1357911 1357911 1357911 1357911
Oil supply Global activity Oil-specific demand Oil supply Global activity Oil-specific demand
ILS1y1y ILS5y5y
-0.03
-0.02
-0.01
0.00
0.01
0.02
0.03
0.04
0.05
0.06
0.07
123456789101112 123456789101112 123456789101112 123456789101112 123456789101112 123456789101112
Oil supply Global activity Oil-specific demand Oil supply Global activity Oil-specific demand
ILS 1y1y –adjusted for the inflation risk premia ILS 5y5y –adjusted for the inflation risk premia
ECB Occasional Paper Series No 264 / September 2021
56
Across the employed models, it appears that only short-term expectations react to an economically
meaningful extent to shocks driving the oil price, particularly in the case of global demand shocks
(see Chart A). For short-term inflation expectations, significant reactions can also be seen in
response to oil-specific demand shocks, with these reactions being smaller in size and less
persistent. For longer-term expectations, all reactions to all shocks are fairly muted and within a range
of 1 to 2 basis points at most. Furthermore, the somewhat larger reactions visible in the case of 1y1y
and 5y5y ILS rates appear to largely reflect developments in inflation risk premia, as the reactions
become much more muted once these measures have been corrected for changes in the inflation risk
premium.
It should be noted, however, that there is still some uncertainty surrounding these messages. In the
case of oil supply shocks, for instance, the BH19 model yields stronger reactions than the KM14
model. That difference stems from two factors: first, the two models are set up in different ways; and
second, one-on-one mapping of the identified shocks in the two models is an approximation. While in
KM14 only one global demand shock is identified, in BH19 one additional shock is assumed to move
the oil price, namely “oil demand”; this shock, along with the oil-specific shock, has greater
importance for oil price movements in the BH19 historical decomposition, so we chose to focus on
this particular shock as a proxy for the wider demand conditions driving the real price of oil. If one
were to add the impact coming from the other identified demand shock, the reaction of the oil price –
and, implicitly, inflation expectations – to demand shocks would also be greater according to this
alternative model.
3.2 Defining and measuring risks to anchoring
3.2.1 The concept of anchoring and relevant metrics
The concept of anchoring of expectations is complex and multi-faceted, with no
single definition or measure. Traditionally, much of the literature on anchoring has
focused on two main approaches to its assessment:43 (i) examining the level of
inflation expectations, particularly relative to an inflation target or aim, and/or
(ii) examining the responsiveness of longer-term inflation expectations to shorter-term
developments (e.g. actual inflation or other economic news). More recently, there has
also been a focus on higher moments of inflation expectations – i.e. on their variability,
disagreement and uncertainty, and the balance of risks surrounding them, including
tail risks.44 From a theoretical perspective, (un)anchoring is seen as a binary state, in
line with the assertion in Orphanides (2015) that “inflation expectations are anchored
until they are not”. In practice, however, there may be a spectrum running from entirely
43 For an example of work that looks at developments across countries, see Beechey et al. (2011) or
Gürkaynak et al. (2010). For more recent work, see Yetman (2020) or Dovern & Kenny (2020).
44 Additional dimensions that could be considered include analysing the extent to which being in a state of
sustained low inflation contributes to the deterioration of inflation expectations, or unanchoring of
expectations (a duration dependence point of view).
ECB Occasional Paper Series No 264 / September 2021
57
anchored to entirely unanchored expectations, depending on how many metrics point
in the same direction at the same point in time.
While a level-based concept of anchoring is the most straightforward in theory,
it is complicated in practice if there is ambiguity about the benchmark. In the
case of the euro area, the quantitative price stability objective is defined as “below, but
close to, 2%” over the medium term. What is meant by “close to” has never been
definitively communicated. Box 4 provides evidence on this from the perspective of
responses to a special question in the SPF, suggesting that respondents generally
interpreted the ECB’s price stability objective as lying between 1.7% and 2.0%.45
Another question relates to the horizon at which inflation expectations should be
anchored, since the length of the “medium-term” horizon has never been definitively
clarified by the Governing Council either. It may be state-dependent.46 The analysis in
Section3.1 points, for both the euro area and the United States, to a lengthening of the
period of time that it takes for inflation forecasts to converge to form steady-state
inflation expectations. This implies, for instance, that inflation expectations five years
ahead may not necessarily be an unambiguous benchmark reference for the “below,
but close to, 2%” inflation aim. Even if there was agreement on the precise level and
horizon, there could still be differing interpretations of what constitutes level-based
(un)anchoring. Consider, for example, a hypothetical scenario with a precise inflation
target of 2.0%. While most would argue that long-term inflation expectations of 0.5%
were not anchored, some might argue that expectations of 1.9% were unanchored. In
other words, while it would appear uncontroversial to contend that expectations that
are further away from the target are less likely to be anchored, it is far from
straightforward to assess how far away they have to be before they start to be
regarded as “unanchored”.
The concept of responsiveness starts from the premise that well-anchored
longer-term expectations should not be sensitive to shorter-term
developments. Such shorter-term developments encompass movements in actual
inflation, changes in short-term inflation expectations, and economic surprises or
shocks. Longer-term expectations should not show responsiveness to such
developments, as long as their impact can be expected to fade out on its own before
the “longer term”, and as long as monetary policy appropriately counteracts those
shocks that are not expected to (fully) fade before the “longer term”. While this
responsiveness-based approach has its merits, it cannot be considered independently
of the level-based assessment. For instance, expectations cannot be considered
anchored if they stand at a level that is not considered desirable, even if they do not
respond to shorter-term developments. In fact, the responsiveness-based concept is
arguably best thought of as a stricter test of anchoring, conditional on inflation
expectations having already “passed” the level-based test.47 Conversely,
45 The median reported figure for the lower end of the range was 1.7%, the median for the upper end was
2.0%, and the median span of ranges (upper-lower) was 0.3 p.p.
46 See, for instance, Schnabel (2020): “[T]he medium-term horizon over which the ECB pursues the
sustainable alignment of inflation with its aim is considerably longer than in the past”.
47 An exception applies in times of market stress, when the level of market-based inflation indicators may be
affected by technical factors (see Box 2). In such circumstances, concentrating on the responsiveness to
economic surprises over a short time period may still deliver valuable information about anchoring,
despite the level of ILS rates being uninformative.
ECB Occasional Paper Series No 264 / September 2021
58
expectations moving back towards the target following a period away from that target
may be considered a welcome consequence of largely one-sided responsiveness in
the right direction.
A more recent concept looks at higher moments of inflation expectations,
focusing on uncertainty surrounding expectations and the balance of risks.
Whereas the level-based assessment concentrates on the first moment of inflation
expectations (i.e. their average or mean expected value), the assessment of higher
moments extends right up to the fourth moment of inflation expectations. The second
moment captures the uncertainty surrounding expectations, the third moment
captures their skewness or the balance of risks, while the fourth moment captures the
degree of kurtosis or tail risks. One can also consider the risk of deflation or
“lowflation”.48 For example, in the context of the SPF (where the entire probability
distribution is surveyed), deflation (lowflation) risks can be measured by the probability
of longer-term inflation being negative (below a specific threshold, such as 1%). These
metrics have the advantage that although forecasters’ modal expectations might be in
line with the inflation target, they can signal increased risks to their modal forecast that
could be a harbinger of future level-based unanchoring. Finally, disagreement among
forecasters may be seen as an additional metric. No single measure is likely to
encompass all the relevant information, and consideration of a wider range of
measures is likely to give the best signal.
Box 4
What levels of inflation are consistent with the ECB’s definition of price stability, according to
the SPF?
The ECB’s Governing Council adopted a quantitative definition of price stability in 1998 and
clarified its objective in 2003. In 1998 the ECB’s Governing Council defined price stability as “a
year-on-year increase in the Harmonised Index of Consumer Prices (HICP) for the euro area of below
2%”. In 2003, the Governing Council clarified that the aim was to maintain inflation rates below, but
close to, 2% over the medium term. Some evidence on how this is interpreted by the public is
provided by the SPF.
SPF participants have twice (in the Q4 2020 and Q3 2019 survey rounds) been asked the
question: “What is the level or range of inflation that, according to your view, is in line with
the ECB’s price stability objective?” In the Q4 2020 round, quantifiable replies were received from
18 of the 66 participants (with 46 responding to the question on longer-term inflation expectations).
And in the Q3 2019 round, replies were received from 21 of the 52 participants (with 39 responding to
the question on longer-term inflation expectations). Although the number providing quantifiable
information is relatively small, it can be seen as broadly representative of the panel as a whole, given
48 See Banerjee & Mehrotra (2018): “We find some evidence that expectations become less well anchored
during deflations. Deflations are associated with a downward shift in inflation expectations and a
somewhat higher backward-lookingness of those expectations. We also find that deflations are
correlated with greater forecast disagreement. Delving deeper into such disagreement, we find that
deflations are associated with movements in the left-hand tail of the distribution.”
ECB Occasional Paper Series No 264 / September 2021
59
the strong co-movement between their longer-term expectations and the aggregate SPF longer-term
expectations.49
On both occasions, the median responses implied an ECB price stability objective targeting a
range of 1.7-2.0%. Nearly all respondents reported a range, rather than a single point value (see
Chart A), implying some difficulty in benchmarking against a specific value. The median figure for the
lower end of the range was 1.7%, the median for the upper end was 2.0%, and the median span of
ranges (upper-lower) was 0.3 p.p. Chart B presents this information from a different perspective.50
That chart suggests that all respondents answering the special question regarded 1.9% as being
consistent with the ECB’s definition of price stability. The other modal values were 1.8% (97% of
respondents), 1.7% (69%) and 2.0% (69%). Outside of these values, the percentage dropped
substantially: for example, 45% implicitly regarded 1.6% as being in line with the ECB’s definition of
price stability. One interesting feature is that, despite the objective being clearly defined as “below, but
close to, 2%”, some respondents’ answers suggested that they believed small overshoots of 2%
would, de facto, be deemed acceptable.
Chart A
“What is the level or range of inflation that, according to your view, is in line with the ECB’s price
stability objective?”
Source: ECB (SPF).
Notes: The bars depict the inflation rate or range of inflation rates that, according to the respondent, is in line with the ECB’s price stability objective. The
patterned bars represent the one respondent in each round who provided the range 0-2%.
49 Ten respondents provided quantifiable responses in both rounds. Of these, six provided the same answer
in both rounds.
50 The chart shows the replies in the Q3 2019 and Q4 2020 rounds, as well as the combined replies. For the
latter, where a respondent reported different values in the two rounds (which was the case for four of the
ten respondents who responded to the question in both rounds) the Q4 2020 response has been used.
Q3 2019 Q4 2020
(x-axis: respondent; y-axis: annual percentage changes) (x-axis: respondent; y-axis: annual percentage changes)
1.4
1.6
1.8
2.0
2.2
2.4
2.6
12345678910 11 12 13 14 15 16 17 18 19 20 21
Range
1.4
1.6
1.8
2.0
2.2
2.4
2.6
12345678910 11 12 13 14 15 16 17 18
Range
ECB Occasional Paper Series No 264 / September 2021
60
Chart B
“What is the level or range of inflation that, according to your view, is in line with the ECB’s price
stability objective?”
(x-axis: annual percentage changes; y-axis: percentage of respondents)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: The bars cumulate the one-decimal inflation rates covered by the ranges provided by respondents. For instance, all respondents regard 1.9% as being in
line with price stability.
While the responses to this special question should be viewed as point-in-time snapshots,
the answers are in line with the histogram of reported longer-term inflation expectations. On
average, over the period since 1999, the most commonly reported individual longer-term expectation
(i.e. the mode) was 2.0% (30%), followed by 1.9% and 1.8% (19% each). Thus, these three values
(1.8-2.0%) account for over two-thirds of all responses.
3.2.2 Are euro area inflation expectations anchored?
Euro area inflation expectations were considered to be relatively well anchored
until well after the financial crisis. This finding was common in the literature, with
expectations also being well anchored compared with the United States.51 With the
onset of the euro area sovereign debt crisis and the decline in growth and inflation
outcomes in the euro area, longer-term inflation expectations in the euro area came
under downside pressure. In this context, more recent literature is less sanguine and
generally suggests, at the very least, that the risk of unanchoring has increased. Some
papers argue that expectations have become unanchored,52 others contend that they
remained anchored,53 while the findings of a third group of papers are more
nuanced.54 The range of assessments stems, in part, from the different measures and
concepts that are considered, but also from the differing interpretation of signals
51 See, for example, Beechey et al. (2011) and Autrup & Grothe (2014).
52 See Byrne & Zekaite (2019), Corsello et al. (2021), Henckel et al. (2019), Garcia & Werner (2018), Łyziak
& Paloviita (2017) and Natoli & Sigalotti (2018).
53 See Grishchenko et al. (2019), Mehrotra & Yetman (2018) and Speck (2017).
54 See, for example, Dovern & Kenny (2020), Apokoritis et al. (2019), Carvalho et al. (2019) and Stevens &
Wauters (2018).
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
≥2.6
2.52.42.32.22.12.01.91.81.71.61.5
≤1.4
Q4 2020
Combined
Q3 2019
ECB Occasional Paper Series No 264 / September 2021
61
derived from similar concepts. This section reports on some of the metrics that are
used for the different approaches to (un)anchoring.
3.2.2.1 Level-based assessments of anchoring
During 2019, survey-based measures of longer-term inflation expectations have
moved to the very bottom of their historical range. The distribution of average
longer-term point inflation expectations in the SPF peaks at 1.9% (for more than half of
all rounds), with 1.8% and 2.0% being outcomes for almost a quarter of all rounds. As
Chart 23 shows, average point expectations only moderated somewhat between the
financial and sovereign debt crises. A downward movement then started in 2013, with
average expectations reaching a low of 1.77% by Q1 2015. Although this was followed
by a brief but modest rebound until the end of 2018, a renewed decline led to
longer-term expectations reaching new lows of 1.67% in Q4 2019 and 1.648%
(i.e. 1.6% after rounding) in Q3 2020.55 This moved average expectations to the lower
end of the median range that SPF respondents associate with price stability (see
Box 4). Up until the sovereign debt crisis, the upper end of the interquartile range of
individual longer-term inflation expectations was generally at 2.0%, while the lower
end of the interquartile range saw more changes. The decline in 2019 was then also
associated with a clear downward shift in the interquartile range. Looking at the overall
cross-sectional distribution, rather than just the interquartile range, we can see that
since 2019 the distribution has not only become more skewed towards lower values,
but has also become flatter in terms of more kurtosis.
Chart 23
Longer-term euro area inflation expectations from the SPF
(percentages; quarterly data; latest observations Q4 2020)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: The orange shaded area shows the range 1.7-1.9% (actually 1.65-1.94% before rounding). “5qcma” refers to the five-quarter
centred moving average.
55 The Q4 2020 round showed average expectations standing at 1.656% (i.e. 1.7% after rounding).
1.4
1.5
1.6
1.7
1.8
1.9
2.0
2.1
1999 2001 2003 2005 2007 2009 2011 2013 2015 2017 2019 2021
Average point estimate
Average point estimate (5qcma)
Inter-quartile range of point estimates
ECB Occasional Paper Series No 264 / September 2021
62
Market-based measures of longer-term inflation expectations have also moved
downwards, falling below a range consistent with the ECB’s inflation aim as
early as 2014 (see Chart 24). Like survey-based measures, market-based measures
also proved relatively resilient during the GFC and remained broadly in line with a
range that could be considered consistent with the ECB’s inflation aim. However,
since about 2014, those market-based indicators have declined significantly. While
this was initially driven by a decline in the estimated inflation risk premium, it
subsequently became visible also in measures adjusted for premia. Overall, survey
and market-based measures of longer-term inflation expectations have showed
similarly broad movements but have also differed substantially in terms of their levels
and the scale of their movements. In particular, market-based measures – also when
adjusted for risk premia – have moved more clearly below the range 1.7-1.9% in the
period since the sovereign debt crisis.
Chart 24
Cross-check of survey and market-based longer-term inflation expectations
(annual percentage changes)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: The shaded area shows the range 1.7-1.9% (actually 1.65-1.94% before rounding).
Formal time series tests can complement the largely descriptive analysis
provided. Dovern & Kenny (2020) apply “break-point tests” to SPF inflation
expectations data. Updating their metrics using data up to 2020 points to break points
in the mean of the probability distribution in 2013 and 2019. However, a break in the
modal (point) expectation could only be detected for Q1 2019 (see Chart 25).56
Overall, while using formal break-point tests as a metric for level-based assessment of
unanchoring points to some uncertainty in the timing of shifts, depending on which
level series they are applied to, the break around 2019 appears to be more universal
across different series.
56 A break in the average point expectation (not shown in the chart) is also found in Q1 2019.
0.5
1.0
1.5
2.0
2.5
3.0
2005 2007 2009 2011 2013 2015 2017 2019 2021
SPF average point inflation
SPF average mean of probability distribution 1y4y ILS raw data
1y4y ILS 'genuine' expectations
ECB Occasional Paper Series No 264 / September 2021
63
Chart 25
Break-point tests for SPF longer-term inflation expectations
Mean expectations Modal expectations
Sources: Eurosystem staff updates to tests originally carried out by Dovern & Kenny (2020).
Notes: Selection of break points based on Bai & Perron (1998) and (2003). The solid blue lines refer to the average moments of the
density forecasts of individual SPF participants. The yellow lines show the implied unconditional means for different sub-periods, with
breaks in AR(1) models for average moments selected using the LWZ statistic. The minimum distance between two break points was set
to eight quarters. The last observations relate to Q3 2020.
It could be argued that inflation expectations which deviate from the target can
still be anchored if they are in line with the central bank’s own forecasts for the
horizon in question. Łyziak & Paloviita (2018) use the idea proposed by Domit et al.
(2015) to construct an anchoring index that captures whether private sector inflation
expectations stay within the perceived ECB communication range.57 If they are inside
this range, being either consistent with the ECB inflation target range as defined in
Box 4 or between the ECB inflation projection and the target range, the anchoring
index is zero. If, instead, forecasted inflation is above or below this range, the value of
the index is defined as being equal to the difference between the forecasted inflation
rate and the closer limit of the range.58 Chart 26 presents this anchoring index for
inflation expectations one and two years ahead based on the SPF. Inflation
expectations one year ahead have, on balance, been more consistent with ECB
communication than inflation expectations two years ahead. For the latter, the
anchoring index was negative in the period from 2015 to early 2017 and has been
negative again since Q2 2019. These episodes broadly correspond to those indicated
by other level-based unanchoring metrics and may reflect some risks of unanchoring.
57 The perceived ECB communication range is defined in Annex B, which describes in detail the heat maps
for inflation expectations.
58 This notion of anchoring is somewhat different from some of the other concepts, which are based more
on the anchoring of inflation expectations to a price stability objective. Łyziak & Paloviita (2018) examine
the anchoring of short and longer-term private sector forecasts to Eurosystem projections, which may be
more closely associated with the notion of “path credibility” or “path anchoring”.
1.50
1.55
1.60
1.65
1.70
1.75
1.80
1.85
1.90
1.95
2.00
2.05
2.10
2001 2004 2007 2010 2013 2016 2019
Mean of distribution
Trend
1.50
1.55
1.60
1.65
1.70
1.75
1.80
1.85
1.90
1.95
2.00
2.05
2.10
2001 2004 2007 2010 2013 2016 2019
Mode of distribution
Trend
ECB Occasional Paper Series No 264 / September 2021
64
Chart 26
Risk of unanchoring inflation expectations based on survey measures
Source: Łyziak & Paloviita (2018), based on the SPF and other ECB data.
Notes: This chart presents instances of SPF forecasts deviating from the perceived ECB communication range. The last observations
are for July 2020.
3.2.2.2 Responsiveness-based assessments of anchoring
Responsiveness-based metrics relate longer-term inflation expectations to
shorter-term developments. Empirical studies often measure short-term
developments, such as (i) changes in short-term inflation expectations,
(ii) movements in actual inflation or (iii) macroeconomic surprises. Given the higher
frequency of such data, those studies often use market-based measures of inflation
expectations for their analysis.
Studies on the responsiveness of long-term expectations to changes in
short-term expectations or actual inflation do not provide a conclusive picture.
Stevens & Wauters (2018) estimate a VAR model in order to investigate the impact
that shocks to short-term inflation expectations have on longer-term expectations as
derived from ILS rates.59 Chart 27below presents an update using the most recent
euro area data, suggesting statistically significant responsiveness. However, this
might, at least in part, reflect co-movement of inflation risk premia and technical
factors across horizons. The results stand in contrast to those obtained by Ciccarelli et
al. (2017), who used a stochastic volatility methodology to estimate the
responsiveness of longer-term SPF inflation expectations (five years ahead) to
short-term expectations (one year ahead) and actual inflation. Although in both
instances, the estimated responsiveness coefficient is positive, the estimates are not
always precise enough to show up as statistically significant (with the exception of
brief periods such as 2009 and 2014-16; see Chart 28).
59 More specifically, the VAR model estimates the dynamic interactions between weekly measures of
short-term (1y) and long-term (5y5y) inflation expectations. The identification of structural shocks is
based on a Cholesky decomposition, where the ordering of expectation readings runs from the short to
the long run. The assessment of time variation is based on a rolling window regression of moving
101-week periods (i.e. approximately two years). Medians and 16th/84th percentiles are reported.
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
2005 2007 2009 2011 2013 2015 2017 2019
SPF forecasts one year ahead
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
2005 2007 2009 2011 2013 2015 2017 2019
SPF forecasts two years ahead
ECB Occasional Paper Series No 264 / September 2021
65
Chart 27
Responsiveness of medium and long-term expectations to short-term ILS rates
(regression coefficients)
Source: Eurosystem staff calculations.
Notes: This chart shows time-varying estimates of the response of 5y5y ILS inflation expectations to a 1 p.p. shock in short-term inflation
expectations. It is derived from a structural VAR model of weekly measures of inflation expectations, where the Cholesky ordering of
expectation readings runs from the short to the long run. Coefficients are dated at the end of each rolling sample. The first sample starts
with the first week of 2005 and the final sample ends with the second week of 2020.
Chart 28
Pass-through coefficients to longer-term inflation expectations from
Actual inflation Shorter-term inflation forecasts
(one year ahead)
Sources: Eurosystem staff calculations based on Ciccarelli et al. (2017).
Note: The latest observations are for Q2 2020.
Assessing the responsiveness of ILS forward rates to macroeconomic
surprises provides some indication of periods with less well-anchored
expectations, but results are sensitive to the specification of the underlying
regression. First of all, the responsiveness metrics proposed by Speck (2017) are
updated, estimating the time-varying responsiveness of the five-year forward ILS rate
five years ahead to macroeconomic (inflation and corporate sentiment) surprises
-0.1
0.0
0.1
0.2
0.3
0.4
2007 2010 2013 2016 2019
5y5y ILS
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
2000 2003 2006 2009 2012 2015 2018
Median
Bayesian confidence interval
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
2000 2003 2006 2009 2012 2015 2018
Median
Bayesian confidence interval
ECB Occasional Paper Series No 264 / September 2021
66
(see Chart 29).60 The response of the 5y5y forward rate was positive and statistically
significant during 2017 and 2018. As this was a period when longer-term inflation
expectations were firming up somewhat after a temporary trough, the positive
coefficient might be interpreted as a strengthening of the inflation anchor, rather than
indications of unanchoring. Second, similar analysis specified slightly differently
(based on surprises in a set of flash releases on HICP inflation, GDP and PMI data for
Germany, France, Italy, Spain and the euro area as a whole) suggests that
market-based indicators of inflation expectations at the 5y5y horizon have recently
become unresponsive to macroeconomic surprises again, after a statistically
significant response coefficient between 2014 and 2019 ( Chart 30). However,
considering only a subset of the aforementioned indicators yields notably different
results.61 The different timings highlight the problem that the responsiveness metric
can, in principle, capture the effects of very persistent shocks, unanchoring and/or
re-anchoring.
Chart 29
Responsiveness of longer-term expectations to macroeconomic surprises
(index)
Source: Eurosystem staff calculations based on Speck (2017).
Notes: Estimates of responsiveness and 95% confidence intervals based on heteroskedasticity-adjusted standard errors over a
nine-month rolling window. A value of one represents the average responsiveness of the two-year ILS in the pre-GFC period. The last
observations are for 7 January 2021.
60 The original findings in Speck (2017) suggested that, compared with the pre-crisis period, surprises had
a much stronger effect on short-term spot ILS rates during the crisis, but that longer-term forward ILS
rates (such as the 5y5y ILS rate) remained insensitive to news most of the time – a finding that he
considered to imply inflation anchoring. He found only short periods of sensitivity on the part of
medium-term forward ILS rates at times of low inflation or recession and argued that sensitivity is lower
over more distant horizons, such that medium-term sensitivity represents an inflation adjustment process
in response to severe and persistent shocks and provides no evidence of unanchoring of inflation
expectations or a loss of credibility for the Eurosystem’s policy target.
61 More specifically, considering only HICP, GDP or PMI releases in isolation, or any two of those series
(rather than all three at the same time), suggests a visibly smoother evolution of euro area 5y5y ILS rates
relative to surprises. In particular, the swift unanchoring and re-anchoring in Chart 30 around 2014 and
2020 respectively is not confirmed by these alternative specifications. Moreover, it is often the case that
estimates of sensitivity over different time periods also differ across specifications.
-4
-2
0
2
4
6
8
10
2005 2007 2009 2011 2013 2015 2017 2019 2021
δ 2 years
δ 5→10 years
ECB Occasional Paper Series No 264 / September 2021
67
Chart 30
Responsiveness of 5y5y ILS rates to macroeconomic surprises
(basis points per standard deviation)
Sources: Bloomberg and Eurosystem staff calculations.
Notes: The chart depicts rolling beta estimates for the following model: Δ_daily ILS^5y5y= α_t+β_t Surprises+ε_t. The size of the rolling
window is three years. Surprises for each reporting period only include the first flash release for inflation, GDP and PMI data for Germany,
France, Italy, Spain and the euro area. The last observations are for August 2020.
Another responsiveness-based approach involves considering whether
longer-term expectations react differently depending on the “direction” of
surprises. Corsello et al. (2021) test for the responsiveness of SPF longer-term
expectations to surprises in euro area inflation releases, distinguishing between
positive and negative surprises in actual inflation outcomes. Only negative surprises
are found to have had a significant impact on longer-term inflation expectations,
especially after 2013, when inflation was persistently over-predicted (see Chart 31).
However, the timing of the periods of significant responsiveness differs depending on
whether one looks at the average point forecast, the mean of the probability
distribution or the median of the cross-sectional distribution. Overall, those findings
provide some support for the view that euro area inflation expectations have become
unanchored on the downside in recent years, with repeated downside surprises in
inflation outcomes triggering a decline in longer-term expectations.
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
0.8
2011 2013 2015 2017 2019
Three-year rolling beta estimate with 95% confidence interval
ECB Occasional Paper Series No 264 / September 2021
68
Chart 31
Responsiveness of mean longer-term inflation expectations to positive and negative
inflation surprises: recursive estimates
Negative surprise: mean
expectations Negative surprise: median
expectations Negative surprise: mean
aggregate expectations
Positive surprise: mean
expectations Positive surprise: median
expectations Positive surprise: mean
aggregate expectations
Source: Eurosystem staff calculations.
Note: The latest observations are for July 2020.
3.2.2.3 Assessments of anchoring based on higher moments
Higher moments in the distribution of inflation expectations may provide early
indications of a risk of unanchoring. Section 2.2.3 introduced the idea that shifts in
the BoRI might anticipate shifts in longer-term point inflation expectations. Since 2008,
there has been strong co-movement between the BoRI for five-year expectations and
the gap between expectations two and five years ahead (see Chart 33). However,
quantitative econometric analysis has found little evidence of a reliable lead/lag
relationship.
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
2008 2010 2012 2014 2016 20182020
Mean expectations
Confidence bands
-0.50
-0.25
0.00
0.25
0.50
0.75
20082010 20122014201620182020
Mean expectations
Confidence bands
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
2008 2010 2012 2014 2016 20182020
Mean expectations
Confidence bands
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
2008 2010 2012 2014 2016 20182020
Mean expectations
Confidence bands
-0.50
-0.25
0.00
0.25
0.50
0.75
20082010 20122014201620182020
Mean expectations
Confidence bands
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
2008 2010 2012 2014 2016 20182020
Mean expectations
Confidence bands
ECB Occasional Paper Series No 264 / September 2021
69
Chart 32
Point forecasts and mean longer-term expectations from the SPF
(annual percentage changes)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: The shaded area shows the range 1.7-1.9% (actually 1.65-1.94% before rounding). The latest observations are for Q3 2020.
Chart 33
BoRI and gap between expectations two and five years ahead
(percentage points)
Sources: ECB (SPF) and Eurosystem staff calculations.
Notes: The latest observations are for Q3 2020.
Break-point tests can also be used for higher moments of inflation
expectations. Dovern & Kenny (2020) also conduct break-point tests for higher
moments of survey-based expectations, including the second (uncertainty
surrounding average longer-term expectations), third (skewness or balance of risks
surrounding average longer-term expectations) and fourth (kurtosis or tail risks)
moments (see Chart 34). In addition to the level-based break-point tests presented
above, breaks in forecast uncertainty and kurtosis are found to have occurred in early
2019.
1.5
1.6
1.7
1.8
1.9
2.0
2.1
2001 2003 2005 2007 2009 2011 2013 2015 2017 2019
Average point forecast
Mean of probability distribution
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
2001 2004 2007 2010 2013 2016 2019
SPF –BoRI for five-year expectations
SPF –gap between expectations two and five years ahead (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
70
Chart 34
Break-point tests on the basis of Dovern & Kenny (2020)
Forecast uncertainty Skewness Kurtosis
Sources: Eurosystem staff updates to tests originally carried out by Dovern & Kenny (2020).
Notes: Selection of break points based on Bai & Perron (1998) and (2003). The solid blue lines refer to the average moments of the
density forecasts of individual SPF participants. The yellow lines show the implied unconditional means for different sub-periods, with
breaks in AR(1) models for average moments selected using the LWZ statistic. The minimum distance between two break points was set
to eight quarters. The last observations relate to Q3 2020.
Box 5
Inflation expectations in advanced economies: are there signs of unanchoring?
This box analyses developments in inflation expectations over the last two decades for a
sample of seven advanced economies outside the euro area. The economies in question are
Canada, Japan, Norway, Sweden, Switzerland, the United Kingdom and the United States. The box
addresses the following three key questions: Are there commonalities in the evolution of
survey-based inflation expectations? Are there signs of any unanchoring of longer-term inflation
expectations? And are there common factors driving longer-term inflation expectations?
Longer-term Consensus Economics inflation expectations have stabilised since about 2000
amid country-specific heterogeneity. This stability follows the broad-based decline observed
during the 1990s (see panel a of Chart A).62 The cross-country median of longer-term inflation
expectations fluctuates narrowly around 2% (which is widely regarded as representing price stability,
as increasingly reflected in formal inflation targets over the past two decades). While this
cross-country measure of central tendency suggests stability, inflation expectations have been fairly
heterogeneous across countries, in terms of both their levels and their volatility. For instance, in
Canada and Sweden the Consensus Economics measure of longer-term inflation expectations has
fluctuated narrowly around the relevant target, while in Norway and Switzerland it has showed a
higher degree of volatility. Japan stands out as having experienced mild but persistent deflation over
most of the review period, which has weighed on the level of longer-term inflation expectations and
contributed to higher volatility (see panel b of Chart A). The cross-country median of short-term
inflation expectations is consistently below 2% (as disinflationary forces seem to have dominated
since the early 2000s) and it exhibits greater variability than its longer-term counterpart.
62 Castelnuovo et al. (2003) observed that by the early 2000s longer-term inflation expectations in
advanced economies had converged on point inflation targets, or the mid-points of target ranges.
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
2001 2005 2009 2013 2017
Standard deviation
Trend
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
2001 2005 2009 2013 2017
Skewness
Trend
-0.80
-0.70
-0.60
-0.50
-0.40
-0.30
-0.20
-0.10
0.00
0.10
2001 2005 2009 2013 2017
Kurtosis
Trend
ECB Occasional Paper Series No 264 / September 2021
71
Chart A
Survey-based longer-term inflation expectations have been broadly stable since 2000
Sources: Consensus Economics and Eurosystem staff calculations.
Notes: Panel (a) reports the median inflation expectations of professional forecasters one year ahead and six to ten years ahead, as compiled by Consensus
Economics for the following countries: Canada, Japan, Norway, Sweden, Switzerland, the United Kingdom and the United States. The last observations are for
April 2020. Panel (b) reports country-specific summary statistics for inflation expectations six to ten years ahead.
There is limited evidence of an unanchoring of longer-term inflation expectations in advanced
economies. The pass-through from changes in short-term inflation expectations to longer-term
expectations as a metric of unanchoring is tested using a linear regression approach
(see Castelnuovo et al. (2003) and Yetman (2020)). For most advanced economies, we find no
evidence of pass-through from short to longer-term expectations. For Japan, we find a positive and
statistically significant pass-through coefficient, which probably reflects the country’s history of mild
deflation, but that coefficient is relatively small (see Chart B). Overall, our finding that inflation
expectations are well anchored in advanced economies is in line with Yetman (2020), who draws
similar conclusions for a broader set of countries (including several emerging markets).
Cross-country medians Country-specific summary statistics
(annual percentage changes) (annual percentage changes)
0
2
4
6
1990 1995 2000 2005 2010 2015 2020
One year ahead
Longer term
0
1
2
3
US JP UK CA SE NO CH EA
Median
25th to 75th percentiles
Min/max range
ECB Occasional Paper Series No 264 / September 2021
72
Chart B
Evidence of unanchoring of longer-term inflation expectations remains limited
(estimated pass-through coefficients with 95% confidence intervals)
Sources: Consensus Economics and Eurosystem staff calculations.
Notes: Estimated country-specific pass-through coefficient , as derived from the following regression model (see Castelnuovo et al. (2003) or Yetman (2020)):
,=+
,+ , where
, and
, refer to changes in longer-term and short-term inflation expectations between two consecutive periods (April
and October of each year from 2000 to 2020).
The finding that expectations remain anchored appears to hold also when using
market-based inflation expectations instead of survey-based measures and when allowing for
time variation in the pass-through coefficient. The latter is important, since the degree of
co-movement between short-term and longer-term expectations may have been affected by several
developments during the sample period, including monetary policy approaching the effective lower
bound and the introduction of non-standard measures. We estimate a time-varying model proposed
by Ciccarelli et al. (2017) for the United States and the United Kingdom using market-based inflation
expectations derived from financial instruments.63 In line with previous studies, we find that the
relationship between short-term and longer-term inflation expectations is not stable over time, but
episodes of instability are generally short-lived and policies have gradually succeeded in reversing
unanchoring. For the United States, the estimated coefficient remains in a relatively narrow band and
is statistically insignificant for most of the sample period (see panel a of Chart C). For the United
Kingdom, the results suggest somewhat greater time variation in the estimated coefficient. We find
several episodes (including the GFC and a prolonged period from 2015 to 2017) when the
pass-through may have become significantly positive (see panel b of Chart C). However, these
episodes seem to have been transitory.
63 For the other economies in our sample, financial instruments providing inflation protection are either
unavailable or traded in thin markets. Consequently, this analysis is limited to the United States and the
United Kingdom.
-0.4
-0.2
0.0
0.2
0.4
JP UK EA NO SE CA CH US
ECB Occasional Paper Series No 264 / September 2021
73
Chart C
Signs of unanchoring remain limited on the basis of market-based inflation expectations and allowing
for time variation
Sources: Bank of England, Board of Governors of the Federal Reserve and Eurosystem staff calculations.
Notes: For the United States, the time-varying coefficient captures the pass-through from inflation compensation on two-year Treasuries and TIPSs
(“break-even” rates) to inflation compensation on five-year forward rates five years ahead. For the United Kingdom, the time-varying coefficient captures the
pass-through from the inflation compensation implied by the 30-month spot rate to the inflation compensation implied by the ten-year forward rate. In line with
Ciccarelli et al. (2017), we estimate the following equation at a monthly frequency using Bayesian methods: ,= + , +
,
where is computed as changes in short and long-term inflation expectations over the previous six months, the pass-through measure is modelled as a
time-varying parameter and is a stochastic volatility term to account for potential changes in market conditions and volatility since the beginning of the GFC.
For both countries, the last observations are for July 2020.
Since 2000, common factors in longer-term inflation expectations among advanced
economies do not appear to have played a significant role. The country-specific dynamics of
longer-term inflation expectations are fairly heterogeneous. Principal component analysis shows how
much of the variation in the data can be explained by a single common factor, revealing that the first
principal component explains 32% of variation, which implies limited co-movement across
country-specific expectations. Regressing longer-term inflation expectations for each country on the
first principal component suggests that this global factor has very limited explanatory power and
implies that longer-term inflation expectations predominantly reflect domestic developments.
We conclude that developments in longer-term inflation expectations are heterogeneous
across advanced economies, primarily reflecting idiosyncratic factors, with evidence of
unanchoring remaining limited. More specifically, longer-term inflation expectations have
stabilised since 2000, following a secular decline during the 1990s. Evidence of pass-through from
short-term inflation expectations to longer-term expectations is limited, suggesting that inflation
expectations remain anchored. This finding also holds when using market-based inflation
expectations and allowing for time variation in the pass-through coefficient. Finally, global factors
appear to play only a limited role in driving longer-term inflation expectations.
United States United Kingdom
(estimated time-varying pass-through coefficients) (estimated time-varying pass-through coefficients)
-1.0
-0.5
0.0
0.5
1.0
2005 2008 2011 2014 2017 2020
Median
16th-84th percentiles
-1.0
-0.5
0.0
0.5
1.0
2005 2008 2011 2014 2017 2020
Median
16th-84th percentiles
ECB Occasional Paper Series No 264 / September 2021
74
3.2.3 Overall assessment and its measurement using a heat map
Overall, it is clear that longer-term inflation expectations in the euro area have
become less well anchored over the years. This can be seen in both survey and
market-based measures and across different unanchoring metrics (levels,
responsiveness and higher moments). While it is too early to conclude that inflation
expectations have become completely unanchored, the risk of this happening as a
result of the additional negative shock caused by the COVID-19 pandemic and its
aftermath is high. More generally, the fact that different unanchoring metrics point to
very different conclusions as regards the degree and timing of unanchoring
tendencies underscores the importance of considering and cross-checking all metrics,
as each of them has both strengths and weaknesses. Attempting to reconcile such
differences is a challenging endeavour, so it may be useful to provide a general visual
overview of the different dimensions of (un)anchoring in the form of a heat map.
Heat maps are an illustrative method that can be used to assess whether or not
expectations are consistent – on the basis of a given criterion – with the
concept of anchoring. Heat maps have the clear advantage of providing a quick
overview of a wide range of indicators, while inevitably abstracting from technical and
numerical detail. Moreover, they all treat individual indicators the same and do not
allow for any consideration of possible time variation in the reliability of individual
measures. They are naturally dependent on the design of the benchmark. For
instance, if proxies for anchoring are considered relative to their historical variation
(volatility), a relatively small change in a proxy that has been stable over time can
potentially indicate substantial unanchoring. In contrast, if a proxy has been fairly
volatile in the past, even a relatively large change in expectations will not necessarily
point to considerable unanchoring. Algorithms can be designed to normalise across
different indicators (see Annex B for more details).
To facilitate graphical interpretation and narrative storytelling, heat maps use
different colours and shades. For example, in the case of a level-based indicator of
inflation expectations, red could be used to indicate that expectations are on the high
side (i.e. over-heating), blue could be used to indicate that expectations are on the low
side (i.e. too cool), and white could be used to show that they are at neutral levels
(neither too high nor too low). Furthermore, shading can be using to communicate the
degree to which expectations are running hot or cold (i.e. using more or less intense
shades of red/blue). The heat maps below show the risk of unanchoring for
longer-term inflation expectations (with Annex B also showing heat maps for
shorter-term expectations), aiming to summarise the main findings via different
metrics capturing the level of inflation expectations, their sensitivity (responsiveness)
to macroeconomic developments and economic agents’ uncertainty about those
expectations. Each row of the heat map refers to a different metric, while columns
show time periods. The colours of cells reveal whether expectations are – on the basis
of a given criterion – consistent with the concept of anchored expectations (white) or
not (red, blue, purple or grey). Moreover, the intensity of the colours shows how
severe any potential unanchoring is, with darker shades indicating more severe
unanchoring.
ECB Occasional Paper Series No 264 / September 2021
75
The colour intensity of the heat maps confirms that there have been some signs
of expectations becoming less firmly anchored over the past decade. This is
particularly true of market-based expectations, but it also applies, to a lesser extent, to
survey-based measures. Shorter and more medium-term market-based measures
have stayed substantially below the perceived inflation target range (i.e. 1.7-2.0%).
There is also some evidence that the responsiveness of inflation expectations to
short-term developments has increased since the beginning of the GFC. Risks to
anchoring are also signalled by increases in the aggregate uncertainty of longer-term
SPF forecasts, especially in two recent years.
Chart 35
Heat map for longer-term inflation expectations
Legend
All rows: Rows I.1-I.7: Row I.8: Rows II.1-II.7: Row III.1:
Colours show how far
from the perceived ECB
target range inflation
expectations are.
Deviations are measured
in standard deviations
(sd) of the inflation
expectations series.
Grey cells indicate
periods where probability
is more than 1 sd below
its historical average.
Colours show whether
pass-through coefficients
are statistically significant
and more than 1 sd larger
than zero.
Grey cells indicate
periods where uncertainty
is more than 1 sd above
its historical average.
Notes: (1) Macro surprises include inflation and corporate sentiment over a nine-month rolling window. (2) Macro surprises include
inflation, GDP and PMI over a 36-month rolling window.
Part I. Levels of inflation expectations
123412341234123412341234123412341234123412341234123412341234123
I.1 SPF, long term, mean vs. perceived ECB inflation target range
I.2 SPF, long term, mean of distr. vs. perceived ECB inflation target range
I.3 Consensus Economics, 6-10y ahead, mean vs. perceived ECB target range
I.4 Market data, 1y rate 4y ahead vs. perceived ECB target range
I.5 Market data, 1y rate 4y ahead, exp. component vs. perceived ECB target range
I.6 Market data, 5y rate 5y ahead vs. perceived ECB target range
I.7 Market data, 5y rate 5y ahead, exp. component vs. perceived ECB target range
I.8 SPF, long term, probability of inflation between 1.5% and 1.9%
II.1 SPF, long term, responsiveness to negative inflation surprises
II.2 SPF, long term, responsiveness to positive inflation surprises
II.3 SPF, long term, responsiveness to actual inflation
II.4 SPF, long term, responsiveness to short-term inflation forecast
II.5 Market data, 5y rate 5y ahead, responsiveness to 1y rate
II.6 Market data, 5y rate 5y ahead, responsiveness to macro surprises(1)
II.7 Market data, 5y rate 5y ahead, responsiveness to macro surprises(2)
III.1 SPF, long term, aggregate uncertainty
Part I. Levels of inflation expectations
Part II. Responsiveness of inflation expectations
Part III. Uncertainty (relative to long-term mean s)
2020
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2008
2005
2006
2007
no data
>2 sd above
1.5 to 2.0 sd above
1.0 to 1.5 sd above
0.5 to 1.0 sd above
within 0.5 sd
0.5 to 1.0 sd below
1.0 to 1.5 sd below
1.5 to 2.0 sd below
>2 sd below
close to average
1 sd to 2 sd below
more than 2 sd below
less than 1 sd above zero
1 to 2 sd above zero
2 to 3 sd above zero
3 sd or more above zero
not more than 1 sd above average
1 to 2 sd above average
2 sd or more above average
ECB Occasional Paper Series No 264 / September 2021
76
4 Inflation expectations in macroeconomic
forecasting
Central banks can use observed data on inflation expectations in two ways for
forecasting purposes. First, for medium-term horizons, survey and market-based
measures probably reflect forecasts which take account of the latest developments
and information and can thus provide a quantitative benchmark for central banks’ own
projections. Second, survey and market-based measures can be used as explanatory
variables in central banks’ models and potentially improve their forecasting
performance. This chapter discusses these two purposes from an empirical
perspective.
4.1 Indicators of inflation expectations as standalone
forecasts
A useful test of inflation expectations as direct, standalone forecasts is whether
they have predictive power for actual inflation. There is mixed evidence in the
literature on the question of whether measures of inflation expectations are reliable
and accurate forecasts of future inflation. For instance, Ang et al. (2007) showed that
survey-based measures outperformed ARIMA models, Phillips curve models and term
structure models, while Gil-Alana et al. (2012) reported, for the United States, that
survey-based expectations outperformed standard time series models. However, in a
more recent study, Trehan (2015) showed that the forecast accuracy of surveys of
households and professional forecasters had deteriorated. Bauer & McCarthy (2015)
showed that, for the United States, market-based measures had lower accuracy than
survey-based measures and simple forecasting rules such as the random walk, which
was probably related to their inclusion of risk premia. In a comprehensive overview of
available forecasting models, Faust & Wright (2013) compared the forecasting
performance of various approaches, showing that survey based measures
outperformed model-based forecasts They also suggested that the forecast accuracy
of market-based measures could be impaired by the inclusion of time-varying risk
premia. More generally, they argued that if monetary policy smooths deviations of
inflation from some slowly moving target, it will be difficult to beat forecasts that simply
take account of nowcasting and secular changes in the local mean inflation rate.
4.1.1 Comparison of market and survey-based inflation expectations as
forecasts
When assessing the forecast accuracy of market and survey-based measures
of inflation expectations, it is important to bear in mind their different features.
Chapter 2 discussed a number of features, such as (i) the differences in target
variables (e.g. the HICP for surveys, versus the HICP excluding tobacco for
ECB Occasional Paper Series No 264 / September 2021
77
market-based measures) and (ii) the role of risk premia and technical factors in
market-based indicators. For forecast accuracy, other features also matter (Meyler &
Grothe, 2015). For example, in the case of market-based indicators, there is an
indexation lag for swaps, which implies, for instance, that information included in a
one-year inflation swap rate reflects three months of actual inflation data and
expectations over a nine-month horizon. For survey data, meanwhile, the frequency
and timing (e.g. quarterly in the case of the SPF: mid-January, mid-April, mid-July and
mid-October) may not coincide with the forecast schedule of the panel members,
which implies that their reported forecasts may not reflect the latest available
macroeconomic data.64 Moreover, survey expectations reflect averages across
(unbalanced) panels, so might, in principle, suffer unduly from “bad” forecasters in the
panel. However, Genre et al. (2013) for the euro area and D’Agostino et al. (2012) for
the United States have shown that it is generally difficult to identify individual
forecasters that consistently outperform the average forecast.
A comparison of forecast performance can be conducted on the basis of
different statistical tests. Table 1 summarises these statistics for inflation swap
rates, SPF survey expectations and simple statistical benchmark forecast rules such
as the random walk (RW), an autoregressive process of order 1 (AR) and a constant
expectation with an assumed level of 2%. In order to acknowledge the indexation lag,
the comparisons are presented in different columns for similar horizons. In all cases
(also for swap rates) the comparison is vis-à-vis the HICP.
64 However, available evidence suggests that SPF responses are fairly timely, particularly because (a) for
the one and two-year-ahead horizons considered here, there is a relatively high frequency of regular
updates and (b) survey respondents also adjust their forecasts in exceptional circumstances. Meyler &
Rubene (2009) surveyed respondents participating in the SPF and found that the majority of respondents
(84%) reported that their forecasts were updated on a regular calendar basis, while around one-third
indicated that they updated their forecasts following data releases or other events relevant to their
forecasts. In this context, there appears to be some correlation between the length of the forecast horizon
and the frequency with which forecasts are revised. On average since 1999, approximately 80% of SPF
respondents have revised their forecasts for inflation one year ahead from one round to the next,
compared with 70% for inflation two years ahead and 30% for inflation five years ahead.
ECB Occasional Paper Series No 264 / September 2021
78
Table 3
Forecast performance statistics – survey and market-based measures
9ma 4qa 21ma 8qa
Mean error (ME)
Swaps 0.10 -0.12
SPF -0.11 -0.30
RW -0.09 -0.12 -0.17 -0.19
AR 0.03 0.03 0.08 0.08
2% -0.52 -0.54 -0.57 -0.58
Root mean squared error (RMSE)
Swaps 0.79 1.03
SPF 0.96 1.11
RW 1.09 1.26 1.49 1.53
AR 0.99 1.08 1.20 1.17
2% 1.16 1.17 1.19 1.21
Theil’s U
Swaps vs RW 0.73 0.69
SPF vs RW 0.76 0.72
Swaps vs AR 0.80 0.86
SPF vs AR 0.89 0.95
Swaps vs 2% 0.68 0.86
SPF vs 2% 0.82 0.92
Diebold-Mariano (DM) statistic
Swaps vs RW 1.35 1.82
SPF vs RW 1.12 1.66
Swaps vs AR 1.42 1.16
SPF vs AR 0.94 0.44
Swaps vs 2% 1.73 0.92
SPF vs 2% 1.25 1.08
Sources: Eurosystem staff calculations based on Meyler & Grothe (2015).
Note: “9ma” = nine months ahead; “4qa” = four quarters ahead; “21ma” = 21 months ahead; “8qa” = eight quarters ahead.
The mean error points to a relatively modest forecast bias for both market and
survey-based measures. For swaps, the 9ma horizon under-forecasted slightly (with
actual inflation 0.10 p.p. higher than forecast inflation). However, when compared with
HICPxT, the mean error was essentially zero. For the 21ma horizon, the swaps slightly
over-forecasted (with actual inflation 0.12 p.p. lower than forecast inflation), and the
gap was larger in the case of HICPxT (0.2 p.p.). For the SPF, the 4qa and 8qa
horizons both over-forecasted HICP inflation (i.e. actual inflation was lower than
forecast inflation). However, it should be noted that when computed over the entire
sample (1999-2020), the 4qa mean error for the SPF changed sign to +0.08 p.p., and
the equivalent figure for the 8qa horizon fell to -0.08 p.p. Both SPF and swap data
have a larger bias than the AR model, but a smaller one than the RW and the constant
2%. Thus, although a specific measure may look good (either in absolute or in relative
terms) for a given time period, performance can and does change over time.
The RMSE and Theil’s U suggest that both ILS and SPF data are more accurate than
the statistical benchmarks. In the case of the ILS 9ma, for instance, its RMSE of 0.79
ECB Occasional Paper Series No 264 / September 2021
79
compares favourably with those of the AR (0.99), the RW (1.09) and the 2%
benchmark (1.16) and consequently the Theil’s U statistic is below unity in each case
(ranging from 0.68 to 0.73). Using the HICPxT as opposed to the HICP makes only a
marginal difference. In fact, notwithstanding the fact that the ILSs are priced on the
basis of the HICPxT, using the HICPxT actually increases the RMSE marginally (from
0.79 to 0.81). The SPF 4qa also had an RMSE (0.96) which was lower than those of
the statistical benchmarks, and the same was true of the ILS 21ma and the SPF 8qa.
Decomposing the RMSE into bias (or the mean error) and variance shows that, for
both ILS and SPF data, the variance component is substantially larger than the bias
component, meaning that, relatively speaking, although they capture the average level
relatively well, both are less good at capturing movements in inflation over time. This
also helps to explain why, for ILS data, the RMSE remains largely unchanged when
using the HICPxT as opposed to the HICP (i.e. because the variance component does
not change so much).
The Diebold-Mariano statistic allows a comparison of forecasts (Diebold, 2013;
Diebold & Mariano, 1995). Although both ILS and SPF data outperform the statistical
benchmarks, the Diebold-Mariano statistic is not generally statistically significant. The
only exceptions to this (at the 10% level) are the ILS 9ma (against the 2% forecast),
the ILS 21ma (against the RW forecast) and the SPF 8qa (against the RW forecast).
Outperformance vis-à-vis statistical benchmarks implies some general
credibility for survey and market-based expectations as forecasts, but how they
compare with Eurosystem projections is also relevant. When conducting such
comparisons, we need to bear in mind that those projections are based on
conditioning assumptions and can thus lack some degree of freedom. The
Eurosystem and ECB staff projections forecast HICP inflation over horizons of 1 to
12 months in the Narrow Inflation Projection Exercise (NIPE)65 and 1 to 9 quarters in
the (Broad) Macroeconomic Projection Exercise ((B)MPE).66 Comparing, for instance,
the ILS 9ma with the NIPE 9ma over the period 2005-2020 suggests that the NIPE
performs slightly better, with an RMSE of 0.70 vs 0.79. Comparing the SPF 4qa/12ma
with the NIPE 12ma suggests that the SPF performs marginally better, with an RMSE
of 0.86 vs 0.89.67 However, using rolling windows suggests that no single measure or
projection outperforms others all of the time. The left-hand panel of Chart 36 shows
that, for the shorter horizon (9ma/12ma), the SPF tends, on average, to perform better
than the NIPE and ILS data. Although there are times when the Theil’s U for the SPF
12ma relative to the NIPE 12ma is (a) above unity and/or (b) is above the Theil’s U for
the ILS 9ma relative to the NIPE 9ma. The right-hand panel shows that, for the longer
horizon (21ma/24ma), the picture is more balanced. At the beginning and end of the
65 Since 2015, the maximum horizon in the NIPE has been 11 months. As the RMSE increases with the
forecast horizon, this actually makes it a more challenging benchmark. See ECB (2016) for more
information.
66 The forecast horizon for the Broad Macroeconomic Projection Exercises varies according to the round.
Currently, the longest horizons are 10qa in the September round, 11qa in the June round, 12qa in the
March round and 13qa in the December round.
67 Over the period 2005-20, the NIPE appears to perform better than the SPF (RMSE of 0.84 vs 0.96).
However, some of this apparently superior performance may be due to the fact that the longest NIPE
horizon is, as of 2015, 11ma. In the period when both 11ma and 12ma NIPE forecasts were available, the
difference between the RMSEs for the two was 0.08 p.p.
ECB Occasional Paper Series No 264 / September 2021
80
sample, the relative Theil’s U for the SPF is below the figure for ILS data, but in the
middle it is above. On average over the sample, the average Theil’s U is lower for ILS
data (albeit slightly above unity at 1.03) than it is for the SPF (1.08). Overall, a
reasonable conclusion from these comparisons is that all measures should be
monitored and analysed as benchmarks for Eurosystem projections.
Chart 36
Comparison of market and survey-based measures against Eurosystem projections
Theil’s U relative to NIPE Theil’s U relative to (B)MPE
Source: Eurosystem staff calculations.
Note: This chart shows the Theil’s U relative to Eurosystem projections (based on RMSE statistics calculated over four-year rolling
windows).
4.1.2 Assessment of probability distributions/density forecasts using the
probability integral transform (PIT) approach
SPF survey expectations can also be assessed on the basis of the probability
distribution surrounding their point forecasts. At first glance, the problem of
evaluating density forecasts seems to be very challenging, as there appears to be no
observable benchmark that could facilitate a test of how closely reported densities
correspond to the unobservable true density of the variable under consideration.68
One possible way of dealing with this is the PIT approach. The basic idea behind this
approach is to check whether actual figures for the variable are, on average,
consistent with the forecasted densities. To check this, Diebold et al. (1998) propose
the estimation of the probability integral transform (Zt), which gives the estimated
probability of the variable being less than or equal to the actual outcome (observed
68 This section draws on Annex 3 to Bowles et al. (2010).
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
2009 2011 2013 2015 2017 2019
Theil's U (SPF 1ya vs NIPE 1ya)
Theil's U (ILS 9ma vs NIPE 9ma)
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
2009 2011 2013 2015 2017 2019
Theil's U (SPF 2ya vs (B)MPE 2ya)
Theil's U (ILS 21ma vs (B)MPE 21ma)
ECB Occasional Paper Series No 264 / September 2021
81
only ex post) according to the predictive forecast density.69 For example, suppose that
forecasters correctly assess the mean of future inflation, but mistakenly assume a
normal (i.e. Gaussian) distribution, whereas the true distribution actually has higher
than normal probabilities for extreme outcomes (i.e. “fat tails”). Under such
circumstances, there will be many more realisations of inflation taking on extreme
values than were predicted by respondents’ distributions. One limitation of the PIT
framework is that it can only be applied to a relatively long time sample, as it is only
possible to assess averages of ex post outcomes over considerable periods of time.
The PIT framework suggests that SPF panellists have generally underestimated
the uncertainty surrounding their forecasts. Ex post, they too often end up in the
tails (lower/upper quintiles) of their ex ante probability distributions.70 The left-hand
panel of Chart 37 shows the realised PIT scores for the reported aggregate probability
distributions surrounding inflation forecasts one, two and five years ahead. For the
early years of EMU (1999-2004), inflation outcomes were consistently at the upper
end of the reported probability distributions for each horizon. Thereafter, there was
more variation, and also occasions (e.g. the 2009 and 2010 rounds) when the PIT
score for shorter horizons was the opposite of that seen for longer horizons. The
right-hand panel of Chart 37 shows how often the realised PIT score was in each
quintile relative to what one would expect if they were truly random (i.e. 20% in each
quintile). In practice, outcomes have tended to be in the lower and upper quintiles too
often, and they have tended to be in the middle quintiles too seldom. Following the
GFC, there was a step increase in the standard deviation of the aggregated SPF
probability for HICP inflation. This increase in ex ante uncertainty was visible across
horizons (both shorter and longer-term horizons) and was also seen for other
economies (being observed, for example, in the results of the Bank of England’s
Survey of External Forecasters). Nonetheless, it still appears to be the case that actual
outcomes tend to be towards the lower extremity of the reported ex ante probability
distribution and thus imply an underestimation of the true degree of uncertainty.
Overall, analysis of the probability distributions suggests that, although informative, as
they represent the reported beliefs of professional forecasters, they are likely to
understate the actual degree of uncertainty surrounding reported inflation
expectations. It may also be that, rather than their absolute level, it is their evolution
over time that is most informative.
69 Zt is defined as the cumulative probability distribution function as evaluated at the time of the actual
outcome, Xt, for the forecast period in question – i.e. =()
. The properties of Zt depend on
how closely the reported densities approximate the true underlying density. If survey respondents
accurately assess the true underlying probabilities, then Zt will be a uniformly, independent and
identically distributed random variable bounded between zero and unity. Conversely, if forecasters have
not accurately assessed the shape and location of the true density, the Zt series will display
non-uniformities that highlight the discrepancies between the reported and true densities.
70 This finding would be even stronger if applied to distributions at individual level, as the standard deviation
of the aggregate distribution combines both average individual uncertainty (i.e. the standard deviation of
the individual distributions) and disagreement between individuals (i.e. the standard deviation of point
forecasts). Therefore, the standard deviation of the aggregated probability distribution is always at least
as large as the average standard deviation of the individual distributions, and the greater the
disagreement between forecasters, the more this will be the case.
ECB Occasional Paper Series No 264 / September 2021
82
Chart 37
PIT scores for SPF density forecasts
PIT – raw PIT – standardised
(percentages) (percentages)
Source: Eurosystem staff calculations.
Notes: The left-hand panel shows the raw PIT scores for the one, two and five-year-ahead horizons. The right-hand panel shows how
often the realised PIT score was in each quintile relative to what one would expect if they were truly random (i.e. 20% in each quintile). If,
for example, the score was in a given quintile 30% of the time, that would be 10 p.p. more than would be expected if it were truly random.
4.1.3 How to fruitfully use survey expectations in combination with
Bayesian VARs: a performance assessment
SPF survey expectations are forecasts in their own right and thus comparable
with pure model-based forecasts. Bayesian vector autoregression models (BVARs)
have become a standard tool for forecasting and scenario analysis in the central
banking community. SPF expectations – for medium-term horizons, at least – are
formed on the basis of models, but they also include judgement and may thus go
beyond time series models’ extrapolation of historical data. This exercise explores
whether the two information sources (i.e. pure model-based forecasts and SPF
inflation expectations) can be brought together in an optimal combination that exploits
their comparative advantages and beat each individual source. Optimally combining
forecasts from multiple models to more robustly predict future paths of
macroeconomic variables is a methodology which has been advocated for some time
in the economic literature – see, for example, Timmermann (2006) and Genre et
al. (2013).
The value of SPF information is tested via the pooling of forecasts in
successive stages.71 The first step optimally pools real-time forecasts from several
types of BVAR, which differ in terms of modelling choices (e.g. dataset size and
composition, data transformation, degree of time variation, prior specification or
inclusion of off-model information). In this linear optimal pooling, weights are
time-varying and selected in order to maximise forecast accuracy. The second step
71 Results are based on Bańbura et al. (2021a).
0
10
20
30
40
50
60
70
80
90
100
1999 2002 2005 2008 2011 2014 2017 2020
One year ahead
Two years ahead
Five years ahead
0.05
-0.09
-0.13
0.00
0.17
0.11
-0.13
-0.08 -0.05
0.14
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
0.20
1 2 3 4 5
One year ahead
Two years ahead
ECB Occasional Paper Series No 264 / September 2021
83
includes the SPF forecasts in the optimal pool and also investigates alternative
approaches based on “tilting” the model forecasts to the SPF expectations (on the
assumption that these should provide extra relevant information).72
This exercise evaluates the accuracy of both point and density forecasts. It
compares (i) the optimal linear pooling combination, (ii) the optimal linear pooling
combination including the SPF and (iii) entropic tilting (whereby we tilt either the
individual models before combining them, or the model combination, to either the first
moment of the SPF or both the first and second moments of the SPF). The evaluation
is carried out over the period 2000-19 at the one and two-year-ahead horizons. The
performance measures are the root mean squared forecast error (RMSFE), the log of
predictive scores (LPS) and the continuous ranked probability score (CRPS). The
results are summarised in Table 4. Looking at RMSFEs shows that the SPF has a
higher forecast accuracy than the optimal combination of models and that this
increases with the forecast horizon. Further, the accuracy of the models increases if
they are individually tilted to the SPF mean (more so if the individual models are tilted
ex-ante).
Table 4
Forecasting HICP inflation: absolute accuracy scores and uniformity test results for the
main combinations considered
Opt. pool SPF
Opt. pool
with SPF
Opt. pool
mean-tilted
Opt. pool
mean and VAR-tilted
ex-ante ex-post ex-ante ex-post
4Q RMSFE 0.689 0.004 0.698 0.672 0.687 0.670 0.672
CRPS 0.503 0.469 0.499 0.461 0.471 0.474 0.475
LPS 1.306 1.330 1.303 1.188 1.250 1.312 1.388
Berkowitz 0.839 0.002 0.704 0.218 0.156 0.000 0.000
8Q RMSFE 0.823 0.736 0.824 0.739 0.757 0.746 0.744
CRPS 0.567 0.538 0.579 0.523 0.534 0.547 0.546
LPS 1.429 1.470 1.431 1.347 1.397 1.693 1.713
Berkowitz 0.552 0.000 0.961 0.368 0.232 0.000 0.000
Source: Eurosystem staff calculations.
Notes: This table shows results for: (i) including the simulated SPF density in the BVAR pool and combining individual models and the
SPF by means of optimal pooling (“opt. pool with SPF”); and (ii) using entropic tilting, including moments from the SPF, in the following
four ways: tilting each individual model to the SPF mean, then performing optimal pooling (“opt. pool mean-tilted ex-ante”); tilting the
optimal pool of combined models to the SPF mean (“opt. pool mean-tilted ex-post”); tilting each individual model to the SPF mean and
variance, then performing optimal pooling (“opt. pool mean and VAR-tilted ex-ante”); and tilting the optimal pool of combined models to
the SPF mean and variance (“opt. pool mean and VAR-tilted ex-post”). A p-value for the Berkowitz test that is smaller than 0.10 indicates
that the null hypothesis of good calibration can be rejected at the 10% confidence level.
The results also suggest that while the SPF improves forecasts when its first
moment (average) is used, this does not hold with regard to the second
moment (standard deviation). In terms of the accuracy of density forecasts, the
relative performance of the SPF is worse for both horizons and does not, therefore,
help when accurate measures of uncertainty around the point forecasts are needed.
This is also evident from the distribution and the PIT scores: the SPF always tends to
be over-confident, delivering narrow distributions and U-shaped PITs (see Chart 38
72 In order to be able to combine densities from the SPF and the models, we construct a continuous
distribution from the discrete bins in the SPF histograms using kernel densities.
ECB Occasional Paper Series No 264 / September 2021
84
for an example). When tilting to both the first and second moments of the SPF, there is
a general worsening of the performance of the combined models for both horizons. It
is therefore counterproductive to include too much survey information. The overall
conclusion is that incorporating survey information improves forecast performance
and calibration, albeit only when the first moment is used.
Chart 38
PIT scores for selected models: HICP inflation forecasts one year ahead
(percentages)
Source: Eurosystem staff calculations.
Notes: These panels show PIT scores for the model combinations in question. The x-axis shows the ten deciles, and the y-axis indicates
the percentage of outcomes that fall within each decile. If the forecasts were well calibrated, one would expect approximately 10% in
each decile.
4.2 Predictive power of indicators of inflation expectations
Beyond their role as standalone forecasts, inflation expectations can serve as
inputs for inflation forecasting models. The literature argues that using
expectations in this way can play an important role in predicting inflation, but there is
no consensus on which expectations are most relevant. Two aspects are key in this
respect: first, the question of which parts of the inflation expectations term structure to
use (as in some models, the inflation trend is linked to long-term expectations, while in
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10
Opt pool
0
2
4
6
8
10
12
14
16
18
20
12345 6 7 8 9 10
SPF
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10
Mean-tilted opt pool
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10
Mean and VAR-tilted opt pool
ECB Occasional Paper Series No 264 / September 2021
85
other models, short-term inflation expectations are considered a driver of the cyclical
component of inflation);73 and second, the question of whose expectations to use, as
there is no consensus in the literature on whose expectations matter more. This
section explores these issues using two types of real-time forecast evaluation
exercise.
Does the inclusion of inflation expectations in time series models improve the
performance of inflation forecasting? The answer is yes. An extensive real-time
forecast evaluation covering both unconditional and conditional forecasts and
encompassing a diverse set of models and several economies suggests that
indicators of inflation expectations do bring some gains to the accuracy of inflation
forecasts, but they are typically modest. The available evidence does not point to one
type of model being superior to another in terms of forecasting performance, favouring
a comprehensive approach.
The first exercise evaluates the forecast accuracy of a battery of alternative
time series models including short or long-term SPF inflation expectations
(Bańbura et al., 2021b). The evaluation covers various ways of linking inflation
expectations to inflation, such as detrending inflation, linking the unobserved inflation
trend in a Phillips curve model to long-term inflation expectations, informing the
conditional mean of inflation in a VAR model, or disciplining the long-run priors for
inflation (see Box 7 for more details). Chart 39 and Chart 40 show the RMSEs for
unconditional forecasts one year ahead for all expectation-augmented models relative
to their counterparts without expectations. All models include long-term SPF
expectations, with the exception of models 5, 6aS and 6bS. For both headline inflation
and HICP inflation excluding energy and food (HICPX), most models derive forecast
accuracy gains from incorporating survey-based measures of expectations, but these
gains are not typically large. In this respect, the results confirm the findings in
Section 4.1, showing that the augmented models do not outperform the survey
expectations as such. The predictive gains from including survey expectations tend to
be slightly higher for the medium-term horizon (two years ahead) than for the
short-term horizon (one year ahead). This is particularly true of the HICPX indicator
(see Chart 41 and Chart 42). It is worth noting that differences in the forecasting
performance of models with long and short-term expectations tend to be small.74
73 For the United States, some results point to a role for long-term expectations (Clark & Davig, 2008), while
others highlight the importance of short-term measures (Canova & Gambetti, 2010; Fuhrer, 2012). For
the euro area, too, the existing evidence is mixed, with some studies pointing to the importance of short to
medium-term term inflation expectations (Ciccarelli et al., 2017; Stevens & Wauters, 2018;
Moretti et al., 2019) and others showing the significance of long-term survey-based inflation expectations
(Jarociński & Lenza, 2018; Bańbura & Bobeica, 2020).
74 Models 1, 2, 3 and 4 use long-term measures of expectations; model 5 uses short-term measures; and
model 6 considers both short and long-term expectations in turn.
ECB Occasional Paper Series No 264 / September 2021
86
Chart 39
Headline inflation forecasts one year ahead
RMSFEs of models incorporating SPF expectations relative to their counterparts without
expectations
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q4 2001 to Q4 2019. The numbers denote model classes: 1 = ADL models with
time-varying trend inflation proxied by an exponentially weighted moving average (E) or the historical mean (M) (in the version without
expectations); 2 = ADL models with time-varying trend inflation, time-varying coefficients and stochastic volatility; 3: = Bayesian VARs
with democratic priors, with standard (S) and tight (T) priors; 4 = Bayesian VARs with time-varying trends; 5 = Phillips curves with
constant coefficients; 6 = Bayesian VARs with Minnesota priors and long(L)- and short(S)-term inflation expectations. The letters “a” and
“b” denote univariate and multivariate models respectively.
Chart 40
HICPX inflation forecasts one year ahead
RMSFEs of models incorporating SPF expectations relative to their counterparts without
expectations
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q4 2001 to Q4 2019. See notes to Chart 39.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1aM 1aE 1bM 1bE 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1aM 1aE 1bM 1bE 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
ECB Occasional Paper Series No 264 / September 2021
87
Chart 41
Headline inflation forecasts two years ahead
RMSFEs of models incorporating SPF expectations relative to their counterparts without
expectations
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q4 2002 to Q4 2019. See notes to Chart 39.
Chart 42
HICPX inflation forecasts two years ahead
RMSFEs of models incorporating SPF expectations relative to their counterparts without
expectations
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q4 2002 to Q4 2019. See notes to Chart 39.
Robustness checks point to a relatively good performance by the SPF
measures in models, but also some time variation in this performance.
Long-term Consensus Economics inflation expectations or 5y5y ILS do not improve
the forecast accuracy of the models relative to using SPF expectations. Also, using
expectations collected by the European Commission from consumers and firms
instead of short-term SPF expectations does not improve the forecast accuracy of the
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1aM 1aE 1bM 1bE 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1aM 1aE 1bM 1bE 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
ECB Occasional Paper Series No 264 / September 2021
88
models.75 Also, the inflation forecast gains coming from the inclusion of survey-based
measures of expectations vary substantially over time and deteriorate in the aftermath
of the sovereign debt crisis relative to models without expectation measures. This is
also confirmed by Bańbura & Bobeica (2020), who find that including long-term
Consensus Economics expectations in Phillips curve models strongly improves the
forecast performance from the early 2000s until the low-inflation period, after which
including those measures actually impairs performance. Models that embody a lower
inflation trend would have performed better over this recent past. In the context of the
models presented here, an interesting case is the autoregressive distributed lag (ADL)
model, in which an exponentially weighted moving average of inflation appears to
capture trend inflation at least as well as – and, in many cases, better than – long-term
SPF expectations.
Conducting this type of exercise at country level points to some nuances in the
usefulness found for the euro area as whole. The ECB/Eurosystem staff projection
exercises forecast euro area inflation on a bottom-up basis using country-specific
forecasts. It is hence interesting to see whether forecast performance changes due to
the consideration of inflation expectations are also visible at the level of individual euro
area countries. The real-time forecast evaluation of including inflation expectations
has hence also been applied to a set of individual countries (Germany, France, Italy,
Spain, the Netherlands, Belgium, Austria and Finland). As SPF data are only available
for the euro area, the country-specific evaluations use inflation expectations from
Consensus Economics. Also, owing to the more limited availability of real-time data for
some countries, the forecast evaluation only starts in 2005. Chart 43 and Chart 44
show the RMSFEs of the model versions including expectations relative to their
corresponding alternatives without expectations, providing data separately for each
country. Overall, and in contrast to the findings for the euro area, forecasting gains
seem to be somewhat larger for the short-term horizon (one year ahead) than they are
for the medium-term horizon (two years ahead). Moreover, and again in some contrast
to the euro area, adding expectations helps more in forecasting headline inflation than
core inflation. With the exception of certain model versions (particularly in the case of
Germany, France and Finland), models including expectations almost always yield
better forecasts in the short run. In the medium run, including expectations
substantially worsens the results for Italy (and, to a lesser extent, Finland, France and
Belgium). Finally, similar to the euro area, forecasting performance varies significantly
over time, particularly for headline inflation. From 2005 to 2009, adding expectations
leads to better forecasts for almost all models and countries. From 2010 to 2014, gains
from expectations become smaller, but they tend to increase again after 2015. When
compared with the country results in the (B)MPE, models including expectations tend
to perform worse (with the exception of Germany, the Netherlands and Austria).
However, this comparison favours the (B)MPE somewhat, as the NCBs already
incorporate information available in the current quarter, such as monthly inflation data.
75 This stands in contrast to the results of Álvarez & Correa-López (2020), who argue that expectations
“from consumers and firms are better at predicting inflation if compared to those from experts and,
especially, those from financial markets”. However, their analyses are based on “(pseudo) out-of-sample
conditional forecasts”, whereby they condition on actual realised expectations, which in the case of
consumers tend to be highly contemporaneously correlated with actual inflation – thus explaining the
apparent good performance of consumers’ expectations.
ECB Occasional Paper Series No 264 / September 2021
89
Chart 43
Country-specific headline inflation forecasts
RMSFEs of models incorporating Consensus Economics expectations relative to their
counterparts without expectations – one year ahead
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q1 2005 to Q4 2019. See notes to Chart 39.
Chart 44
Country-specific HICPX inflation forecasts
RMSFEs of models incorporating Consensus Economics expectations relative to their
counterparts without expectations – one year ahead
Source: Bańbura et al. (2021b).
Notes: RMSFEs are computed over the period from Q1 2005 to Q4 2019. See notes to Chart 39.
The second exercise uses the “thick” Phillips curve approach that is regularly
employed in the Eurosystem’s macroeconomic projection exercises to
cross-check underlying inflation projections and confirms modest forecast
gains from including expectations. Applying the theoretical (New Keynesian)
Phillips curve idea to the data is not trivial. Most commonly, the expectation term is
proxied by the one-quarter-ahead forecast, which can be proxied by survey-based
inflation expectations (as concluded by Mavroeidis et al. (2014) and applied, for
instance, by Nunes (2010)). However, some might argue that the assumption that
0.7
0.8
0.9
1.0
1.1
1.2
1aE 1aM 1bE 1bM 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
DE
FR
IT
ES
BE
AT
FI
NL
0.6
0.8
1.0
1.2
1.4
1.6
1.8
1aE 1aM 1bE 1bM 2a 2b 3aS 3aT 3bS 3bT 4a 4b 56aL 6aS 6bL 6bS All
DE
FR
IT
ES
BE
AT
FI
NL
ECB Occasional Paper Series No 264 / September 2021
90
such a short-term inflation expectation is exogenous is too strong and that it might be
correlated with the cost-push shocks affecting actual inflation, for instance. In this
respect, the “thick modelling” approach takes a more agnostic view and considers a
wide range of inflation expectations () across horizons and agents (Ciccarelli &
Osbat, 2017; Bobeica & Sokol, 2019; Kulikov & Reigl, 2019; Moretti et al., 2019;
Álvarez & Correa-López, 2020). In addition to expectations, this framework also
considers several measures of slack () and imported inflation () and forecasts
inflation () conditioned on the future paths for the explanatory variables projected in
each (B)MPE round:
=+++++
Chart 45
Performance of Phillips curve conditional forecasts for HICPX inflation
RMSFEs of models incorporating various survey expectations extended via an autoregressive
process
Source: Eurosystem staff calculations.
Notes: RMSFEs are based on a pre-COVID sample ranging from the December 2008 BMPE to the December 2019 BMPE. Figures are
averages across specifications with various measures of slack: GDP growth, the unemployment rate, the unemployment gap and the
output gap. Models have both fixed and time-varying parameters.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Backward
Consensus 1Q
Consensus 2Q
Consensus 3Q
Consensus 4Q
Consensus 5Q
Consensus 6Q
Consensus 2Y
Consensus 3Y
Consensus 4Y
Consensus 5Y
Consensus LT
SPF 1Y
SPF 2Y
SPF 5Y
SPF 5Y mean
Consumers 1Y
One-year ahead
(B)MPE
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Backward
Consensus 1Q
Consensus 2Q
Consensus 3Q
Consensus 4Q
Consensus 5Q
Consensus 6Q
Consensus 2Y
Consensus 3Y
Consensus 4Y
Consensus 5Y
Consensus LT
SPF 1Y
SPF 2Y
SPF 5Y
SPF 5Y mean
Consumers 1Y
Two-year ahead
(B)MPE
ECB Occasional Paper Series No 264 / September 2021
91
Chart 46
HICPX Phillips curve conditional inflation forecasts based on SPF one year ahead
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: Expectations have been extended via a random walk (RW), an autoregressive process (ARMA(4)), an assumption that inflation
will revert to 2% at the end of the forecast horizon (2%) and a scaled-down model of inflation expectation formation (behavioural).76
The real-time forecast accuracy evaluation confirms that specifications
involving inflation expectations tend to slightly outperform backward-looking
ones, as well as the (B)MPE HICPX inflation projection. At the same time, there is
not very much difference, in terms of results, between using short-term measures and
medium-term measures for expectations (see Chart 45). Given the conditional nature
of this forecasting exercise, one important element to consider is how to extend the
measures of inflation expectations over the projection horizon.77 In the workhorse
set-up, all measures are extended via an autoregressive process, as the real-time
forecast evaluation revealed that this is superior to other alternatives. In order to
illustrate the sensitivity of inflation forecasts to the method used to extend inflation
expectations, Chart 46 looks at inflation forecasts based on a Phillips curve model with
an unemployment gap and SPF expectations one year ahead, which are prolonged in
different ways. The point forecast for inflation varies quite markedly depending on the
method for extending expectations, so the judgement as to what process should
govern the evolution of inflation expectations is an important element.
4.3 The use of inflation expectations in E(S)CB projection
models
Most euro area NCBs include a proxy for inflation expectations in the models
used for the Eurosystem staff projections, but they are not generally directly
76 The model consists of a system of three equations for (1) actual inflation (a Phillips curve): = +
+; (2) short-term inflation expectations: = + (1 )+ and (3) long-term
inflation expectations linked to a three-year moving average of headline inflation. In the original version of
Nishino et al., (2016), long-term expectations were linked to their own lag and the price stability target.
77 Another aspect to consider is the fact that these results are for an evaluation at euro area level, and the
results might be different at country level.
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4
2018 2019 2020 2021 2022 2023
RW
ARMA(4) 2%
Behavioural
ECB Occasional Paper Series No 264 / September 2021
92
observed measures. In most cases, these expectations are modelled in a
backward-looking manner, with only a few NCBs using hybrid models – i.e. a mixture
of backward and forward-looking elements. If expectations need to be carried forward
over the projection horizon, then this is partly based on the extrapolation of past
inflation values and partly based on additional indicators (such as market-based swap
rates, labour cost growth or other variables), and there are also some cases where
expectations are assumed to gradually converge towards the inflation objective.
Expectations are mostly modelled via Phillips curves, which in some cases are
satellite models and in some instances are also part of bigger macro models. In
structural models, the inclusion of inflation expectations allows views to be formed on
the monetary policy transmission and creates complex feedback loops, but the
inflation expectations that are used are based on past values for actual inflation.
These experiences are also consistent with the stocktaking exercise conducted
by the work stream on Eurosystem modelling. In this context, a large number of
NCBs consider that the general treatment of expectations – and inflation expectations
in particular – within their semi-structural projection models (where expectations are,
in many cases, modelled as backward-looking) is not fully satisfactory. The
introduction of model-consistent expectations in large semi-structural models would
be a promising avenue for the further development of modelling, to the extent that this
preserves the computational tractability and empirical performance of projection
models. Conversely, in the case of structural models, the Eurosystem modelling work
stream set out to also adapt the main DSGE models in order to allow for simulation
modalities under alternative expectation formation mechanisms. Indeed, structural
models tend to embed explicit expectation formation mechanisms, which critically
affect their dynamic properties and the propagation of structural shocks.
Within the two workhorse euro area-wide models used in the ECB projection
process – ECB-BASE and the New Area-Wide Model II (NAWM II) (see Angelini
et al. (2019) and Cönen et al. (2018) respectively) – agents’ longer-term inflation
expectations can temporarily deviate from the central bank’s inflation objective.
In both models, longer-term inflation expectations can be influenced by persistent
fluctuations in actual inflation rates. In NAWM II, the perceived inflation objective
follows a simple adaptive learning scheme, whereby private sector agents’
expectations about longer-term inflation are partly based on past actual inflation
deviating from past perceptions of the objective:
=
+,
()
+,
with being the perceived inflation objective and ,
() being the weighted average
of consumer price inflation over the last four quarters. The parameter controls the
“learning speed”, which is related to actual inflation outcomes.78 Furthermore, the
78 The adaptive learning scheme is akin to an exponentially weighted moving average, as considered in the
empirical illustrations in the previous section. Incidentally, the parameter value of 0.058 that is assumed
in the implementation of the adaptive learning scheme is fairly close to the value of 0.05 that is used for
the EWMA model.
ECB Occasional Paper Series No 264 / September 2021
93
term is a serially uncorrelated shock to the perceived objective.79 In the estimation
of NAWM II, the data series for longer-term inflation expectations is drawn from the
SPF. In ECB-BASE, long-term inflation expectations are also informed by past
inflation, but in this model it is the difference between actual inflation and the official
target (rather than the perceived target) that governs the adjustment process.
Specifically, the process is described by the following equation:
=
+(1)[+(1)]+,
where represents long-term inflation expectations; the central bank’s annual
inflation objective is assumed to be around 1.9%; is the previous period’s
annual GDP deflator inflation; and is a residual. The baseline model specification
for ECB-BASE assumes a partial anchoring to the objective with a weight of = 0.6
and stickiness which is captured by the autoregressive coefficient = 0.75.80 In the
estimation of the ECB-BASE model, the data series for longer-term inflation
expectations is drawn from Consensus Economics.
In the context of such (semi-)structural models, observed inflation expectations
can be used to assess risks stemming from potential unanchoring. However,
there are various different ways of operationalising such an unanchoring scenario. For
instance, a pure loss of credibility could be assumed, which would be reflected in a
sudden drop in long-term expectations. This kind of shock could be calibrated using
information on the evolution and/or empirical distribution of longer-term survey or
market-based inflation expectations. For instance, a quantitative risk could be
generated by assuming that long-term inflation expectations will shift from the central
tendency to lower percentiles of the probability distribution or (in the case of the SPF)
the cross-sectional distribution. For example, Chart 47 illustrates the path of long-term
expectations as implied by ECB-BASE when considering the economic variables
projected in the September 2020 MPE (blue line). The model implies a persistent
deviation of long-term inflation expectations from the inflation target in 2021 and 2022,
with the unanchoring effect amounting to roughly 0.25 p.p. at the end of the horizon.
Chart 48 illustrates the implied sensitivity of the baseline inflation projection to shifts in
long-term inflation expectations. First, a sudden and persistent 0.1 p.p. decrease in
long-term inflation (yellow line) implies a downward revision to the baseline inflation
projection totalling 0.08 p.p. at the end of the horizon (i.e. an almost complete
pass-through). Second, an unanchoring scenario can be operationalised by
79 A hat, , denotes logarithmic deviations from the central bank’s invariant long-run inflation objective.
The equation can be rewritten as =(1)
+,
()+, which, at a first glance, looks
comparable to the process used in ECB-BASE. One important difference between the two learning
schemes is that in the ECB-BASE model inflation enters in levels, while NAWM II draws on (logarithmic)
deviations from a steady-state value. For this reason, the inflation objective does not enter the equation.
80 The baseline calibration of these parameters stems from the in-sample forecast performance over the
period 2000-17. This sample period, however, exhibits two distinct regimes: the period before 2008 (with
average inflation slightly above 2%); and the period after 2008 (with average inflation close to 1%).
Consequently, adopting a constant anchoring weight over the entire sample may not provide sufficient
time variation in trend inflation and may result in a biased forecast. Specifically, in the recent period the
forecasting performance of the model improves if we assume that long-term inflation expectations were
more unanchored than in the baseline scenario (< 0.6). For details, see Angelini et al. (2019),
pp 49-50.
ECB Occasional Paper Series No 264 / September 2021
94
considering off-model forecasts of long-term inflation expectations obtained by
conditioning on macroeconomic scenarios for the euro area (see Box 7). Instead of
assuming an arbitrary decline, such data-informed operationalisation would use the
empirical density relating to forecasts of long-term inflation expectations. The figure of
0.1 p.p., for example, would be close to the 5th percentile of the SPF five-year
forecast’s distribution (green line), suggesting that this is a fairly extreme adverse
scenario.
Chart 47
ECB-BASE-implied path for long-term inflation expectations and scenario
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: The blue line represents the ECB-BASE-implied path for long-term inflation expectations, derived recursively using the ECB
BASE model and data for the September 2020 MPE baseline from an initial point in Q1 2019 set according to Consensus Economics
inflation expectations data. The yellow line represents sensitivity to a persistent 0.1 p.p. decrease in long-term inflation expectations. The
red and green lines represent paths for long-term inflation expectations derived from a forecast distribution coming from a satellite BVAR.
1.4
1.5
1.6
1.7
1.8
1.9
2.0
01/19 04/19 07/19 10/19 01/20 04/20 07/20 10/20 01/21 04/21 07/21 10/21 01/22 04/22 07/22 10/22
ECB-BASE long-term inflation expectations (September BMPE)
0.1 p.p. drop in ECB-BASE-implied path
25th percentile of empirical densi ty of inflation expectations
5th percentile of empirical density of inflation expectations
ECB Occasional Paper Series No 264 / September 2021
95
Chart 48
Sensitivity of headline inflation to a change in long-term inflation expectations in
ECB-BASE
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: The blue line represents baseline HICP inflation consistent with the macroeconomic projection from the September 2020 BMPE.
The yellow, red and green lines represent the ECB-BASE-implied sensitivity of the baseline inflation forecast in response to alternative
paths for long-term inflation expectations.
Unanchoring risks can also be gauged by comparing revisions to conditional
forecasts of SPF inflation expectations five years ahead (SPF5y) across
projection rounds. Although it is, in principle, generally difficult to forecast
longer-term inflation expectations, as they should be largely governed by central bank
credibility, it is possible for them to also reflect some conjunctural and structural
influences emerging in the macroeconomic outlook. Chart 49 illustrates the changes
in point forecasts for the SPF5y when conditioned first on the macroeconomic
variables projected within the December 2020 BMPE and second on the previous
forecast round (i.e. the September 2020 MPE). This exercise suggests that in the
December round the interplay between the macroeconomic environment and own
dynamics led to a downward revision to the conditional future path of the SPF5y
relative to the previous projection round. However, the uncertainty surrounding
model-based forecasts of longer-term inflation expectations is fairly large, implying a
range of almost ¼ percentage point at the end of the projection horizon
(see Chart 51).
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
01/19 04/19 07/19 10/19 01/20 04/20 07/20 10/20 01/21 04/21 07/21 10/21 01/22 04/22 07/22 10/22
September 2020 BMPE
Response of HICP to 0.1 p.p. drop in long-term inflation expectations
HICP related to 25th percentile of empirical distribution
HICP related to 5th percentile of empirical density of long-term inflation expectations
ECB Occasional Paper Series No 264 / September 2021
96
Chart 49
Changes in conditional forecasts of SPF5y inflation expectations across two
consecutive forecast rounds
(annual percentage changes)
Sources: ECB (SPF) and Eurosystem staff calculations.
Note: Conditional BVAR-based forecast for the SPF5y.
Chart 50
Conditional forecasts of SPF5y inflation expectations
(annual percentage changes)
Sources: ECB (SPF) and Eurosystem staff calculations.
Note: Conditional BVAR-based forecast for the SPF5y.
A scenario regarding future developments in long-term expectations impacts
not only the future inflation path, but also the uncertainty around it. The
uncertainty surrounding the path of inflation expectations is only one source (albeit a
notable one) contributing to the overall uncertainty of the inflation forecast within the
ECB-BASE model. The illustration in Chart 51 suggests that, taking account only of
the uncertainty surrounding the conditional forecast for longer-term inflation
expectations generated using the satellite BVAR model, the predicted expectation
would move within a range of 0.25 p.p., which is reflected in a range of around
0.17 p.p. for baseline inflation projections (red lines). This range is compatible with the
scenario where the empirical density forecasts of long-term inflation expectations
1.50
1.55
1.60
1.65
1.70
1.75
1.80
1.85
1.90
06/17 06/18 06/19 06/20 06/21 06/22
Observed
December 2020 BMPE
September 2020 MPE
1.4
1.5
1.6
1.7
1.8
1.9
2.0
2.1
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020 2022
Observed
10th and 90th percentiles
Median
ECB Occasional Paper Series No 264 / September 2021
97
obtained via the BVAR model are used (yellow lines). If, instead, uncertainty is
considered from the perspective of the entire ECB-BASE model (green lines), the risk
relating to the baseline long-term inflation expectation path is estimated as being
within a range of roughly 0.65 p.p. For inflation, the uncertainty becomes considerable,
with possible future inflation paths spanning a range of more than 3.5 p.p.
(see Chart 52).
Chart 51
ECB-BASE long-term inflation expectations
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: This chart shows the confidence bands constructed around the long-term inflation expectations path implied by the baseline for
the September 2020 MPE using the ECB-BASE model. The red bands are constructed via bootstrapped draws based only on residuals
related to ECB-BASE long-term inflation expectations. The green bands are constructed via bootstrapped draws based on all model
residuals.
Chart 52
Annual HICP inflation in ECB-BASE
(annual percentage changes)
Source: Eurosystem staff calculations.
Notes: This chart shows the confidence bands constructed around the long-term inflation expectations path implied by the baseline for
the September 2020 MPE using the ECB-BASE model. The red bands are constructed via bootstrapped draws based only on residuals
related to ECB-BASE long-term inflation expectations. The green bands are constructed via bootstrapped draws based on all model
residuals.
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2
01/19 05/19 09/19 01/20 05/20 09/20 01/21 05/21 09/21 01/22 05/22 09/22
ECB-BASE long-term inflation expectations (September 2020 MPE)
99% BdE BVAR density
95% confidence interval –ECB-BASE long-term inflation expectation draws
95% confidence interval –all model draws
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
01/19 05/19 09/19 01/20 05/20 09/20 01/21 05/21 09/21 01/22 05/22 09/22
September 2020 MPE
1% BdE BVAR density
95% confidence interval –ECB-BASE long-term inflation expectation draws
95% confidence interval –all model draws
ECB Occasional Paper Series No 264 / September 2021
98
Box 6
Conditional forecasts of inflation expectations
This box proposes an empirical framework for producing forecasts of long-term inflation
expectations that are conditional on the macroeconomic scenarios for the euro area. The
BVAR model used provides both point and density forecasts of long-term inflation expectations for
different horizons which are conditional on the paths of the other variables as derived in the (B)MPE
context. The density forecasts can then be used to quantify the risks associated with unanchoring
episodes.
The conditional BVAR forecasts are derived using the approach employed by Waggoner &
Zha (1999). Forecasts are decomposed into two components: first, the unconditional forecast (in the
absence of shocks); and second, the dynamic impact of future shocks. Conditioning an endogenous
variable also implies imposing restrictions on its future innovations. These restrictions, together with
the impulse responses of the constrained variables, explain the difference between the path of the
constrained variables and the corresponding unconditional forecasts. At each iteration of the
algorithm, draws for the parameters and the conditional forecasts are generated. The collection of
draws for the conditional forecasts constitutes the posterior predictive density.
Since the target variable to be forecasted represents the expectations of agents about price
changes in the long run, we also, besides controlling for macroeconomic and financial
aspects, include not one but several measures of inflation in the model. The reasoning for this
strategy is that agents are assumed to form their expectations on the basis of a wide range of
indicators of inflation, rather than a single measure or the most representative one. Table A lists the
variables included in the proposed BVAR model.
Accordingly, the BVAR model is estimated for the euro area using the variables listed in
Table A. The data are quarterly and span the period from Q2 1999 to Q3 2020. Forecasts of
long-term inflation expectations are computed for the period from Q4 2020 to Q4 2022. In addition,
projections from the (B)MPE for all the other variables in the BVAR, spanning the period from
Q4 2020 to Q4 2022, are used as conditioning scenarios when computing the forecasts.
Owing to the Bayesian nature of the estimation procedure, empirical distributions can be
obtained for the forecasts of long-term inflation expectations. These distributions can then be
applied to other workhorse models in order to provide (i) measures of uncertainty about the future of
long-term inflation expectations, and (ii) risk assessments about the strength of unanchoring
episodes.
ECB Occasional Paper Series No 264 / September 2021
99
Table A
List of variables
Box 7
Econometric models for inflation forecasting
This box describes the set of macro-econometric models that are used in the real-time
forecast evaluation exercises to test whether observed measures of inflation expectations
help forecast accuracy. A set of six widely used econometric time series models is employed. For
each model, two versions are tested: one with inflation expectations and one without. The models can
be robustly applied to different euro area countries, different measures of inflation and different
measures of inflation expectations. Moreover, these models evaluate not only average point
forecasts, but also density forecasts. They are all estimated using Bayesian methods.
Forecasting models
Let =400 ×ln
denote the annualised quarter-on-quarter inflation rate, where is the
appropriate price index, expressed at a quarterly frequency. The models below are used to provide
forecasts of :
1. ADL models with time-varying trend inflation:
Let = denote the inflation gap, where is the inflation trend. Model 1 can be thought of
as a backward-looking Phillips curve for the inflation gap:
=++,~(0, )
where denotes a measure of the output gap.81
81 Two versions of the model are considered: one that contains an output gap and one that does not.
Real activity Real GDP
Real investment
Price dynamics Compensation per employee
HICP excluding food and energy
GDP deflator
Consumption deflator
Financial conditions Short-term interest rate
Loans (non-financial corporations)
External factors EUR/USD exchange rate
Oil price in USD
Real US GDP
Target variable SPF inflation expectations five years ahead
ECB Occasional Paper Series No 264 / September 2021
100
(a) When excluding inflation expectations:
The inflation trend is assumed to be either constant, proxied by the sample mean, or based on
exponentially weighted moving averages.
(b) When including inflation expectations:
The inflation trend is given by a measure of long-term inflation expectations.
2. ADL models with time-varying trend inflation, time-varying coefficients and stochastic
volatility:
Model 2 is a generalisation of the first model where both slope coefficients and variance in residuals
are allowed to exhibit changes over time:
()=()++, ~0, ,
where the slope coefficients and log volatilities of residuals are assumed to follow random walks.
Also, the inflation trend follows a random walk:82
=+, ~0, ,
(a) When excluding inflation expectations:
No further equations are included.
(b) When including inflation expectations:
The inflation trend is also linked to long-term inflation expectations via a measurement equation with
time-varying coefficients:
=++, ~0, ,
3. Bayesian VARs with democratic priors and stochastic volatility:
Model 3 consists of a vector autoregression where the priors are chosen in order to line up the
model’s forecasts with long-term inflation expectations:
=
(
)+
,
~(0,
)
where denotes the unconditional mean (sometimes referred to as the “steady state”).83 Two
versions of the model are considered. In the first one, only contains data on inflation. In the
second, contains information on real GDP, inflation and the short-term interest rate.
82 As with the first model, versions including and excluding the output gap are used.
83 The log volatilities of residuals are also assumed to follow random walks.
ECB Occasional Paper Series No 264 / September 2021
101
(a) When excluding inflation expectations:
The priors used to estimate are loose.
(b) When including inflation expectations:
The mean of the prior used to estimate is equal to long-term inflation expectations.
4. Bayesian VARs with time-varying trends and stochastic volatility:
This VAR model is specified for the variables as deviations from their “local” mean, which is allowed to
evolve over time as a random walk:
=
(
)+
,
~(0,
)
Similar to Model 3, two versions are considered. The first one only includes inflation in , while the
second includes real GDP growth, inflation and the short-term interest rate.
(a) When excluding inflation expectations:
No further equations are included.
(b) When including inflation expectations:
The local mean is linked to long-term expectations:
=+,~(0, )
5. Phillips curves with constant coefficients:
Model 5 is similar to Model 1, but instead of letting long-term inflation expectations influence the
inflation trend, short-term inflation expectations are incorporated as an additional regressor:
=+++
+,~(0, )
where
denotes short-term inflation expectations (one year ahead).
(a) When excluding inflation expectations:
The slope coefficient is set to zero.
(b) When including inflation expectations:
No additional modifications are made.
ECB Occasional Paper Series No 264 / September 2021
102
6. Bayesian VARs with “Minnesota” priors and stochastic volatility:
A standard Bayesian VAR model is also included in the set of models:
=+
+
,
~(0,
)
where the intercept and autoregressive coefficients are assumed to remain constant, while the log
volatilities of residuals vary over time following random walks.84
(a) When excluding inflation expectations:
No further variables are included.
(b) When including inflation expectations:
Data on either short or long-term inflation expectations are included in .
Lastly, a couple of widely used inflation forecasting models are employed as benchmarks. The first
benchmark is the unobserved components stochastic volatility model, and the second benchmark is
the random walk.
84 Similar to Models 3 and 4, two versions are considered. The first one only includes inflation in , while
the second includes real GDP growth, inflation and the short-term interest rate.
ECB Occasional Paper Series No 264 / September 2021
103
5 Conclusions and implications
This section summarises the main findings of the work in relation to the group’s
mandate and points to some implications that it may have for the ECB’s economic and
monetary analysis going forward. This includes the need for (i) more and better data
on households’ and firms’ expectations, (ii) more empirical evidence on how such
expectations influence real decisions and (iii) modelling that explicitly incorporates
observed inflation expectations and their interaction with actual inflation and real
activity using methods other than reduced-form relationships.
5.1 Conclusions
This paper starts by looking at the availability and use of observed measures of
inflation expectations. It finds that there is a dichotomy between the conceptual and
practical relevance of agents’ inflation expectations. The theory of monetary
transmission looks at agents’ inflation expectations mainly through the lens of what
they imply for expected real interest rates and for price/wage setting. This suggests a
strong focus on the inflation expectations of households and firms as the main actors
in the economy. However, corresponding data for the euro area are scarce (especially
in terms of covering different horizons), and for households in particular they also pose
questions of understanding, as they have an upward bias relative to actual inflation. In
its economic and monetary analysis, the ECB has generally focused on the inflation
expectations of forecasters and those implied by financial market prices. Unless
households and firms consider these prominently discussed expectations when
making their decisions, there is a risk of monetary policy using as benchmarks inflation
expectations that have little actual bearing on the actual expectations channel.
A recurring issue in economic and monetary analysis is the differing signals in
headline survey and market-based inflation expectations. These differences can
occur across all moments of the data. This paper finds that when the two sources are
compared properly, the differences between survey and market-based measures
largely dissipate or can be reconciled in terms of the differences in their underlying
nature. Professional forecasters provide figures for the expected level of inflation and
the physical probabilities surrounding it. Market-based measures are inflation
expectations as implied by the prices that market participants pay in hedging against
inflation risks. By definition, this implies that inflation risk premia and the greater
emphasis on tails in risk-neutral distributions are important elements to bear in mind in
comparisons. Differences in the level of inflation shrink when comparing market-based
measures adjusted for inflation risk premia with survey data in terms of the means of
their probability distributions (rather than average point estimates). Similarly, the
presence of risk premia cautions against comparing the magnitude of percentiles of
risk-neutral distributions in market data with those of physical probabilities in survey
data, but it does not preclude comparisons in terms of changes in probabilities.
However, given that risk premia have their own information content, the possibility of
reconciling market and survey-based data still suggests that analysts and
ECB Occasional Paper Series No 264 / September 2021
104
policymakers should look at both sources in parallel. This is also suggested by the
finding that there is no robust evidence of causal or lead/lag relationships between
market and survey-based expectations.
Inflation expectations are both an important target and a tracking device in
monetary policy (re)actions and guidance. Understanding what drives inflation
expectations is therefore important. This paper confirms that the drivers of inflation
expectations vary across the different horizons, with shorter-term expectations
responding to diverse macroeconomic shocks and longer-term expectations being
determined by the degree of confidence in the central bank’s inflation aims. One
finding implied by the evolution of the term structure of the inflation expectations curve
(using Consensus Economics data up to ten years ahead) is that the horizon over
which shocks are expected to fade out has lengthened at the same time as
data-implied steady-state inflation expectations have softened. This leaves open the
possibility that movements in longer-term inflation expectations do not necessarily
reflect forecasters’ beliefs about the ability or willingness of the central bank to achieve
its aim, but also recognises that adjustment mechanisms have (unavoidably) become
more protracted following the shocks entailed by the financial and sovereign debt
crises.
Inflation expectations can potentially be influenced by a wide range of different
factors and shocks. This paper looks selectively at oil prices and monetary policy as
two factors that often tend to be discussed in the context of shifting inflation
expectations. It finds that oil price shocks – as they should – tend to influence only
shorter-term expectations. As with other analyses, the finding here is that the
magnitude of the impact depends on the underlying nature of the shock and that, in the
case of oil prices, the impact is strongest if it reflects global activity, rather than oil
market-specific shocks. There is some co-movement between oil prices and
longer-term market-based measures, but this is largely explained by the impact that oil
price changes have on the inflation risk premium in market-based measures. The role
of monetary policy shocks is of key interest, as they speak not only to the task of
steering shorter-term inflation expectations towards the inflation anchor, but also, in a
situation where longer-term inflation expectations have declined, to the task of
providing a re-anchoring channel. Market-based measures have responded to pure
monetary policy shocks since the PSPP was established, but the impact has mainly
been on spot rates, rather than longer-term forward expectations. The analysis
suggests that when assessing a monetary policy shock, it is important to distinguish
between such pure monetary policy shocks and information shocks that are triggered
by the central bank signalling an unexpected change in its macroeconomic outlook. An
exercise based on Consensus Economics data suggests that after controlling for news
and data-related surprises, Eurosystem forecasts have still had some impact on
changes in forecasters’ expectations.
Assessments of the risk of longer-term inflation expectations becoming
unanchored are a key input for monetary policy deliberations. The practical
challenge here is that (un)anchoring is a complex and multi-faceted concept and
cannot be captured in clear-cut binary states. Policymakers can look at different
metrics, which will not necessarily provide consistent signals at a given point in time.
ECB Occasional Paper Series No 264 / September 2021
105
Mostly, (un)anchoring is assessed in terms of the level of long-term expectations, the
uncertainty surrounding it, and its responsiveness to short-term inflation
(expectations). The level of longer-term SPF inflation expectations has displayed
some variation over time, but it has generally remained within the 1.7-2.0% range that
panellists regard as consistent with the price stability aim. More recently, however, it
has fallen to the bottom of this range. Longer-term market-based measures seem to
have become more clearly unanchored, given their low levels also when adjusting for
inflation risk premia. The implied suggestion that, on balance, risks of unanchoring
have increased is corroborated by metrics based on higher moments, such as the
increase in (cross-sectional) uncertainty, the fact that constructed indicators capturing
the balance of risks have seen declines since the financial crisis, and the fact that the
probability associated with low inflation and deflation has increased. Furthermore,
there is some evidence that longer-term survey-based measures have been
responsive to shorter-term developments – although this result is not always
statistically significant. This paper finds that the different metrics are complementary to
each other and need to be assessed in conjunction, as – for instance – it might well be
that responsiveness of long-term expectations to short-term expectations is found
before (but not after) a shift in the level of long-term expectations. An anchoring “heat
map” can provide a general visual cross-check of the different metrics, but using it
requires awareness that the colouring is the result of applying specific benchmarking
criteria to the individual anchoring metrics.
International comparisons suggest that the risk of longer-term inflation
expectations becoming unanchored in the euro area is not part of a global
phenomenon. Developments in longer-term inflation expectations have been
heterogeneous across advanced economies and primarily reflect idiosyncratic factors.
Evidence of widespread unanchoring risks remains limited. Following a secular
decline during the 1990s, levels of longer-term inflation expectations stabilised as of
2000. With the exception of Japan, there is also limited evidence of a statistically
significant pass-through from short-term inflation expectations to longer-term
expectations. This finding also holds when using market-based inflation expectations
and allowing for time variation in the pass-through coefficient. The finding that there is
no generalised unanchoring risk in advanced economies is consistent with the finding
that global factors appear to play only a limited role in driving longer-term inflation
expectations.
Observed measures of inflation expectations can be useful benchmarks and
cross-checks for Eurosystem projections. Survey and market-based measures
both have an average forecasting performance for actual inflation which is similar to
that of Eurosystem projections. They are hence credible benchmarks in forecast
comparison tables, as they are regularly presented to policymakers. While the central
tendencies of observed measures of inflation expectations can thus inform
Eurosystem projections, the information they provide on the uncertainty surrounding
these baseline views needs to be assessed with more caution. For instance, PIT
analysis of the probability distributions in the SPF confirms that forecasters tend (with
the exception of very short horizons) to underestimate the uncertainty surrounding
their point forecasts. The different degrees of usefulness of observed inflation
expectations for benchmarking point forecasts and uncertainties is confirmed by
ECB Occasional Paper Series No 264 / September 2021
106
pooling the forward-looking information in the SPF with that generated by a number of
Bayesian models. Tilting the model predictions towards the SPF improves point
forecasts but worsens density forecasts. Thus, injecting too much information from
survey forecasts can be counterproductive.
Observed measures of inflation expectations can usefully be incorporated in
time series models used for forecasting. This can take different forms, such as
detrending actual inflation, linking the unobserved inflation trend in a Phillips curve
model to long-term inflation expectations, informing the conditional mean of inflation in
a VAR model or disciplining the long-run priors for inflation. This paper finds that
informing time-series models with survey expectations as regressors provides some –
albeit not major – forecast gains. The gains cannot consistently be assigned to either
short or long-term expectations, and they can vary over time. This argues in favour of
employing a battery of model specifications that makes use of the full expectations
curve. Comparing forecast accuracy gains across different sources of inflation
expectations, it appears to be hard to improve on SPF expectations. Using
expectation measures as contemporaneous regressors and using the models for
conditional forecasting (as in the case, for example, of the thick Phillips curve
modelling tool) requires an extension of the expectation series over the projection
horizon. Such extensions can take different forms – e.g. involving a random walk,
autoregressive processes or an assumed convergence with the inflation aim. Point
forecasts for inflation vary quite markedly across different extension methods and
forecasters should thus be aware of the judgement they are making in choosing a
particular method.
Unanchoring has been a key component of risk analysis in macroeconomic
projections. Making use of observed measures of inflation expectations, such
analysis can take two different forms. First, macro models can calibrate their implicit
inflation anchor using developments in empirically observed longer-term inflation
expectations or use changes in longer-term inflation expectations to calibrate a shock
to an otherwise constant anchor. This allows us to quantify the implications for actual
inflation (and other variables) in the model. Second, observed long-term inflation
expectations can be made an endogenous variable in a satellite model that uses the
main macroeconomic variables featuring in the workhorse forecasting models. Using
the satellite model to produce conditional forecasts of longer-term inflation
expectations and comparing the evolution of these forecasts across projection
vintages can then point to unanchoring risks.
5.2 Implications
The work of the EGIE largely supports the current use of expectations in the
ECB’s economic and monetary analysis. More specifically, it confirms the dual
information content of observed measures of inflation expectations. Short-term
expectations provide information on how agents regard shocks as shaping the
inflation forward curve. This provides input and a cross-check for the ESCB’s own
projections. Longer-term expectations provide information on agents’ belief in the
central bank’s willingness and ability to achieve its inflation aim. This may inform and
ECB Occasional Paper Series No 264 / September 2021
107
provide input for policy calibration and forward guidance. The lengthening of the
horizon over which agents typically believe that (unavoidable) shocks will fade out and
monetary policy credibility will kick in has important implications for what we can learn
about unanchoring risks particularly if such lengthening is accompanied by a
simultaneous lowering of longer-term inflation expectations.
Monetary policy can use inflation expectations as a checking device, but also
as an instrument. The evidence suggests that monetary policy actions influence
private sector inflation expectations. However, using observed measures of inflation
expectations as a checking device for implied effectiveness would be greatly
enhanced if more use could be made of data on households’ and firms’ expectations,
rather than relying mainly on the expectations of informed professional forecasters
and financial market participants. Having better information on different agents’
expectations would also facilitate a better understanding of whether and how these
expectations shape real economic decisions and thus also a better evaluation of the
important real interest rate channel. Central banks influence private sector
expectations not only through their monetary policy actions, but also through the
public information they provide via their projections. Thus, those projections play an
important role in the provision of consistent forward guidance.
More consistent inclusion of observed inflation expectation measures in
forecasting requires additional modelling efforts. This relates primarily to the
need, in the main macro models, to account in a consistent manner for possible
movements in the inflation anchor and tie this to observed longer-term inflation
expectations. It also relates to the need to allow for possible interaction between
longer-term (steady-state) inflation expectations and shorter-term developments in
actual inflation and inflation expectations. In this respect, it is likely that a re-anchoring
channel will not operate via one-off shifts in longer-term inflation expectations, but
through long-term expectations gradually reacting to observed trends in actual
inflation.
ECB Occasional Paper Series No 264 / September 2021
108
Bibliography
Aastveit, K. A., Bjørnland, H. C., & Cross, J. C. (2020). Inflation expectations and the
pass-through of oil prices. Centre for Applied Macro- and Petroleum economics
(CAMP) Working Papers, 2020 (3).
Ahn, H. J., & Fulton, C. (2021). Index of Common Inflation Expectations. FEDS Notes.
March.
Álvarez, L. J., & Correa-López, M. (2020). Inflation expectations in euro area Phillips
curves. Economics Letters, 195. October.
Andrade, P., Breckenfelder, J., De Fiore, F., Karadi, P., & Tristani, O. (2016). The
ECB's asset purchase programme: an early assessment. ECB Working Paper Series,
1956.
Andrade, P., Gautier, E., & Mengus, E. (2020). What Matters in Households’ Inflation
Expectations? Banque de France Working Paper, 770. June.
Ang, A., Bekaert, G., & Wei, M. (2007). Do macro variables, asset markets or surveys
forecast inflation better? Journal of Monetary Economics, 54, 1163-1212.
Angeletos, G.-M., Huo, Z., & Sastry, K. A. (2020). Imperfect Macroeconomic
Expectations: Evidence and Theory. NBER Working Paper, 27308. June.
Angelini, G., Bacchiocchi, E., Caggiano, G., & Fanelli, L. (2019). Uncertainty across
volatility regimes. Journal of Applied Econometrics, 34(3), 437-455. April.
Apokoritis, N., Galati, G., Moessner, R., & Teppa, F. (2019). Inflation expectations
anchoring: new insights from micro evidence of a survey at high-frequency and of
distributions. BIS Working Papers, 809.
Arioli, R., Bates, C., Dieden, H., Duca, I., Friz, R., Gayer, C., Kenny, G., Meyler, A., &
Pavlova, I. (2016). EU consumers’ quantitative inflation perceptions and expectations:
an evaluation. DG-ECFIN European Economy Discussion Papers 38. November.
Autrup, A. S., & Grothe, M. (2014). Economic surprises and inflation expectations: Has
anchoring of expectations survived the crisis? ECB Working Paper Series, 1671.
Bachmann, R., Berg, T. O., & Sims, E. R. (2015). Inflation expectations and readiness
to spend: Cross-sectional evidence. American Economic Journal: Economic Policy,
7(1).
Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple
structural changes. Econometrica, 66(1), 47-78.
Bai, J., & Perron, P. (2003). Computation and analysis of multiple structural change
models. Journal of Applied Econometrics, 18(1), 1-22.
ECB Occasional Paper Series No 264 / September 2021
109
Bańbura, M., & Bobeica, E. (2020). Does the Phillips curve help to forecast euro area
inflation? ECB Working Paper Series No. 2471. September.
Bańbura, M., Brenna, F., Paredes, J., & Ravazzolo, F. (2021a). Combining Bayesian
VARs with survey density forecasts: does it pay off? ECB Working Paper Series,
2543.
Bańbura, M., Leiva-León, D., & Menz, J. O. (2021b). Do inflation expectations improve
model-based inflation forecasts? ECB Working Paper Series, forthcoming.
Banerjee, R., & Mehrotra, A. (2018). Deflation expectations. BIS Working Papers, 699.
Bartiloro, L., Bottone, M., & Rosolia, A. (2019). The Heterogeneity of the Inflation
Expectations of Italian Firms along the Business Cycle. International Journal of
Central Banking, 15(5), 175-205. December.
Bauer, M. D., & McCarthy, E. (2015). Can we rely on market-based inflation forecasts?
Federal Reserve Bank of San Francisco Economic Letter, 30.
Baumeister, C., & Hamilton, J. D. (2019). Structural Interpretation of Vector
Autoregressions with Incomplete Identification: Revisiting the Role of Oil Supply and
Demand Shocks. American Economic Review, 109(5), 1873-1910.
Beechey, M. J., Johannsen, B. K., & Levin, A. T. (2011). Are Long-Run Inflation
Expectations Anchored More Firmly in the Euro Area than in the United States?
American Economic Journal: Macroeconomics, 3(2), 104-129.
Bloom, N., Bunn, P., Mizen, P., Smietanka, P., Thwaites, G., & Young, G. (2017).
Tracking the views of British businesses: evidence from the Decision Maker Panel.
Bank of England Quarterly Bulletin, 2017(Q2).
Bobeica, E., & Sokol, A. (2019). Drivers of underlying inflation in the euro area over
time: a Phillips curve perspective. Economic Bulletin, 2019 (4).
Böninghausen, B., Kidd, G., & de Vincent-Humphreys, R. (2018). Interpreting recent
developments in market based indicators of longer term inflation expectations.
Economic Bulletin, 2018(6).
Bottone, M., & Rosolia, A. (2019). Monetary policy, firms' inflation expectations and
prices: causal evidence from firm-level data. Temi di Discussione (Banca d’Italia
Working Paper), 1218.
Bowles, C., Friz, R., Genre, V., Kenny, G., Meyler, A., & Rautanen, T. (2010). An
evaluation of the growth and unemployment forecasts in the ECB Survey of
Professional Forecasters. Journal of Business Cycle Measurement and Analysis, 2,
1-28.
Bruine de Bruin, W., Manski, C. F., Topa, G., & van der Klaauw, W. (2011). Measuring
consumer uncertainty about future inflation. Journal of Applied Economics, 26,
454-478.
ECB Occasional Paper Series No 264 / September 2021
110
Bruine de Bruin, W., van der Klaauw, W., Downs, J. S., Fischhoff, B., Topa, G., &
Armantier, O. (2010). Expectations of Inflation: The Role of Demographic Variables,
Expectation Formation, and Financial Literacy. Journal of Consumer Affairs, 44,
381-402.
Bryan, M., & Venkatu, G. (2001). The Demographics of Opinion Surveys. Federal
Reserve Bank of Cleveland.
Byrne, D., & Zekaite, Z. (2019). Euro area longer-term inflation expectations revisited.
Central Bank of Ireland Economic Letter, 2019(12).
Camba-Méndez, G., & Werner, T. (2017). The inflation risk premium in the
post-Lehman period. ECB Working Paper Series, 2033.
Candia, B., Coibion, O., & Gorodnichenko, Y. (2020). Communication and the Beliefs
of Economic Agents. NBER Working Papers, 27800.
Canova, F. (2005). The transmission of US shocks to Latin America. Journal of
Applied Econometrics, 20, 229-251.
Canova, F., & Gambetti, L. (2010). Do Expectations Matter? The Great Moderation
Revisited. American Economic Journal: Macroeconomics, 2(3), 183-205.
Carvalho, C., Euseppi, S., Moench, E., & Preston, B. (2019). Anchored inflation
expectations. CEPR Discussion Paper, 13900.
Castelnuovo, E., Rodriguez-Palenzuela, D., & Nicoletti Altimari, S. (2003). Definition
of price stability, range and point inflation targets: the anchoring of long-term inflation
expectations. ECB Working Paper Series, 273.
Cecchetti, S., Fantino, D., Notarpietro, A., Riggi, M., Tagliabracci, A., Tiseno, A., &
Zizza, R. (2021). Inflation expectations in the euro area: indicators, analyses and
models used at the Banca d’Italia. Questioni di Economia e Finanza (Banca d’Italia
Occasional Paper), 612.
Chmielewski, T., Kocięcki, A., Łyziak, T., Przystupa, J., Stanisławska, E., Walerych,
M., & Wróbel, E. (2020). Monetary transmission mechanism in Poland. What do we
know in 2019? NBP Working Papers, 329.
Christelis, D., Geogarakos, D., & Jappelli, T. v. (2020). Trust in the Central Bank and
Inflation Expectations. International Journal of Central Banking, 65, 1-37.
Christensen, J., & Gillan, J. (2019, December). Does quantitative easing affect market
liquidity? Federal Reserve Bank of San Francisco Working Paper(26).
Ciccarelli, M. and Osbat, C. (2017), Low inflation in the euro area: Causes and
consequences, ECB Occasional Paper Series, No 181, January.
Ciccarelli, M., Garcia, J. A., & Montes-Galdón, C. (2017). Unconventional monetary
policy and the anchoring of inflation expectations. ECB Working Paper Series, 1995.
ECB Occasional Paper Series No 264 / September 2021
111
Clark, T. E., & Davig, T. A. (2008). An empirical assessment of the relationships
among inflation and short- and long-term expectations. Federal Reserve Bank of
Kansas City Research Working Paper, 2008(5).
Cönen, G., Karadi, P., Schmidt, S., & Warne, A. (2018). The New Area-Wide Model II:
an extended version of the ECB’s micro-founded model for forecasting and policy
analysis with a financial sector. ECB Working Paper Series. November.
Coibion, O., Gorodnichenko, Y., & Kamdar, R. (2018). The Formation of Expectations,
Inflation, and the Phillips Curve. Journal of Economic Literature, 56(4), 1447-1491.
Coibion, O., Gorodnichenko, Y., & Ropele, T. (2020). Inflation Expectations and Firm
Decisions: New Causal Evidence. The Quarterly Journal of Economics, 135(1),
165-219.
Conflitti, C., & Cristadoro, R. (2018). Oil prices and inflation expectations. Banca
d’Italia Occasional Papers, 423.
Conflitti, C., & Zizza, R. (2020). What’s behind firms’ inflation forecasts? Empirical
Economics, forthcoming.
Corsello, F., Neri, S., & Tagliabracci, A. (2021, October). Anchored or de-anchored?
That is the question. European Journal of Political Economy, 516.
Crump, R., Eusepi, S., Tambalotti, A., & Topa, G. (2015). Subjective Intertemporal
Substitution. Federal Reserve Bank of New York Staff Reports, 734.
Cunningham, R., Desroches, B., & Santor, E. (2010). Inflation Expectations and the
Conduct of Monetary Policy: A Review of Recent Evidence and Experience. Bank of
Canada Review, 2010 (Spring).
D’Agostino, A., McQuinn, K., & Whelan, K. (2012). Are some forecasters really better
than others? Journal of Money, Credit and Banking, 44(4), 715-732.
D’Acunto, F., Malmendier, U., Ospina, J., & Weber, M. (2021). Exposure to Grocery
Prices and Inflation Expectations. Journal of Political Economy, forthcoming.
Darracq Paries, M., & Zimic, S. (2021). Macroeconomic implications of heterogenous
long-term inflation expectations: illustrative simulations through the ECB-BASE.
Economic Bulletin, 3, Box 4.
Darvas, Z., & Hüttl, P. (2016). Oil prices and inflation expectations. Bruegel. January.
de Vincent-Humphreys, R., Dimitrova, I., Falck, E., Henkel, L., & Meyler, A. (2019).
Twenty years of the ECB Survey of Professional Forecasters. Economic Bulletin,
2019(1).
Di Iorio, A., & Fanari, M. (2020). The break-even inflation rates: the Italian case. Banca
d’Italia Occasional Papers, 578.
ECB Occasional Paper Series No 264 / September 2021
112
Diamond, J., Watanabe, K., & Watanabe, T. (2020). The Formation of Consumer
Inflation Expectations: New Evidence from Japan’s Inflation Experience. International
Economic Review, 61, 241-281.
Diebold, F. X. (2013). Comparing Predictive Accuracy, Twenty Years Later: A
Personal Perspective on the Use and Abuse of Diebold-Mariano Tests. Department of
Economics, University of Pennsylvania.
Diebold, F. X., & Mariano, R. S. (1995). Comparing Predictive Accuracy. Journal of
Business and Economic Statistics, 13(3), 253-263.
Diebold, F., Gunther, T., & Tay, A. (1998). Evaluating Density Forecasts, with
Applications to Financial Risk Management. International Economic Review, 39,
863-883.
Diegel, M., & Nautz, D. (2020). The role of long-term inflation expectations for the
transmission of monetary policy shocks. Free University of Berlin Discussion Paper,
2020 (19).
Döhring, B., & Mordonu, A. (2007). What drives inflation perceptions? A dynamic
panel data analysis. European Economy - Economic Papers 2008 - 2015, 284.
Domit, S., Jackson, C., & Roberts-Sklar, M. (2015). Do inflation expectations currently
pose a risk to inflation? Bank of England Quarterly Bulletin.
Dovern, J., & Kenny, G. (2020). Anchoring Inflation Expectations in Unconventional
Times: Micro Evidence for the Euro Area. International Journal of Central Banking.
Driessen, J., Nijman, T., & Simon, Z. (2017). The missing piece of the puzzle: liquidity
premiums in inflation indexed markets. SAFE Working Paper 183.
Duca, I. A., Kenny, G., & Reuter, A. (2018). Inflation expectations, consumption and
the lower bound: micro evidence from a large euro area survey. ECB Working Paper
Series, 2196.
ECB (2020). The ECB Survey of Professional Forecasters - Fourth quarter of 2020,
October.
ECB (2021). The ECB’s Consumer Expectations Survey: A First Evaluation, ECB
Occasional Paper Series Forthcoming.
Elliott, D., Jackson, C., Raczko, M., & Roberts-Sklar, M. (2015). Does oil drive
financial market measures of inflation expectations? Bank of England. October.
Eser, F., Lemke, W., Nyholm, K., Radde, S., & Vladu, A. (2019). Tracing the impact of
the ECB’s asset purchase programme on the yield curve. ECB Working Paper, 2293.
Faust, J., & Wright, J. H. (2013). Forecasting inflation. In Elliott & Timmermann (Eds.),
Handbook of Economic Forecasting. Elsevier.
Flecksenstein, M., Longstaff, F., & Lustig, H. (2014). The TIPS-treasury bond puzzle.
Journal of Finance, 69(5), 2151-2197.
ECB Occasional Paper Series No 264 / September 2021
113
Fuhrer, J. C. (2012). Real expectations: replacing rational expectations with survey
expectations in dynamic macro models. Federal Reserve Bank of Boston Working
Papers, 12-19.
Fujiwara, I. (2005). Is the central bank’s publication of economic forecasts influential?
Economics Letters, 89(3), 255-261.
Garcia, J. A., & Werner, S. (2018). Inflation News and Euro Area Inflation
Expectations. IMF Working Paper, 18/167.
Georgarakos, D. & Kenny, G. (2020). Euro Area Household Sector Adjustment during
the Global Pandemic: Insights from a New Consumer Survey. European Central Bank
Mimeo.
Genre, V., Kenny, G., Meyler, A., & Timmermann, A. (2013). Combining expert
forecasts: Can anything beat the simple average? International Journal of
Forecasting, 29(1), 108-121.
Gil-Alana, L., Moreno, A., & Perez de Gracia, F. (2012). Exploring survey-based
inflation forecasts. Journal of Forecasting, 31(6), 524-539.
Gosselin, M.-A., & Kahn, M. (2015). A Survey of Consumer Expectations for Canada.
Bank of Canada Review.
Grishchenko, O., Mouabbi, S., & Renne, J. P. (2019). Measuring Inflation Anchoring
and Uncertainty: A U.S. and Euro Area Comparison. Journal of Money, Credit and
Banking, 51(5), 1053-1096.
Grosse-Steffen, C., Fratzscher, M., & Rieth, M. (2020). Inflation Targeting as a Shock
Absorber. Journal of International Economics, 123, 1-16.
Gürkaynak, R. S., Levin, A., & Swanson, E. (2010). Does inflation targeting anchor
long-run inflation expectations? Evidence from the US, UK, and Sweden. Journal of
the European Economic Association, 8(6), 1208-1242.
Hanoma, A., & Nautz, D. (2019). The information content of market-based measures
for the long-term inflation expectations of professionals: evidence from a midas
analysis. Applied Economics, 51(51), 5623-5636.
Hartmann, P., & Smets, F. (2018). The first twenty years of the European Central
Bank: Monetary policy. ECB Working Paper, 2219.
Hattori, M., Kong, S., Packer, F., & Sekine, T. (2016). The effects of a central bank's
inflation forecasts on private sector forecasts: Recent evidence from Japan. BIS
Working Papers, 585.
Haubrich, J., & Pennacchi, G. R. (2012). Inflation expectations, real rates and risk
premiums: Evidence from inflation swaps. Review of Financial Studies, 25(5),
1588-1629.
ECB Occasional Paper Series No 264 / September 2021
114
Henckel, T., Menzies, G. D., Moffatt, P., & Zizzo, D. J. (2019). Three dimensions of
central bank credibility and inferential expectations: The Euro zone. Journal of
Macroeconomics, 60, 294-308.
Hördahl, P., & Tristani, O. (2015). Inflation Risk Premia in the Euro Area and the
United States. International Journal of Central Banking, 10(3), 1-47.
Hubert, P. (2015, May 28). Do Central Bank Forecasts Influence Private Agents?
Forecasting Performance versus Signals. Journal of Money, Credit and Banking,
47(4), 771-789.
Jarociński, M., & Lenza, M. (2018, September). An Inflation‐Predicting Measure of the
Output Gap in the Euro Area. Journal of Money, Credit and Banking, 50(6),
1189-1224.
Jorda, O. (2005). Estimation and Inference of Impulse Responses by Local
Projections. American Economic Review, 95(1), 161-182.
Joslin, S., Singleton, K. J., & Zhu, H. (2011). A New Perspective on Gaussian Dynamic
Term Structure Models. Review of Financial Studies, 24(3), 926-970.
Kerßenfischer, M. (2019). Information effects of euro area monetary policy: New
evidence from high-frequency futures data. Bundesbank Discussion Paper, 2019(7).
Kilian, L. (2009). Not All Oil Price Shocks Are Alike: Disentangling Demand and
Supply Shocks in the Crude Oil Market. American Economic Review, 99(3),
1053-1069.
Kilian, L., & Murphy, D. (2014). The Role of Inventories and Speculative Trading in the
Global Market for Crude Oil. Journal of Applied Econometrics.
Kulikov, D., & Reigl, N. (2019). Inflation Expectations in Phillips Curve Models for the
Euro Area. Eesti Pank Working Paper, 8.
Kumar, S., Coibion, O., Afrouzi, H., & Gorodnichenko, Y. (2015). Inflation targeting
does not anchor inflation expectations: Evidence from firms in New Zealand. NBER
Working Papers, 21814.
Kupfer, A. (2018). Estimating inflation risk premia using inflation-linked bonds: A
review. Journal of Economic Surveys, 32(1), 1326-1354.
Lamla, M. J., & Vinogradov, D. V. (2019). Central bank announcements: Big news for
little people? Journal of Monetary Economics, 108(C), 21-38.
Leiva-León, D., Ortega, E., & Martinez-Martin, J. (2021). Exchange rate shocks and
inflation co-movement in the euro area. International Journal of Central Banking.
Lewis, M., McDermott, J., & Richardson, A. (2016). Inflation expectations and the
conduct of monetary policy in New Zealand. Reserve Bank of New Zealand Bulletin,
79(4), 1-12. March.
ECB Occasional Paper Series No 264 / September 2021
115
Lusardi, A. (2008). Financial Literacy: An Essential Tool for Informed Consumer
Choice? NBER Working Paper, 14084. June.
Łyziak, T. (2013). Formation of Inflation Expectations by Different Economic Agents.
Eastern European Economics, 51(6), 5-33.
Łyziak, T. (2016a). The Phillips Curve is still alive. Interpretation of low inflation
episode in Poland. Studia Ekonomiczne, 2, 161-198.
Łyziak, T. (2016b). Survey measures of inflation expectations in Poland: Are they
relevant from the macroeconomic perspective? Baltic Journal of Economics, 16(1),
33-52.
Łyziak, T., & Paloviita, M. (2017). Anchoring of inflation expectations in the euro area:
Recent evidence based on survey data. European Journal of Political Economy, 46,
52-73.
Łyziak, T., & Paloviita, M. (2018). On the formation of inflation expectations in
turbulent times: The case of the euro area. Economic Modelling, 72, 132-139.
Manski, C. F. (2004). Measuring Expectations. Econometrica, 72(5), 1329-1376.
Mavroeidis, S., Plagborg-Møller, M., & Stock, J. H. (2014). Empirical Evidence on
Inflation Expectations in the New Keynesian Phillips Curve. Journal of Economic
Literature, 52(1), 124-188.
Mehrotra, A., & Yetman, J. (2018). Decaying Expectations: What Inflation Forecasts
Tell Us about the Anchoring of Inflation Expectations. International Journal of Central
Banking, 2018 (Q4).
Mertens, K., Lewis, D. J., & Makridis, C. (2020). Do Monetary Policy Announcements
Shift Household Expectations? CEPR Discussion Paper, DP14360.
Meyler, A. (2020). Forecast performance in the ECB SPF: ability or chance? ECB
Working Paper Series, 2371. February.
Meyler, A., & Grothe, M. (2015). Inflation forecasts: Are market-based and
survey-based measures informative? ECB Working Paper Series, 1865.
Meyler, A., & Reiche, L. (2021). Making sense of consumers’ inflation perceptions and
expectations - the role of (un)certainty. Economic Bulletin, 2021(2).
Meyler, A., & Rubene, I. (2009). Results of a special questionnaire for participants in
the ECB Survey of Professional Forecasters. MPRA Paper, 20751.
Moore, A. (2016). Measures of Inflation Expectations in Australia. Reserve Bank of
Australia Bulletin. December.
Moretti, L., Onorante, L., & Saber, S. Z. (2019). Phillips curves in the euro area. ECB
Working Paper Series, 2295.
ECB Occasional Paper Series No 264 / September 2021
116
Mumtaz, H., & Surico, P. (2009). The Transmission of International Shocks: A
Factor-Augmented VAR Approach. Journal of Money, Credit and Banking, 41(1),
71-100.
Naggert, K., Rich, R., & Tracy, J. (2021). Flexible Average Inflation Targeting and
Inflation Expectations: A Look at the Reaction by Professional Forecasters. Federal
Reserve Bank of Cleveland Economic Commentary 2021 (09).
Nakamura, E. (2021). Is the Phillips Curve Getting Flatter? Princeton Bendheim
Center for Finance Presentation.
Nakata, T. (2020). Raising the Inflation Target: Lessons from Japan. FEDS Notes.
Natoli, F., & Sigalotti, L. (2018). Tail Co-movement in Inflation Expectations as an
Indicator of Anchoring. International Journal of Central Banking, 14(1), 35-71.
Nishino, K., Yamamoto, H., Kitahara, J., & Nagahata, T. (2016). Developments in
inflation expectations over the three years since the introduction of QE. Bank of Japan
Review. October.
Nunes, R. (2010). Inflation dynamics: The role of expectations. Journal of Money,
Credit and Banking, 42, 1161-1172.
Orphanides, A. (2015). Fear of Liftoff: Uncertainty, Rules, and Discretion in Monetary
Policy Normalization. Federal Reserve Bank of St. Louis Review, 97(3).
Ortega, E., & Osbat, C. (2020). Exchange rate pass-through in the euro area and EU
countries. ECB Occasional Paper 241.
Paloviita, M., Haavio, M., Jalasjoki, P., & Kilponen, J. (2021). What does "below, but
close to two percent" mean? Assessing the ECB’s reaction function with real time
data. International Journal of Central Banking, forthcoming.
Pedersen, T. Q. (2015). Predictable Return Distributions. Journal of Forecasting,
34(2), 114-132.
Reis, R. (2020). The people versus the markets: A parsimonious model of inflation
expectations. London School of Economics Mimeo. December
Romer, C. D., & Romer, D. H. (2000). Federal Reserve Information and the Behavior
of Interest Rates. American Economic Review, 90(3), 429-457.
Rondinelli, C., & Zizza, R. (2020). Spend Today or Spend Tomorrow? The Role of
Inflation Expectations in Consumer Behaviour. Temi di Discussione (Banca d’Italia
Working Paper), 1276.
Rostagno, M., Altavilla, C., Carboni, G., Lemke, W., Motto, R., Guilhem, A., &
Yiangou, J. (2019). A tale of two decades: The ECB’s monetary policy at 20. ECB
Working Paper, 2346.
Schnabel, I. (2020, February 27). How long is the medium term? Monetary policy in a
low inflation environment. Barclays International Monetary Policy Forum.
ECB Occasional Paper Series No 264 / September 2021
117
Speck, C. (2017, January). Inflation anchoring in the euro area. ECB Working Paper,
1998.
Speck, C. (2019). Break-even inflation rates for the euro area. mimeo.
Speck, C. (2020). Monetary Policy, Market Imperfections and Inflation Expectations.
Deutsche Bundesbank Internal Note.
Speck, C. (2021). Market Microstructure of Inflation-Indexed Bonds and Technical
Biases in Market-Based Inflation Indicators. Deutsche Bundesbank Internal Note.
Stanisławska, E., Łyziak, T., & Paloviita, M. (2019). Assessing reliability of aggregated
inflation views in the European Commission Consumer Survey. NBP Working Papers,
313.
Stevens, A., & Wauters, J. (2018). Is euro area lowflation here to stay? Insights from a
time-varying parameter model with survey data. Nationale Bank van België/Banque
Nationale de Belgique Working Paper, 355.
Szafranek, K. (2017). Flattening of the New Keynesian Phillips curve: evidence for an
emerging, small economy. Economic Modelling, 63, 334-348.
Timmermann, A. (2006). Forecast Combinations, Chapter 4 in Handbook of Economic
Forecasting, Editor(s): Elliott, G., Granger, C.W.J. & Timmermann, A., Elsevier,
Volume 1, Pages 135-196.
Trehan, B. (2015). Survey measures of expected inflation and the inflation process.
Journal of Money, Credit and Banking, 47(1), 207-222.
Vellekoop, N., & Wiederholt, M. (2019). Inflation Expectations and Choices of
Households. SAFE Working Paper, 250.
Venditti, F., & Veronese, G. (2020). Global financial markets and oil price shocks in
real time. ECB Working Paper, 2472.
Waggoner, D. F., & Zha, T. (1999). Conditional Forecasts In Dynamic Multivariate
Models. The Review of Economics and Statistics, 81(4), 639-651.
Yetman, J. (2020). Pass-through from short-term to long-term inflation expectations,
and the anchoring of inflation expectations. BIS Working Papers, 895.
ECB Occasional Paper Series No 264 / September 2021
118
Annex A – Households’ and firms’
inflation expectations
In practice, it is the expectations of professional forecasters and those of
market participants that are implied in inflation-linked products that play the
key role in monetary policy discussions – both at the ECB and in other central
banks. This largely reflects the availability and easy accessibility of these
expectations. In contrast, the use of households’ or firms’ inflation expectations is not
as common in policy discussions, despite their importance on a conceptual level. As
regards firms, businesses along the production chain set what will ultimately end up as
the consumer prices that monetary policy aims to control. Firms’ inflation expectations
can be expected to guide their pricing strategies and thus the forward component of
the Phillips curve. Besides, they have the potential to be relevant for nearly all firms’
economic choices – borrowing, saving, hiring and investing – to the extent that they
involve an intertemporal dimension whereby contemporaneous decisions hinge on
expected future outcomes. Similarly, there are at least two conceptual reasons why
monetary policy should, in principle, be looking at households’ inflation expectations.
The one relating to price setting is less strong for households, as the prices they set
are wages and the corresponding negotiations are typically conducted by unions
(which may form their inflation expectations in a different way from households).
Meanwhile, the one relating to inflation expectations’ relevance (via real interest rates)
for nearly all economic choices – borrowing, saving, consuming and investing – in the
sector as a whole applies also to households.
The lack of reliable and extensive data on households’ and firms’ expectations
means that there are many open questions regarding these expectations. How
do firms/households form their expectations? How and to what extent do inflation
expectations affect the decisions of households/firms? Can central banks influence
firms’ and households’ inflation expectations? How important are firms’ and
households’ inflation expectations for the inflation process? Those questions are still
open, as existing research does not provide unambiguous results. This annex
provides some insights from recent literature and gives an overview of the information
that can be distilled from available data on households’ and firms’ expectations.
A.1 What do we understand of firms’ observed inflation
expectations?
A.1.1 European Commission Business Survey data for the euro
area/EU
The business surveys conducted by the European Commission and national
statistical partners provide price expectation data for EU countries that can be
ECB Occasional Paper Series No 264 / September 2021
119
used to compile a euro area aggregate. However, the surveys provide only
qualitative information (i.e. details of the expected direction of the change in prices,
rather than a quantitative inflation forecast), they ask about firms’ own selling prices
rather than consumer prices, and they are only for the next three months. For the euro
area as a whole, manufacturing firms’ expectations regarding selling prices are
positively correlated with firms’ production expectations and the 12-month growth
rates of producer and consumer price indices (see Chart A.1).
Chart A.1
Euro area manufacturing firms’ selling price and production expectations over the next
three months
(percentage balances and percentage changes)
Source: Monthly European Commission surveys of the manufacturing industry.
Notes: Balances are calculated as the percentage of respondents expecting an increase minus the percentage expecting a decrease.
Data are seasonally adjusted.
The evolution of qualitative selling price expectations tends to be fairly similar
across the sectors for which these data are available – namely manufacturing,
retail and services (see Chart A.2).
-4
-3
-2
-1
0
1
2
3
4
5
-30
-20
-10
0
10
20
30
2013 2014 2015 2016 2017 2018 2019 2020 2021
Selling price expectations: three-month moving average (left-hand scale)
Production expectations: three-month moving average (left-hand scale)
HICP: year-on-year change (right-hand scale)
Producer Price Index: year-on-year change (right-hand scale)
ECB Occasional Paper Series No 264 / September 2021
120
Chart A.2
Firms’ selling price expectations over the next three months by sector
(percentage balances)
Source: Monthly European Commission surveys of manufacturing, retail and services firms.
Notes: Balances are calculated as the percentage of respondents expecting an increase minus the percentage expecting a decrease.
Data are seasonally adjusted.
When assessing these qualitative expectations, it should be borne in mind that
they are percentage balances, constructed as the difference between the
percentage of firms expecting an increase in selling prices minus the
percentage expecting a decrease. While this ignores the percentage of firms that
expect no change in prices, disaggregate data suggest that in recent years,
particularly for the manufacturing and retail sectors, the percentage expecting no
change has generally risen (see Table A.1). It might be the case that, in an
environment of persistently low inflation more generally, retail firms in particular have
elected to keep prices unchanged (rather than increasing or reducing them). If so, this
would be indicative of some degree of menu or coordination costs, and also a
self-fulfilling feedback mechanism whereby low inflation pressures give rise to low
expected and actual retail inflation.
-20
-15
-10
-5
0
5
10
15
20
25
30
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Manufacturing
Retail
Services
ECB Occasional Paper Series No 264 / September 2021
121
Table A.1
Decomposition of percentage balances across broad sectors
(percentages where indicated and percentage points otherwise)
Increase Same Decrease Balance
Manufacturing 2004-15 13.9% 76.7% 9.4% 4.5
2016-20 12.4% 80.0% 7.6% 4.8
Change -1.5 +3.3 -1.8 +0.3
Retail 2004-15 17.1% 71.3% 11.6% 5.5
2016-20 13.3% 78.2% 8.5% 4.8
Change -3.8 +6.9 -3.1 -0.7
Services 2004-15 11.8% 79.4% 8.8% 3.0
2016-20 12.3% 81.3% 6.4% 5.9
Change +0.5 +1.9 -2.4 +2.9
Sources: European Commission Business Surveys and Eurosystem staff calculations.
The absence of large-scale historical surveys of firms’ aggregate inflation
expectations makes it difficult to study their properties and evaluate alternative
models that could represent them. Within the European Union, the Banca d’Italia’s
regular Survey of Growth and Inflation Expectations (SIGE) and the regular data
collected by Narodowy Bank Polski represent exceptions in this regard, allowing
information to be gathered on firms’ expectations concerning consumer price inflation
(see Cecchetti et al. (2021) for an overview of the results of the Banca d’Italia survey).
A.1.2 The business surveys conducted by the Banca d’Italia and
Narodowy Bank Polski
Both central banks’ business surveys cover many areas of firms’ activity,
looking both backwards and forwards. The sections regarding prices look not only
at the prices for firms’ own output, but also at consumer prices. The SIGE survey on
inflation and growth expectations has been conducted on a quarterly basis since 1999
and is aimed at firms with 50 employees or more in (i) industry excluding construction,
(ii) non-financial private services and (iii) construction (since 2013). It spans more than
1,000 firms, with the sample being stratified according to economic sector, firm size
and geographical area. The question on firms’ inflation expectations is worded as
follows: “What do you think consumer price inflation in Italy, measured by the
12-month change in the harmonised index of consumer prices, will be in 6, 12 and
24 months?” Since Q3 2012, about two out of three respondents (“informed/anchored
firms”) have been provided with a nominal anchor – the latest available official figure at
the time that the questionnaire is sent out – while the remaining firms are not given that
information.
The Polish central bank launched its annual and quarterly business surveys in
1995. The first survey was sent to less than 200 selected companies from all over
Poland. The sample has gradually expanded over the years, with the most recent
ECB Occasional Paper Series No 264 / September 2021
122
editions of the quarterly survey being sent to more than 2,700 entities all over the
country. The respondents include enterprises from all non-financial sectors of the
economy with the exception of farming, fishing and forestry. Since Q3 2008, the
survey question concerning changes in consumer prices has been qualitative,
whereas previously a quantitative question was asked. The qualitative question
provides the respondents with the latest available official CPI figure and is phrased in
the following way: “In … CPI inflation was …% in annual terms. During the next 12
months, will prices, in your opinion, (1) rise faster than at present, (2) rise at the same
rate, (3) rise more slowly, (4) stay at their present level, (5) fall, or (6) difficult to say?”
Expected inflation is quantified on the basis of this question using the probability
method.
A.1.3 Formation of firms’ inflation expectations
Both in Italy and in Poland, firms’ inflation expectations – in terms of their
averages and volatility – seem to be more similar to the expectations of experts
than they are to the expectations of consumers (see Table A.2). Although the
volatility of firms’ expectations is somewhat greater than that of experts’ forecasts, it is
less than the volatility of inflation, especially in Poland. Empirical research suggests
that firms in both economies form their predictions on the basis of a variety of factors,
including monetary policy. Wage increases determined by contract renewals and the
prices of raw materials appear to be significant drivers of Italian firms’ expectations,
while the latest official inflation data also influence firms’ beliefs (Conflitti & Zizza,
2020). In addition to current inflation, the drivers of Polish firms’ inflation expectations
include short-term interest rates, central bank inflation projections, industrial
production data and, more recently, wage growth (Chmielewski et al., 2020).
Table A.2
Selected features of enterprises’ short-term expectations (12 months ahead)
regarding consumer inflation
Italy (2004-19) Poland (2004-19)
Data source Mean (%) Standard
deviation Data source Mean (%) Standard
deviation
Inflation expectations of
enterprises SIGE 1.7 1.0 Narodowy
Bank Polski 2.21 1.3
Inflation expectations of
consumers (quantified) GUS 3.6 1.9
Inflation expectations of
consumers (quantitative)
Commission
consumer
survey 4.5 1.8
GUS
11.8 3.6
Inflation expectations of
experts
Consensus
Economics
(quarterly) 1.6 0.5
Refinitiv
2.3 0.5
HICP (Poland: CPI) inflation Eurostat 1.6 1.1 GUS 2.1 1.7
The similarity between firms’ and experts’ inflation expectations is, at least in
part, a reflection of the fact that in both economies experts’ inflation forecasts
seem to play a significant role in driving firms’ inflation expectations. In Italy this
ECB Occasional Paper Series No 264 / September 2021
123
conclusion is based on survey evidence suggesting that news media and the reports
of professional forecasters are the most important sources of information for Italian
firms (Conflitti & Zizza, 2020), while in Poland it is based on the results of sticky
information models (Łyziak, 2013). In both economies, firms seem to incorporate new
information in their expectations quickly: Italian firms learn the most recent inflation
rate within one quarter of its release (Bartiloro et al., 2019), while Polish firms update
professional forecasts every six months on average (Łyziak, 2013). According to the
Italian survey, about half of the dispersion of inflation expectations is attributable to a
lack of information about the most recent inflation developments, with the remaining
cross-sectional dispersion related to developments in selected economic aggregates
(Bartiloro et al., 2019).
From a monetary policy perspective, it is important to point to two factors
determining firms’ inflation expectations. First, they seem to be influenced by
monetary policy actions and communication. In the case of Italy, Bottone &
Rosolia (2019) exploit a confidential version of the SIGE data and compare inflation
expectations reported by firms just before ECB Governing Council meetings with
those of firms surveyed just after them. These differences are then related to standard
market-based measures of unanticipated monetary policy news, based on daily
changes in major market interest rates on the days of ECB Governing Council
meetings. Italian firms’ inflation expectations react directly to monetary policy and do
so in a way that is consistent with its orientation. Specifically, over the entire period
under scrutiny, from 2002 to 2017, an unanticipated increase of 1 p.p. in the
three-month overnight index swap rate on Governing Council meeting days is
associated with a 0.5 p.p. decline in inflation expectations one year ahead for firms
interviewed immediately afterwards relative to those interviewed just before the
meeting. The reaction is even stronger in the case of unconventional monetary
policies. The survey conducted by Narodowy Bank Polski consistently confirms that
enterprises’ inflation expectations respond more strongly to changes in the short-term
interest rates set by the central bank than the inflation expectations of consumers and
financial sector analysts, despite finding that Polish firms attach less importance to the
central bank’s inflation target than financial sector analysts (Chmielewski et al., 2020).
Second, firms’ inflation expectations are influenced by news about current inflation.
This has implications for monetary policy communication. Chart A.3 uses SIGE data
on expectations to show that firms’ inflation expectations at different horizons tend to
co-move strongly with actual inflation (red line). This finding holds both for firms who
received information about the latest inflation figure (blue line) and for firms who did
not receive such information (yellow line).
ECB Occasional Paper Series No 264 / September 2021
124
Chart A.3
Firms’ inflation expectations at different horizons
6 months ahead 12 months ahead
(annual percentage changes) (annual percentage changes)
24 months ahead 48 months ahead
(annual percentage changes) (annual percentage changes)
Sources: Eurosystem staff calculations based on SIGE data.
Firms’ inflation expectations also prove to be useful in explaining actual price
developments. Polish data suggest that the best fit and forecasting accuracy for
estimated versions of the New Keynesian Phillips Curve (NKPC) are displayed by
specifications in which expected inflation is proxied by expectations elicited in firm or
consumer surveys (Łyziak, 2016a; Szafranek, 2017). In exercises based on a
complete small-scale New Keynesian model, it appears that, regardless of the
forecasting horizon, a version of the model in which survey-based measures of firms’
inflation expectations are used fares substantially better than other versions of the
model in terms of predictive accuracy, applying either model-consistent (rational)
expectations or direct measures of consumers’ or analysts’ expectations
(Łyziak, 2016b). Interestingly, survey-based measures of the inflation expectations of
financial sector analysts and firms demonstrate smaller errors than the forecasts of the
New Keynesian model with model-consistent expectations (see Chart A.4). However,
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
2014 2015 2016 2017 2018 2019 2020
Firms given latest i nflation figures
Firms given no information
Actual inflation
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
2014 2015 2016 2017 2018 2019 2020
Firms given latest i nflation figures
Firms given no information
Actual inflation
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
2014 2015 2016 2017 2018 2019 2020
Firms given latest i nflation figures
Firms given no information
Actual inflation
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5
2014 2015 2016 2017 2018 2019 2020
Firms given latest i nflation figures
Firms given no information
Actual inflation
ECB Occasional Paper Series No 264 / September 2021
125
for all types of agent – but especially firms – the forecasting properties of the models
using survey-based measures of expectations exceed those of both (i) the
survey-based measures of expectations themselves and (ii) the model with rational
expectations.
Chart A.4
Forecasting accuracy gains from using survey-based measures in the New Keynesian
model
MAE RMSE
(percentage points) (percentage points)
Source: Łyziak (2016b, p. 46).
Notes: These charts compare four-quarter-ahead forecasting errors (MAEs and RMSEs) for the MMPP-RE model with the errors of raw
survey-based measures for different groups of economic agents and errors based on an MMPP model with survey-based measures of
inflation expectations.
A.2 What do we understand of households’ observed
inflation expectations?
A.2.1 European Commission Consumer Survey data for the euro
area/EU
The European Commission Consumer Surveys provides price expectation data
for EU countries that can be used to compile a euro area aggregate. Originally,
those surveys tended to report only qualitative information (i.e. details of the expected
direction of the change in prices, rather than a quantitative inflation forecast) for the
next 12 months. Those data extend right back to 1985. Systematic collection of
quantitative data on the magnitude of inflation did not start until 2004, and those data
were first made public by the European Commission in early 2019. As a result, much
of the existing literature has, hitherto, focused on qualitative measures of inflation
expectations.
0.0
0.5
1.0
1.5
2.0
2.5
Enterprises Financial sector
analysts Consumers
Error of the MMPP model with rational expectations
Error of the survey-based measure of expectations
Error of the MMPP model with survey-based measure of
expectations
0.0
0.5
1.0
1.5
2.0
2.5
3.0
Enterprises Financial sector
analysts Consumers
Error of the MMPP model with rational expectations
Error of the survey-based measure of expectations
Error of the MMPP model with survey-based measure of
expectations
ECB Occasional Paper Series No 264 / September 2021
126
Quantitative expectations show a substantial upward bias relative to actual
inflation. While consumers’ qualitative and quantitative inflation expectations have
tended to display broad co-movement with actual inflation, the aggregate quantitative
expectations are significantly higher than actual inflation (see Chart A.5).85 The mean
since 2004 has been 5.7% (with a median of 3.8%), compared with 1.5% for actual
inflation. The lower quartile of quantitative expectations has averaged 2.0%
(i.e. approximately 75% of consumers have reported inflation expectations higher than
2%). Moreover, the peak correlation between quantitative expectations and actual
inflation has tended to be contemporaneous, whereas if consumers were able to
anticipate inflation, one would expect to see the peak correlation with actual inflation
coming some months in advance. The strong correlation with inflation perceptions
suggests that consumers’ inflation expectations are based primarily on their current
observations.
Chart A.5
Consumers’ qualitative and quantitative inflation expectations and actual HICP
inflation
(left-hand scale: percentage balances; right-hand scale: annual
percentage changes) (annual percentage changes and (for interquartile range)
percentage points)
Sources: European Commission (DG-ECFIN) and Eurostat.
A.2.2 Making sense of the bias in consumers’ quantitative inflation
expectations
Individual responses to the European Commission surveys point to apparent
rounding in consumers’ quantitative inflation expectations. Numbers are often
multiples of 5 and 10, as can be seen in the noticeable peaks at 0%, 5%, 10%, 15%
and 20% in the distribution (see Chart A.6). A smaller percentage of respondents
report in single digits, and the modal response for this group is around 2-3% (i.e. not
85 The upward bias in consumers’ expectations could be partly ascribed to the way in which households’
expectations are gathered. Using the Survey of Household Income and Wealth, Rondinelli & Zizza (2020)
show that when inflation expectations are collected using probabilistic questions – instead of asking for a
point estimate – households provide values that are more in line with official releases.
-2
-1
0
1
2
3
4
5
6
7
8
-20
-10
0
10
20
30
40
50
60
70
80
1985 1989 1993 1997 2001 2005 2009 2013 2017 2021
Qualitative inflation perceptions
Qualitative inflation expectations
Actual HICP inflation (right-hand scale)
-2
0
2
4
6
8
10
12
14
2004 2006 2008 2010 2012 2014 2016 2018 2020
HICP
HICP food
HICP FROOPP
Mean expectation
Median expectation
Interquartile range
ECB Occasional Paper Series No 264 / September 2021
127
as biased as the aggregate figures). The rounding can be interpreted as consumers
intentionally signalling that the figure should be regarded as imprecise. Thus, the use
(or not) of rounding can provide information on how (un)certain consumers are about
the inflation outlook. The use of rounding seems to be independent of the level of
inflation, but in periods of higher inflation (such as mid-2008) peaks can be observed
at 10%, 20% and even 40%, while at times of low inflation (such as mid-2009) peaks
can be observed at 0%.
Chart A.6
Histogram of responses (data from 2004 to 2020)
frequency of (or precent giving) each response
Sources: European Commission (DG-ECFIN) and Eurosystem staff calculations.
This (un)certainty framework allows us to compile a metric that helps to
interpret movements in consumers’ quantitative inflation expectations. The
portion of uncertain consumers in a given month can be calculated as those reporting
in multiples of five and ten divided by the total number of responses. On average,
approximately two-thirds are more uncertain on that basis (see Chart A.7). This
fluctuates, with noticeable increases in the percentage of uncertain respondents being
visible around the time of the global financial crisis and, more recently, the COVID-19
pandemic. While uncertain respondents may not quantify inflation with precision, they
appear able to capture broad developments in inflation, given that the modal
expectations of uncertain consumers co-move very closely with the modal
expectations of certain consumers. The bias in quantitative expectations relative to
actual inflation does not, therefore, imply that uncertain consumers are generally
unable to assess broader developments in inflation. However, as more certain
consumers also overestimate inflation to some extent, the certainty channel cannot be
considered independently of other hypothesised reasons for bias, including
psychological aspects of loss aversion, and the idea that consumers might have
different and highly heterogeneous baskets in mind when estimating inflation.
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
-10 -8 -6 -4 -2 0246810 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Perceptions
Expectations
ECB Occasional Paper Series No 264 / September 2021
128
Chart A.7
Percentage of “uncertain” respondents
(percentage of respondents reporting in multiples of five)
Sources: European Commission (DG-ECFIN) and Eurosystem staff calculations.
A.2.3 Implications for monetary policy
Genuine inflation expectations allow for the idea that individuals adjust
consumption plans in response to policy. If, in contrast, expectations were fully
explained by current inflation perceptions, it would be more challenging for
policymakers to change how people thought about the future, which could have an
impact on the effectiveness of forward guidance as a policy tool. The empirical finding
(based on ECCS data) that respondents’ individual characteristics (such as age,
gender or economic sentiment) help to explain inflation expectations also after
controlling for inflation perceptions suggests that there are indeed differences
between expectations and perceptions that can be “exploited” by monetary policy.
The (un)certainty framework helps to interpret the empirical finding of a
negative correlation between changes in inflation expectations and economic
activity. Candia et al. (2020) argue that the negative correlation reflects the way in
which households interpret the news about inflation. They caution that if news on
higher inflation gets an unambiguous supply-side interpretation (“inflation is bad for
the economy”) then this can lead to negative income effects, which can depress
economic activity. If this supply-side interpretation is dominant, it may be risky for a
central bank to communicate an increase in expected inflation to the public. The
(un)certainty framework offers an alternative perspective on this negative correlation
between inflation and economic growth expectations: negative economic sentiment
increases uncertainty, and uncertainty increases inflation expectations. Thus, it could
be that during economic downturns individuals are more uncertain and their resorting
to rounding leads to higher aggregate inflation expectations. Therefore, an increase in
expected inflation accompanied by an improved economic situation might not have an
adverse impact on economic sentiment.
0.60
0.65
0.70
0.75
0.80
0.85
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
ECB Occasional Paper Series No 264 / September 2021
129
A.2.4 De Nederlandsche Bank’s household survey
Evidence on the anchoring of consumers’ long-term inflation expectations is
still scarce, with most available surveys only investigating a relatively short
horizon up to a year ahead. A satellite survey conducted by De Nederlandsche Bank
in the context of the Dutch Household Survey (DHS) helps to fill this information gap.
All (2,482) DHS panel members were asked questions about the levels and probability
distributions of their short and long-term inflation expectations for the Netherlands and
the euro area. Respondents were randomly assigned to four different groups. Half of
the respondents are asked about inflation one year ahead and ten years ahead in the
Netherlands, while the other half were asked the same questions about euro area
inflation. Half of the respondents in each group were provided with information about
actual inflation and the ECB’s inflation aim. These sample splits were used to test
whether consumers have distinct views on inflation in the Netherlands and in the euro
area, and to what extent expectations are driven by (mis)perceptions about actual
inflation. The study also explored the effects of consumers’ characteristics (gender,
education, age and income) on their long-term inflation expectations. The empirical
results below relate to expected euro area inflation, since there were no notable
differences between the findings for euro area inflation expectations and Dutch
inflation expectations.
This empirical exercise finds that Dutch households’ long-term inflation
expectations are not consistent with level anchoring at the ECB’s inflation aim,
in the sense that median long-term inflation expectations are 4% – i.e. 2 p.p.
above the ECB’s inflation aim. There is also considerable disagreement among
respondents, as indicated by the large interquartile range (8 p.p.) for long-term
inflation expectations. Moreover, Dutch households tend to be more concerned about
the risk of higher – rather than lower – inflation in the long run. This can be seen in the
fact that mean long-term inflation expectations are higher than median expectations,
reflecting a positively skewed distribution of inflation expectations.
The study also generates other important insights for the analysis of anchoring.
First, short-term inflation expectations are considerably lower (and show less
disagreement among respondents). Thus, high long-term inflation expectations are
not simply a reflection of high short-term inflation expectations. Second, the provision
of information about actual inflation and the ECB’s inflation aim reduces consumers’
median long-term inflation expectations by 1 p.p., from 5% to 4%, but as a result of the
large cross-sectional variation, the difference in mean long-term inflation expectations
is statistically insignificant. Third, looking at the aggregate probability distributions of
long-term inflation expectations, the probability of future euro area inflation being in a
range that is consistent with the ECB’s inflation aim (which is assumed here to be
between 1% and 3%) is relatively low at 35% on average (see Chart A.8). Inflation
rates which are 2 p.p. or more above the ECB’s inflation aim have a much higher
probability, at 28% on average, than inflation rates that are 2 p.p. or more below it
(i.e. deflationary), at 12% on average. This distribution reflects individual uncertainty,
as well as disagreement among respondents. It corroborates the findings for the level
of inflation expectations to the extent that they suggest that the unanchoring of
consumers’ long-term inflation expectations is mainly due to expectations of higher
ECB Occasional Paper Series No 264 / September 2021
130
inflation, rather than expectations of lower inflation (or deflation). This contrasts with
recent concerns about a possible unanchoring of long-term inflation expectations on
the downside, rather than the upside.
Chart A.8
Average of Dutch consumers’ probability distributions for expected euro area inflation
(percentages)
Source: De Nederlandsche Bank household survey.
We find that long-term inflation expectations are better anchored for men, older
respondents and people with higher levels of education and net household
income. This assessment is made based on three measures of anchoring calculated
directly from the probability distribution of individual consumers’ long-term inflation
expectations – namely, the probability of inflation being close to the target, the
probability of inflation being far above the target, and the probability of deflation. Older,
male, better educated and higher-income respondents are more likely to report higher
probabilities for long-term inflation expectations that are close to the inflation target
and less likely to report probabilities for expectations that are far above or far below
that target.
A.3 The role of inflation expectations for households’
and firms’ choices
There is broad agreement in the academic literature that the way in which
agents react to news on expected inflation is heavily dependent on how they
interpret its source and its implications for the broader economic outlook
(Candia et al., 2020). The literature suggests that households’ beliefs about inflation
are mostly consistent with a supply-side narrative (“inflation is bad for the economy”),
while for firms this evidence is weaker, suffering from a lack of large-scale historical
surveys (Candia et al., 2020).
Looking specifically at firms, Coibion et al. (2020) suggest, on the basis of SIGE
data, that prior to the effective lower bound on policy interest rates, Italian firms
had a supply-side view of inflation which was akin to that of households and
0
10
20
30
40
<-4.0 [-4,-3[ [-3,-2[ [-2,-1[ [-1,0[ [0,1[ [1,2[ [2,3[ [3,4[ >4
Short-term (one year ahead)
Long-term (ten years ahead)
ECB Occasional Paper Series No 264 / September 2021
131
different from the view of professional forecasters. Indeed, firms with higher
inflation expectations tended, if anything, to be more pessimistic about the economic
outlook. Accordingly, they raised their prices, reduced their employment, used more of
their credit lines, applied for loans from new financial institutions, increased their
leverage and slightly reduced their liquidity, reflecting fears of reduced access to funds
in the future. However, when focusing solely on the period characterised by an
effective lower bound on policy interest rates following the start of the global financial
crisis, the effects that inflation expectations have on prices and credit utilisation
become stronger, while the effects on employment disappear. This is consistent with
firms perceiving a stronger demand-side channel for inflation at the effective lower
bound, in line with New Keynesian models.
Turning to households, theory suggests that we would expect to see an
increase in their intended spending if they expect higher inflation in the future
and all other factors held constant. The empirical evidence is quite mixed. Duca et
al. (2018) show that a higher expected change in inflation is associated with an
increase in the probability that a given consumer will make major purchases.
Rondinelli and Zizza (2020) find that in a high inflation regime (early 1990s)
consumers tend to bring spending forward, as higher inflation expectations lead to
lower real interest rates, supporting the functioning of an intertemporal substitution
mechanism. Conversely, in a low-inflation period (here, 2016), as higher expected
inflation translates into a loss in purchasing power, consumers’ readiness to buy
durables tends to react negatively, in line with the income effect argument raised by
Candia et al. (2020) and the estimates in Coibion et al. (2020). This is also consistent
with Andrade et al. (2020), who argue that it is the general inflation regime, rather than
the precise magnitude of expected inflation, that matters for consumption. Bachmann
et al. (2015) show, for the United States, that the impact which higher inflation
expectations have on reported readiness to spend on durables is generally small,
often being statistically insignificant outside the zero lower bound and typically being
significantly negative inside of it. According to their estimates, a 1 p.p. increase in
expected inflation during the recent zero lower bound period reduces households’
probability of having a positive attitude towards spending by about 0.5 p.p. In contrast,
Crump et al. (2015) provide evidence for the United States showing that households
increase their consumption if they expect higher inflation in the future. Also, Vellekoop
& Wiederholt (2019) find a strong and positive correlation between a household’s
inflation expectations and its propensity to buy vehicles.
Since the effect that a change in inflation expectations has on agents’ choices
depends on the way in which they interpret its source and its implications, the
provision of information about inflation to households and firms could
sometimes potentially “backfire” in terms of their subsequent decisions. As
stressed by Candia et al. (2020), providing more holistic messages which describe the
broader outcomes that policy decisions aim to achieve might be more effective than
just communicating information on inflation.
Recent theoretical and empirical research suggests that communication should
be different for professional market participants/experts and households/firms.
The former appear to react relatively fast, while households may exhibit a significant
ECB Occasional Paper Series No 264 / September 2021
132
lag in their reaction to such communication or even fail to pay attention to it (Lamla &
Vinogradov, 2019). In contrast, Mertens et al. (2020) provide evidence showing that
households react quickly to news about monetary policy. To counter the issue of
inattention, Angeletos et al. (2020) suggest that central banks’ policy communication
should be centred on unemployment, as households are more attentive to such news
and readily incorporate it into their decision-making, while exhibiting inattention to
news about interest rate changes.
Considering the aforementioned challenges, there remains a lot of research to
be done on households’ and firms’ inflation expectations. Methodological
advances in survey-based research and improvements in data availability suggest
that households’ and firms’ inflation expectations can become more relevant for
monetary policy. However, several data gaps still need to be closed in order to gain
consistent evidence on such expectations for the euro area and member countries.
Only then can households’ and firms’ observed expectations start playing a more
important role in policy deliberations.
ECB Occasional Paper Series No 264 / September 2021
133
Annex B – Constructing a heat map of
inflation expectation anchoring
B.1 Introduction
Analysing whether inflation expectations are anchored should ideally cut
across expectations formed by different economic agents and across
expectations at different horizons. Although the notion of anchoring inflation
expectations relates mainly to the longer term, shorter-term expectations can also
signal risks of unanchoring of long-term inflation expectations if economic agents
anticipate that inflation will, after a shock, return to the target level more slowly than
projected by the central bank (Domit et al., 2015). Anchored inflation expectations
have several dimensions, such as the level of inflation expectations and the level of
disagreement or uncertainty surrounding it (Kumar et al., 2015; Łyziak &
Paloviita, 2017).
Against that background, this annex constructs illustrative graphics – “heat
maps” – in order to assess the anchoring of inflation expectations in the euro
area according to a given criterion. Heat maps are especially useful for comparing
alternative proxies for inflation expectations, as they consider their historical variation
(volatility). If a given proxy has been stable over time, a relatively small change can
potentially indicate substantial unanchoring. If, on the other hand, the proxy has been
quite volatile historically, even a relatively large change will not necessarily signify
considerable unanchoring.
We analyse inflation expectations using data from professional forecasters, the
European Commission Consumer Surveys and financial market-based
measures. Both longer and shorter-term expectations are examined, using quarterly
data for the period 2005-20. The design of our heat maps is largely inspired by the
experiences of the Bank of England (Domit et al., 2015, pp. 165-180) and Narodowy
Bank Polski, but we also contribute to the existing literature by defining neutral levels
consistent with anchoring for (i) longer-term inflation expectations (the perceived
inflation target range) and (ii) shorter-term inflation expectations (the perceived central
bank communication range, which also takes into account central bank projections at
the time).
B.2 How the heat maps are constructed
In general terms, the heat maps use colours of various shades to show
whether, at a given time t, various metrics of inflation expectations, ,, stay at
their neutral levels, ,
, consistent with the concept of anchored expectations,
or deviate from them. The metrics’ deviations from their neutral levels are expressed
ECB Occasional Paper Series No 264 / September 2021
134
in terms of the volatility of the variable under consideration – i.e. by their standard
deviation, ,.86 More specifically, the colour of the heat map for each of the metrics
considered depends on the number ,, which shows how far (i.e. by how many
standard deviations) a given metric has deviated from its neutral level:
,
=,,
,
If , is close to zero, that means that the ith metric of inflation expectations at time t is
near its neutral level, which corresponds to a white colour on the heat map. The further
away from zero (above or below) , is, the darker the colour used becomes,
signalling deviations from the metric’s neutral level. We construct separate heat maps
based on (i) levels of inflation expectations (Heat Map A), (ii) an assessment of their
responsiveness to macroeconomic developments (Heat Map B) and (iii) the level of
disagreement or uncertainty surrounding them (Heat Map C).
Heat Map A – Levels of inflation expectations
The first heat map shows levels of expectations for various types of agent: the
median response to the quantitative question on inflation expectations 12 months
ahead in the European Commission Consumer Survey; the average point forecast
and the mean of the aggregated distribution from the SPF for different horizons; the
long-term expectations reported by Consensus Economics; and financial market ILS
rates for different horizons (both raw rates and the – IRP-adjusted – expectations
component). In addition, we look at the probability of SPF inflation 12 and 24 months
ahead being in the range 1.5-1.9%.87
The neutral levels are defined differently, as a single value or a range,
depending on the metric of inflation expectations under consideration.88 In the
case of longer-term inflation expectations, we assume that their neutral level is given
by the perceived ECB inflation target range. In our heat maps, that range comprises
values between 1.7% and 2.0%, which are regarded by most SPF participants as
being in line with the ECB’s price stability objective (see ECB, 2020). That range is
also broadly consistent with empirical studies analysing the inflation aim of the ECB’s
Governing Council (Paloviita et al., 2021; Hartmann & Smets, 2018; Rostagno et al.,
2019).
86 In line with the real-time data approach, the formula uses cumulative standard deviation, calculated up to
the previous period, (,), and the neutral level known at time t.
87 In the SPF questionnaire, survey participants are asked how they assess the probability of the forecasted
inflation outcome falling within predefined intervals. For heat map analysis, the most logical interval –
i.e. the range closest to the perceived ECB inflation target range – is 1.5% to 1.9%.
88 When the neutral level is given by a range, ,
is an upper or lower bound of that range.
ECB Occasional Paper Series No 264 / September 2021
135
Heat Map A
Levels of expectations
Notes: The perceived ECB inflation target range comprises values between 1.7% and 2.0%, which are regarded by most SPF
participants as being in line with the ECB’s price stability objective. The perceived ECB communication range comprises values between
the ECB inflation projection and the upper or lower bound of the perceived ECB inflation target range.
Legend
– no data
Rows 1-4a and 7-10a:
– metric is above its neutral level by 2.0 standard deviations (sd) or more
– metric is above its neutral level by 1.5 sd or more but less than 2.0 sd
– metric is above its neutral level by 1.0 sd or more but less than 1.5 sd
– metric is above its neutral level by 0.5 sd or more but less than 1.0 sd
– metric is close to its neutral level, deviating from it by no more than by 0.5 sd above or below
– metric is below its neutral level by 0.5 sd or more but less than 1.0 sd
– metric is below its neutral level by 1.0 sd or more but less than 1.5 sd
– metric is below its neutral level by 1.5 sd or more but less than 2.0 sd
– metric is below its neutral level by 2.0 sd or more
Rows 5, 6 and 11:
– probability of inflation between 1.5% and 1.9% is above its neutral level or just below it (i.e. within 1.0 sd)
– probability of inflation between 1.5% and 1.9% is below its neutral level by 1.0 sd or more but less than 2.0 sd
– probability of inflation between 1.5% and 1.9% is below its neutral level by 2.0 sd or more
For shorter-term inflation expectations, we assume, in line with Domit et al.
(2015), that their neutral level is given by the perceived ECB communication
range. If the Eurosystem/ECB staff inflation projections (hereinafter, “ECB inflation
projections”) stay within the perceived ECB inflation target range, the ECB
communication range is identical to the perceived ECB inflation target range. If not,
A. Levels of inflation expectations
123412341234123412341234123412341234123412341234123412341234123
[1] Consumers, quantitative, 12 months ahead vs. long-term mean
[2] SPF, 12 months ahead, mean vs. perceived ECB communication range
[2a] SPF, 12 months ahead, mean of distr. vs. perceived ECB communication range
[3] SPF, 24 months ahead, mean vs. perceived ECB communication range
[3a] SPF, 24 months ahead, mean of distr. vs. perceived ECB communication range
[4] Market data, 1y rate 1y ahead vs. perceived ECB communication range
[4a] Market data, 1y rate 1y ahead, exp. component vs. perceived ECB communication range
[5] SPF, 12 months ahead, probability of inflation between 1.5% and 1.9%
[6] SPF, 24 months ahead, probability of inflation between 1.5% and 1.9%
[7] SPF, long term, mean vs. perceived ECB inflation target range
[7a] SPF, long term, mean of distr. vs. perceived ECB inflation target range
[8] Consensus Economics, 6-10y ahead, mean vs. perceived ECB target range
[9] Market data, 1y rate 4y ahead vs. perceived ECB target range
[9a] Market data, 1y rate 4y ahead, exp. component vs. perceived ECB target range
[10] Market data, 5y rate 5y ahead vs. perceived ECB target range
[10a] Market data, 5y rate 5y ahead, exp. component vs. perceived ECB target range
[11] SPF, long term, probability of inflation between 1.5% and 1.9%
Shorter-term inflation expectations
2012
2013
2005
2006
2007
2008
2014
Longer-term inflation expectations
2015
2016
2017
2018
2019
2020
2009
2010
2011
ECB Occasional Paper Series No 264 / September 2021
136
the upper and lower limits of the ECB communication range are bounded by the lowest
and highest of three values: the ECB inflation projection and the upper and lower limits
of the perceived ECB inflation target range. If the views of professionals are outside
the range determined by the ECB’s projections and the perceived ECB inflation target
range, their inflation expectations signal risks of unanchoring. For example, if the
ECB’s inflation projection 12 months ahead is below 1.7%, inflation expectations
12 months ahead are considered to be consistent with the ECB’s communication if
they are between the ECB inflation projection and the upper limit of the perceived ECB
inflation target range (i.e. 2.0%).
For illustrative purposes, Chart B.1 presents the perceived ECB inflation target
range and the perceived ECB communication range for two forecasting
horizons: 12 months ahead and 24 months ahead. Since the ECB’s shorter-term
inflation projections have been relatively volatile, the perceived ECB communication
range for professionals is, in this case, often quite wide (with the lowest value for the
lower bound standing at 0.4% and the highest value for the upper bound standing at
2.9%). As we can see, the ECB’s longer-term inflation projections have been more
stable, so the neutral range for professionals never extends above 2.0% or below
1.3%.
ECB Occasional Paper Series No 264 / September 2021
137
Chart B.1
Perceived ECB inflation target range and perceived ECB communication range
Perceived ECB inflation target range and ECB
inflation projections 12 months ahead
Perceived ECB inflation target range and ECB
inflation projections 24 months ahead
(percentages) (percentages)
Perceived ECB communication range
12 months ahead Perceived ECB communication range
24 months ahead
(percentages) (percentages)
Source: Łyziak & Paloviita (2018).
Note: The last observations are for October 2020.
In the case of consumers, whose quantitative expectations display substantial
bias (Arioli et al., 2016; Stanisławska et al, 2019) and, as such, are not directly
comparable with the ECB’s inflation target range or projections, the neutral
level for their expectations is proxied using the historical (cumulative) mean
(calculated up to period t-1). Similarly, the historical (cumulative) mean, calculated
up to the previous period, is also used as the proxy for the neutral levels of the other
metrics of inflation expectations, which relate to the level of disagreement and
uncertainty in inflation expectations and the probability of inflation staying between
0.0
0.5
1.0
1.5
2.0
2.5
3.0
2004 2006 2008 2010 2012 2014 2016 2018 2020
ECB inflation projection, 12 months ahead
Perceived ECB inflation target range (1.7-2.0%)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
2004 2006 2008 2010 2012 2014 2016 2018 2020
ECB inflation projection, 24 months ahead
Perceived ECB inflation target range (1.7-2.0%)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
2004 2006 2008 2010 2012 2014 2016 2018 2020
0.0
0.5
1.0
1.5
2.0
2.5
3.0
2004 2006 2008 2010 2012 2014 2016 2018 2020
ECB Occasional Paper Series No 264 / September 2021
138
1.5% and 1.9%.89 In these cases, the colour white in the heat map indicates that the
level of disagreement or uncertainty is not high relative to historical values.
Heat Map B – Responsiveness of inflation expectations
The second heat map illustrates the responsiveness of inflation expectations to
macroeconomic developments (such as changes in current inflation rates,
shorter-term expectations and other macroeconomic news). The metrics of
responsiveness are model-based (see Section 3.2.2.2) and represent estimates of
pass-through coefficients.90
Anchored inflation expectations should not respond to any short-term factors,
so the neutral level for the metrics in this heat map is equal to zero. We measure
the distance from the responsiveness parameter to zero in (cumulative) standard
deviations of parameter estimates91 (). However, we assume that the
responsiveness metric is equal to zero if the responsiveness coefficient is not
statistically significant:
=
0.1
0
> 0.1
where is a time-varying estimate of the responsiveness of inflation expectations to
specific macroeconomic factors and denotes corresponding p-values for this
estimate. Pass-through estimates that are close to zero or statistically insignificant are
shown in white, suggesting a lack of unanchoring risks. Larger and statistically
significant pass-through estimates are shown using darker purple colours.
89 In the SPF questionnaire, survey participants are asked how they assess the probability of the forecasted
inflation outcome falling within predefined intervals. For heat map analysis, the most logical interval –
i.e. the range closest to the perceived ECB inflation target range – is 1.5% to 1.9%.
90 Negative pass-through coefficients are ignored (i.e. treated as zero).
91 Note that is not the standard error of the parameter estimate in a given period. Instead, it is the
cumulative standard deviation of series of time-varying pass-through coefficients over time. This provides
information about the variability of estimated pass-through coefficients over time.
ECB Occasional Paper Series No 264 / September 2021
139
Heat Map B
Responsiveness of inflation expectations
Notes: (1) Macro surprises include inflation and corporate sentiment. (2) Macro surprises include inflation, GDP and PMI.
Legend
– no data
– responsiveness is not statistically significant or is above zero by 1.0 standard deviation (sd) or less
– responsiveness is above zero by 1.0 sd or more but less than 2.0 sd
– responsiveness is above zero by 2.0 sd or more but less than 3.0 sd
– responsiveness is above zero by 3.0 sd or more
Heat Map C – Disagreement and uncertainty
The final heat map shows subjective assessments of the level of
disagreement/uncertainty about future inflation. The list of metrics shown includes
a measure of the dispersion of consumers’ inflation expectations (disagreement) and
the aggregate uncertainty of professional forecasters. Aggregate uncertainty
combines forecasters’ disagreement about future inflation as well as their individual
uncertainty. The neutral levels for these metrics are proxied by the historical long-run
cumulative mean, calculated up to the previous period. In this case, the use of white in
the heat map indicates that the level of disagreement or uncertainty is not high relative
to historical values.
B. Sensitivity of inflation expectations
123412341234123412341234123412341234123412341234123412341234123
[1] Market data, 2y rate, sensitivity to macro surprises(1)
[2] Market data, 2y rate 3y ahead, sensitivity to 1y rate
[3] SPF, long term, sensitivity to negative inflation surprises
[4] SPF, long term, sensitivity to positive inflation surprises
[5] SPF, long term, sensitivity to actual inflation
[6] SPF, long term, sensitivity to short-term inflation forecast
[7] Market data, 5y rate 5y ahead, sensitivity to 1y rate
[8] Market data, 5y rate 5y ahead, sensitivity to macro surprises(1)
[9] Market data, 5y rate 5y ahead, sensitivity to macro surprises(2)
2008
2009
2010
2017
2005
2006
2007
2018
2019
2020
Shorter-term inflation expectations
Longer-term inflation expectations
2011
2012
2013
2014
2015
2016
ECB Occasional Paper Series No 264 / September 2021
140
Heat Map C:
Disagreement and uncertainty (relative to long-term means)
Legend
– no data
– the disagreement or uncertainty metric is below its neutral level or just above it (i.e. by less than 1.0 sd)
– the disagreement or uncertainty metric is above its neutral level by 1.0 sd or more but less than 2.0 sd
– the disagreement or uncertainty metric is above its neutral level by 2.0 sd or more
Findings
The heat maps provide some evidence that inflation expectations in the euro
area have become less well anchored in the low-inflation period. While measures
of responsiveness and uncertainty also deviated from normal values during the GFC,
level-based metrics more clearly showed risks of unanchoring on the downside during
the low-inflation period.
As regards levels of longer-term inflation expectations, signs of unanchoring
are most visible in financial market data and – to a slightly smaller extent –
survey-based measures. Market-based indicators of expectations have recently
been substantially below the perceived ECB inflation target range. Point inflation
expectations in the SPF for shorter forecast horizons (i.e. 12 and 24 months ahead)
have stayed in line with the ECB inflation target range or inflation projections.
However, looking at the distribution of SPF experts’ forecasts, we can see that the
perceived probability of HICP inflation being between 1.5% and 1.9% has recently
reached relatively low levels across forecasting horizons.
As regards the responsiveness of inflation expectations, we can see that two
market-based measures of inflation expectations (two-year ILS rates three
years ahead and five-year ILS rates five years ahead) have become more
responsive to developments in shorter-term expectations (one year ahead)
since the beginning of the GFC. In the case of the latter measure, this process
intensified in 2020. In addition, five-year ILS rates five years ahead displayed elevated
responsiveness to macroeconomic surprises in the period 2015-19.
Unanchoring risks are also signalled by metrics measuring disagreement
between consumers in terms of their inflation expectations and the uncertainty
surrounding SPF experts’ inflation forecasts. In particular, the aggregate
C. Disagreement and uncertainty (relative to long-term means)
123412341234123412341234123412341234123412341234123412341234123
[1] Consumers (quantitative), 12 months ahead, robust coefficient of variation
[2] SPF, 12 months ahead, aggregate uncertainty
[3] SPF, 24 months ahead, aggregate uncertainty
[4] SPF, long term, aggregate uncertainty
2010
2011
2012
2013
2005
2006
2007
2008
2009
2015
2016
2017
2018
2019
2020
Shorter-term inflation expectations
Longer-term inflation expectations
2014
ECB Occasional Paper Series No 264 / September 2021
141
uncertainty of longer-term SPF forecasts has been elevated in recent years, while the
uncertainty surrounding shorter-term forecasts has reached historically high levels
during the COVID-19 pandemic.
Robustness checks
The simplicity and transparency of these heat maps has been achieved at the
cost of some technical (ad hoc) assumptions. We analyse the robustness of our
heat maps with respect to two such assumptions. First, we modify the way in which we
define the neutral values for (i) consumers’ inflation expectations and the probability of
inflation being between 1.5% and 1.9% in Heat Map A and (ii) the measures of
disagreement or uncertainty in Heat Map C. More specifically, instead of applying
long-run cumulative means, we use 32-quarter moving averages. By using moving
averages, we account for the possibility of structural breaks in some characteristics of
inflation expectations. For example, Dovern & Kenny (2020) point out that there has
been an increase in the uncertainty surrounding SPF forecasts since the global
financial crisis. Second, instead of excluding sensitivities that appear statistically
insignificant in Heat Map B, we apply a less restrictive approach. The responsiveness
metric is calculated with the correction for the degree of statistical significance
measured using the p-value, i.e.:
=
(1)
For highly significant estimates, the metric converges to the metric proposed in the
benchmark option, while if expectations’ responsiveness is highly insignificant, the
metric converges to zero.
In general, heat maps modified along the above lines (available on request)
result in conclusions similar to those of our main analysis. However, relatively
large and persistent differences can be observed for the responsiveness of
longer-term SPF forecasts to negative inflation surprises and short-term inflation
forecasts. In the case of the former, there is a significant part of the sample period
where the alternative heat map signals larger unanchoring risks than the original heat
map. In the case of the latter, differences of the same kind are visible in the periods
2007-09 and 2013-14. In addition, for the level of consumer inflation expectations, the
alternative heat map signals stronger unanchoring risks in the most recent period than
the corresponding benchmark heat map.
© European Central Bank, 2021
Postal address 60640 Frankfurt am Main, Germany
Telephone +49 69 1344 0
Website www.ecb.europa.eu
All rights reserved. Any reproduction, publication and reprint in the form of a different publication, whether printed or produced
electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.
This paper can be downloaded without charge from the ECB website, from the Social Science Research Network electronic library or from
RePEc: Research Papers in Economics. Information on all of the papers published in the ECB Occasional Paper Series can be found on
the ECB’s website.
PDF ISBN 978-92-899-4471-7, ISSN 1725-6534, doi: 10.2866/878799, QB-AQ-20-027-EN-N
