ArticlePDF Available

Discussion of the Fisher Effect Puzzle: A Case of Non-Linear Relationship

Discussion of The Fisher Effect Puzzle: A Case of Non-Linear Relationship
Peter N Smith (University of York)
1. Assessing the Fisher equation
Analysis of the Fisher equation or, equivalently, of the determination of the real interest rate
has been the subject of many studies over the years. It provides a significant challenge because
the two key variables to be modelled both appear in expectation and so the impact of how
expectations are modelled has to be allowed for along with the other aspects of the problem.
The Fisher equation is usually understood to connect the ex-ante real interest rate e
r to ex-ante
(or expected) inflation e
and the observed nominal interest rate it(which many authors regard
as fixed in advance of the determination of inflation), so ee
. The difference between
the ex-ante and ex-post real interest rate t
r is the (assumed to be) random inflation
expectational error:
ee e
tt t t t tt t
rr r i
=+ − =+= (1)
where the expectational error is e
=−. The Fisher equation is therefore an accounting
identity which defines the ex-post real interest rate. The approach of Hall et al is to focus on
the relationship between the nominal interest rate and ex-post inflation i a b u
=+ + where
the hypothesis of interest is b=1 and the real interest rate is a constant a. As the authors say,
analysis of this particular equation, either through cointegration analysis or otherwise has often
returned estimates of b substantially (and often significantly) below one. It is impressive that
the method of estimation employed in this paper returns a (time varying average) value of b
close to one. There is, however, a serious limitation to the approach that the authors adopt in
that they assume that the real interest rate is constant, so:
tt t t t t t
ri ab uaw
=− =+ + =+ (2)
As is demonstrated below, it is unlikely in most modelling situations that this will be true. Put
differently, the error term t
u is likely to be predictable. An alternative approach is to test for
cointegration between t
i and t
based on the finding that they are integrated I(1), non-
stationary, variables. This approach has the appeal that it allows the equilibrium (or long-run)
real interest rate to be time varying and stationary and which could be a function of long-run
stationary features of the macro economy such as technology growth.
Having said this, Hall et al use a result in Swamy at al (2008) which shows that their
estimation method is robust to mispecification of various types. Their estimate of b is therefore
reliable and it is interesting that they obtain estimates of b which are close to unity on average
over time. The time-varying nature of the estimates of a and b must depend in part on the
elements of the determination of the real interest rate which have been omitted. It would have
been interesting to see the impact of introducing some of these variables on the estimates of a
and b. The authors offer the possibility of asymmetry in the behaviour of inflation in the
parameters after 1983 and more could be learnt about this by further structural analysis.
2. Evaluating the persistence and stationarity of the elements of the real interest rate
The time-series properties of inflation and the nominal and real interest rates have occupied
numbers of researchers. Representative is the work of Rose (1988) who found that the nominal
interest rate and prices appear to both be I(1) implying that inflation is stationary whilst the
nominal interest rate is non-stationary. He then goes on to show how this prime facia evidence
of non-stationarity in the ex-ante real interest rate is inconsistent with the implications of the
consumption-based capital asset pricing model (CCAPM) given the stationarity of
consumption growth and the real returns of a number of other financial assets. Subsequently,
evidence has been presented that price inflation maybe be non-stationary (see Ball and
Cecchetti (1990), for example). This, however, has not resolved the problems because further
analysis, supported by the results in Table 1 of Hall et al, show no evidence in favour of
cointegration between inflation and the nominal interest rate and therefore stationarity of the
ex- ante real interest rate.
Further evidence, mirroring developments in the modelling of persistence in time series
processes, has supported the argument that inflation and the nominal interest rate are not best
described as integrated I(1) or stationary I(0) processes but rather as persistent fractionally
integrated I(d) processes where 0<d<1. Whether they are fractionally stationary or not depends
on estimates of the parameter d. Sun and Phillips (2004) initiated much of this work and found
estimates of d0.9 for t
i, t
and t
r for a sample of quarterly data from 1934-1999 implying
fractional non-stationarity for all three variables. Their conclusion then followed that the real
interest rate or Fisher equation is not a cointegrating long-run relationship. This analysis,
however, does not take into account the potential structural breaks and non-linearity in the
behaviour of the variables concerned. Recent research suggests that, having accounted for one
or more of these factors, the evidence might be that the three variables concerned are, in fact,
fractionally stationary. That would mean that they are stationary but persistent which would
lead to behaviour of the real interest rate which is more consistent with the implications of
economic theory. Charfeddine and Guegan (2007), for example, find breaks in inflation in the
US in April 1967, January 1973 and July 1981 using Bai and Perron (1998) sequential testing.
Their estimates of d for untransformed inflation are between 0.408 (standard error = 0.05) and
0.89 (0.10). When they remove the structural breaks in inflation from the series and re-estimate
d they obtain estimates between 0.046 (0.039) and 0.228 (0.074) which are clearly significantly
smaller than 0.5 and border on failing to reject I(0) stationarity against a more persistent
alternative. Baillie and Morana (2009) allow for non-linearity in estimation of d. They estimate
an adaptive ARFIMA model which allows the constant term to be time varying (they choose a
flexible fourier form which can approximate a range of non-linear functions). Their results are
that estimation of d produces an estimate of between 0.42 (0.042) and 0.69 (0.067) for
unadjusted US CPI inflation from January 1948 to October 2006. The estimate of d for a
generalisation of the ARFIMA model which allows for non-linearity is 0.346 (0.04). Therefore
allowing for non-linearity and structural breaks reduces the estimated persistence of the
inflation rate significantly. If similar results can be found for the nominal interest rate series
there would then be the prospect of the three series t
i, t
and t
r being regarded as mildly
persistent stationary series. Whilst further analysis is required, similar results for consumption
growth and other asset returns would re-establish the consistency of the time-series behaviour
of key variables and the equilibrium properties of standard macroeconomic models such as the
3. The macroeconomic determinants of the real interest rate
In parallel to the investigation of the time series properties of the real interest rate there has
been development in analysis of the macroeconomic equilibrium determinants. In a standard
monetary model (eg Gali (2008), Ch 2) the ex-ante real interest rate is given by:
=+ (3)
where ρ=-lnβ the discount rate of utility, σ the coefficient of relative risk aversion (CRRA),
is the reduced-form parameter relating equilibrium output to technology and 1
is the
technology shock. In this equation the real interest rate is determined by the real side of the
economy in equilibrium and is stationary for a stationary growth rate in technology. In a
stochastic model, Canzoneri and Dellas (1998) show that monetary factors can also play a role
in the equilibrium real interest rate. They show that when decomposed into a risk-free rate and
a risk premium, the determination of the ex-ante real interest rate can be affected by central
bank operating procedures such as the choice between interest rate, money and nominal
income targeting. In particular, they show that real interest rates should be lower under money
than interest rate targetting for a given set of shocks.
What have we learned about the determination and behaviour of the real interest rate? Many
studies have contributed to finding that the real interest rate is a function of a number of,
probably stationary, variables including technology growth. In addition, changes in monetary
policy regimes are likely to be important source of variation in the real interest rate. These
effects might appear as structural change and parameter variation in estimates of the Fisher
equation. Hall et al show us that, armed with an estimation method which is robust to these
developments, we can obtain an estimate of the slope of the Fisher equation which is much
closer to our expectations than when we employ more limited methods.
4. References
Bai, J. and P. Perron (1998), "Estimating and testing linear models with multiple structural
changes", Econometrica, 66, pp 47-78.
Baillie, R.T and C. Morana (2009), "Investigating inflation dynamics and structural change
with an adaptive ARFIMA approach", mimeo Queen Mary University London.
Ball, L. and S.G. Cecchetti (1990), "Inflation and uncertainty at short and long horizons",
Brookings Papers on Economic Activity, pp 215-254.
Canzoneri, M.B. and H. Dellas (1998), "Real interest rates and central bank operating
procedures", Journal of Monetary Economics, 42, pp 471-494.
Charfeddine, L. and D. Guegan (2007), "Which is the best model for the US inflation rate: A
structural changes models or a long memory process?", CES Working Paper No. 2007.61,
Universite Paris.
Gali, J.(2008), Monetary Policy, Inflation and the Business Cycle: An Introduction to the New
Keynesian Framework, Princeton University Press, Princeton NJ.
Rose, A. (1988), "Is the real interest rate stable", Journal of Finance, 43, pp 1095-1112.
Sun, Y. and P.C.B. Phillips (2004), "Understanding the Fisher equation", Journal of Applied
Econometrics, 19, pp 869-886.
Swamy, P.A.V.B., Tavlas, G., Hall, S.G. and G. Hondroyannis (2008), "Estimation of
parameters in the presence of model mispecification and measurement error", mimeo,
University of Leicester.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Previous models of monthly CPI inflation time series have focused on possible regime shifts, non-linearities and the feature of long memory. This paper proposes a new time series model, named Adaptive ARFIMA; which appears well suited to describe inflation and potentially other economic time series data. The Adaptive ARFIMA model includes a time dependent intercept term which follows a Flexible Fourier Form. The model appears to be capable of succesfully dealing with various forms of breaks and discontinities in the conditional mean of a time series. Simulation evidence justifies estimation by approximate MLE and model specfication through robust inference based on QMLE. The Adaptive ARFIMA model when supplemented with conditional variance models is found to provide a good representation of the G7 monthly CPI inflation series.
Full-text available
Misspecifications of econometric models can lead to biased coefficients and incorrect interpretations of error terms, which in turn can lead to incorrectly estimated models and incorrect inference. There are specific techniques such as instrumental variables, which are used in the economics literature to deal with some individual forms of model misspecification, only addressing one problem at a time. The joint and separate solutions to the problems of unknown functional forms, omitted variables and measurement errors, discussed in this paper, prove that instrumental variables do not exist. Therefore, the specific techniques used in the literature are not feasible. This paper proposes a general method for estimating underlying parameters in the presence of a range of model misspecifications. It is argued that this method can consistently estimate the direct effect of an independent variable on a dependent variable with all of its other determinants held constant even in the presence of an unknown functional form, measurement error and omitted variables.
: This paper analyzes the dynamics of the US inflation time series using two classes of models: structural change models and long memory processes. For the first class, the Markov Switching Autoregressive (MS-AR) model of Hamilton (1989) and the Structural Change-Autoregressive (SCH-AR) model developed by Bai and Perron (1998 and 2003) are used. For the second class, the Autoregressive Fractionally Integrated Moving Average (ARFIMA) process developed by Granger and Joyeux (1980) is used. Moreover, the nature (true or spurious) of the observed long memory behavior is also investigated. The empirical results provide evidence for changes in mean. Break dates coincide with some economic and financial events, such as the Vietnam War and the two oil price shocks. Moreover, the results reveal that the observed long memory behavior is spurious and is due to the presence of breaks in the data.
We use consumption-based CAPM models (with fixed and flexible wages) to analyze the effect of central bank operating procedures on the ex ante real rate of return on a one-period nominal bond. We show that operating procedures affect both the risk-free rate and the risk premium. Nominal interest rate targeting produces the highest real interest rates; money targeting produces the lowest rates; and nominal income targeting comes out somewhere in between. Our simulations suggest that the central bank's choice of operating procedure may make as much as 50 or 100 basis points difference in the real rate of interest. The role of monetary aggregates has diminished in most central banks' operating procedures, and this provides one explanation for the common perception that real interest rates may have risen. Nevertheless, our analysis also demonstrates that there is no presumption that monetary targeting dominates from a welfare point of view.
The New Keynesian framework has emerged as the workhorse for the analysis of monetary policy and its implications for inflation, economic fluctuations, and welfare. It is the backbone of the new generation of medium-scale models under development at major central banks and international policy institutions, and provides the theoretical underpinnings of the inflation stability-oriented strategies adopted by most central banks throughout the industrialized world. This graduate-level textbook provides an introduction to the New Keynesian framework and its applications to monetary policy. Using a canonical version of the New Keynesian model as a reference framework, Jordi Galí explores issues pertaining to the design of monetary policy, including the determination of the optimal monetary policy and the desirability of simple policy rules. He analyzes several extensions of the baseline model, allowing for cost-push shocks, nominal wage rigidities, and open economy factors. In each case, the implications for monetary policy are addressed, with a special emphasis on the desirability of inflation targeting policies.
URL des Documents de travail : Classification JEL : C13, C32, E3.
It is argued that univariate long memory estimates based on ex post data tend to underestimate the persistence of ex ante variables (and, hence, that of the ex post variables themselves) because of the presence of unanticipated shocks whose short-run volatility masks the degree of long-range dependence in the data. Empirical estimates of long-range dependence in the Fisher equation are shown to manifest this problem and lead to an apparent imbalance in the memory characteristics of the variables in the Fisher equation. Evidence in support of this typical underestimation is provided by results obtained with inflation forecast survey data and by direct calculation of the finite sample biases. To address the problem of bias, the paper introduces a bivariate exact Whittle (BEW) estimator that explicitly allows for the presence of short memory noise in the data. The new procedure enhances the empirical capacity to separate low-frequency behaviour from high-frequency fluctuations, and it produces estimates of long-range dependence that are much less biased when there is noise contaminated data. Empirical estimates from the BEW method suggest that the three Fisher variables are integrated of the same order, with memory parameter in the range (0.75, 1). Since the integration orders are balanced, the ex ante real rate has the same degree of persistence as expected inflation, thereby furnishing evidence against the existence of a (fractional) cointegrating relation among the Fisher variables and, correspondingly, showing little support for a long-run form of Fisher hypothesis. Copyright © 2004 John Wiley & Sons, Ltd.
Univariate time series models for consumption, nominal interest rat es, and prices each appear to have a single unit-root before 1979. If nominal interest rates have a unit-root , but inflation and forecast errors do not, then ex ante real interest rates have a unit-root and are therefore nonstationary. The unit-root characteristic of real interest rates is puzzling: man y models imply that the growth rate of consumption and the real interest rate should have similar time-series characteristics; also, nominal returns for other assets appear to have very different time-series properties. Copyright 1988 by American Finance Association.