PreprintPDF Available

The Composite Leading Indicator of Mongolia

Authors:
Preprints and early-stage research may not have been peer reviewed yet.

Abstract and Figures

Mongolia's first composite leading indicator (CLI) is designed here to give early signals of turning points in economic activity in the near future. This information is of prime importance for economists, businesses and policymakers to enable a timely analysis of the current and short term economic situation. Mongolia's CLI uses monthly GDP as a proxy measure for economic activity. It focuses on the business cycle, defined as the difference between the smoothed GDP data and its long-term trend. Mongolia's CLI aims to predict turning-points in this business cycle estimate. The CLI is composed from a set of selected economic indicators whose composite provides a robust signal of future turning points. Out of 51 monthly time series covering the real economy, financial markets, international trade and the government sector that pass these criteria the quantity of imported diesel, M2, FDI, total import, international gold price and new real estate loans were selected on the basis of their predictive precision of turning points. The composite leading indicator based on these 6 components not only successfully predicts the turning points but also is highly correlated with the cyclical movements of the GDP growth.
Content may be subject to copyright.
Munich Personal RePEc Archive
The Composite Leading Indicator of
Mongolia
Bataa, Erdenebat
National University of Mongolia
10 December 2012
Online at https://mpra.ub.uni-muenchen.de/72415/
MPRA Paper No. 72415, posted 07 Jul 2016 05:13 UTC
1
The Composite Leading Indicator of Mongolia
Erdenebat Bataa1
National University of Mongolia
10 December 2012
Abstract
Mongolia’s first composite leading indicator (CLI) is designed here to give early signals of
turning-points in economic activity in the near future. This information is of prime importance
for economists, businesses and policy makers to enable timely analysis of the current and short
term economic situation. Mongolia’s CLI uses monthly GDP as a proxy measure for economic
activity. It focuses on the business cycle, defined as the difference between the smoothed GDP
data and its long-term trend. Mongolia’s CLI aims to predict turning-points in this business cycle
estimate. The CLI is composed from a set of selected economic indicators whose composite
provides a robust signal of future turning points. Out of 51 monthly time series covering the real
economy, financial markets, international trade and the government sector that pass these criteria
the quantity of imported diesel, M2, FDI, total import, international gold price and new real
estate loans were selected on the basis of their predictive precision of turning points. The
composite leading indicator based on these 6 components not only successfully predicts the
turning points but also is highly correlated with the cyclical movements of the GDP growth.
JEL classification: E32, E37.
Keywords: macroeconomic forecasting, Mongolia, composite leading indicator, structural
changes.
1 Department of Economics, National University of Mongolia, Baga Toiruu 4, Ulaanbaatar, Mongolia,
tsors79@yahoo.com. Acknowledgement: Much of this work was carried out as a part of an international consulting
service on sectoral analysis, economic modeling and risk assessment in Mongolia, Component A1 (Improving
Budget Preparation) of a MultiSectoral Technical Assistance Project (MSTAP) funded by a credit received by the
International Development Agency (IDA) of the World Bank and implemented at the Ministry of Finance Mongolia.
The author is very grateful for the opportunity and financial assistance received from all the above-mentioned
parties, but any errors or omissions are the responsibility of the author. The views expressed in the paper are
personal and should not be attributed neither to the Bank nor the Minsitry.The author gratfully acknowledges the
opportunity and financial assistance from the Bank but the views expressed and any errors are the author’s.
2
Executive summary
Mongolia’s first composite leading indicator (CLI) is designed here to give early signals of turning-points in
economic activity in the near future. This information is of prime importance for economists, businesses and policy
makers to enable timely analysis of the current and short term economic situation. Mongolia’s CLI uses monthly
GDP as a proxy measure for economic activity. It focuses on the business cycle, defined as the difference between
the smoothed GDP data and its long-term trend. Mongolia’s CLI aims to predict turning-points in this business cycle
estimate. The CLI is composed from a set of selected economic indicators whose composite provides a robust signal
of future turning points. The criteria in choosing this set of indicators were economic relevance, breadth of coverage,
frequency, absence of revision, timeliness, and length. Out of 51 monthly time series covering the real economy,
financial markets, international trade and the government sector that pass these criteria the quantity of imported
diesel, M2, FDI, total import, international gold price and new real estate loans were selected on the basis of their
predictive precision of turning points. The composite leading indicator based on these 6 components not only
successfully predicts the turning points but also is highly correlated with the cyclical movements of the GDP
growth. However it has to be emphasized that Mongolia’s CLI is optimised to identify turning points and not for
judging the speed or strength of a recovery or downturn in the business cycle. A very high or low CLI for example
cannot be interpreted as an indication of very high or low levels of economic activity or growth. It merely provides a
strong signal of the phase the country is likely to be in its business cycle in the near future. At the same time a value
above 100 in the de-trended GDP means a positive output gap.
Mongolia’s CLI is therefore an event forecast, where the forecasted event is the turning-point in economic
activity measured by the de-trended GDP. This forecast is calculated without modelling the interaction between
variables and it is based solely on historical data, without invoking any expert judgments. The following graph
presents the CLI and the estimated business cycle for Mongolia. The two series show strong co-movements, with all
the turning points of the CLI consistently preceding those of the business cycle.
Mongolia’s CLI (continuous red line) and economic activity (blue dashes); (long term trend=100)
Estimation of the business cycle based on data covering the period up to August 2012 indicates that Mongolia’s
economic activity reached its peak in December 2011 while the CLI’s peak occurred five months before that. On
average Mongolia’s CLI predicts peaks and trough 5 and 9 months in advance respectively. Currently Mongolia’s
economic activity is proceeding towards its trough but the CLI shows no turning points yet, meaning that the
economic situation will worsen at least for 9 months before improving. This however may change if the copper
production of Ouy Tolgoi alters completely the past relationship among economic variables.
The system of Mongolia CLI includes pre-programmed Gauss codes and Excel data files. It is designed to
require a minimal intervention from the user. Intervention is restricted by updating the Excel data files of
Mongolia’s and international financial and economic variables every month. All the rest should be done by the
program.
3
Contents
1. INTRODUCTION
2. PRE-SELECTION
A. Reference series
B. Candidate component series
3. DECOMPOSITION AND BUSINESS CYCLE EXTRACTION IN THE
PRESENCE OF STRUCTURAL BREAKS
A. Rationale
B. Decomposition methodology
C. Estimation of long term trend and short term noise
4. EVALUATION
A. Cyclical conformity
B. Bry-Boschan algorithm of turning point detection
C. Length and consistency of the lead
5. IMPLEMENTATION GUIDELINE
6. APPENDIX 1. SUMMARY RESULTS FROM THE ITERATIVE
DECOMPOSITION
7. APPENDIX 2. MONTHLY TIME SERIES AND ITS ESTIMATED LONG TERM
TREND
8. APPENDIX 3. CROSS CORRELEGRAM OF THE BUSINESS CYCLE
MOVEMENTS OF THE GDP AND PROSPECTIVE COMPONENTS OF THE CLI
9. APPENDIX 4.TURNING POINTS IN BUSINESS CYCLE MOVEMENTS OF THE
GDP AND PROSPECTIVE COMPONENTS OF THE CLI
10. APPENDIX 5. COMPONENTS OF CLI IN DIFFERENT COUNTRIES
11. APPENDIX 6. TWO COMPUTER PROGRAMS (procedures are excluded as they
are approximately 500 pages)
4
1. Introduction
Information on where the economy is heading in the near future is of paramount importance for
economists, businesses and policy makers to enable timely analysis of the current and short term
economic situation. One of the key interests of the early analysts of business cycle was to
identify series that moved in and out of recession before the rest of the economy. Identification
of such series was attempted based on simple extrapolation, sophisticated macro econometric
models, the so-called “technical analysis” and surveys of consumers and entrepreneurs (e.g.
Purchasing Managers’ Index).
System of leading indicators has been one of the most widely used methods of
anticipation of future economic activity in the sense Mitchell and Burns (1938) originally
defined. They developed its earlier versions of this technique at the National Bureau of
Economic Research during the 1930s and 1940s. The idea behind this approach is simple and
clear: there should be an “early warning” system to forecast when the economy will shift from
expansion to recession (or on the contrary, from recession to expansion). In other words, we have
to choose the indicators, which get to their turning points earlier than the economy in general.
Then, whenever the leading indicator gets to its peak or to its trough, we are able to predict a
forthcoming peak or trough in the business activity in general. However it must be emphasized
that this concept is totally different from forming linear forecasts by minimizing the mean
squared error of a forecast for the level of the variable. The most prominent examples of the CLI
include the United States’ composite leading index that originally based on 12 most promising
leading indicators and has been systematically released since 1968 and the OECD’s system of
composite leading indicators that was developed in the 1970’s and 1980’s for its member
countries. In the 1990’s, leading indicators for Turkey, Korea, Hungary, and Poland and other
emerging economies were developed under the supervision of the OECD. In addition to
commonly used “official” indices, some “designers’ indicators” were also introduced, with
subtle variations in handling the initial statistical data.
Unfortunately such information is currently unavailable for Mongolia. The composite
index presented in the following sections is a variant of the one elaborated by the OECD’s
Statistics Division. The modifications incorporate statistical techniques that facilitate the
5
estimation process in the presence of possible structural changes that are happening in the
country. This document can also serve as a user manual for a GAUSS computer language based
system on composite leading indicator (CLI) to forecast economic activity in Mongolia. The
cycle here is the result of deviations of the economy from its long term trend. A contractionary
phase of the cycle means a decline in the rate of growth of the economy, and not necessarily an
absolute decline in economic activity.
For this analysis it is necessary to select a reference economic activity measure such as
Index of Industrial Production or GDP growth. I chose to use the latter on the basis of its
availability and also because the industrial production might be losing its importance in the
current economic structure. Then a group of candidate variables are selected from a large pool of
data to predict the turning points in the reference series. The selection of the components of the
leading indicator is based on the turning point forecast efficiency and on their economic
significance. Once selected and cleared from seasonality and outliers allowing for possible
endogenous structural breaks, using a methodology to be detailed in what follows, the relevant
variables are adjusted, de-trended and aggregated into a single CLI that forecasts the de-trended
reference series turning points. For de-trending the series, the Hodrick-Prescott (HP) Filter
method is applied. This method is a smoothing technique that decomposes seasonally adjusted
series into cyclical and trend components. I also use the same filter to remove high frequency
noise components from the series.
Out of 51 monthly time series covering the real economy, financial markets, international
trade and the government sector that pass initial economic criteria 6 of them were selected on the
basis of their predictive precision of turning points. The resulting leading indicator not only
successfully predicts the turning points but also is highly correlated with the cyclical movements
of the GDP growth. The methodology can be outlined as follows:
1. Choose the reference series to be forecast (GDP growth in this case).
2. Adjust the reference series for seasonal, outlier and trend effects, after taking into account
of possible structural changes. Quarterly series is linearly interpolated to produce
monthly data.2
2 In fact every quarterly data such as those from the Quarterly Loan Report of the Mongol Bank
are treated the same way.
6
3. Selection of the components of the CLI. This is based on economic relevance, breadth of
coverage, frequency, absence of revision and timeliness. Estimation of seasonally
adjusted, outlier corrected versions of the selected series after taking into account of
possible structural breaks. Quarterly series are linearly interpolated to produce monthly
data.
4. Detrending the seasonally-adjusted-reference and component series of the CLI with a
double Hodrick-Prescott Filter. The first filter removes the long term trend while the
second one gets rid of the noise with frequencies higher than 12 months.
5. Identifying turning points in the reference and component series using Bry and Boschan
(1971).
6. Selecting those component series that predict the turning points in the reference series
and aggregate them into a single CLI.
7. Comparison of the business cycle of the reference series and the CLI.
This document is accompanied by a pre-programmed Gauss code and a set of Excel files that
need to be updated every month. Although this document considers a methodology of predicting
turning points in the overall economic activity this could also be used for other reference series
as well.
7
2. Pre-selection
A. Reference series
The reference variable is the benchmark that indicates fluctuations in the economic activity, and
is the variable whose turning points are to be forecast. Usually chosen is the Industrial
Production Index, IIP, which has the advantage of being reported on a monthly basis, available
for most countries, and measures the real sector of the economy. The GDP growth data is chosen
as a reference series for Mongolia since the industrial production index is not available. I use
seasonally adjusted quarterly data reported in monthly Statistical Bulletins and they cover the
period of the 1st quarter of 2000 to the second quarter of 2012. Following the OECD’s leading
indicator methodology I linearly interpolate them into monthly frequency by assuming the
growth rate in each month is the same in a given quarter.
B. Component series
As previously mentioned, the essential feature taken into account for selecting a component of
the CLI is that it leads the reference series with a similar cyclical profile. Other salient features
are: the consistency of the lead of the indicator over the reference cycle at turning points, the
absence of extra or missing cycles, smoothness, freedom from excessive revisions by the
authorities, timeliness in delivery from the authorities, and availability of a long run of data of
satisfactory reliability with no breaks. Monthly series are preferred. A general requirement
applied here for a candidate component series is to satisfy at least one of the following:
- to display the expectations of economic agents;
- to respond to changes in business activity earlier than the overall economy;
- to have gained recognition as a leading indicator in other countries.
Moreover, from a practical point of view, series must also meet the following:
8
- to have cyclical fluctuations (that is, there must be a succession of growth and decline
periods);
- to be sufficiently reliable and comparable during the whole period in question;
- to be renewed on a timely basis, preferably each month.
The candidate component series for the composite indicator are 51 monthly variables that cover
ten spheres of the Mongolian economy: production input, energy use, mining production,
international trade, monetary environment, finance, government finances, new loans to economic
sectors, international commodities markets and Chinese economic situation measured by its CLI
provided by the OECD. The following table provides the breakdown.
Category Name of the candidate component series Data coverage Source
1
Input
Qua
n
tity of imported
tires
(thousand
s
)
2000.01
-
2012.08
Customs
2
Price of imported
tires
(USD)
2000.01
-
2012.08
Customs
3
Qua
n
tity of imported trucks (thousand
s
)
2000.01
-
2012.08
Customs
4
Price of imported
trucks (thousand USD)
2000.01
-
2012.08
Customs
5
Energy use
Quantity of imported diesel, thousand ton
1997.10
-
2012.08
NSO
6
Price of imported diesel, USD/ton
1997.10
-
2012.08
NSO
7
Quantity of imported petrol, thousand ton
1997.10
-
2012.08
NSO
8
Pric
e of imported petrol, USD/ton
1997.10
-
2012.08
NSO
9
Mining
Coal production
, thousand ton
2000.01
-
2012.08
NSO
10
Copper concentrate production (35%), thousand ton
2000.01
-
2012.08
NSO
11
Molybdenum
concentrate production, ton
2000.01
-
2012.08
NSO
12
Tra
de
Copper concentrate export price, USD/ton
1998.01
-
2012.08
NSO
13
Molybdenum
concentrate export price, USD/ton
1998.01
-
2012.08
NSO
14
Total trade, mln USD.
1997.10
-
2012.08
Mongol Bank
15
Import, mln. USD
1997.10
-
2012.08
Mongol Bank
16
Export, mln
. USD
1997.10
-
2012.08
Mongol Bank
17
Export to China, mln. USD
1997.10
-
2012.08
Mongol Bank
18
FDI, mln. USD
1997.10
-
2012.08
Mongol Bank
19
Net trade, mln. USD
1997.10
-
2012.08
Mongol Bank
20
Export to import ratio, %
1997.10
-
2012.08
Mongol Bank
21
Monetary
Consumer Price Index
1997.10
-
2012.08
Mongol Bank
9
22
environment
M2, mln. MNT
1997.10
-
2012.08
Mongol Bank
23
M1, mln. MNT
1997.10
-
2012.08
Mongol Bank
24
Loan to deposit ratio, %
1997.10
-
2012.08
Mongol Bank
25
Weighted average loan rate in M
NT, %
1997.10
-
2012.08
Mongol Bank
26
Weighted average loan rate in foreign currency, %
1997.10
-
2012.08
Mongol Bank
27
Central Bank Bill rate, %
1997.10
-
2012.08
Mongol Bank
28
Finance
Top20
1997.10
-
2012.08
Mongol Bank
29
USD exchange rate (MNT/USD)
997.10
-
2012.08
Mongol Bank
30
Yuan exchange rate (MNT/Yuan)
1997.10
-
2012.08
Mongol Bank
31
Budget
Current income
1997.10
-
2012.08
Mongol Bank
32
Current expenditure
1997.10
-
2012.08
Mongol Bank
33
Capital expenditure
1997.10
-
2012.08
Mongol Bank
34
L
oans
Total new loans, thousand MNT
2000Q1
-
2012Q2
Mongol Bank
35
New loans to administrative service
2000Q1
-
2012Q2
Mongol Bank
36
New agricultural loans
2000Q1
-
2012Q2
Mongol Bank
37
New construction loans
2000Q1
-
2012Q2
Mongol Bank
38
New consumer lo
ans
2002Q1
-
2012Q2
Mongol Bank
39
New loans to electricity, steam generation
2002Q1
-
2012Q2
Mongol Bank
40
New loans to financial and insurance activities
2000Q1
-
2012Q2
Mongol Bank
41
New loans to health organizations
2000Q2
-
2012Q2
Mongol Bank
42
New
industrial loans
2000Q1
-
2012Q2
Mongol Bank
43
New mining loans
2000Q1
-
2012Q2
Mongol Bank
44
New real estate loans
2000Q3
-
2012Q2
Mongol Bank
45
New loans to transport sector
2000Q1
-
2012Q2
Mongol Bank
46
New wholesale and retail trade loans
2002Q1
-
2
012Q2
Mongol Bank
47
International
International oil price
1997.10
-
2012.08
IndexMundi
48
International copper price
1997.10
-
2012.08
IndexMundi
49
International coal price
1997.10
-
2012.08
IndexMundi
50
International gold price
1997.10
-
2012.08
IndexMu
ndi
51
China
Composite Leading Indicator of China
1997.10
-
2012.08
OECD
Table 1. Prospective component series for the Mongolia’s CLI.
Although some of the preceding series may not be included in the final index, all exhibit a close
cyclical relation with GDP and were tested for inclusion in the CLI (see Figures in Appendix 3).
There is an ongoing official attempt to include publically non-reported statistics such as the new
10
car registration number, building permits, mining equipment imports. I think they need to be
added to the above list once the necessary data from the relevant government authorities are
obtained. It has also to be said that it’s important to enrich the above list as much as possible
since the system sifts through all the available data and checks their relevance for predicting the
turning points in the reference series. Appendix 5 includes some component indicators that are
being used in other countries that could guide to expand the list in the future.
3. Decomposition and business cycle
extraction in the presence of structural
breaks
A. Rationale
The absence of leading economic indicators in former communist countries, except in a few
eastern European countries and Russia is often associated with the possible structural breaks
happening in those economies and consequently a lack of sufficient data to make a distinction
between a long-term trend and cyclical or short-term fluctuations. Mongolia is no exception. Its
economy is experiencing important structural changes, first related with the break-up of the
centrally planned economy and the emergence of a market based system and more lately with the
rise of the mining sector in the economy. So it’s crucial to take into account possible structural
breaks in any analysis involving Mongolia macroeconomic variables.
On the other hand it is vital for the composite leading indicator methodology to
decompose any time series and remove its seasonal fluctuations and aberrant observations before
estimating its long term trend and high frequency fluctuations. Conventional technique to
remove seasonality and identifying outliers is often based on a filtering operation, and the
smoothing implied in such adjustment may reduce the magnitude of changes in the mean and
persistence properties (Ghysels and Perron, 1996); I return to this issue below. Furthermore, the
seasonal pattern itself may also exhibit structural change, which is not handled adequately by
seasonal adjustment filters and thus should be modelled explicitly for a country like Mongolia.
11
Such a methodology for countries possibly experiencing structural changes were
developed recently in Bataa, Osborn, Sensier and Dijk (2012). Their methodology first tests for
any structural breaks in the time series components and that information is later used in
identifying seasonality, outliers and the rest. The following section explains their methodology.
B. Decomposition methodology
In this section I first detail the iterative decomposition used to identify and distinguish between
breaks in mean, seasonality, persistence and (conditional) volatility of the series, while also
accounting for the possible presence of outliers. This is followed by an outline of Qu and
Perron’s (2007) multiple break testing procedure, which is my main econometric tool.
B1. Iterative procedure for structural break and outlier detection
Consider decomposing a stationary time series Yt into components capturing level (Lt),
seasonality (St), outliers (Ot) and dynamics (yt), where level and seasonality are deterministic and
only the last component is stochastic and represented by means of an autoregressive (AR)
process (although this could include stationary stochastic seasonality, if appropriate). This differs
from the usual unobserved components approach, as employed by Harvey (1989) and others,
which is designed to capture nonstationarity in both the levels and seasonal components.
However, the presence of such stochastic components would imply that a time series has both a
zero frequency unit root and the full set of seasonal unit roots, a conclusion which has not found
support in previous analyses; see, for example, Canova and Hansen (1995) for the US or Osborn
and Sensier (2009) for the UK.
As indicated in the Rationale, the possibility of changes to the Mongolia’s macroe
economic process is important for the conduct and understanding of monetary and fiscal policies.
The model I consider allows for structural change in each of the level, seasonal and dynamic
components, where breaks in the latter may occur in the AR coefficients or in the conditional
volatility. A crucial feature of the model is that the numbers of structural breaks in these
12
components do not have to be the same and nor do their temporal locations. The general model
specification is given by
ttttt yOSLY +++= (1)
1
kt
L
µ
= 11
111 ,...,1 kk TTt +
++
+=
==
=
, 1,...,1 11 += mk (2)
=
=
s
l
ltlkt DS
1
2
δ
22
122 ,...,1 kk TTt +
++
+=
==
=
; 1,...,1 22 += mk (3)
t
p
i
itikt uyy +=
=
1
,
3
φ
33
133 ,...,1 kk TTt +
++
+=
==
=
; 1,...,1 33 += mk (4)
2
4
)var( kt
u
σ
= 44
144 ,...,1 kk TTt +
++
+=
==
=
; 1,...,1 44 += mk (5)
where mj denotes the number of breaks of type j that occur at observations
(kj = 1, ..., mj),
with
= 0 and
=  (where T denotes the total sample size), and for s seasons per year (s =
12 for monthly data), Dlt (l = 1, …, s) are seasonal dummies equal to unity if the observation at
time t falls in season l and zero otherwise. Note that the coefficient lk2
δ
represents the deviation
of the unconditional mean of Yt in the l-th season (month) from the overall mean level j
µ
and,
for identification purposes, we impose the restriction
=
s
llk
12
δ
= 0 for all seasonality regimes k2 =
1, …, m2+1. Hence, except for outlier effects, the decomposition implies
lkkt lYE 21
][
δµ
+
++
+=
==
= (6)
when t falls into regime k1 for the mean and regime k2 for seasonality, with l being the season
corresponding to the specific observation.
13
Although our principal interest is the possibility of breaks in the components (2) to (5),
outliers are corrected to prevent these distorting inference concerning other components.
Outliers, Ot in (1), are observations that are abnormally distant from the overall level, defined
using the procedure of Stock and Watson (2003) and, when detected, are replaced with the
median of the six neighbouring non-outlier observation3. However, the particular outlier
methodology is not our central concern, and other procedures could be employed, such as that of
Tsay (1988).
Returning to our focus of interest, namely (2) to (5), it is difficult, if not impossible, to
test satisfactorily for multiple structural breaks in all these components in a simultaneous
procedure if breaks may occur at different dates. Complications arise because the structural break
testing methodologies developed so far assume a pre-specified minimum distance between
consecutive break dates, thus limiting the possible number of breaks. For example, using 15%
trimming in the methodology of Bai and Perron (1998, 2003a) or Qu and Perron (2007) would
allow at most 5 breaks in the parameters in any of (2) to (5) over the 35-year sample period we
analyse. However, potential changes in seasonal patterns due to (say) changes in data collection
methods or to tax effects, considered alongside mean and/or persistence shifts arising from
changes in monetary policy and volatility changes due to good luck renders plausible the
existence of more than a total of five parameter changes over our sample. Nevertheless,
estimating and testing for breaks sequentially also poses problems, since testing for structural
breaks in one component can be affected by the presence of breaks in other components. For
example, Marques (2005) shows empirically that conclusions concerning changes in inflation
persistence crucially depend on the specification of mean inflation, with much more evidence for
a decline in inflation persistence obtained if mean inflation is assumed constant than if the level
is allowed to exhibit structural breaks. In addition, a further complication is that the presence of
outliers can affect persistence estimates; see Franses and Haldrup (1994).
3 The procedure was downloaded from Mark Watson’s website http://www.princeton.edu/~mwatson/publi.html. The
results presented define an outlier as being more than five times the interquartile range from the median; except in
the initial loop, this is measured after the removal of mean, seasonal and dynamic effects. Neighbouring
observations are also considered in this context. Outlier correction is sequential (one at a time), until no more are
detected. Outlier correction is applied in the inflation series, unless two sequential approximately off-setting outliers
are detected. In this latter case, a single correction is made to the underlying CPI series, rather than two corrections
for inflation.
14
For this reason, I employ an iterative approach to examine breaks in each of the
components of the economic series alongside outlier detection and removal. First, an initial
identification of outliers is made. Outliers are considered first in this procedure for the same
reason that they are conventionally removed prior to other empirical analysis, namely so that
these do not distort inference on the components of primary interest. This procedure, however,
later re-considers outliers within each iteration (see below). Second, assuming a constant mean, I
test for breaks in the seasonal component4, which then yields (using the appropriate sub-samples)
our first estimate of the seasonal component allowing for structural instability. Third, the
deterministic seasonal component is removed and level shifts are examined in the outlier-
corrected series. The break dates are recorded, leading to the first estimate of mean of the series,
adjusted for any breaks uncovered. In the fourth step of the loop, having removed outliers,
seasonal and level components from the original series, we test for breaks in the autoregressive
(AR) coefficients of the dynamic component. The choice to consider breaks in deterministic
components prior to those for dynamics is based on the analyses of Cecchetti and Debelle
(2006), Levin and Piger (2004), Marques (2005).
In this initialisation, heteroskedasticity and autocorrelation (HAC) robust inference is
employed when testing seasonal and level changes, since (6) may be subject to dynamic effects
and possible volatility changes. Similarly, heteroskedasticity consistent (HC) inference is
employed in the initial analysis of changes in dynamics5, to take account of possible volatility
breaks. Although the within loop component order outliers, deterministic components, dynamics
follows recent empirical practice, I build on this by iterating the loop until convergence is
achieved, in the sense that the dates of outliers and all structural breaks do not change. In each
iteration, the latest estimates of the components of (1) are removed, except for that under study.
Since dynamics are taken into account, HC (not HAC) inference is employed for mean and
seasonal break tests subsequent to the initialization.
4 When initial mean breaks are considered prior to initial seasonal breaks, qualitatively very similar results are
obtained in relation to the numbers and dates of breaks. In practice, however, we prefer the analysis considering
seasonality first because it yields better convergence overall for our data series and we conjecture this is because of
the larger role of seasonality to changing mean effects for inflation in (6).
5 Since we lose p observations in order to estimate the AR(p) model, and consequently our dynamic component will
be p observations short of others, we set missing dynamic components to their unconditional mean of zero.
15
Once convergence is achieved in the iterative procedure just described for the
components of (1), we could in principle subtract these four components from the original series
and proceed to test for (conditional) volatility breaks in the residuals t
u
ˆ. However, Pitarakis
(2004, page 44) notes that It is a notoriously difficult problem to design good test procedures
about the equality of regression slopes while not necessarily maintaining the equality of variance
assumption”. Indeed, Hansen (2000) shows that structural changes in the marginal distribution of
regressors render the Andrews (1993) type structural break tests asymptotically invalid.
To account for this possibility I incorporate an additional ‘inner loop’ that iterates
between testing for breaks in the AR coefficients of the dynamic component yt and its
conditional volatility. To be precise, after removing outliers, mean and seasonal components, the
sub-loop tests for breaks in dynamics (assuming serially uncorrelated disturbances); in the first
sub-loop iteration this employs HC inference, but subsequently a constant variance assumption is
used. If any break is detected, the AR model is estimated allowing for these breaks, with
variance breaks then investigated using the resulting residuals. If volatility breaks are detected,
the residual standard deviations are estimated over the implied volatility segments, which are
then used to apply generalized least squares (GLS) estimation. Specifically, the test for breaks in
dynamics is applied to the GLS-transformed data, with the volatility break test repeated, and so
on until convergence is achieved in the dynamics/volatility break dates. Once this ‘inner loop’
has converged, we return to the main loop and proceed as above.
The procedure employed in the inner loop is based on the findings of Pitarakis (2004),
who uncovers very large size distortions for coefficient break tests in the presence of unmodelled
volatility change, but who also provides evidence on improvements offered by a feasible GLS
transformation in that context.
In the implementation of this procedure, the maximum number of iterations is set to 20
for each of the main and inner loops. In a small number of cases, the procedure does not
converge to a unique set of break dates, but rather converges to a cycle between two sets of
dates. In such cases, we select between these based on the minimization of the Hannan-Quinn
(HQ) information criterion, computed in this context as
16
[
[[
[ ]
]]
]
1)1()1(121
))ln(ln(2
)(
ˆ
ln
4321
1
1
1
1
2
4
4
44
4
1
4
+
++
++
++
++
++
++
++
++
++
++
++
++
++
+
+
++
+
=
==
=
+
++
+
=
==
=
+
++
+=
==
=
mmpmm
T
T
TTuHQ
m
k
kk
T
Tt t
k
k (7)
where the superscript has been omitted from T to ease notation.
B.2. Testing for multiple structural breaks
The iterative procedure outlined above is implemented using the Qu and Perron (2007) test for
multiple structural breaks6. At each step, following the recommendation of Bai and Perron
(2006), I first test the null hypothesis of no breaks against an unknown number of breaks. If the
null of no breaks is rejected I use a sequential testing procedure to estimate the number and
locations of breaks.
All tests relating to the components of (1) are examined in a regression framework, with
the form of the regression varying according to the component being tested for structural breaks.
Specifically:
(i) To test for breaks in the seasonal component St, we regress tttt yOLY ˆ
ˆˆ on a set
of centred seasonal dummies jt = Djt Dkt, j = 1, …, s but excluding j=k (where Djt is
a conventional zero/one seasonal dummy variable for season j) that is
t
s
kj
j
jt
j
itttt uyOLY +=
=1
ˆ
ˆ
ˆ
δ
, with the omitted season k coefficient retrieved
using
=
=
s
kj
j
j
i
k
i
1
δδ
. The estimated AR coefficients for the dynamic component yt,
are used to form t
y
ˆ.
6 Although the Qu and Perron (2007) procedure is developed for multivariate systems, it is adopted here because of
its attractive features, including the possibility of using volatility break information when testing for coefficient
breaks, as well as an explicit handling of volatility breaks that allows us to avoid using a volatility proxy in the
context of the Bai and Perron (1998, 2003a) methodology.
17
(ii) For the level (mean) component Lt, we employ the regression
titttt uyOSY +=
µ
ˆ
ˆˆ .
(iii) To test for breaks in the dynamic component ttttt OSLYy ˆˆˆ , we use an AR(p)
model tti uyL =)(
φ
, with AR polynomial p
ipiii LLLL
φφφφ
= ...1)( 2
21 in the lag
operator L.
In all cases, the most recent estimates are employed when constructing the dependent variable
for the regression.
In the three cases listed above, to test the null hypothesis 00 :
µµ
=
i
H )1...,,1( 1+= mi ,
00 :δδ =
i
H )1...,,1( 2+= mi where )',...,( 1isii
δδ
=δ, or 00 :
φ
φφ
φφ
φφ
φ
=
i
H )1...,,1( 3+= mi where
φ
φφ
φ
i
= )',...,,( 21 ipii
φφφ
, against m M breaks (for a specified maximum M) we employ the statistic
[
]
),,(supmaxmax 1
ε
qmFaWD Tm
Mm
=, (8)
where 1
1=a and for 1
>
m, m is m1, m2 or m3, as appropriate, ),(/)1,( mccam
αα
=, in which
),( mc
α
is the asymptotic critical value of the supremum statistic ),,(sup
ε
qmFT at significance
level
α
, in which7
(
((
(
)
))
)
+
++
+
=
==
=
β
ββ
ββ
ββ
ββ
ββ
β
λ
λλ
λλ
λλ
λ
ˆ
])
ˆ
(
ˆ
[
ˆ
)1(
sup),,(sup 1
),...,( 1
RRVRR
T
qmT
εqmF
m
T
Λ
, (9)
is a Wald-type test statistic for structural change at m unknown break dates,
β
ˆ is the vector of
coefficients, that is,
µ
ˆ
, δ
ˆ or
φ
φφ
φ
ˆ, for m breaks at given dates with estimated covariance matrix
7 The statistic as given by Qu and Perron (2007), specifically their equation (20), differs from (9) in omitting the
denominator term T, which is irrelevant for obtaining the supremum. However, (9) is the form employed in their
computer code used for inference.
18
)
ˆ
(
β
V, R is a non-stochastic matrix such that )...,,()( 121 +
=
mm
R
β
ββ
ββ
ββ
ββ
ββ
ββ
ββ
ββ
ββ
β
, q is the number
of coefficients that are allowed to change, λi (i = 1, …, m) indicate the break dates as fractions of
the sample size, that is, 1...0 1<<<< m
λλ
with ][ ii TT
λ
=and finally
ε
Λ denotes all
permissible sample partitions. When HAC inference is employed, this uses the quadratic spectral
kernel with automatic bandwidth selection as in Andrews (1991).
If the WDmax test of (8) rejects the null of no breaks at the 5% significance level, a
sequential F-type test is used to determine the number of breaks and their locations. In particular,
the test statistic is defined as
=+
+)
ˆ
,...,
ˆ
()
ˆ
,...,
ˆ
,,
ˆ
,...,
ˆ
(supmax)1(sup 111
11 ,
lTljjT
lj
TTTFTTTTFllSEQ
j
τ
ε
τ
Λ
ΛΛ
Λ
, (10)
where })
ˆˆ
(
ˆ
)
ˆˆ
(
ˆ
;{ 111,
ετετ
ε
++=Λ jjjjjjj TTTTTT for l = 1,2,…, and FT is given by (9).
The test statistic in (10) is applied for l = 0, 1, …, M until the test fails to reject the null
hypothesis of no additional structural breaks. Note that, for each value l, the estimates of all
break dates are re-estimated to find those corresponding to the global maximum of the likelihood
function.
The null hypothesis of no break in conditional volatility, 2
0
2
0:
σσ
=
i
H )1...,,1( 4+= mi ,
is tested using a likelihood ratio test statistic. In particular, the SupF statistic of (8) is replaced by
the SupLR statistic defined as
19
=
ΛT
mT
TL
TTL
qmLR
m
~),...,(
ˆ
ln2sup),,(sup 1
),...,( 1
λλ
ε
, (11)
where 2
1
1
1
1ˆ
ln
2
)12(ln
2
),...,(
ˆ
ln j
m
j
jj
mT
TT
T
TTL
σπ
+
=
+= and
+=
=
j
1j
T
1Tt
2
t
j-1j
2
ju
TT
1ˆˆ
σ
with t
u
ˆ (t
= 1, …., T) the residual series from (4), while ~ represents the corresponding quantities
computed under the null hypothesis of no volatility breaks. Similarly, the sequential test of (10)
is replaced by
=+
Λ
+),...,(
ˆ
),...,,,,...,(
ˆ
lnsupmax)1(sup
1
11
11 ,lT
ljjT
lj
TTTL
TTTTL
llSEQ
j
τ
ε
τ
. (12)
Having obtained the number of structural breaks using (10) or (12), as appropriate, the
break dates are estimated as those that maximise the corresponding F-type or LR-type statistic.
From a practical point of view the maximum number of breaks, M, needs to be specified,
as well as the minimum fraction
ε
of the sample in each regime. Critical values of the tests
depend on both the number of coefficients allowed to change and
ε
. In general
ε
has to be chosen
large enough for tests to have approximately correct size and small enough for them to have
decent power. Moreover, when the errors may be autocorrelated and/or heteroskedastic,
ε
has to
be larger than when these features are absent. In order to balance these issues, our empirical
analysis of Section IV sets
ε
= 0.2 and M = 2 when the tests are applied for the seasonal
component8 and 15.0
=
ε
with M = 5 otherwise. Critical values for all tests employed are
obtained from Bai and Perron (2003b) for a 5 percent level of significance.
8 Of course, only one observation per year is available on monthly seasonal effects, implying that relatively few
structural changes can be realistically allowed in these.
20
Figure 1. Notes: Panels show: a) observed time series, b) dynamic component, persistence (red line) and
volatility break dates (green vertical lines); c) regime means, d) deterministic seasonal component for regime 1
in blue, regime 2 in red and regime 3 in pink, e) outliers and f) correlogram for the estimated dynamic
component, with 95% confidence intervals shown as dashed lines. If relevant, the correlogram is shown over
sub-periods identified by dynamic breaks, with regime 1 in blue and regime 2 in red.
Summary results that are obtained from the application of the iterative decomposition
methodology are shown in graphical form in Figure1 (for the remaining of the series see Figures
1-42 of the Appendix 1). Our principal empirical results concern the presence of structural
breaks in different characteristics of monthly and quarterly time series that are considered
potential candidates for the CLI component over the period October 1997 to August 2012. These
charts provide: (i) the original unadjusted growth rates of the relevant time series; (ii) the
estimated dynamic component yt (constructed by removing outliers, mean and seasonal
components) together with its estimated persistence, defined as the sum of the autoregressive
coefficients in (4) and corresponding
2
±
standard error bands (in red), and volatility break dates
(vertical green lines); (iii) the level component Lt with
2
±
standard error bands; (iv) the
estimated seasonal component for each seasonal regime (again with
2
±
standard error bands);
(v) outliers Ot that are removed; and (vi) the correlogram of the dynamic component yt within
21
each dynamic regime j, together with an approximate 95% confidence interval of
)(/2 1
± jj TT . With the exception of the correlogram, all standard errors are obtained using
the White (1980) HC covariance matrix in the corresponding regression over the regime defined
by the appropriate estimated break dates. Where relevant, the graphs showing the seasonal
components and the correlograms for the dynamic component are colour-coded with the first
regime (that is, the sub-sample to the first break date) in blue, the second in red and the third in
pink.
More detailed numerical results are printed out into a Gauss output file.
C. Estimation of long term trend and short term noise
Once seasonality and outliers in the growth rates of the economic time series have been removed,
as described in Section B, I convert them back into their levels. For quarterly data I linearly
interpolate them into monthly frequency by assuming the growth rate in each month is the same
in a given quarter, following the OECD’s leading indicator methodology. But there will still be a
trend in the most of the time series and a high frequency noise in all of the series. The presence
of a trend in economic activity can distort evaluations of cyclical events. Removing these trends
(de-trending) provides a better measure of underlying movements in the business cycle. Policy
designed to affect trend growth is very different from policies reacting to business cycle
fluctuations.
Importantly, and this is particularly relevant in the context of the current economic
climate where the recent developments associated with the mining has created uncertainty about
trends, if estimates of trend economic activity changes, the output gap changes and, so, values of
the GDP greater or lower than 100 can arise during periods of trend change. A trend change
means possibilities of temporary and permanent level changes in the growth changes. Temporary
changes are treated in Section B as outliers and permanent changes are modeled as the growth
mean changes.
22
Since the outliers detected are removed a priori to the analysis and there was no mean
breaks as can been seen from the graphs in the Appendix 1, I use the conventional Hodrick-
Prescott filter which is a model-free approach to decompose a time series into its trend and
cyclical components. The Hodrick-Prescott filter is in effect an algorithm that “smoothes” the
original time series t
y to estimate its trend component, t
τ
. The cyclical component is the
difference between the original series and its trend, i.e.,
ttt cy +=
τ
where t
τ
is constructed to minimize:
∑ ∑
+ +
T T
tttttt
y
1
1
2
2
11
2)]()[()(
ττττλτ
The first term is the sum of the squared deviations of t
y from the trend and the second term,
which is the sum of squared second differences in the trend, is a penalty for changes in the
trend’s growth rate. The larger the value of the positive parameter
λ
, the greater the penalty and
the smoother the resulting trend will be. If for example, 0
=
λ
, then tt
y
τ
=, t = 1,…,t, while
λ→∞ means t
τ
is the linear trend obtained by fitting t
y to a linear trend model by OLS.
Following Hodrick and Prescott (1998) I use λ = 14400 when extracting the long term
trend from the monthly time series. The results are plotted in Figures 1-6 of the Appendix 2.
The cyclical component obtained by subtracting this trend term from the seasonality and
outlier removed series contain both fluctuations related to the business cycle and high frequency
noise. This can be inferred from the above figures. To remove the noise I again follow the OECD
CLI methodology. The default setting there is to remove cyclical components that have a cycle
length shorter than 12 months. This is equivalent to setting λ = 13.93. Going from frequencies to
λ parameter is achieved by substituting into the formula
(
)
1
2
)0cos(1(4
= w
λ
where ω0 is the
frequency expressed in radians, and τ denotes the number of periods it takes to complete a full
cycle. The two parameters are related through ω0=2π/τ. So the λ values above correspond to
τ=12 months. The OECD suggests first de-trending and then smoothing (using the larger and
then smaller smoothing parameter, respectively). After the first application of the HP filter (de-
23
trending, larger parameter), one is left with a cyclical and a trend component. The original series
is de-trended by dividing it by the trend component, thus implying a multiplicative approach.
Multiplicative methods seem to be the most popular and the Bank of Spain explains this in its
TRAMO/SEATS literature. This de-trended series is used in the second application of the HP
filter (smoothing, smaller parameter) and one is left with a smoothed and de-trended series which
fluctuates around 1.
Given the wide variability in amplitudes across series, forming the CLI with a simple
average of non-standardized variables may generate distortions in the construction of the final
index. To avoid this problem a normalization is applied to the business cycle of the prospective
CLI component series. I normalize each series by subtracting its mean and multiplying by 100
and then adding 100; so that a value of 100 represents its long term state.
4. Evaluation
A. Cyclical conformity
Although not a necessary condition for a CLI its cyclical similarity with the reference series is a
useful feature. If the cyclical movements between them are highly correlated, the indicator will
provide a signal, not only to approaching turning points, but also to developments over the whole
cycle. The cross correlation function between the reference series and the candidate components
(or the composite leading indicator itself) provides invaluable information on cyclical
conformity. The location of the peak of the cross correlation function is a good alternative
indicator of average lead time. Whereas the correlation value at the peak provides a measure of
how well the cyclical profiles of the indicators match, the size of correlations cannot be the only
indicators used for component selection. As a cross-check the average lead of the cyclical
indicator, measured by the lag at which the closest correlation occurs, should not be too different
from the median lag if the composite leading indicator is to provide reliable information about
approaching turning points and the evolution of the reference series. The cross correlations
24
between the reference and some candidate component series are plotted in Figure 2 (See
Appendix 3 for the remaining cross-correlations).
Cross correlegram of the business cycle movements of the GDP and prospective
components of the Composite Leading Indicator
Figure 2. Note: Cross correlegram of the business cycle movements of the GDP and prospective components of the
Composite Leading Indicator. Leads of the prospective component are on the horizontal line so that the correlegram
is between its past and the current GDP cycle.
In fact one can select those component series that have similar cyclical fluctuations with the
reference series using some criteria and form quick CLI’s at each lead, i.e. the component series
qualitatively predicting movements in the reference series. Table 2 provides a list of components
that have correlations of at least 70% with the reference series while Figure 3 illustrates CLI’s
based on them at each lead.
25
Lead time
Series name (and correlation, %)
Contem
-
poraneous
Import (78), M1 (73)
1 month
Import (82), M1 (78), M2 (75), Industrial loan (73)
2 months
Import (83), M2 (82), M1 (81), Wholesale loan
(77), FDI (77), Industrial loan (75), Electricity loan (74),
Real estate loan (73), Imported trucks (72)
3 months
M2 (86), M1 (83), Wholesale loan (83), FDI (81), Import (81), Real estate loan (77), Electricity loan (77),
Industrial loan (75), Imported trucks (73), Consumer loan (73)
4 months
M2 (89), Wholesale loan (86), FDI (83), M1 (82), Electricity loan (78), Real estate loan (77), Import (77),
Consumer loan (73), Imported trucks (72), Budget current income (71), Industrial loan (71), Total loan(70)
5 months
M2 (89), Wholesale loan (88), FDI (81), M1 (80), Electricity loan (77), Budget current income (76), Real
estate loan (72), Loan rate in MNT (72), Consumer loan (71) Import (70),
6 months
M2 (88), Wholesale loan (87),
Budget current income (7
9), M1 (77),
FDI (
76
),
Loan rate in MNT (74),
Electricity loan (74)
7 months
M2 (84), Wholesale loan (84), Budget current income (80), Loan rate in MNT (76), M1 (72)
8 months
Budget current income (79), M2 (79), Wholesale loan (78), Loan rate in MNT (7
6),
Table 2. Name of the series whose cross-correlations with the reference series are at least 70%.
Two series are selected for their close contemporaneous relationship with the GDP while budget
current income, M2, new wholesale and retail loan and loan rate in Mongolian togrog have the
strongest 8-month-leading relationship between the GDP. Although alternative methods could be
applied in the construction of the composite index the CLI’s shown in Figure 3 is based on
simple averages. The variables in the Table 2 appear to be effective in predicting recent
movements in the reference series both contemporaneously and in advance. It is important to
note that these components are by no means definitive. Their relevance must be re-evaluated in
terms of their predictive ability of turning points and should be reevaluated from time to time. In
other words inclusion in the above table does not guarantee a series to be included into the
formal CLI.
26
Figure 3. Mongolian GDP and quick CLI’s based on cyclical conformity at each lead.
B. Bry- Boschan algoritm of turning point detection
The Bry and Boschan (1971) technique for determining business cycle turning points (or
simplifications of it) is used by both the Conference Board and the OECD. It consists of consists
of the following steps9:
I. Determination of extremes and substitution of values.
II. Determination of cycles in 12-month moving average (extremes replaced).
A. Identification of points higher (or lower) than 5 months on either side.
B. Enforcement of alternation of turns by selecting highest of multiple peaks (or
lowest of multiple troughs).
III. Determination of corresponding turns in Spencer curve (extremes replaced).
A. Identification of highest (or lowest) value within ±5 months of selected turn in 12-
month moving average.
9 Since the methodology as a textbook can be freely downloaded from the NBER website I refrain from discussing it
in length.
27
B. Enforcement of minimum cycle duration of 15 months by eliminating lower peaks
and higher troughs of shorter cycles.
IV. Determination of corresponding turns in short-term moving average of 3 to 6 months,
depending on MCD (months of cyclical dominance).
A. Identification of highest (or lowest) value within ±5 months of selected turn in
Spencer curve.
V. Determination of turning points in unsmoothed series.
A. Identification of highest (or lowest) value within ±4 months, or MCD term,
whichever is larger, of selected turn in short-term moving average.
B. Elimination of turns within 6 months of beginning and end of series.
C. Elimination of peaks (or troughs) at both ends of series which are lower (or higher)
than values closer to end.
D. Elimination of cycles whose duration is less than 15 months.
E. Elimination of phases whose duration is less than 5 months.
VI. Statement of final turning points.
A turning-point in the CLI is generally expected to signal a turning-point in the business cycle in
6-9 months. However lead times sometimes fall outside of this range and turning points are not
always correctly identified, sometimes missing the reference series turning points and sometimes
flagging false turning points. The composition of the CLI should be so that to avoid such pitfalls
as much as possible. The Bry-Boschan procedure is applied to each series coming from the
analysis in Section 3 and the identified peaks and troughs are in Panels A and B of Table 3 in
index form: 1 is for April 2000 while 149 is for August 2012.
The first column in Panel A of the table contains peak dates in the GDP at 24, 51, 101
and 141 which translate into calendar dates of March 2002, June 2004, August 2008 and
December 2011. An ideal CLI component’s peak should have happened 6-8 months before these
dates. The first column in Panel B has trough dates at August 2001, November 2002, June 2006
and November 2010. An ideal component’s turning points should also have happened before
28
these dates. Visual illustrations of how well they are doing are provided in Figures 1-6 of the
Appendix 4.
A. Peak dates
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
24
46
13
17
10
32
48
50
29
20
24
28
47
19
27
11
16
43
10
29
7
45
15
34
14
34
51
98
57
48
36
62
95
94
95
40
49
48
76
50
42
66
53
81
28
61
48
74
52
74
43
49
101
116
74
73
61
102
135
138
116
94
82
80
102
98
105
88
86
115
50
92
76
96
100
99
94
90
141
142
100
95
101
122
136
97
95
118
136
125
116
77
132
97
138
132
133
132
115
119
140
135
137
99
131
138
133
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
53
28
22
51
6
67
49
16
18
13
18
28
46
35
8
32
23
52
14
47
39
27
20
26
35
36
85
55
60
76
67
110
95
38
58
40
44
44
90
69
24
61
50
96
36
67
71
52
56
51
56
70
104
101
112
109
89
129
57
97
67
80
79
117
95
95
89
70
136
54
94
104
88
98
98
96
100
121
104
133
108
122
122
135
115
115
96
93
135
135
141
135
118
130
134
133
138
136
135
140
136
135
B. Trough dates
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
17
24
5
31
28
27
20
17
78
10
15
15
32
13
33
40
29
8
21
21
40
13
39
21
27
5
32
59
43
61
47
50
70
79
108
28
43
36
66
43
87
76
62
57
41
39
64
64
81
44
68
44
75
106
66
79
87
86
112
112
126
81
62
65
85
72
131
111
105
103
65
77
83
87
110
83
108
66
128
126
88
105
110
115
123
88
87
108
111
136
127
83
110
107
105
107
123
108
124
134
111
109
108
129
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
4
19
36
14
32
22
12
30
12
24
34
16
20
15
17
10
11
35
24
41
21
10
27
17
41
56
62
44
90
59
81
82
57
49
31
58
65
36
60
62
56
41
33
78
48
56
54
45
86
42
66
80
90
85
134
90
102
133
108
74
76
84
104
66
104
77
107
81
60
110
72
75
91
69
120
75
112
109
126
120
130
132
116
112
114
100
109
122
107
79
112
110
116
110
110
136
125
110
126
Table 3. Business cycle turning point dates in the reference and prospective CLI component series (in indexes).
Numbers in bold represent each series, where 0 means the reference series while numbers 1-51 are those ordered in
Table 1. The numbers immediately below them (2- 6 numbers) represent the peak dates index: 1 is April 2000 while
149 is August 2012.
29
C. Length and consistency of the lead
Lead times are measured in months, reflecting the time that passes between turning points in the
component and reference series. Of course lead times vary from turning point to turning-point
but the aim is to construct leading indicators whose lead times are on average between 6 to 9
months and that have relatively small variances. To evaluate the length of leads, both mean and
median leads are used, because the mean lead on its own can be strongly affected by outliers.
The consistency of leads is measured by the standard deviation from the mean lead. One can see
from the graphs 1-6 in Appendix 4 how well each series perform in predicting turning points in
the reference series (GDP).
Based on the results of the turning point dates in Table 3 the system then identifies cases
where a turning point in the reference series preceded by a turning point in the components series
2- 8 months in advance. Because of the counter-cyclicality of some series here a peak in the
component series could precede a trough in the reference series and vice versa. If a component
series successfully predicts the turning points in such manner a certain proportion of the time
(currently it is 40%) then that series is further considered for inclusion in the CLI. It has to be
noted that as this proportion increases a fewer and fewer series will satisfy the restriction. On the
other hand a combination of component series, however imperfect, can result in a CLI that
perfectly predict turning points of the reference series. The quantity of diesel imported, M2, FDI,
total import, international gold price and new real estate loan pass this criteria (Numbers 3, 6, 9,
13, 21 and 48 in Table 3). It is interesting to note from this result and the ones in Table 2 that
although international gold price is not highly correlated with Mongolia’s business cycle its
turning points precede those of the business cycle.
Notice also that a casual observation of Table 3 reveals that none of them perfectly
predicts the turning points in business cycle. But the Composite Leading Indicator that is
constructed by simple averaging of these series has peaks at 19, 47, 96 and 136 (24, 51, 101 and
141 in the reference series) and troughs at 12, 30, 63 and 109 (17, 32, 75 and 128 in the reference
series). Figure 4 illustrates this final CLI (in continuous red line) together with the business cycle
(in blue dashes).
30
Figure 4. Mongolia’s CLI (continuous red line) and economic activity (blue dashes); (long term trend=100)
Cyclical conformity of the CLI with Mongolia’ business cycle is extraordinary fine. Figure
illustrates cross correlations between the CLI (contemporaneous at zero and leading at positive
horizontal value) and the business cycle of economic activity measured by the GDP. The peak
correlarion of 91% occurs at lead time of 4 months but it stays above 75% up to 7 months.
The CLI has four peaks and four troughs as can be seen from Figure 4. That is the same
number of turning points in Mongolia’s business cycle meaning that there are no missing or extra
cycles. The average lead time of the peak is 4.75 months while that of the trough is 9.5 months.
As can be seen from the Figure Mongolia’s economic activity reached its recent peak in
December 2011 and currently is proceeding towards its trough based on the second quarter of
2012 GDP growth. The CLI estimated using data up to August 2012 shows no turning points yet,
meaning that the economic situation will worsen at least for 9 months before improving. This
however may change if the copper production of Ouy Tolgoi alters completely the past
relationship among economic variables.
31
Figure 5. Note: Cross correlegram of the business cycle movements of the GDP and the Composite Leading
Indicator. Leads of the CLI are on the horizontal line so that the correlegram is between its past and the current GDP
cycle.
5. Implementation guideline
The system of Mongolia’s Composite Leading Indicator comes with two GAUSS run files and a
dozen sets of procedures that are capable of producing results in both Mongolian (using Latin
alphabet) and English languages. Two run files could be modified if and when the system is
adjusted, for example to add more prospective component series. The procedures should not be
modified unless there is a very good reason for it. If the procedures are to be modified internal
consistency of the system should be maintained. The main steps involved in the execution and
update of the CLI system are sketched below:
Step 1. It is necessary to update databases in MoF_CLI_database.xls (this contains all monthly
data), LoanReport.xls (these are quarterly data) and gdp2005SA_2000Aug.txt (that contains
Mongolian real GDP data since 2000 in 2005 prices, seasonally adjusted by the NSO) from the
sources detailed in Table 1 of Section 2. The updating process consists of adding the most recent
value of each variable to the file.
32
Step 2. Then CLI_processing.prg program is run. It performs analysis described in Section 3.
Some of the outputs of the program are reported in Appendices in 1-3. Whole output of the
program is written to an output file experiment_big.out and what the program is capable of
producing is detailed in Bataa et al. (2012b). Another output of the program is trends.out which
contains business cycle fluctuations of all the relevant series.
Step 3. Finally Mongolian_CLI.prg program is run. The program reads in trends.out carries out
the steps detailed in Section 4 and provides with the Mongolian CLI.
33
References
Andrews, D.W.K. (1991). 'Heteroskedasticity and autocorrelation consistent covariance matrix
estimation', Econometrica, Vol. 59, pp. 817–858.
Andrews, D.W.K. (1993). 'Tests for parameter instability and structural change with unknown change
point, Econometrica, Vol. 61, pp. 821-856.
Bai, J.S. and Perron, P. (1998). 'Estimating and testing linear models with multiple structural changes',
Econometrica, Vol. 66, pp. 47-78.
Bai, J.S. and Perron, P. (2003a). 'Computation and analysis of multiple structural change models', Journal
of Applied Econometrics, Vol. 18, pp. 1-22.
Bai, J.S. and Perron, P. (2003b). 'Critical values for multiple structural change tests', Econometrics
Journal, Vol. 6, pp. 72-78.
Bai, J.S. and Perron, P. (2006). 'Multiple structural change models: A simulation analysis', in D. Corbea,
S. Durlauf and B. E. Hansen (eds), Econometric Theory and Practice: Frontiers of Analysis and
Applied Research, Cambridge University Press, Cambridge.
Bataa, E., Osborn, D.R., Sensier, M. and van Dijk, D. (2012). 'Structural breaks in the international
dynamics of inflation', Review of Economics and Statistics, forthcoming.
Bataa, E., Osborn, D.R., Sensier, M. and van Dijk, D. (2012). 'Identifying Changes in Mean, Seasonality,
Persistence and Volatility for G7 and Euro Area Inflation', Oxford Bulletin of Economics and
Statistics, forthcoming.
Beziz P. and G. Petit (1997), “The 1994 Mexican Crisis: Were Signals Inadequate?” OECD,
Statistics Directorate.
Bikker J.A. and N.O. Kennedy (1999), “Composite Leading Indicators of Underlying Inflation
for Seven EU Countries,” Journal of Forecasting, 18, 225-258.
Bry G. and C. Boschan (1971), Cyclical Analysis of Time Series: Selected Procedures and
Computer Programs, NBER.
Burns A. and W. Mitchell (1946), Measuring Business Cycles, NBER.
Canova, F. and Hansen, B.E. (1995). 'Are seasonal patterns constant over time: A test for seasonal
stability', Journal of Business and Economic Statistics, Vol. 13, pp. 237-252.
34
Cullity J. And A. Banerji (1996), “Procedures for Constructing Composite Indexes: A Re-
Assessment,” OECD, mimeo.
Franses, P.H. and Haldrup, N. (1994). 'The effects of additive outliers on tests for unit
roots and cointegration', Journal of Business and Economic Statistics, Vol. 12, pp. 471-478.
Ghysels, E. and Perron, P. (1996). 'The effect of linear filters on dynamic time series with structural
change', Journal of Econometrics, Vol. 70, pp. 67-97.
Gómez, V. and Maravall, A. (1996). 'Programs TRAMO and SEATS: Instructions for the User', Working
Paper 9628, Bank of Spain.
Hansen, B.E. (2000). 'Testing for structural breaks in conditional models', Journal of Econometrics, Vol.
97, pp. 93-115.
Harvey, A.C. (1989). Forecasting, Structural Time Series Models and the Kalman Filter. Cambridge
University Press, Cambridge.
Hodrick R. and J. Prescott (1997), “Postwar U.S. Business Cycles: An Empirical Investigation,”
Journal of Money, Credit, and Banking, vol.29, No.1.
Levin, A.T. and Piger, J.M. (2004). 'Is inflation persistence intrinsic in industrial economies?', Working
Paper No. 334, European Central Bank.
Marques, C.R. (2005). 'Inflation persistence: Facts or artefacts?', Bank of Portugal Economic Bulletin,
Summer, pp. 69-79.
OECD (1987), OECD Leading Indicators and Business Cycles in Member Countries: 1960-
1985, Department of Economics and Statistics, Sources and Methods No. 39.
Pitarakis, J.-Y. (2004). 'Least squares estimation and tests of breaks in mean and variance under
misspecification', Econometrics Journal, Vol. 7, pp. 32-54.
Qu, Z.J. and Perron, P. (2007). 'Estimating and testing structural changes in multivariate regressions',
Econometrica, Vol. 75, pp. 459-502.
Stock, J.H. and Watson, M.W. (2003). 'Forecasting output and inflation: The role of asset prices', Journal
of Economic Literature, Vol. 41, pp. 788-829.
Tsay, R.S. (1988). 'Outliers, level shifts and variance changes in time series', Journal of Forecasting, Vol.
7, pp. 1-20.
White, H. (1980). 'A heteroskedasticity-consistent covariance matrix estimator and a direct test for
hetereskodasticity', Econometrica, Vol. 48, pp. 817-838.
35
Appendix 1. Summary results from the iterative decomposition of Section 3.B.
Figure 1. Notes: Panels show: a) observed time series, b) dynamic component, persistence (red line) and
volatility break dates (green vertical lines); c) regime means, d) deterministic seasonal component for regime 1
in blue, regime 2 in red and regime 3 in pink, e) outliers and f) correlogram for the estimated dynamic
component, with 95% confidence intervals shown as dashed lines. If relevant, the correlogram is shown over
sub-periods identified by dynamic breaks, with regime 1 in blue and regime 2 in red.
36
Figure 2. Notes: See Figure 1.
Figure 3. Notes: See Figure 1.
37
Figure 4. Notes: See Figure 1.
Figure 5. Notes: See Figure 1.
38
Figure 6. Notes: See Figure 1.
Figure 7. Notes: See Figure 1.
39
Figure 8. Notes: See Figure 1.
Figure 9. Notes: See Figure 1.
40
Figure 10. Notes: See Figure 1.
Figure 11. Notes: See Figure 1.
41
Figure 12. Notes: See Figure 1.
Figure 13. Notes: See Figure 1.
42
Figure 14. Notes: See Figure 1.
Figure 15. Notes: See Figure 1.
43
Figure 16. Notes: See Figure 1.
Figure 17. Notes: See Figure 1.
44
Figure 18. Notes: See Figure 1.
Figure 19. Notes: See Figure 1.
45
Figure 20. Notes: See Figure 1.
Figure 21. Notes: See Figure 1.
46
Figure 22. Notes: See Figure 1.
Figure 23. Notes: See Figure 1.
47
Figure 24. Notes: See Figure 1.
Figure 25. Notes: See Figure 1.
48
Figure 26. Notes: See Figure 1.
Figure 27. Notes: See Figure 1.
49
Figure 28. Notes: See Figure 1.
Figure 29. Notes: See Figure 1.
50
Figure 30. Notes: See Figure 1.
Figure 31. Notes: See Figure 1.
51
Figure 32. Notes: See Figure 1.
Figure 33. Notes: See Figure 1.
52
Figure 34. Notes: See Figure 1.
Figure 35. Notes: See Figure 1.
53
Figure 36. Notes: See Figure 1.
Figure 37. Notes: See Figure 1.
54
Figure 38. Notes: See Figure 1.
Figure 39. Notes: See Figure 1.
55
Figure 40. Notes: See Figure 1.
Figure 41. Notes: See Figure 1.
56
Figure 42. Notes: See Figure 1.
Figure 43. Notes: See Figure 1.
57
Figure 44. Notes: See Figure 1.
Figure 45. Notes: See Figure 1.
58
Figure 46. Notes: See Figure 1.
Figure 47. Notes: See Figure 1.
59
Figure 48. Notes: See Figure 1.
60
Appendix 2. Seasonality and outlier adjusted monthly time series and its long term trend
Figure 1. Note: Red dashed curves are the seasonality and outlier adjusted monthly time series. Their detailed description, coverage and source are
detailed in Table 1. Blue continuous lines are the long term trends obtained using the Hodrick and Prescott filter described in Section 3.C.
61
Figure 2. Note: See Figure 1.
62
Figure 3. Note: See Figure 1.
63
Figure 4. Note: See Figure 1.
64
Figure 5. Note: See Figure 1.
65
Figure 6. Note: See Figure 1.
66
Appendix 3. Cross correlegram of the business cycle movements of the GDP and prospective components of the Composite
Leading Indicator
Note: Cross correlegram of the business cycle movements of the GDP and prospective components of the Composite Leading Indicator. Leads of the prospective
component are on the horizontal line so that the correlegram is between its past and the current GDP cycle.
67
68
69
70
71
72
Appendix 4. Turning points in business cycle movements of the GDP and prospective components of the Composite Leading
Indicator
Figure 1. Business cycle movements of the GDP and prospective components of the Composite Leading Indicator along with their estimated turning points
determined using Bry and Boschan (1971) routine. The plotted are petrol import quantity, petrol import price, diesel import quantity, diesel import price,
consumer price index, M2, M1 and Top20 stock market index.
73
Figure 2. See Notes to Figure 1. The plotted are Foreign Direct Investments, total foreign trade, export, export to China, import, loan to deposit ratio, coal
production and copper production.
74
Figure 3. See Notes to Figure 1. The plotted are molybdenum production, export copper concentrate price, export molybdenum concentrate price, international
copper price, international gold price, international coal price, international oil price and General government current income.
75
Figure 4. See Notes to Figure 1. The plotted are General government current expenditure, General government capital expenditure, Central Bank bill rate,
commercial banks average MNT loan rate, commercial banks average foreign exchange loan rate, USD exchange rate (MNT/USD), Chinese Yuan exchange rate
and imported tyre quantity.
76
Figure 5. See Notes to Figure 1. The plotted are import tyre price, import truck quantity, import truck price, net trade (export minus import), export to import
ratio, Chinese CLI (OECD), Total new loan and new agricultural loan.
77
Figure 6. See Notes to Figure 1. The plotted are new mining loan, industrial loan, loans to electricity generation, consumer loan, wholesale and retail loan, new
transport loan, new loans to financial sector and new real estate loan.
78
Figure 6. See Notes to Figure 1. The plotted are new loans to administrative service sector, new loans to health sector and new consumer loan.
79
Appendix 5. Current components in the CLI of the World countries
Australia (OECD)
Dwelling permits issued (number)
Orders inflow (manufact.): tendency (bs) (% balance)
Production (manufacturing): tendency (bs) (% balance)
Employment (manufacturing): tendency (bs) (% balance)
Share price index (all industrials) (2000=100)
Terms of trade (2000=100)
Yield 10-year commonwealth government bonds (% per
annum) Inverted
Chile (
OECD
)
Production of copper sa (tones)
Consumer confidence indicator – Retail trade sa (% balance)
Production (manufacturing): tendency sa (% balance)
Monetary aggregate: M1 sa (CLP)
Share prices (2005=100)
Net trade (f.o.b. - c.i.f.) sa (CLP)
Australia (Conference Board)
building approvals,
rural goods exports
money supply
stock prices
yield spread
sales to inventories ratio
gross operating surplus
Canada (OECD)
Deflated money supply (m1) sa (1995 cad)
Housing starts large cities sa (number)
USA business climate indicator (pmi) sa (normal=50)
Consumer confidence sa (2000y)
Spread of interest rates (% a.r.)
Ratio of inventories to shipments (ratio), inverted
Share prices (S&O/TSX composite index) (2005y)
Czech (OECD)
BOP Capital account, debit (czk)
Demand evolution (Services): future tendency (% balance)
Production (Manufact.): tendency (%)
CPI Harmonised All items inverted
Consumer confidence indicator (% balance)
ITS Exports f.o.b. total
Share prices: PX-50 index (2005=100)
Austria (OECD)
Production: future tend. (manufact.) (% balance)
Order books: level (manufacturing) (% balance)
Ifo business climate indicator for Germany (normal=100)
Consumer confidence indicator (% balance)
Unfilled job vacancies (persons)
Spread of interest rates (% per annum)
Belgium (OECD)
New passenger car registrations (number)
Employment (manufacturing): future tend. (% balance)
Export orders inflow (manufact.): tendency (% balance)
Demand (manufact.): future tendency(% balance)
Production (manufacturing): tendency(% balance)
Consumer confidence indicator(% balance)
Denmark (OECD)
Total volume of retail sales (2000=100)
New passenger car registrations (number)
Employment: future tendency (manufact.) (% balance)
Production: future tendency (manufact.)(% balance)
Official discount rate (% per annum), inverted
Deflated money supply m1 (dkk)
Petrol exports deflated by consumer price index (dkk)
Consumer confidence indicator (% balance)
Estonia (OECD)
Manufacturing - Export order books: level sa (% balance)
Total retail trade (Value) sa (2005=100)
Passenger car registrations sa (2005=100)
CPI Food excl. Restaurants (2005=100) inverted
Share prices: OMX Tallin index (2005=100
ITS Net trade (f.o.b. - c.i.f.) (Eur)
Finland (OECD)
CPI All items (2005=100), Inverted
Consumer confidence indicator(% balance)
Spread of interest rates (% a.r.)
Production tendency (manufacturing)(% balance)
PPI Total (2005=100)
Finished goods stocks (manufacturing)(% balance) inverted
Share prices (HEX All Share index) (2005=100)
France (OECD)
New passenger car registrations (number)
Consumer confidence indicator (% balance)
Production: future tendency (manufact.) (% balance)
SBF 250 share price index (2005=100)
CPI Harmonised All items (2005=100) inverted
Export order books: level (manufact.) (% balance)
Selling prices: future tendency (Construction) (% balance)
its issued for dwellings (2005=100)
PermExpected level of life in France (CS) (% balance)
France (Conference Board)
yield spread
building permits (residential)
inverted new unemployment claims
industrial new orders
production expectations
stock price index
ratio of the deflator of manufacturing value added to unit
labor cost in manufacturing.
Germany (OECD)
Germany (Conference Board)
80
Ifo business climate indicator (normal=100)
Orders inflow/demand: tendency (manufact.) (% balance)
Export order books : level (manufacturing) (% balance)
Total new orders (manufacturing) (2000 = 100)
Finished goods stocks: level (manufacturing)(% balance)
inverted
Spread of interest rates (% per annum)
yield spread
inventory change
gross enterprises and properties income stock prices new
residential construction orders
new orders in investment goods industries
consumer confidence.
Greece (OECD)
Cost of residential construction (2005=100), inverted
Bank credit to the manufact. sector deflated(eur 2001)
Services - employment: future tendency (% balance)
Production: future tendency (% balance)
Production of manufactured non-durable consumer goods
(2000y)
Volume of retail sales sa (2000y)
Retail trade - volume of stocks: level (% balance), inverted
Wholesale prices: all items (2005=100), inverted
Hungary (OECD)
Production (manufacturing): future tendency (% balance)
Unemployement registered (numbers)
Monthly hours of work in manufacturing (hours)
Money supply m1 (huf)
Share prices: budapest stock exchange (2000=100)
Central bank base rate (% per annum)
Total imports c.i.f (huf)
Ireland (OECD)
Value of exports to Northern Ireland (EUR)
Exports: agricultural products to other EU Member States
other than UK (EUR)
Passenger cars registration (2005=100)
Total PPI mining and quarrying activities (2005=100)
ISEQ share price index (2005=100)
Reel effective exchange rate based on CPI (2005=100),
inverted
Money supply M2 to Euro area (EUR)
Israel (OECD)
Business confidence - all businesses - Expected net balance
sa
Total retail trade (Volume) (2005=100)
Domestic PPI Manufacturing (2005=100)
Share prices: The TA-Composite Index (2005=100)
Tourism - Total Departures of Israelis sa (number)
Exports in manufacturing - diamonds
Italy (OECD)
Component Series (Unit)
Consumer confidence indicator (% balance)
Production: future tendency (manufacturing) (% balance)
Deflated net new orders (2005 = 100)
Order books: level (manufacturing) (% balance)
CPI All items (2005=10) inverted
Imports from Germany Cif (USD)
Luxemburg (OECD)
OECD CLI for Germany (trend restored)
OECD CLI for Belgium (trend restored)
Japan (OEC
D)
Inventories to shipments ratio (mining and manufacturing)
(2005=100) Inverted
Ratio imports to exports (2000=100),
Ratio loans to deposits (%) Inverted
Monthly overtime hours (manufacturing) (2000=100)
Construction: dwellings started (2000=100)
Share price index (TOPIX) Tokyo (2000=100)
Spread of interest rates (% annual rate)
Small business survey: Sales tendency (% balance)
Japan (Conference Board)
dwelling units started
interest rate spread
real money supply
Tankan business conditions survey
index of overtime worked
six-month growth rate of labor productivity
stock prices,
(inverted) business failures
real operating profits
new orders for machinery and construction
Korea (OECD)
Business situation (manufact.): future tendency (%
balance)
Share prices KOSPI index
Stocks of total investment manufactured goods
(volume)Inverted
Inventory circulation indicator (manufacturing)
Interest rate spread (3 year treasury bonds less overnight
rate)
Net Barter Terms of trade (2005=100) sa
Korea (Conference Board)
Stock Prices
Value of Machinery Orders
Letter of Credit Arrivals
Index of Shipments to Inventories
Export FOB
Yield of Government Public Bonds
Private Construction Orders
81
Mexico (OECD)
Monthly changes in manufacturing employment (%)
Employment: tendency (manufact.) (BS) (% balance)
Finished goods stocks: tendency (mfc.) (BS) (% balance)
Inverted
Production: tendency (manufacturing) (BS) (% balance)
Yield >10-year US federal government securities
(composite) (% per annum) inverted
Cost managing deposits for banks (% per annum) inverted
Real effective exchange rate (2000=100)
Mexico (Conference Board)
stock prices
net insufficient inventories
US refiners’ acquisition cost of domestic and imported crude
oil
(inverted) real exchange rate
industrial production construction
(inverted) federal funds rate
Netherlands (OECD)
Order books: level (manufacturing) (% balance)
Production: future tendency (manufact.) (% balance)
Finished goods stocks: level (manufact. (bs) (% balance)
inverted
Orders inflow: tendency (manufact.) (% balance)
Ifo business climate indicator for Germany (normal=100)
Share prices: total index (2005=100)
New Zealand (OECD)
Business situation (manufac.): future tend. (% balance)
Consumer confidence indicator (% balance)
Total retail sales (value) (nzd)
Unemployed persons less than 1 month (persons) inverted
Monetary aggregate m1 (2005=100)
Yield of 90-day bank bills (% per annum) inverted
Norway (OECD)
Exports to UK (USD)
Stocks of orders for exports (manuf., mining, quarrying)
(% balance)
Production (manuf.): tendency (% balance)
General judgement of the outlook for the enterprise in next
quarter (manuf., mining, quarrying) (% balance)
CPI All items (2005-100)
Share price index (industrials) oslo (2005=100)
Poland (OECD)
Real effective exchange rate (2000=100) Inverted
Interest rate: 3-month wibor (% per annum) Inverted
Production (manufacturing): tendency (% balance)
Unfilled job vacancies (number)
Production of coal (tonnes)
Portugal (OECD)
Industrial production:electricity, gas & water (2005=100)
Production:future tendency (manufact.) (% balance)
Order books/demand:level (manufact.) (% balance)
Export order books/demand: tendency (mfk) (% balance)
Share prices: BVL general share price index
Unfilled job vacancies (number)
Slovak (OECD
)
Confidence indicator (Retail trade) (% balance)
Total retail trade (Volume) (2005=100)
Expected economic situation (CS Consumer) (% balance)
Share prices: SAX index
Imports f.o.b. total (USD)
Slovenia (OECD)
Manufacture Basic Metals sa (2005=100)
Manufacturing - Order books: level sa (bs) (% balance)
Production: tendency mfg. sa (bs) (% balance)
CPI All items (2005=100) inverted
EUR/USD exchange rate monthly average
Expected economic situation sa
Spain (OECD)
Rate of capacity utilisation (BTS manufact.) (% balance)
Production of total construction (2005=100)
CPI Services less housing inverted (2005=100)
Share prices: IGBM general index (2005=100)
Passenger car registrations (2005=100)
Sweden (OECD)
5-year government bonds yields (% p.a.) inverted
Overtime hours worked mining & mfg. (%)
new orders mining & mfg. sa (2005=100)
Order books: level (% balance)
Finished goods stocks: level (% balance) inverted
AFGX share price index sa (2005=100)
Switzerland (OECD)
Finished goods stocks: level (manufact.) (% balance)
inverted
Orders inflow: tendency (manufact.) (% balance)
Production: tendency (manufacturing) (% balance)
UBS 100 share price index (2000=100)
Consumer surveys: expected economic situation sa
Silver prices CHF/kj
Turkey (OECD)
Production amount of electricity (gw hours)
Finished goods stocks: level (manufacturing) (% balance)
inverted
new orders from domestic market: future tendency
(manufacturing) (% balance)
Employment: future tendency (manufact.)(% balance)
UK (OECD)
Business climate indicator (% balance)
new car registrations sa (number)
Consumer confidence indicator sa (% balance)
3-month eligible bank bills (% p.a.) inverted
Production: future tendency (% balance)
Finished goods stocks: tendency (% balance) Inverted
82
Prospects for exports (manufacturing) (
bs) (% balance)
Discounted treasury auction interest rate (% per annum)
inverted
Imports of intermediate goods (usd)
FTSE
-
A non financial share price index (2000y)
USA (OECD)
Dwellings started (number)
Net new orders for durable goods (us dollar - million)
Share prices: NYSE composite (2005=100)
Consumer sentiment indicator (normal = 100)
Weekly hours of work : manufacturing (hours)
Purchasing managers index (BS) (% balance)
Spread of interest rates (% per annum)
USA (Conference Board)
Average weekly hours, manufacturing
Average weekly initial claims for unemployment insurance
Manufacturers’ new orders, consumer goods and materials
ISM Index of New Orders
Manufacturers' new orders, nondefense capital goods
excluding aircraft orders
Building permits, new private housing units
Stock prices, 500 common stocks
Leading Credit Index
Interest rate spread, 10-year Treasury bonds less federal
funds
Average consumer expectations for business conditions
Brazil (OECD)
Share price index: all shares (2005=100)
Manufacturing - Production: future tendency (BS) (%
balance))
Manufacturing - Order books: level (BS) (% balance)
Monetary Aggregate M2 inverted
Discount rate inverted
Net Trade (f.o.b. - f.o.b.) with EU
India (OECD)
Industrial production of durable goods (2005=100)
Prodcution of manufactured non metalic mineral products
(2005=100)
Passenger car sales (number)
Monetary aggregate m1 (inr)
Share prices: BSE dollex (2005=100)
call money rate (% per annum) inverted
China (OECD)
Production of chemical fertilizer (tonnes)
Monetary aggregate m2 (cny)
Production of manufactured crude steel (tonnes)
5000 Industrial Enterprises: Diffusion Index: Overseas
order level (%)
Production of buildings (m2)
Production of motor vehicles (number)
Shanghai Stock Exchange: Turnover (cny)
China (Conference Board)
Total Loans Issued by Financial Institutions Enterprises
Diffusion Index
Raw Materials Supply Index
NBS Manufacturing PMI Sub-Indices: PMI Supplier
Deliveries
Consumer Expectations Index
Total Floor Space Started
NBS Manufacturing PMI Sub-Indices: Export Orders
Indonesia (OECD)
Share prices: jsx index (2005=100)
Central Bank Discount rate (% per annum) inverted
IDR/USD exchange rate (end of period) (idr/usd) inverted
Producer Price Index/Wholesale Price Index (2005=100)
inverted
Consumer confidence index (normal = 100)
Production of paper and paper products (2000=100)
Production of paper and paper products (2005=100)
Russia (OECD)
US imports from Russia inverted
Order books: level (BS) (% balance)
Share prices: rts index (2005=100)
World market price of crude oil (2005=100)
Production: trend observed in recent month (BS) (%
balance)
Assessment of export order books: present level (BS) (%
balance)
South Africa (OECD)
Orders inflow (manufacturing): tendency (% balance)
Industrial confidence indicator (% balance)
Permits issued: dwellings (2005=100)
Sales of motor cars (2005=100)
ShaShare prices: ftse/jse index (2005=100)
Spread of interest rate (% per annum)
Euro Area (Conference Board)
Economic Sentiment Index
Index of Residential Building Permits Granted
EURO STOXX Index
Money Supply (M2)
Interest Rate Spread
Eurozone Manufacturing Purchasing Managers’ Index
Eurozone Service Sector Future Business Activity
Expectations Index
Source: OECD and Conference Board.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
This paper proposes an iterative procedure to discriminate between structural breaks in the coefficients and the disturbance covariance matrix of a system of equations, with recursive procedures then identifying individual coefficient shifts and separating volatility from correlation breaks. Structural breaks in short-term cross-country inflation relations are then examined for major G-7 economies and within the Euro area. There is evidence that the Euro area leads inflation in North America, while changing short-term interactions apply within the Euro area. Co-variability generally increases from the late 1990s, while Euro area countries move from essentially idiosyncratic contemporaneous variation to co-movement in the 1980s.JEL classifications: C32, E31
Article
In a recent paper, Bai and Perron (1998) considered theoretical issues related to the limiting distribution of estimators and test statistics in the linear model with multiple structural changes. In this companion paper, we consider practical issues for the empirical applications of the procedures. We first address the problem of estimation of the break dates and present an efficient algorithm to obtain global minimizers of the sum of squared residuals. This algorithm is based on the principle of dynamic programming and requires at most least-squares operations of order OOT 2 for any number of breaks. Our method can be applied to both pure and partial structural change models. Second, we consider the problem of forming confidence intervals for the break dates under various hypotheses about the structure of the data and the errors across segments. Third, we address the issue of testing for structural changes under very general conditions on the data and the errors. Fourth, we address the issue of estimating the number of breaks. Finally, a few empirical applications are presented to illustrate the usefulness of the procedures. All methods discussed are implemented in a GAUSS program.
Article
“Are economists tempted to sit overmuch in offices and think, supported by basic principles, rather than trudge the often privately unrewarding empirical highways and byways?”. This question raised in a general context by Theodore Morgan (see [1]) is worth examining in the particular case of the 1994 Mexican crisis. The purpose of this article is to show that statistical data available at that time would have given discerning economists early signals of the crisis. More precisely, using a practical forecasting tool called “leading indicators”, we show that such signals existed in a number of forms. Composite leading indicators are aggregate time series which show a leading relationship with the growth cycles of key macro-economic indicators. Typically, they are constructed to anticipate the cycles of Total industrial production or Gross domestic product (GDP), which are chosen as reference series for the aggregate economy. Composite leading indicators are calculated by combining component time series. Component series cover a wide
Article
This paper presents short- and long-term composite leading indicators (CLIs) of underlying inflation for seven EU countries, namely Belgium, Germany, France, Italy, the Netherlands, Sweden and the UK. CLI and CPI reference series are calculated in terms of both growth rates and in deviations from its trend. The composite leading indicators are based on leading basic series, such as sources of inflation, series containing information on inflation expectations and prices of intermediate goods and services. Neftci's decision rule approach has been applied to transfer movements in the CLIs into a measure of the probability of a cyclical turning point, which enables the screening out of false turning point predictions. Finally, CLIs have been used to analyse the international coherence of price cycles. The forecast performance of CLIs of inflation over the past raises hope that this forecast instrument can be useful in predicting future price movements. Copyright © 1999 John Wiley & Sons, Ltd.
Article
Outliers, level shifts, and variance changes are commonplace in applied time series analysis. However, their existence is often ignored and their impact is overlooked, for the lack of simple and useful methods to detect and handle those extraordinary events. The problem of detecting outliers, level shifts, and variance changes in a univariate time series is considered. The methods employed are extremely simple yet useful. Only the least squares techniques and residual variance ratios are used. The effectiveness of these simple methods is demonstrated by analysing three real data sets.