Munich Personal RePEc Archive

Rethinking error correction model in

macroeconometric analysis : A relevant
review

PINSHI, Christian

University of Kinshasa

January 2020

Online at https://mpra.ub.uni-muenchen.de/98322/
MPRA Paper No. 98322, posted 27 Jan 2020 05:05 UTC
Rethinking error correction model in macroeconometric analysis :
A relevant review

Christian P. Pinshi1

Abstract
The cointégration methodology has bridged the growing gap between economists and
econometricians in understanding dynamics, equilibrium and bias on the reliability of
macroeconomic and financial analysis, which is subject to non-stationary behavior. This
paper proposes a comprehensive literature review on the relevance of the error
correction model. Econometricians and economists have shown that error-correction
model is a powerful machine that provides the economic system and macroeconomic
policy with a refinement in the econometric results2.

Keys words : Cointegration, Error correction model, Macroeconomics

JEL Classification : C32, E0

1
University of Kinshasa, Researcher, christian.pinshi@unikin.ac.cd

2
I am so indebted to Jennifer Louise Castle for many helpful conversations and comments that helped
further refine and scrutinize this research paper. I also thank Valerio Scalone.

1
1| Introduction
The advent of time series analysis in econometrics3 and economics has transformed
economic thinking (especially macroeconomic thinking, which is why the time series are
called “macroeconometrics4”), Sophisticated processes radically transformed the
landscape of research sector and added rigor in macroeconomic analysis. Although its birth
is the result of the great battle between keynesians and monetarists (Johnston and
Dinardo, 1999), the analysis of time series is therefore at the heart of macroeconomics and
has emerged as the essential tool of the economic policy assessment.
The pioneering work of Box and Jenkins (1976) was based on the ARIMA (pdq)5 Auto
Regressive Integrated Moving Average, introduced to model process behavior based on past
values subjected to random shocks over time. A random event called noise or disturbance
affects the temporal behavior of this process and thus modifies time series6 values. This
model develops the forecast by exploiting statistical characteristics (mean, variance,
autocorrelation function, autocovariance function, ...). To fill the shortcomings of the
univariate models (ARIMA, ...), which only described the behavior of a series, not to
explain, the analyzes of the multivariate time series were born, of which the most known
and used is the VAR family. The father of this process is Christopher Sims, Nobel Prize
winner in 2011.
It is clear that these models had become the cornerstone of any macroeconomic analysis.
However, these are analyzes that require the stationarity of the series, which, its mean and
its variance must be constant over time. For more understanding, a process is said to be
stationary if it tends to return to equilibrium (its mean value or variance) after suffering the
effect of a shock over time (mainly over long periods). However, in economics several
phenomena make that macroeconomic variables over time have non-stationary
characteristics, such as, GDP, exchange rate, inflation, stock prices and so on, and the price
to pay is to apply the difference filter or the transformation by the regression on the trend.
This could have the consequence of moving away from reality and proposing strategies
and policies based on erroneous or unreal results.

3
The study of time series is a discipline that appeared relatively before econometrics, since already around
1905 they were used in astronomy and a little further in statistics and meteorology. Econometrics, on the
other hand, is a discipline that was born around the 1930s by the Alfred Cowles Research Institute called the
Cowles Commission and the learned econometrics society founded by Ragnar Frisch and his colleagues
(Fisher, Ross , Schumpeter…).
4
According to Greene, macroeconometrics is a discipline that focuses on the analysis of time series that are
typically aggregates such as GDP, money supply, prices, exchange rate, investment, and so on. (2011).
5
Where p is the order of the autoregressive process AR (p), d the degree of integration of a process I (d),
and q the order of the moving average MA (q)
6
This model is generally in the following form : 𝑍𝑡 = ∆𝑑 𝑋𝑡 ; it is a development of the form : ∆𝑑 𝑋𝑡 = 𝛾 +
∅1 ∆𝑑 𝑋𝑡−1 + ⋯ + ∅𝑝 ∆𝑑 𝑋𝑡−𝑝 + 𝜀𝑡 − 𝜃1 𝜀𝑡 − ⋯ − 𝜃𝑞 𝜀𝑡−𝑞 . For more details see Lardic and Mignon (2002).

2
In view of all these difficulties, econometricians refined their research by developing non-
stationary time series analyzes to fit the data, to forecast macroeconomic and financial
series and to apply them to retroactive control systems. Thus the so-called ARCH models
(AutoRegressive Conditionally Heteroscedastics) and the methodology of cointegration
with some models (error-correction model, vector error correction model, and so on.)
were born. These models are relevant because of their closeness to reality and their
powers to produce better short-term forecasts and certainly long-term forecasts
aggregated in economically meaningful ways for macroeconomic policy analysis (Maddala
and In-Moo Kim (1998).
As Granger (1986) put it: “A test for cointegration can thus be thought of as a pre-test to
avoid ‘spurious regression’ situations”. According to Granger, instead of stationaryizing the
series a priori in order to avoid the fallacious regression situation, the best approach would
be to test whether the regression residuals are stationary, so the error-correction model
can be estimated with non-stationary series and give better results in the dynamics of the
short and long-term relationship. Cointegration is the key word in the new econometrics,
referring to the long-term relationship between economic variables.
This paper is intended to illustrate the relevance of the cointegration methodology to the
error correction model, its implications for macroeconomic modeling and forecasting, and
its fundamental role in explaining short- and long-term dynamics.
The organization of this paper is as follows. Section 2 reviews a selection of empirical
studies with the error-correction model contributing to the economic analysis. Section 3
illustrates the cointegration methodology by exploiting the different cointegration tests
and the dynamics of the error correction model. Finally section 4 summarizes the paper
with a conclusion.

2| Relevant literature Review

In 2003, the Nobel Prize in Economics was awarded to two researchers who conducted
their research in the 1980s and 1990s: Robert Engle and Clive Granger for their scientific
contributions to “methods of analysis of economic time series on the one hand with seasonal
volatility for Engle and on the other hand with a common trend for Granger” that have
contributed to the improvement of forecasts of macroeconomic and financial variables
(Lardic and Mignon, 2003). In fact, the winners were awarded for their work relating
respectively to ARCH type models and non-stationary so-called cointegration analysis. The
great merit of Granger was to show that specific combinations of non-stationary time
series can behave “stationarily” and thus make it possible to find statistically correct
results. Granger discovered cointegration by trying to refute Hendry’s (1977) criticism of
his research with Newbold on nonsense regressions between nonstationary data (Granger
and Newbold, 1974; 1977). Although the initial estimation approach has been replaced by a
plethora of methods, the concept of cointegration has led to a fusion of analyzes of long-
term equilibrium relationships with empirical dynamic systems (Castle and Hendry, 2016).

3
However, cointegration could not have happened without Hendry's criticism. Hendry's role
is overlooked in the economic literature, while he is the essential link between Sargan's
(1964) work on the formulation of analysis of stationary time series with error correction
and that of Granger on spurious and nonsense regressions, which will lead to error
correction models. This link will be made by the cointegration analysis (Meuriot, 2015).
These models have been shown to be very effective for short-term dynamic systems and
subject to strong stochastic disturbances, but whose long-term dynamics are also
constrained by existing equilibrium relationships in an economy, for example, the
relationship between exchange rate and inflation rate, both short-term and long-term
(which will be analyzed for illustrative purposes in the next section).
The pioneering work of Davidson and al (1978) on the dynamic relationship between
consumption and disposable income in the United Kingdom is a treating essay on the error
correction model. The literature on the relevance of cointegration on the economic
phenomenon has spectacularly exploded. The work of Engle and Granger (1987) that
cointegration and the error-correction model is a relevant model for analyzing the
relationship between nominal GDP and money supply (verification of monetary neutrality
) and other variables such as inflation, consumption. They have developed and
demonstrated the importance of the error-correction model in economic policy analysis.
The revival of the debate on monetary neutrality led other researchers to test this
relationship with the relevance of the error-correction model, Mehra economist at the
Federal Reserve of Richmond, verified this hypothesis in 1989 using cointegration, its
results validated monetary neutrality hypothesis with the broad money. However, the
hypothesis seemed to be invalid for money (M1).
Kremers and al. (1992) analyzed the power of cointegration by showing that the error-
correction model gives more efficient results. They found that when there is a
cointegration relationship, the error-correction model is usually more powerful. Several
empirical studies of money demand demonstrate this power of the error-correction model
and its strategic implications for monetary policy making (Hendry and Ericsson 1991, Mehra
1991).
After studies on monetary neutrality and money demand, the tendency was towards
efficient financial markets (Van Quang, 2007) and exchange rate behavior. Godbout and
Van Norden (1997) conduct three case studies. The first on cointegration and the projection
of nominal exchange rates. The second is based on work related to the long-term validity
of the monetary model for determining the exchange rate. The third study presents the
existence of stochastic trends common to international stock markets.
Eslamloueyan and Darvishi (2007) used an unrestricted error correction model and the test
approach of limits proposed by Pesaran and al (2001) to study the short and long-term
effects of bank credit on inflation in Iran. Their result indicates that there is a long-term
relationship between inflation and its main determinants, namely bank credit, import price,
real GNP and the parallel exchange rate. However, bank credit has no short-term effect on
the movement of price levels in Iran. In addition, they show that the nationalization of

4
banks and the implementation of interest-free or interest-free banking system in Iran have
caused a structural change in the behavior of inflation.
Researchers investigated the behavior of long-term growth rate determinants using the
error-correction model (Morales, 1998 ; Özmen and Şanli, 2018). A major empirical interest
in growth studies is whether permanent changes in the fundamentals of the economy
affect the long-run growth rate. However, a direct time series analysis of this hypothesis is
not always feasible because the permanence of many such changes is rather debatable.
For example, Lau (2008) explains why examining the long-term effects of temporary
changes in the share of investment on per capita output indirectly provides the answer to
the effects of permanent (possibly hypothetical) investment changes. . Applying the error-
correction model, he finds that a disruption in investment does not produce a positive long-
term effect in each of the three countries - France, Japan and the United Kingdom - in which
GDP per capita and investment are cointegrated.
Adouka and al. (2013) modeled the Algerian public expenditure demand function using
error correction and vector error correction (VECM) models from 1970 to 2010. They sought
to study the sensitivity of the economic activity in the face of changes in public spending
and to measure the effect of income and productivity on the growth of public spending.
They found that all the coefficients of the variables that explain the growth of public
expenditures are not significant and that there is therefore no short-term relationship
between public expenditure and GDP. But in the long run they captured the effect of
spending on activity, and thus the relationship was stable and significant in the long run.
Pinshi and Sungani (2018) analyze the relevance of the pass-through effect of the exchange
rate in the DRC and its implications on monetary policy regime for the period from January
2002 to March 2017. The main idea is to measure the degree of transmission of exchange
rate variations to the change in the general price level in a context of macroeconomic
instability that is unfavorable to the Congolese economy. Indeed, a strong and / or weak
degree of pass-through would suggest that changes in the exchange rate have more / less
effect on inflation. This could alter the central bank's predictions of the future reaction of
inflation, which are decisive for monetary policy strategies and tactics. Based on the
cointegration approach with the error-correction model, the main conclusion is that a
change in the exchange rate will affect inflation more than proportionally, the degree of
pass-through being relatively high. A depreciation of 1% causes a rise in the general price
level of 0.38% in the short term. This effect is even wider in the long run, where the increase
in the general price level is 1.66%. In addition, the adjustment to balance will take time (12
months and 2 weeks). Their study suggests Central Bank, on the one hand, to be vigilant
and closely monitor exchange rate movements in order to take quick action and contain
inflationary pressures and secondly, reflect on the strategies of the political economy by
adopting a hybrid regime (monetary targeting and implicit and flexible targeting of the
exchange rate).
In 2019, Ntungila and Pinshi analyze the short and long-term sensitivity of the Congolese
economy to fluctuations in commodity prices and verify the resource curse hypothesis in
the DRC. They use the method of Fully modified least squares (FM-OLS) to estimate the

5
error-correction model. They find that the Congolese economy is adversely affected by
commodity price shocks in the short and long term. The readjustment of the economy is
slow and persistent. The short and long term relationship seems to validate the hypothesis
(or paradox) of resource curse. They conclude that if there is not an ambitious launch of
the structural reform process, the economy would remain in an eternal whirlwind of curse.
The authors have contributed in some way to the research landscape and to
macroeconomic and financial understanding through the powerful error correction model
and the cointegration approach. The relevance of this model to the problem of fallacious
regression is a remarkable advance. What is extraordinary is that these econometricians
know how to analyze macroeconomics and economic policy in a world characterized by
the non-stationarity of variables, many have circumvented this problem with a price to pay
and a risk of deviating from the real analysis, Engle, Granger and the other econometricians
exploited this econometric weakness and were able to analyze against the current in the
storm and swirl the error by using non-stationary variables to find relevant results.

3| Methodology : Relationship between inflation and nominal exchange rate

Consider two time series 𝜋𝑡 and 𝑒𝑡 , inflation rate and exchange rate, which are 𝐼(𝑑), that
is, they have compatible properties in the long term. In general, any linear combination of
𝜋𝑡 and 𝑒𝑡 will also be 𝐼(𝑑). However, if there is a linear combination such that
𝜋𝑡 − 𝜃𝑒𝑡 − 𝛼 = 𝑧𝑡 ≈ 𝐼(𝑑 − 𝑏), 𝑏 > 0 (1)
With (1 − 𝛼 − 𝜃) called cointegration vector. The relationship between inflation and the
exchange rate is cointegrated in the sense of Engle and Granger (𝜋𝑡 , 𝑒𝑡 ≈ 𝐶𝐼(𝑑, 𝑏)).
The concept of cointegration attempts to determine the existence of a long-run
equilibrium towards which an economic system converges over time. If, for example,
economic theory suggests the following long-term relationship between 𝜋𝑡 and 𝑒𝑡 in
logarithm
𝜋𝑡 = 𝛼 + 𝜃 𝑒𝑡 + 𝑧𝑡 (2)

Where 𝜃 is an elasticity measuring the effect of a unit change in the exchange rate 𝑒𝑡 on
the inflation rate 𝜋𝑡 . This relation defines the behavior of inflation is a function of the
fluctuations of the exchange rate. Thus 𝑧𝑡 is the distance at which the system is far from
equilibrium at all times, that is, the equilibrium error (Dolado and al., 1990).
The statistical significance of the cointegrating coefficient θ is an indication of the
existence of a long-term relationship between the rate of inflation and the exchange rate
and that these have a common stochastic tendency whose fundamental characteristic is
that the term estimated residual 𝑧̂𝑡 does not have a unit root (stationarity). However, this
cointegrated relationship requires that each of two variables is not stationary in level, but
that they become them after differential filtering (Pinshi and Sungani, 2018).

6
3|1 Integration tests
The aim of the integration or stationarity test is to examine empirically whether each series
contains a unit root. These tests are mainly a descriptive tool used to classify the series into
stationary and non-stationary. Since the integrated variables lead to nonstandard
distributions and perhaps to fallacious regression results, the recommendation is as
follows : “If a data set appears to be non-stationary, assume that it is nonstationary and
integrated . Once you have been able to categorize your variables as integrated steady-state
trends, you are able to solve the long-term and short-term effects in your model” (Sjö, 2008).
The « t statistic » of Dickey-Fuller is based on the model estimation
𝛥𝜋𝑡 = 𝛼 + 𝛽𝑡 + 𝜗𝜋𝑡−1 + 𝜈𝑡 (3)
In case of autocorrelation in the observed series, estimate the Augmented Dickey-Fuller
(ADF Test) based on the following equation estimate:

𝛥𝜋𝑡 = 𝛼 + 𝛽𝑡 + 𝜗𝜋𝑡−1 + 𝛼𝜋𝑡−1 + ∑𝑝−1

𝑗=1 𝜙𝑗 𝛥𝜋𝑡−𝑗 𝜈𝑡 (4)

The null hypothesis is that 𝜋𝑡 = 𝜋𝑡−1 + 𝜈𝑡 where 𝜈𝑡 ≈ 𝑁𝐼𝐷(0, 𝜎 2 ). According to the null
hypothesis, 𝜗 will be negatively biased in a limited sample, so only one test is necessary to
determine 𝐻0 : 𝜗 = 0[𝜋𝑡 ≈ 𝐼(1)] against 𝐻1 : 𝜗 < 0[𝜋𝑡 ≈ 𝐼(0)]. This model is less restricted
because it takes into account a deterministic trend.
It follows a distribution associated with that of the statistic t whose critical values are
presented by Dickey and Fuller. The decision rule is such that we accept the null hypothesis
of the existence of a unit root since: 𝑡 ̂
𝜗1 ≥ 𝑡𝑡𝑎𝑏𝑢𝑙é𝑒 (critical values).

3|2 Cointegration test : Engle-Granger approach

Once the variables have been classified as integrated of order 𝐼(𝑑), it is possible to
establish models leading to stationary relations between the variables. There are several
cointegration tests. Engle and Granger (1987) formulated one of the first cointegration
tests. This test has the advantage of being intuitive, easy to achieve. The intuition
underlying the test motivates him to play his role as the first cointegration test.
In order to test the existence of cointegration between the two series, it is imperative to
use Engle-Granger's Augmented Dickey-Fuller cointegration test, or EG-ADF test (Stock and
Watson, 2012). The first step consists of a relation (2) from which the residual process 𝑧̂𝑡
must be extracted. The second step is to look for a unit root in the residual process of the
cointegration regression above. To this end, configure an ADF test like:

∆𝑧̂𝑡 = 𝛼 + 𝜗𝑧̂𝑡−1 + ∑𝑘𝑗=1 ∅𝑗 𝛥𝑧̂𝑡−𝑗 + 𝑣𝑡 (5)

Where 𝑘 is the shift chosen according to the criteria of Akaike and Schwartz. The
assumptions are as follows:
𝐻0 : 𝜗 = 0 (𝑁𝑜 𝑒𝑥𝑖𝑠𝑡𝑒𝑛𝑐𝑒 𝑜𝑓 𝑎 𝑐𝑜𝑖𝑛𝑡𝑒𝑔𝑟𝑎𝑡𝑖𝑜𝑛 𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛)
𝐻1 : 𝜗 < 0 (𝐸𝑥𝑖𝑠𝑡𝑒𝑛𝑐𝑒 𝑜𝑓 𝑎 𝑐𝑜𝑖𝑛𝑡𝑒𝑔𝑟𝑎𝑡𝑖𝑜𝑛 𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛)

7
The decision rule is such that the null hypothesis of non-cointegration will be rejected if the
calculated McKinnon statistic is greater than the corresponding critical value. Otherwise,
there would be no long-term link between the variables considered. The existence of a
long-term relationship paves the way for the estimation of the Error Correction Model
(ECM).
It should be noted that the Engle-Granger approach poses three main problems. First, since
the approach involves an ADF test in the second step, all ADF test problems are also valid
here, including the choice of the number of delays in the increase is a critical factor.
Secondly, the test is based on the hypothesis of a cointegration vector, captured by the
cointegration regression. Therefore, be careful when applying the test to models with
more than two variables. If two variables are included, adding a third variable built into the
model will not change the result of the test. If the third variable does not belong to the
cointegration vector, the OLS estimate will simply set its parameter to zero, leaving the
residual process unchanged. Two-variable test logical strings are often necessary (or
sufficient) to solve this problem. Third, the test assumes a common factor in the dynamics
of the system. To avoid this problem, it would be more prudent to rewrite the simplest
version of the two-variable test (Sjö, 2008). Another solution in front of more than two
variables, it is better to apply the cointegration approach of “Johansen” which is one of
the most powerful cointegrating tests.
There is a lot of work on managing structural breaks and outliers in the error correction
model, as non-stationarity can result from changes in distribution and not just stochastic
trends. Failure to model the offsets leads to processes which resemble stationary variables
in difference I (1) but can be stationary in level I (0) with breaks.
However, the advantage of the Engle-Granger procedure is that it is easy to implement and
therefore relatively inexpensive compared to other approaches. This may work quite well
for two variables in particular, but remember that the common factor restriction is a severe
restriction since any short-term dynamics are forced into the residual process. In this
regard, one would expect the dynamic model advocated by Hendry, Banerjee and other
econometricians to behave better.

3|3 Error Correction Model

In order to analyze the short-term and long-run dynamics of exchange rate changes on the
behavior of inflation (pass-through) one can use an ECM. The value of ECM formulation lies
in the fact that it combines flexibility in dynamic specification with long-term desirable
properties. It could be perceived as capturing the dynamics of the system while integrating
the equilibrium suggested by economic theory.
The greatest reliability of the ECM is that it does not suffer from serial correlation of
residues; in addition, its regression coefficients offer a good economic interpretation (IMF,
2013).

8
  Engle and Granger Methodology
If all the above conditions are satisfied and the inflation rates and the exchange rate share
a common stochastic trend, ie they are cointegrated (𝜋𝑡 − 𝜃̂𝑒𝑡 − 𝛼̂ ≈ 𝐼(0), the ECM
describing the relationship between the two series is written :
𝛥𝜋𝑡 = 𝜏0 + 𝜏1 ∆𝑒𝑡 + 𝛾(𝜋𝑡−1 − 𝜃𝑒𝑡−1 − 𝛼) + 𝜇𝑡 𝑤𝑖𝑡ℎ 𝛾 < 0 (6)
The term 𝜋𝑡−1 − 𝜃𝑒𝑡−1 − 𝛼 can be symbolized in 𝑧𝑡−1 as the delayed error term, it
represents the magnitude of the imbalance between the level of inflation 𝜋𝑡 and exchange
rate 𝑒𝑡 in the previous period. The ECM indicates that the changes in 𝜋𝑡 depend not only
on the changes in 𝜋𝑡 , but also on the magnitude of the imbalance 𝑧𝑡−1 .
This equation can be rewritten 𝛥𝜋𝑡 = 𝜏0 + 𝜏1 ∆𝑒𝑡 + 𝛾𝑧𝑡−1 + 𝜇𝑡 (7)
The parameter 𝜏1 represents the short-run elasticity of inflation relative to exchange rate
fluctuations. The long-run elasticity is 𝜃 in equation (2).
The mechanism of error correction (the restoring force) or of catching up 𝛾, expresses the
speed of the adjustment towards the long-term equilibrium, must be significantly negative;
otherwise, an ECM specification should be rejected. The slope coefficient of 𝛾 implies that,
if in the preceding period the level of the inflation rate was 1% higher than that predicted
by the long-run equilibrium ratio, there will be an adjustment to reduce the inflation level
of 𝛾 during this period to restore the long-run equilibrium relationship between the
inflation rate and exchange rate fluctuations.

 Methodology at Banerjee and Hendry

In the same way that Engle and Granger identify and estimate the ECM, the methodology
of Banerjee and Hendry (1992) also offers a good interpretation and approximates the
Engle-Granger approach. The major difference is that unlike the first two-step ones,
Banerjee and Hendry's approach proceeds in one step to estimate the ECM:
𝛥𝜋𝑡 = 𝜏0 + 𝜏1 ∆𝑒𝑡 − 𝛾𝜋𝑡−1 + 𝜏2 𝑒𝑡−1 + 𝜇𝑡 (8)
The parameter 𝜏1 represents the dynamics of the short-term pass-through that is to say the
short-term repercussions of the exchange rate variations on inflation and the parameter 𝜏2
characterizes the long-term pass-through equilibrium. , where −𝜏2⁄𝛿 represents the long-
term elasticity that is the long-term impact of exchange rate changes on inflation. The
parameter 𝛾 is the error correction mechanism (error correction coefficient) or the
restoring force, it must be less than unity and negative. This parameter 𝛾 indicates the rate
of adjustment of inflation 𝜋𝑡 to its equilibrium level, ie the way in which inflation adjusts
when there is an imbalance in the foreign exchange market. In addition |1⁄𝛾| represents
the duration by which price volatility is fully absorbed after adjusting the imbalance in the
foreign exchange market.

9
4 | Conclusion
The considerable gap between economists, who have a lot to say about equilibrium but
relatively little about dynamics, and econometricians, whose models focus on the dynamic
adjustment process, has been to some extent fulfilled by the concept of cointegration
(Dolado and al., 1990). Cointegration theory has significantly alleviated the problems of
fallacious regressions due to the non-stationary behavior of macroeconomic and financial
variables. This article provides a relevant review of the power of cointegration and the
error correction model.

10
References

Adouka L., Kchirid EM., Benzekoura A. and Bouguelli Z. (2013). Modélisation de la fonction
de la demande des dépenses publiques algérienne à l’aide de modèle ECM. European
Scientific Journal, vol.9, N°.22, August.

Banerjee A. and Hendry D. V. (1992). Testing integration and cointégration : an overview.

Oxford bulletin of economicsc and statistics.
Box G.E.P. and Jenkins G.M. (1976). Time Series Analysis : Forecasting and Control. Holden-
Day, San Francisco.

Castle J. L. and Hendry, D. F. (2016). Clive W.J. Granger and cointegration economics.
Department and Institute for New Economic Thinking at the Oxford Martin School, University
of Oxford, UK, december.

Davidson J. E. H., Hendry D. F., Srba F. and Yeo S. (1978). Econometric Modelling of the
Aggregate Time-Series Relationship Between Consumers' Expenditure and Income in the
United Kingdom. The Economic Journal, Vol. 88, No. 352, december.

Dolado J. J., Jenkinson T. and Sosvilla-Rivero S. (1990). Cointegra tion and unit roots.
Journal of economic surveys, vol.4, n°3.

Engle R. F. and Granger C. W. J. (1987). Co-integration and error correction: representation,

estimation and testing. Econometrica, vol.55, n°2, march.

Eslamloueyan K. and Darvishi A. (2007). Credit expansion and inflation in Iran: An

unrestricted error correction model. Iranian Economic Review, Vol.12, N°19.

Godbout M. and Van Norden S. (1997). Reconsidering cointegration in international

finance: Three case studies of size distortion in finite samples. Document de travail 97,
Banque de Canada, january.

Granger C. W. J. (1986). Developments in the Study of Co-Integrated Economic Variables.

Oxford Bulletin of Economics and Statistics, vol. 48.
Granger C. W. J. and Newbold P. (1974). Spurious regressions in econometrics. Journal of
Econometrics 2, 111–120.

Granger C. W. J. and Newbold P. (1977). The time series approach to econometric model
building. In New methods in business cycle research, Sims ed. Federal Reserve Bank of
Mineapolis.

Greene W. (2011). Econometric analysis. 7th edition Pearson, Paris.

Hendry D. F. (1977). On the time series approach to econometric model building. Sims ed
(1977). Reprinted in Hendry D. F. Econometrics: Alchemy or Science ? Oxford: Blackwell.

11
Hendry D. F. and Ericsson N. R. (1991). An econometric analysis of UK money demand in
monetary trends in the United States and the United Kingdom by Milton Friedman and
Anna Schwartz. American economic review, vol.81, n°1.

International Monetary Fund (2013). Programmation et politiques financières. Institut du

FMI pour le développement des capacités.

Johnston J. and Dinardo N. (1999). Econometric merhods. 4th edition Economica, Paris.
Lardic S. and Mignon V. (2002). Économetrie des series temporelles macroéconomiques et
financières. Economica, Paris.

Lardic S. and Mignon V. (2003). Robert F. Engle et Clive W.J. Granger prix Nobel d'économie
2003. Revue d'économie politique, Vol. 114.

Lau S. H. P. (2008). Using an error-correction model to test whether endogenous long-run

growth exists. Journal of Economic Dynamics & Control, vol.32, june.

Maddala G. S. and In-Moo Kim (1998). Unit roots, cointegration and structural change.
Cambridge University Press, New York.

Mehra Y. P. (1989). Cointegration and a test of the quantity theory of money. Working
Paper 89-2 Federal Reserve Bank of Richmond, april.

Mehra Y. P. (1991). An error-correction model of US M2 demand. Economic review, Federal

Reserve Bank of Richmond, may.

Meuriot V. (2015), “The concept of cointégration : the decisive meeting between Hendry
and Granger (1975)”, Cahiers d’économie Politique / Papers in Political Economy, 68(1),
doi:10.3917/cep.068.0091.

Morales A. (1998). Determinants of growth in an error-correction model for El Salvador.

IMF Working paper 104, july.

Ntungila F. and Pinshi C. P. (2019). Fluctuations de prix des matières premières et économie
congolaise : manne d’espoir ou de malédiction. MPRA Paper No. 95409, https://mpra.ub.uni-
muenchen.de/id/eprint/95409 , august.

Özmen M. and Şanli S. (2018). Seasonal error correction models for macroeconomic
variables : the case of Turkish economy. International European Journal of Managerial
Research, december.

Pesaran H., Shin Y. and Smith R. J. (2001). Bounds Testing Approaches to the Analysis of
Level Relationship. Journal of Applied Econometrics, vol.16, may.

12
Pinshi C.P. and Sungani E. (2018). The relevance of pass-through effect: should we revisit
monetary policy regime? International Journal of Economics, Business and Management Research,
february.

Sargan J. D. (1964). Wages and prices in the United Kingdom: A study in econometric
methodology. In P.E. Hart, G. Mills, and J.K. Whitaker (eds.) Econometric Analysis for
National Economic Planning vol. 16 of Colston Papers. London: Butterworth.

Sjö B. (2008). Testing for Unit Roots and Cointegration. August.

Van Quang T. (2007). Testing cointegration for Czech stock market. Acta Oeconomica
Pragensia, vo. 15, april.

13