Testing the Long Run Neutrality of Money in Developing Economies: Evidence from EMCCA

We examine the long run neutrality of money (LMN, hereafter) in the Economic and Monetary Community of Central Africa (EMCCA) countries, applying Fisher and Seater (1993) Autoregressive Integrated Moving Average (ARIMA) methodology, using different monetary aggregates, money supply in the strict sense (M1), money supply in the large sense (M2) and domestic credit (credit to private sector) during the period 1978-2008. Tests consistently reject the LMN hypothesis. It is found that monetary aggregates have significant and positive impacts on real Gross Domestic Product (GDP) for all EMCCA countries. The results are robust under various sub-periods and the estimated coefficients are stable under two breakpoints corresponding to the dates of central bank reforms and devaluation of the local currency.


Introduction
The old debate on the effects of monetary policy on real and nominal variables is experiencing a resurgence of interest with the controversy surrounding the role of central banks.The most widely shared position today is the long run neutrality of money (LMN, hereafter).Literature distinguishes the LMN and the super long-run neutrality of money (LMSN, hereafter).Under the LMN hypothesis, a permanent change in the level of money supply has no impact on the level of real variables in the long run; while the LMSN hypothesis claims that a permanent change in the growth rate of money supply does not affect the level of real variables in the long run.
To test the hypothesis of neutrality of money, Fisher and Seater (1993), King and Watson (1997) respectively use an ARIMA model and the VAR methodology with time series for real output and monetary aggregates.According to these works, two important properties must be met for testing LMN: the exogeneity of money and the non-stationarity of real and monetary variables.Based on annual data from the USA for the period 1865-1975 and monthly data for Germany over the period of hyper-inflation after World War I, Fisher and Seater (1993) argue a weak neutrality of money in the case of the USA and refute the hypothesis of LMSN in the case of Germany [1].King and Watson (1997) consider the American experience of post-war period from 1949:Q1 to 1990:Q4.Tests reject neutrality and little evidence of LMSN [2].
Studies on LMN in emerging countries are relatively rare.Wallace (1999), Bae and Ratti (2000), then Sulku (2011) verify the LMN hypothesis, respectively, in the case of Mexico (1932Mexico ( -1992 period) period), Brazil and Argentina (1884-1996and 1912-1995 periods) and Turkey (1987: Q1-2006:Q4), using the methodology of Fisher and Seater (1993) [6][7][8].Similarly, Chen (2007) tested the LMN hypothesis in South Korea and Taiwan using King and Watson (1997) methodology.He uses quarterly data from 1970 to 2004 for South Korea and from 1965 to 2004 for Taiwan.Chen (1997) finds strong support for the LMN in the case of South Korea but poor evidence in favor of the LMN in the case of Taiwan [9].
Although there are relatively numerous studies investtigating the LMN for developed countries as well as emerging economies, the literature on developing countries is emerging and is still at its earlier stage.
In this paper, we would like to contribute to the development of the literature on LMN for African countries, building on the Economic and Monetary Community of Central African (EMCCA) countries case.These countries have in common a Central Bank, the Central African States Bank (CASB) whose primary mission is monetary stability.This means price stability and a sufficient level of foreign exchange reserves, as the local currency (CFA francs) is linked to the European currency (the Euro).Therefore, the LMN is necessary for the success of monetary stability strategy.That is why, testing the LMN for EMCCA countries becomes even more attractive and important.
The study is based on the Fisher and Seater (1993) ARIMA methodology not only because it is appropriate but also for its simplicity.Even though our period of study is relatively short, as explained in Fisher and Seater (1993), since there were sudden changes in money and prices, this data set is qualified to be used for controlling a long run relationship.
The remainder of the paper proceeds as follows: Section 2 established the LMN and LMSN based on the Fisher and Seater's (1993) methodology.In Section 3, our data set is introduced.Empirical results are presented and discussed in Section 4. Some concluding remarks appear in Section 5.

Econometric Methodology
where   m and   y are the orders of integration of the money supply and the real GDP respectively.The vector  ' ,    is independently identically distributed with zero mean and the covariance which is composed of If we consider the specific case when x m   t and such that and are equals to 1 or 0, then the long run effect of permanent change in As stated in Fisher and Seater (1993), if then there are no permanent changes in monetary variables, so that LMN and LMSN are not testable.Therefore, the long run effect is testable if there is long run variation in money which means, It implies that money should be non-stationary in level Fisher and Seater (1993) show that for 1 x  , , z x LRD can be written as, where It is seen that the long run derivative of z with respect to x depends on    

Identification
The identification problem is dealt with by imposing the exogeneity of money in the long run by assuming  .After this assumption, Fisher and Seater The appropriate models for the following special cases:  , the LMN is tested by applying Equation ( 5) to estimate and   1 y  , LMN cannot be rejected and LMSN can be tested by deriving the following equation to estimate, LMSN:  1 k t (8) r Equations ( 5)-( 8) can be computed by appl

Data
ed in Leong and McAleer (2000), the result of porary economic history of EMCCA countri gr

Econometric Results
3) methodology, the nonsher and Seater (1993) analysis provides a st These linea ying Newey and West (1987) [10] approach to obtain consistent estimate of k b .These authors purpose a consistent covariance matrix estimator by applying GMM methodology when error terms are heteroskedastic and autocorrelated.
As indicat LMN test is sensitive to underlying monetary aggregate.We cannot claim LMN only depending on one money aggregate.Therefore, we consider all alternative monetary aggregates: M1, M2 and domestic credit in EMCCA namely, Cameroon, Central African Republic, Chad, Congo Brazzaville and Gabon.Guinea Equatorial was not considered because of data availability.As a real output measure, GDP series at fixed 2000 prices is considered.The CPI with 2000 base year is used to obtain an inflation series.All of the variables are obtained from the Word Bank Data set, the World Development Indicators, covering the periods 1970-2008.They are obtained on yearly basis.
The contem es can be summarized in four main eras: 1960-1980, strong economic expansion; 1981-1994, economic and financial crises, then devaluation of the local currency in 1994; 1995-2005, implementation of stabilization programs with mitigating results; 2005 onward economic stability and growth era.During the 1960-2010 period the average inflation was 5% and the real GDP growth rate was on average 3.5% with 2.5% percent volatility.However, during 1981-1993 period, inflation decreased to one digit, and real GDP growth on average increased to 4.2%, with 4.5% volatility.In EMCCA, from 1970 to 1995 there was high and persistent inflation and unstable economic growth.The 1993 crisis has been followed by the 1994 devaluation of the domestic currency.The inflation reached to 12%.In 1999 financial crisis occurred as an infection of world financial crises and in 2001 the deepest financial melt down took place.However, after 2002 based on IMF and the World Bank stabilization programs, monetary stability has been employed as a framework for monetary policy.Inflation decreased to one digit.The low level inflation has been accompanied by a significant decrease in money supply growth.Indeed, stable economic growth has been achieved.Thus, during the 1960-2010 period EMCCA faced with challenging economic events, economic and financial crises.There were sudden changes in money and prices.Therefore, although our period is relatively shorter, this data set is qualified to be used for controlling a long run relationship.
In this study, we use real GDP and all monetary ag egates in their logarithms.Since all the series are annually, there is no need to test for seasonality.
In the Fisher and Seater (199 stationarity of the variables (money aggregates and real output) is the necessary condition to test LMN.Therefore, the Augmented Dickey-Fuller (1979) (ADF), Phillips and Perron (1988) (PP), and Kwiatkowski et al. (1992) (KPSS) unit root tests are applied to the series involved [11][12][13].The results of the above tests suggest that all of the variables are I(1) for all the countries considered in this analysis.
The Fi raightforward test of LMN.Since real GDP and money aggregates are I(1), the long run derivative of real GDP with respect to money is equal to the slope coefficient of a regression of growth rates of real GDP on growth rates of money.Thus, the LMN hypothesis will be examined by using Equation (5).LMN is sustained if the slope co- The econometric results obtained can ployed t be summarized as l, using M1, M2 and domestic credit, we co tiv is a significant and positive effect of M tim follows: 1) For al uld not find evidence in favor of the LMN hypothesis; in fact, the tests consistently reject the LMN hypothesis; 2) All the estimates of ' k b s are significant and posie for all countries; 3) As a result there 1, M2 and domestic credit on real GDP in the long run.In order to cross-check the above results, the same esations were conducted on different sub-periods and the new estimates of ' k b s were obtained.Again, the results indicate strong support in favor of the positive impacts of money aggregates on real GDP.
efficient   k b goes to zero as k goes to infinity.Confid intervals aroun estimated ' k b s r ence d espectively are determined for M1, M2 and do c credit.The estimates of ' k b s are obtained for 1 12 k   .The 95 percent confidence intervals are cons r the estimated coefficient of money supply using t-distribu- degrees of freedom.The standard errors em o construct the confidence intervals are corrected for autocorrelation and heteroskedasticity by the Newey and West (1987) method.As an example, we present in Figure 1, the graphs of confidence intervals for b k in the case of Cameroon (country leader) and Gabon (EMCCA second economy).

Figure 1 .
Figure 1.Graphs of confidence intervals for the credit, M1 and M2 in the case of Cameroon (left) and Gabon (righ Finally, Chow tests were used to check the stability of th

2.1. LMN and LMSN in Fisher and Seater's (1993) ARIMA Methodology
1 which means that permanent change in the growth rate of money affects the growth rate of output in the same way the level of money affects the level of output.Therefore, this implies that "growth-rate to growth-rate" propositions are not LMSN propositions.To test LMSN in this case, first of all LMN should be held.If LMN holds, then LMSN can be tested by deriving Equation (8) to estimate,