DO WAEMU COUNTRIES EXHIBIT A BUSINESS CYCLE? A SIMULATED MARKOV SWITCHING MODEL FOR A WESTERN AFRICA AREA AKA, Bédia F 1*

Transcription

1 Applied Econometrics and International Development.AEID. Vol. 4-4 (2004) DO WAEMU COUNTRIES EXHIBIT A BUSINESS CYCLE? A SIMULATED MARKOV SWITCHING MODEL FOR A WESTERN AFRICA AREA AKA, Bédia F 1* Abstract This paper examines the issue of existence and identification of a regional business or economic cycle in aggregated West African economic and monetary union (WAEMU) economy and single member countries real GDP as well, by using a Markov regime switching model and the Gibbs Sampling simulation method. We found similarities of business cycle among individual countries. Comparing countries cycles to the aggregated one, the chronology and amplitude of Côte d Ivoire s business cycle appears to be closer to the union s cycle. Using the real GDP data the aggregated WAEMU business cycle can be characterized, according to its mean duration (8 years), as a Juglar type cycle. JEL Classification: C11; C22; E32 Keywords: Growth, Business cycles; Non-linear Time series; Markov Switching model; Gibbs sampling. 1. Introduction The constitution of West African Economic and Monetary Union (WAEMU) launched in place of the former West African Monetary Union (WAMU) created in 1962 has raised several interesting issues. Among them, one of the important concerns the existence and identification of business or economic cycle among the member countries and in aggregated WAEMU s economy. The 1 Assistant Professor, Department of Economics, University of Bouaké, Côte d'ivoire, & Researcher, University of Luxembourg, CREA, 162A, avenue de la Faïencerie, L-1511 Luxembourg, & Research Associate, DIAL, Paris. aka@cu.lu or akbdia@yahoo.fr * I am very grateful for the comments provided by the Editor. All remaining errors are mine. 43

2 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) union is composed of 8 countries including Benin, Burkina Faso, Côte d Ivoire, Guinea Bissau, Mali, Niger, Senegal, and Togo. WAEMU is a part of the Franc zone, which is the monetary cooperative instrument between France and its former African colonies. The central bank of WAEMU is BCEAO (Banque centrale des états de l Afrique de l Ouest), while BEAC (Banque des états de l Afrique centrale) is the central bank of Central Africa Economic Union (CAEU) including 7 countries (Cameroon, Congo, Gabon, Guinea equatorial, Centre African Republic and Chad). WAEMU is an additional step in economic integration of West African countries having the CFA franc as same currency. The new environment raised questions as dissymmetry of cycles in individual countries and in the union or common cycle among member countries. In effect the common monetary policy applied independently by the central bank from member state governments in the union (see Aka (2002)) may cause similarities among national cycles and between aggregated regional cycle and countries individual cycles. But various other reasons like variation in industries and long run growth rate across countries, differences in internal fiscal policies, and in consumer sentiment from one country to another can contribute to differences in cyclical movement among national cycles and between national cycles and the regional one in a zone where the industrial development faces several structural problems. Guisan, Aguayo, and Exposito (2001), have exposed the growth performances of the zone and demonstrated that other African areas have performed generally much better than the western region. Within the union, the contribution of the industrial sector to the GDP ranges from 12% to 27%. Most of the resources in the zone are not manufactured, and exports are mainly composed of raw material. While only 25% of industrial activities are export-oriented, industrial imports account for almost 80% of total imports of the union. Moreover, industrial activities are concentrated in food and textile industries, both representing 2/3 of manufactured value-added of the zone. As exemplified by Guisan and Exposito (2001) for most African countries, WAEMU countries are also concerned by lower 44

3 Aka, B. Do WAEMU countries exhibit a business cycle performance of GDP per capita growth rate due to the higher growth rate in population from 1960 to 2003 mitigating the rate of growth of real GDP. Examining the share of countries GDP in the aggregated WAEMU s GDP we could see that Côte d Ivoire is the major contributor to the union s GDP, with a GDP representing more than 40% of the GDP of the zone during the period under study, making this country the economic leader of the union. In fact Côte d Ivoire is the most industrialized country in the zone although the economic growth of this country has been based on agricultural exports. The specialisation in agriculture has profoundly determined the evolution of its economy. The economic growth of Côte d Ivoire the major economy in the zone, mainly based on exports of cocoa and coffee, has been relatively high from 1960 to 1979 with the GDP per capita mean growth rate being around 5.7%. During this period qualified by «Miracle économique ivoirien», the shares of value added in GDP were 34% for agriculture, 15% for industry and 51% for services. This structure leads to growth in exports and increase in agricultural income and surplus managed by CAISTAB 2 (a public marketing boards), and in investment in all economic sectors. By the end of 1979, there was a slow down of growth due to the drop in international price of agricultural products and oil crises. Côte d Ivoire was not the only one country concerned by these international chocks. During the 1980 s the macroeconomic situation in all the countries of WAEMU zone was worsened and governments adopted Structural Adjustment Programmes (SAP) financed by international institutional partners (mainly International Monetary Fund and World Bank) in order to improve the efficiency of their economies and enhance growth of all member countries in the zone. The structural adjustment programs were combined next with privatisations of public enterprises and liberalization policies at the 2 Caisse de Stabilisation et de Soutien du prix des produits agricoles. 45

4 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) beginning of 1990 s. As these macroeconomic policies outcomes were not sufficiently growth improving, WAEMU s economy experienced the devaluation of CFA franc in The combination of these various macroeconomic policies results unfortunately in an increase of poverty in the zone. The current situation in the union is characterized by a social and political crisis since 2002 in Côte d Ivoire, which is the economic leading country of the union, and no solution have yet been found to definitely overcome this crisis since the political Marcoussis agreements in France analysed by De Gaudusson (2003). Various other analyses trying to understand and explain the origin of the situation including Cogneau and Mesple Somps (2003), Lepape (2003), Roubaud (2003), Hugon (2003) have shown that the crisis in Côte d Ivoire has important impacts on external and intra zone trade, transport, manufactured goods, prices and employment for all member countries. As a result, the GDP growth rate of the zone, which was about 5.7% in 1998 dropped to 2.7% in 2000 and to 0.5% in 2003 as presented in Bossard (2003) and Jumbo (2003). Thus it s of interest to study WAEMU zone business cycle, which behaviour can shed light on differences or similarities between the aggregated business cycle and the individual states specific ones. To analyse business cycles various methods exist in the literature and have diversified since the pioneering work of Burns and Mitchell (1946). Most of the contributions are concerned with linear processes of time series, among them the ARMA and ARIMA models of Beveridge and Nelson (1981), the unobservable components models of Harvey (1985), Watson (1986), Clark (1987), the cointegration approach of King, Plosser, Stock and Watson (1991), and fail at explaining business cycles features about duration of recessions and expansions. Hamilton (1989) has proposed an innovative more general nonlinear Markov regime-switching model (MRSM) to analyse economic fluctuations, which has been widely used in recent years. But this model has been mainly applied on quarterly or monthly data in developed economies recently by Clements and Krolzig (2003), 46

5 Aka, B. Do WAEMU countries exhibit a business cycle Krolzig and Toro (1999 and 2001), Krolzig (2001) and thus, no application exists for developing countries, especially in African countries where only short-range yearly data are available, a problem also point out by Holmes (2003). This paper uses the modern version of business cycles analysis, namely the Markov regime-switching model (MRSM) to investigate the issues of existence and identification of West African Economic and Monetary Union aggregated business cycle and member countries cycle. The remainder of the paper is organized as follows. Section 2 introduces the MRS model innovated by Hamilton (1989). Section 3 presents the Gibbs Sampling simulation method. Section 4 presents the data used in this paper and gives simulation results from the Gibbs Sampling outputs. The last section concludes. 2. The model The non-linear specification introduced by Hamilton is the following Markov switching model: yt = µ s,,, (0, 2 t+ ζt st ζ t st NID σ ) (1) st where s t is an unobserved state variable, which can take the value 1 or 0 depending on whether the economy is in a low growth phase (recession) or a high growth phase (expansion). This unobserved process s t follows a two states first-order Markov process with transition probabilities: p( st = 0 st 1 = 0 ) p( st = 1 st 1 = 0) = p00 p10 = p 1 q (2) p( s = 0 s = 1 ) p( s = 1 s = 1) p p 1 p q t t 1 t t These transition probabilities are restricted so that P ij =P[s=j s=i], with Σ 1 j=0p ij =1 for all i. P ij gives the transition probability that the state i will be followed by state j. More dynamic pattern for y t can be introduced by replacing ζ t with u t in equation (1), with u t following an autoregressive process: 47

6 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) u t = k i= 1 iut i + 2 t, s, ε t t, s ~ NID(0, σ t s ) t φ ε (3) Then the Markov switching model becomes: y t k µ = φ ( y ε (4) st i= 1 i t i µ s ) + t i t, st with φ i indicating the autoregressive parameters 3, and the regime shift in the mean of the time series is: µ 1 if st = 1, µ s t = (5) µ M if st = M. An algorithm has been proposed by Hamilton (1989 and 1991) to estimate the parameters θ = (µ 0,µ 1,φ i,σ 0 ²,σ 1 ²,p,q) of the model. The 1991 algorithm is a Quasi Bayesian Maximum Likelihood including the algorithm of 1989 as a special case with diffuse priors. The parameters µ 0 and µ 1, σ 0 ² and σ 1 ², denote the means, the variances in low growth and high growth state respectively. The transition probabilities p and q determine the business cycle characteristics for dating, thus from this model, the expected duration of an expansion (recession) can be evaluated as follows: k 1 1, k 1 1 k (1 p) k = 1 p ~ (1 p) k = 1 q ~ (1 q) (6) k (1 q) The regime switching from expansion (recession) to recession (expansion) state is given by: ( s 1 Y ) < 0.5, ( P( s = 1Y > 0.5 ) P (7) t = T t T 3 In Hamilton (1989) an AR(4) model is specified for quarterly data of the USA. 48

7 Aka, B. Do WAEMU countries exhibit a business cycle This simple decision rule allows determining the business cycle turning points and requires no prior information on turning points, and conditional probabilities of the regime (expansion and contraction) are constructed post-estimation to suggest turning points. This model can be estimated in various ways but here we proceed by using the simulation method of Gibbs sampling (GS) presented in the next section, which is an alternative to the maximum likelihood estimation method. 3. The Gibbs Sampling simulation method Albert and Chib (1993) have shown that the results from the Gibbs sampling output are quite similar to the output of the maximum likelihood estimation method. The Gibbs sampling tool for estimation presented in Casella and George (1992) is as follows. Suppose we want to simulate the posterior distribution of the parameters θ k, if we arbitrarily set θ i, we can obtain the conditional distribution of all parameters according to the following steps: (i) Arbitrarily specify initial values, for (0) example, = [ ] (0) (0) (0) θ θ1, θ 2,..., θ k, and set i=1; (ii) Repeat the following sampling scheme: ( ) a) for θ from [θ 1 θ (i-1) 2,,θ (i-1) k ] 1 i ( ) b) for θ from [θ 2 θ (i) 1, θ (i-1) 3,θ (i-1) k ] 2 i (i) until k) for θ k from [θ k θ (i) 1,,θ (i) k-1 ] (iii) Set i=i+1 and go to step (ii). After iterating on this cycle N times, we θ 1 2 k, and Geman and Geman (1984) ( N ) obtain = [ ] ( N) ( N) ( N ) θ, θ,..., θ have shown that for high values of N, the distribution of θ (N) converges in distribution towards θ Once we obtain the simulated sample, we can get the marginal posterior moments of all parameters: 49

8 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) N ( i) ( i) [ θ1 θ 2 ] k 1 θ 1 =,...,θ (8) N i= 1 4. Empirical results The results are based on WAEMU countries yearly real GDP data (millions dollars at 1995 prices) and GDP deflator from World Development Indicators database covering the period We first proceed with the individual countries real GDP to analyse the stylised facts of business cycle in each country separately and second we deal with the aggregated 6-countries regional business cycle composed of Benin, Burkina Faso, Côte d Ivoire, Niger, Senegal, and Togo and finally compare the individual cycles to the aggregated one. We excluded the new members Guinea Bissau and Mali because of the data availability and consistency. The retained 6 countries aggregated real GDP represents more than 76% of the real GDP of the zone. The regional indicator is taken as the aggregated real GDP of individual countries GDP; state 1 is related to expansion phase and state 0 to contraction phase. In both cases the variable y t is computed as y t =100*ln(GDP t /GDP t-1 ), thus in the case of Markov switching autoregressive of order p models (MS-AR(p)), the parameter µ 0 indicates the real GDP mean yearly growth rate in the contraction regime, and (µ 0 +µ 1 ) represents the mean yearly growth rate in the expansionary regime. The yearly growth rates of the real GDP and Deflator of GDP are plotted in right panel of Figure 1 while left panel depicts the log level of the variables for each of the 6 countries. From Figure 1 we could see that the level of Côte d Ivoire s real GDP is higher compare to other countries real GDP. We can notice according to the Deflator of GDP that there is a higher correlation among prices in the countries of the WAEMU zone than in their real GDP. 50

9 Aka, B. Do WAEMU countries exhibit a business cycle Figure 1: Log of Real GDP and Deflator of GDP in WAEMU s economy LGDP95BENIN LGDP95CIV LGDP95SENE LGDP95BFASO LGDP95NIGER LGDP95TOGO DLGDP95CIV2 DLGDP95BFASO2 DLGDP95SENE2 DLGDP95BENIN2 DLGDP95NIGER2 DLGDP95TOGO2 5 LIPBenin LIPCiv LIPSenegal LIPBFaso LIPNiger LIPTogo 0.4 DLIPBenin DLIPCiv DLIPSenegal DLIPBFaso DLIPNiger DLIPTogo Business cycles for individual countries The investigations are performed using the Gibbs sampling simulations and the stylised facts of the business cycle in each country from 1960 to 2002 are presented in the following. Simulations results are summarized in Table 1. We use the Gauss code by Kim and Nelson (1999) based on the original code written by Albert and Chib (1993) and incorporating a multi-move Gibbs Sampling. Out of 11,000 Gibbs Sampling simulations, the first 1,000 are discarded; inferences are based on the remaining 10,000. The variable (µ 0 + µ 1 ) indicates growth rate during expansion; (p 00 =p) indicates the probability of staying in a recession state; (p 11 =q) indicates the probability of staying in an expansion state; (ρ=(1- q)/(2-q-p)) indicates the probability to converge in long run towards a high growth rate; (1/(1-q)) represents the expected duration of expansion and (1/(1-p)) the expected duration of recession. 51

10 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) We observe similarities of business cycle pattern among the member countries. The conditional mean of the growth rate in the recession regime (µ 0 ) ranges from the lowest 0.32% (Niger) to 1.4% (Burkina Faso). The growth rate in expansion regime (µ 0 +µ 1 ) ranges from 1.7% (Niger) to 3.9% (Côte d Ivoire) the highest. For each individual country, the duration of expansion is long compared to the duration of recession (see Table 1). These lengths of expansions and recessions found here are consistent with the traditional view of business cycle theory about phases duration. The average mean duration of cycles is quite similar among member countries. Table 1. Gibbs Sampling Simulations Results for Real GDP growth rate Parameters Prior Posterior Mean Means WAEMU Benin Burkina Faso Côte d Ivoire Niger Senegal Togo µ µ p p σ µ 0 + µ p 00 =p p 11 =q p 00 =1-p p 11 =1-q ρ / (1-q) / (1-p) Cycle average mean duration Business cycles in the aggregated economy For the aggregated economy composed of the 6 countries, the growth rate of real GDP in the low growth state (µ 0 ) is 1.27%, while the growth rate in the high growth rate state (µ 0 +µ 1 ) is 3.8%. Phases duration are 10.8 years during expansions, and 5.5 years for 52

11 Aka, B. Do WAEMU countries exhibit a business cycle recessions, meaning that the aggregated WAEMU s business cycle average duration is almost 8 years, which can be characterized in business cycle literature as a Juglar type cycle (lasting from 7 to 11 years, with 8.8 years mean duration) which is called the business cycle in the literature and associated with increases in productivity, aggregate demand and fluctuations in fixed investment. For aggregated WAEMU s economy the probability for the growth process of remaining in an expansionary (recessionary) state is 90.75% (81.97%), meaning that the different phases are persistent. The probability [ρ =(1-q)/(2-q-p)] of the growth process to converge in long run towards a high growth rate is 34% indicating little potential real growth of the aggregate economy, and the probability of an expansionary state to be followed by a recession state is 9.25%, while the probability of a recession to be followed by an expansion is almost 18.03%. The resulting regime probabilities of expansion regime are plotted in Figure 2. The state probabilities can be used to date the business cycle. Finally, comparing individual countries cycles to the union aggregated one, we could find that only Côte d'ivoire has experienced economic growth rate (3.9%) much closer and even exceeds the aggregated WAEMU s economy growth rate (3.8%). All other countries growth rates are bellow the aggregated one, the lowest being Niger (1.70%) and Senegal (2.41%). The average mean duration of cycle for Côte d Ivoire (9), Benin (9), Burkina Faso (10) and Senegal (9) appears to be longer than the mean duration of the aggregated zone business cycle (8), the mean duration for Niger and Togo being similar to the aggregated one. These results confirm the fact that Côte d Ivoire is the central economy in the zone and its business cycles could impact those of member countries cycles and the aggregate cycle as well. 53

12 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) Figure 2: Expansion State probabilities in the WAEMU economy 1.0 Pr(S =1) t BENIN 0.75 Pr(S t =1) BFASO Pr(S =1) CIV t 1.0 Pr(S t =1) WAEMU Pr(S t =1) SENE TOGO 0.8 Pr(S t =1) Pr(S t =1) NIGER 5. Conclusion This paper has investigated the issues of existence and identification of a regional WAEMU-zone business cycle and member countries individual economic cycles using the Hamilton s Markov regime-switching model, which we estimated by the Gibbs sampling simulations. We found that the aggregated WAEMU s business cycle mean duration, which is almost 8 years, suggests a Juglar type cycle characterizing the economy of the zone. We found similarities of business cycles pattern between member countries, and more importantly the chronology and amplitude of Côte d Ivoire s business cycle appears to be closer to the union s 54

13 Aka, B. Do WAEMU countries exhibit a business cycle cycle. In addition to the fact that the state probabilities reveal little potential convergence towards expansion in future for the union, the growth rate should be high enough in order to overcome the high growth rate of the population in the zone. Furthermore the non linear stochastic characterization adopted here implies persistence of the different phases of the cycle and thus any unexpected chock on one individual economy will permanently alter the future level of activity in the zone, and unfortunately that is the case since 2002 the 19 th September with the war in Côte d Ivoire, which is as seen the economic leading country in the WAEMU-zone. The economic growth perspectives of the zone are actually subject to a rapid resolution of the crisis in Côte d Ivoire, as any development initiative could not be undertaken without peace in the region. It s a condition to attract foreign investment and international cooperation necessary to guarantee industrial development of the zone. Further research should investigate firstly the precise dating of the chronology of the economic cycles in the zone, and try to forecast the business cycles of the zone and in member countries as well, and secondly extend the study to the whole Franc zone and in specific sectors of their economies. Bibliography Aka, B.F. (2002). La politique monétaire de la BCEAO est-elle indépendante des politiques fiscales des gouvernements des pays membres? Cas de la Côte d Ivoire, du Niger et du Togo. Revue Africaine de Sciences Economiques et de Gestion, Vol. IV, No. 1, pp Albert, J.H. and Chib, S. (1993). Bayes Inference via Gibbs Sampling of Autoregressive Time Series Subject to Markov Mean and Variance Shifts. Journal of Business and Economic Statistics, Vol. 11, No. 1, pp Beveridge, S. and Nelson, C. (1981). A new Approach to Decomposition of Economic Time Series Into Permanent and 55

14 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) Transitory Components With particular Attention to Measurement of the Business Cycle. Journal of Monetary Economics, Vol. 7, pp Bossard, L. (2003). «Peuplement et migration en Afrique de l Ouest: une crise régionale en Côte d Ivoire», Afrique contemporaine, n 206, pp Burns, A.F. and Mitchell, W.C. (1946). Measuring Business Cycles. New York: National Bureau of Economic Research. Casella G. and George, E.I. (1992). Explaining the Gibbs Sampler. American Statistical Association, Vol. 46, No. 3, pp Clark, P.K. (1987). The Cyclical Component of U.S. Economic Activity. Quarterly Journal of Economics, Vol. 102, pp Clements, M.P. and Krolzig, H.-M. (2003). Business Cycle Asymmetries: Characterisation and Testing based on Markov- Switching Autoregressions. Journal of Business and Economic Statistics, 21, Cogneau, D. and Mesplé-Somps, S. (2003). «Les illusions perdues de l économie ivoirienne et la crise politique», Afrique contemporaine, n 206, pp De Gaudusson, J.D., (2003). «L accord de Marcoussis, entre droit et politique», Afrique contemporaine, n 206, pp Geman, S. and Geman, D. (1984). Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 6, pp Guisan, M.C. and Exposito, P. (2002). Education, Industry, Trade and development of African Countries in Applied Econometrics and International Development, 2-2, pp

15 Aka, B. Do WAEMU countries exhibit a business cycle Guisan, M.C. and Exposito, P. (2001). The economic Development of African and Asian-Pacific Areas in Applied Econometrics and International Development, 1-2-, pp Guisan, M.C., Aguayo, E. and Exposito, P. (2001). Economic growth and Cycles: Cross-Country Models of Education, Industry and Fertility and International Comparison. Applied Econometrics and International Development, 1-1-, pp Hamilton, J.D. (1989). A new Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle. Econometrica, Vol. 57, pp Hamilton, J.D. (1991). A Quasi-Bayesian Approach to Estimating Parameters for Mixture of Normal Distributions. Journal of Business and Economic Statistics, Vol. 9, pp Harvey, A.C. (1985). Trend and Cycles in Macroeconomic Time Series. Journal of Business and Economic Statistics, Vol. 3, pp Holmes, M.J. (2003). Are the Trade Deficits of Less Developed Countries Stationary? Evidence for African Countries. Applied Econometrics and International Development, 3-3-, pp Hugon, P. (2003). «La Côte d Ivoire, plusieurs lecture pour une crise annoncée», Afrique contemporaine, n 206, pp King, R.G., Plosser, C.I. Stock, J.H. and Watson, M.W. (1991). Stochastic Trend and Economic Fluctuations. American Economic Review, Vol. 81, pp Kim, C-J. and Nelson, C.R. (1999). State-space Models with Regime Switching: Classical and Gibbs-Sampling Approaches with Applications. MIT Press. 57

16 Applied Econometrics and International Development. AEID. Vol. 4-4 (2004) Krolzig, H.-M. (2001). Markov-Switching procedure for dating Euro-zone Business Cycle. Vierteljahreshefte zur Wirtschaftsforschung, 70 (3), Krolzig, H.-M. (2001). Business Cycle Measurement in the Presence of Structural Change: International Evidence. International Journal of Forecasting, 17(3), Krolzig, H.-M. and Toro, J. (1999). A new approach to the Analysis of Shocks and the Cycle in a Model of Output and Employment. European University Institute, EUI Working Paper ECO 99/30. Krolzig, H.-M. and Toro, J. (2001). Classical and Modern Business Cycle Measurement: The European Case. University of Oxford, Discussion Paper in Economics 60. Leenhardt, B., E. Fourmann, T. Latreille, and Massujeau, B. (2003). «La crise ivoirienne et son impact régional : regard sur l actualité et scénarios pour l avenir», Afrique contemporaine, n 206, pp Lepape, M. (2003). «Les politiques d affrontement en Côte d Ivoire », Afrique contemporaine, n 206, pp Roubaud, F. (2003). «La crise vue d en bas à Abidjan : ethnicité, gouvernance et démocratie», Afrique contemporaine, n 206, pp Watson, M.W. (1986). Univariate Detrending Methods with Stochastic Trends. Journal of Monetary Economics, Vol. 18, pp Journal published by the Euro-American Association of Economic Development. 58