Resilience capacity of South countries with respect to the global economic crisis: an empirical comparative analysis of Sub-Saharan African and Latin American countries

This paper deals with the problem of vulnerability of developing countries and their resilience capacity with respect to external shocks. The analysis particularly considers the countries of Latin America and Sub-Saharan Africa. Although the transmission risks of the 2007 financial crisis were initially underestimated for Southern countries, it ended up reaching all continents. Aiming at understanding the crisis propagation the paper carries out a comparative analysis between these two groups of countries. The objective is to test the resilience of Latin America and Sub-Saharan Africa countries with respect to the effects of the economic crisis. It accounts for the differences between countries’ behavior with respect to external shocks. Using dynamic panel techniques the paper estimates the growth dynamics for these countries. The estimates results are shown to be relevant and indicate that some groups of countries are more resistant to crisis effects than others. In order to study dynamic economic growth in our sample countries, we use dynamic panel estimation techniques. This allows us to relate economic growth at a given time to that observed at an earlier time [AR (1) model]. The dynamic model that we estimate is as follows: . (1) where is the rate of growth in country (i) at time (t), and is the matrix of the explanatory variables at time (t). It includes according to data availability: public aids in percentage of GDP, external debt, external reserves, domestic savings, net inflows of foreign direct investments, remittances and the current account balance (CAB), two dummy variables (to distinguish Fuel exporting and Metal exporting countries) and the number of countries in recession (to appreciate contagion risks). The stands for the country-specific effects that might explain the differences in growth between countries. These effects are assumed to be fixed and independent of errors ( ). For dynamic models, OLS is quite inefficient particularly because of the endogeneity of the lagged variable relative to the fixed effects. It creates an upward bias in the estimation of the coefficient associated with the lagged endogenous variable. One way that has been suggested to correct this bias is to transform the estimation model so as to eliminate the fixed effects. The first change involves using the Within-Estimator, which subtracts the individual mean at every observation. Since the specific effects are constant over time each observation equals the mean. Nevertheless, Nickell (1981), Judson and Owen (1999), and Bond (2002) have shown that the Within-Estimator is itself not efficient, especially for panels with few time periods. In fact, they showed that in these short-t panels, the transformation results in a substantial negative correlation between the transformed lagged dependent variable and the transformed error term. In this way, according to Bond (2002), any significantly better estimator should find a coefficient for ( ) somewhere between that of the Within-Estimator and that of the non-transformed OLS estimator. Anderson and Hsaio (1981) have suggested a different transformation to correct the endogeneity bias between the lagged variable and the fixed effects. This involves estimating a first-difference model, which by design also eliminates individual effects: . (2) However, this transformation does not make it possible to remove the endogeneity of the transformed lagged dependent variable ( ) in relation to the transformed error term ( ), since in is correlated with in . Anderson & Hsiao (1981) therefore suggest using the instrumental variables method to overcome this hurdle. According to them, for every first-difference observation (beginning in the 2nd period) there are two potential instrumental variables, both already present in the model, namely the level and the first-difference variables of the previous time period. For example, for both and are appropriate instruments since they are highly correlated with but not correlated with , assuming that the errors are time independent and that the initial conditions are predetermined (Bond, 2002). Anderson and Hsiao, on the other hand, prefer levels as instruments for differences, since especially in the case of short-t panels, level instruments offer a better way to use more observations, which is a welcome efficiency gain. However, their method does not allow for the possibility of using potential lags as instruments. This possibility was introduced later by Holtz-Eakin et al (1988) and Arellano and Bond (1991). Their methodology is based on the Generalized Method of Moments (GMM) with additional orthogonality assumptions to ensure the non-endogeneity of the instruments. Arellano and Bond (1991) propose a GMM estimator that is based on the orthogonality of the level variables instruments to the differences of residuals: the condition on the moments is as follows: for and (3) where and stand for the collection of instruments for the first-difference variables. Blundell and Bond (1998), however, show that for very long time series, level variables are very weak instruments for first-difference variables. For efficiency gains, they suggest additional moment conditions that can take into account a wider range of instruments (system GMM). Their suggested transformation is an extension of Arellano and Boyer’s (1995) forward orthogonal deviations to make the instruments exogeneous relative to the fixed effects. The conditions on the additional moments are as follows: , (4) where and stand for the collection of instruments for the level variables, with . For the purpose of this paper in order to estimate our dynamic model, we have chosen to use the GMM (Blundell & Bond, 1998) approach. The efficiency of the GMM method in a dynamic panel, however, must be tested. The two prerequisites are a good identification of instruments (Sargan test) and the absence of autocorrelation among the residuals (Arellano & Bond test). The Sargan test states as a null hypothesis the absence of correlation between instruments and residuals. If this hypothesis is rejected, then the estimations are not efficient. The Arellano & Bond test, on the other hand, states as a null hypothesis the absence of autocorrelation among residuals. Since the test involves a first-difference transformation, there will necessarily be a first-order autocorrelation. On the other hand, the absence of autocorrelation among (level) residuals is guaranteed if there is no second-order autocorrelation among the first-difference residuals. For an efficiency gain, we corrected the standard deviations of the heteroscedacity bias, following Windmeijer’s (2000) guidelines. The transmission of the crisis from developed to developing countries operated through two main channels: the traditional channel of international trade and the international finance (Hugon and Salama, 2010). Theoretically, many factors may justify the vulnerability of economies of Sub-Saharan Africa and Latin America. All these economies did not experience the effects of external shocks in the same way. Aiming at assessing the resilience capacities of these Southern countries with regard to the crisis this paper has performed an econometric investigation using a dynamic panel model methodology. Furthermore, three sample countries have been considered to carry out an empirical comparative analysis. Two samples of Sub-Saharan African countries have been differentiated by the membership to the CFA zone, and one sample of South American countries. The three groups of studied countries share common features of economic structure and terms of trade. Concerning the Sub-Saharan African countries, it has been shown that for the CFA countries the transmission factors listed in the theoretical part are not linked significantly to GDP growth. The only factor of vulnerability for these countries has been shown to be the FDI inflows. In recession periods, we showed that FDI decrease aggravate the crisis. Resilience capacities have not been detected for this group. The Sub-Saharan African countries out of the CFA zone present different results. For them, the results show a significant link between public development aid and GDP growth. This link is significant and positive in recession and expansion periods. This indicates that public development aid constitutes a significant pro-cyclical transmission vector. Domestic savings have been shown to be a pro-cyclical variable too. Indeed, its decline in recession period may worsen the deterioration of the economic situation. The external debt has been shown to be a counter-cyclical variable in recession period. This may contrast with the counter-cyclical property of external reserves shown in expansion periods. These countries have the capacity to use their external assets and debts to adjust their macroeconomic situation. The clear difference of results between CFA zone and non-CFA Sub-Saharan African countries could motivate further research about the role of the strict peg to the euro. In the case of Latin American countries three variables have been shown to be significantly related to the fluctuations of GDP growth. These factors are foreign direct investments (FDI), domestic savings and the current account balance (CAB). FDI and CAB present opposite behaviours relative to GDP. During expansion periods, FDI and CAB fall and increase during recession periods. Indeed, when the growth of the GDP decreases, authorities try to relax their FDI legislation in order to rebalance the economic situation. In addition, they implement competitive devaluation policies affecting the CAB. Countries of Latin America present also the highest risk of propagation in our analysis. Taking into account that the three groups of studied countries share common features of economic structure and terms of trade, we could expect that the resilience capacities would be similar. However, this paper has shown that resilience capacities of the three investigated country groups (African CFA zone, Sub-Saharan African non-CFA zone and Latin America) are not the same. Considering the Sub-Saharan African countries the econometric results show that countries of the non-CFA group better perform in terms of resilience to external shocks. This area has shown to be less vulnerable to the transmission of the crisis effects compared to the two other groups. The econometric regression results reveal also a determining factor of vulnerability common to the Non-CFA zone and Latin America which is domestic savings. This paper highlights interesting factors and mechanisms relative to the capacity of resilience of certain developing countries with respect to external shocks, in particular those of the Latin America and Sub-Saharan Africa. Further research may be carried out to investigate other variables likely to explain the resistance of Southern countries to the crisis effects and their causalities. Data availability will remain the principal limitation.