Abstract
This paper, by applying the recent panel data error-correction modelling developed by Gengenbach et al., Panel error correction testing with global stochastic trends, 2008 University of Maastricht, Working Paper, RM/08/051:1–51 and Gengenbach et al., Journal of Applied Econometrics, 2016, 31, 982–1004 tests the validity of the Feldstein–Horioka puzzle (1980) in a panel of 27 African countries for the period 1965–2015. Unlike the existing studies in the literature, this paper, for the first time, addresses several important panel data econometric issues such as slope heterogeneity, cross-section dependence, unobservable common factors, non-stationarity and endogeneity that would yield inconsistent and biased estimates of the parameters of interest. Unlike the previous residual-based error-correction modelling that has low power, the Gengenbach et al., Panel error correction testing with global stochastic trends, 2008, University of Maastricht, Working Paper, RM/08/051:1–51 and Gengenbach et al., Journal of Applied Econometrics, 2016, 31, 982–1004 estimator explicitly considers structural dynamics in modelling a cointegrated panel that allows for the presence of non-stationary unobserved common factors. As the GUW modelling depends on the Pesaran Common Correlated Effects Mean Group Estimator (CCEMG; 2006) and the Chudik and Pesaran Dynamic Common Correlated Effects Mean Group Estimator (DCCEMG; 2015), we report the results based on these estimators. The econometric exercises performed in the paper show that slope heterogeneity; cross-section dependence and non-stationarity are present in the panel data sample used in the estimation. Estimations yield a relatively low magnitude of the savings retention coefficient for the panel members included in the study with a negative error-correction term that is statistically significant at the individual and panel levels. The empirical evidence rejects the presence of the Feldstein–Horioka puzzle in the panel of 27 African countries studied.