Determinants and Costs of Current Account Reversals under Heterogeneity and Serial Correlation

Empirical evidence suggests that reversing current account balances imply costly adjustment processes. Recent literature analyzes the costs of adjustment processes via treatment regressions and assesses the determinants of reversals via probit models. With larger data sets becoming available, steps to control for unobserved heterogeneity within the data are necessary. This paper contributes a Bayesian analysis, which allows a flexible and efficient handling of heterogeneity and serial correlation and compares the results to those obtained by (simulated) maximum likelihood estimation. While consideration of serial correlation enables to account for the time series character of the data, random coefficients allow an efficient modeling of unobserved heterogeneity across countries. The findings suggest that an efficient consideration of serial correlation and heterogeneity is necessary to assess the determinants and the costs of a reversal correctly. Furthermore, the paper illustrates the dependency of the results on the underlying reversal definition and shows that the costs of a reversal in terms of economic growth are overestimated for some reversal definitions, when heterogeneity and serial correlation are neglected. Bayesian and maximum likelihood based specifications tests provide evidence in favor of models incorporating heterogeneity and correlation. JEL classification: C30; F32; F43