Comparison of the Marshall-Lerner condition in OECD and Asian countries: new evidence from pooled mean group estimation

This paper scrutinizes the well-known Marshall-Lerner (M-L) condition for a group of OECD countries (non-Asian) and Asian countries over the period from 2000 to 2017. We apply a pooled mean group (PMG) estimator to overcome the shortcomings of the other methods, such as the most famous in the recent literature—the autoregressive distributed lag (ARDL), which performs over DOLS, FMLS, and MLE and works despite having endogenous variables in the model. However, the ARDL provides significant but economically implausible coefficients due to country-specific omitted variables, as confirmed by recent empirical studies. In addition, the ARDL ignores the economic convergence assumption that is reasonably included in the PMG approach, assuming homogeneity of the long-run coefficients and heterogeneity of short-run dynamics. Homogeneity of long-run coefficients appears reasonable due to technology and knowledge mobility as a result of globalization. Furthermore, PMG estimator performs better than mean group (MG) and dynamic fixed-effect (DFE) approaches, as it provides more sensible results. We found that the M-L condition holds for Asian countries but not for OECD countries and that there exist signals of J curves for Asian countries in our results. This indicates that currency devaluation works for Asian countries but not OECD countries.