Current account dynamics and degree of capital mobility

Abstract

By modelling the current account balances (as a percentage of GDP) in a dynamic AR(1) model, Taylor (2002) proposed to use speed of mean reversion of the dynamics of the current account as a tool for measurement of capital mobility and confirmed the stylized fact of U-Shape degree of capital mobility through the last two centuries with this new approach. With the assumption that countries obey their Long-Run Budget Constraint (LRBC), the OLS estimate of the degree of current account persistency in Taylor (2002) is downward bias due to small sample and high serial correlation of current account ratios with its lags. By correcting these biases with the Andrews (1993) exact median-unbiased estimation, we find that the confidence intervals for half-life estimates include 1 year (which was used as a benchmark) in 37 out of 75 country–period cases, which cast some doubt on the use of half-life estimate as a practical measure of capital mobility.

Keywords:

• current account adjustments
• capital mobility
• current account persistency

JEL Classification:

• F21
• F32

Notes

¹ Even though the countries are assumed to obey their Long-Run Budget Constraint (LRBC), they do not have to in the short run. Coakley et al. (2004) show that current account-to-GDP ratio may behave in an observationally nonstationary way. Raybaudi et al. (2004) propose a switching ADF model of current account where in one period the LRBC does not hold; hence, current account-to-GDP ratio follows a nonstationary process in some periods.

² In the literature on purchasing power parity puzzle, Murray and Papell (2002) suggest to use Andrews (1993) exact median-unbiased estimator which explicitly accounts for serial correlation, sampling uncertainty and small sample bias.

³ Andrews (1993) shows the robustness of the median-unbiased estimator to non-Gaussian errors. If the error terms are skewed and kurtotic with finite variance, then they demonstrate that the approximation error derived from incorrect assumption of Gaussian errors is small.