In an post in VoxEU, Shang-Jin Wei alluded to work we have undertaken examining whether de facto exchange rate regimes have an impact on current account reversion.
What do we do in this paper? We examine whether the rate at which current account balances revert to mean. We examine the sensitivity of the results to exchange rate regimes, to country groups, and to estimation methods. In addition, we examine if similar patterns hold for real exchange rates.
What do we find? There is no robust evidence that more flexible regimes exhibit faster current account reversion. While current account balance reversion is typically fastest for currencies on a pure or near-pure float, reversion is not necessarily slower as the regime becomes more inflexible. Rather, pure float or pure fix regimes typically exhibit fastest reversion.
More specifically, in a standard pooled regression, no statistically significant effect is detected. In a time fixed effects model, fixed exchange rate regimes exhibit greater persistence, although more flexibility does not necessarily lead to faster reversion.
To be more specific, how do we obtain these results? We estimate autoregressions, interacting dummy variables for de facto exchange rate regimes. We assume AR(1) time series processes, which appear reasonable for annual data (see Chinn and Prasad [pdf]). We also control for trade openness (the sum of exports and imports, normalized by GDP) and financial openness (the Chinn-Ito index).
We test our models on 170 countries worth of current account, trade opennesss, GDP data from World Development Indicators, with the real effective exchange rates from the IMF’s International Financial Statistics. The exchange rate regime data are from Levy-Yeyati/Sturzenegger  (cross-checked with data from Reinhart/Rogoff ).
What are the statistical results? The way to interpret the results is to associate higher AR(1) coefficients with greater persistence. Hence, the lower the AR(1) coefficient, the faster current account balances revert to mean. The priors, associated with conventional wisdom and advice to move to more flexible exchange rates, would suggest monotonically higher AR(1) coefficients as one moves to more rigid regimes. Figure 1 shows the AR(1) coefficients for the non-industrial country sample (blue bars) and non-industrial ex oil country sample (red bars).
Figure 1: AR(1) coefficients for current account to GDP ratios estimated in pooled OLS regressions on non-industrial country sample (blue bars) and non-industrial ex oil sample (red bars), stratified by Levy-Yeyati and Sturzenegger classifications. Source: Chinn and Wei (2008).
We check to see if the results are robust against several modifications. First, size. Larger countries exhibit slower current account reversion, when compared to smaller (as measured by PPP GDP). However, if there is any effect from fixed exchange rates, it is that fixed rate countries exhibit faster reversion. Interestingly, when we stratify by G-7 versus non-G-7 countries, we find a similar pattern for non-G-7 countries: faster reversion under fixed rates.
Astute readers will observe that we take the de facto exchange rate regime as exogenous. However, one could plausibly argue that the selection of exchange rate regime is a function of current account reversion, or alternatively, the exchange rate regime and the rate of reversion are both functions of a common factor. In order to account for this possibility, we use a two-stage instrumental variables procedure to deal with potential endogeneity.
Specifically, what we do is to estimate a probit for exchange rate regime, using as determinants the variables suggested by Levy-Yeyati and Sturzenegger (2003) [pdf]: economic size, land area, island dummy, inital foreign exchange reserves, as well as a regional factor (in LYS, it’s the average exchange rate regime for the region; we just use regional dummies).
In the end, we find that instrumenting does not change the results. Exchange rate regimes do not affect the pace of reversion in a statistically significant fashion.
We also examine real (trade-weighted) exchange rate persistence. Here we do find more fixed regimes exhibit more rate persistence in the fixed effects specification.
Figure 2: AR(1) coefficients for log real effective exchange rates estimated in fixed effects regressions on non-industrial country sample (blue bars) and non-industrial ex oil sample (red bars), stratified by Levy-Yeyati and Sturzenegger classifications. Source: Chinn and Wei (2008).
The details of the exchange rate regressions, I leave for a future post.
Now, let me depart from the empirical results in the paper to discuss how these findings inform the ongoing debate about Chinese adjustment. As I’ve discussed in previous posts , a more rapid pace of real, effective, exchange rate appreciation would be beneficial to China (in terms of inflation stabilization and external balance adjustment), as well as to the world economy . But that is separate from the issue of whether greater exchange rate flexibility induces faster current account reversion.
(I’ll be presenting this paper at the ASSA meetings, subbing for Frankel/Parsley/Wei in the panel “Exchange Rates and Trade Prices in Emerging Markets”.)
real effective exchange rate, href="http://www.technorati.com/tags/exchange+rate+flexibility">exchange rate flexibility,