Global imbalances (current account imbalances) have become an important issue for economists and policy makers. Greater exchange rate flexibility is often suggested as a means to achieve faster and more efficient adjustment in the current account. However, previous empirical studies show little support for this hypothesis. This paper revisits this issue with a large panel dataset and a threshold VAR model and finds that (1) some existing popular exchange rate classifications may not capture actual exchange rate variability as well as expected; (2) Once exchange rate variability is correctly identified, the speed of mean reversion in the current account balance is indeed higher in a regime with greater exchange rate variability.