Foreign Exchange Risk and the Predictability of Carry Trade Returns (Digest Summary)

By empirically testing an advanced-economy dataset and a global dataset, the authors identify three relationships: (1) Periods of high monthly volatility (MV) and very negative carry trade returns are negatively related, (2) MV can be decomposed into an average volatility (AV) portion and an average correlation (AC) portion, and (3) AV is positively related to MV.

Based on their findings related to the first relationship, the authors introduce and test an improved carry trade strategy. Carry trade returns appear to be predictable to some degree with MV.

If the improved, or “augmented,” carry trade strategy is not sample specific, practitioners can benefit from this new trading strategy that basically stops trading by predicting when conditions are particularly bad for carry trades.

The research will be useful to portfolio managers and hedge fund managers who want to better dissect the sources of carry trade risk and understand the conditions under which different types of risk become more apparent.

The authors form two datasets of monthly carry trade returns: an advanced-economy portfolio that consists of 10 countries (January 1985–April 2013) and a global portfolio that consists of 22 countries, including the advanced economies (January 1998–April 2013). The data are from Thomson Reuters Datastream. To produce the daily carry trade return used to generate the monthly return and the MV (i.e., the variance of the daily carry trade returns during the month, incorporating a one-day lag of autocorrelation), the authors sort each dataset into quintiles based on the differential between the current interest rate and the forward rate at the beginning of each month (high differentials in the top quintile and low differentials in the bottom quintile). A portfolio is then formed by going long the top quintile and short the bottom quintile during the month.

The monthly AV is computed as the average of all of the countries’ MVs within a given month. AC is computed as the average of the correlations between the countries’ MVs within a given month. The authors find that AV is highly positively correlated with MV, but AC is only somewhat positively correlated with MV.

Initially, regressing one-month-ahead carry trade returns against MV produces no statistically significant relationship in either dataset. Nor does a statistically significant relationship exist when AV and AC (a decomposition of MV) are used in the regression instead of MV. Using quantile regressions (i.e., looking at specific segments of the monthly carry trade return distribution with associated MV, AV, and AC measures), however, the authors find that MV and AV have statistically significant negative relationships with the one-month-ahead carry trade returns when the carry trade returns are very negative (i.e., in the left tail of the carry trade return distribution).

Based on these findings, the authors determine that trading strategies that shut down the next month’s carry trade when the current month experiences a negative “tail-like” return and the MV is relatively high improve carry trade performance. These strategies are implemented out of sample. The authors use a rolling three-year window to determine whether the carry trade return is in the tail and whether the MV is relatively high. This information is then applied to the next month’s carry trade implementation. Transaction costs are also considered as part of the trading strategy.

This is an interesting piece of empirical work that I would like to see extended to determine under what circumstances other risks could matter within the carry trade return distribution—in particular, whether there could be a particular risk that becomes more apparent in the right tail of the carry trade return distribution that is not relevant in the left tail of the distribution.