countries we examine are Canada, Germany,1 Japan, Korea, New Zealand, Norway, Singapore, Sweden, United Kingdom and the United States. Data availability also dictates that it is more convenient to set the U.S. as the home country for our calculations (since bilateral exchange rates with all currencies with the dollar are easily available), even though we are primarily interested in the Asian economies in our data set. Consequently, we focus primarily on panel data techniques to make up for the short time-series length of our sample.
Our investigation follows two empirical paths. First we try to ascertain the drivers of the dynamics of the carry trade with a novel local projection approach (Jordà, 2005) adapted to panels with cointegration (Chong, Jordà and Taylor, 2009). Second, we pursue a forecasting exercise such as a trader would conduct, and evaluate competing models on the basis of trading loss functions (as opposed to the more conventional forecast loss functions based on root mean squared errors, for example). We use formal statistical tests of predictive ability using these loss functions and rolling regressions with Giacomini and White's (2006) approach.
We find strong corroborating evidence that while exchange rates may be difficult to predict, deviations of the fundamental equilibrium exchange rate help predict large carry trade unwinds of the type described in the opening quote. This can be seen as well when examining the dynamics of the carry trade itself. However, while it is difficult to detect a strong role for volatility using the first of our empirical paths, we find that a more judicious specification (involving a threshold model) can provide useful auxiliary information for carry traders, thus somewhat supporting both Jordà and Taylor's (2009) results and the apparently contradictory results in Brunnermeier et al. (2009).
The paper is structured as follows. We open with a brief description of the data and summary statistics and then continue to discuss the two empirical strategies that we use in this paper, which are somewhat unconventional. The following two sections report the results in each case and offer a commentary on the results. We close with a summary of our results and a brief discussion of their implications for policy makers.
1 We realize that although we use Germany's data on inflation and interest rates, the euro is determined by the European Central Bank and hence may reflect factors affecting other countries. However, as the largest economy in the euro area, we feel this approximation error is small.