The Rise and Fall of the Carry Trade: Links to Exchange Rate Predictability
52 Pages Posted: 27 Sep 2019
Date Written: September 17, 2019
We investigate out-of-sample exchange rate predictability in a high-dimensional panel predictive regression model that includes numerous country characteristics and their interactions with a variety of global variables. To avoid the overfitting problem that plagues conventional estimation of high-dimensional models, we estimate the panel predictive regression model via the elastic net, a machine learning technique based on penalized regression. The elastic net forecasts significantly outperform the no-change benchmark forecast that has proven difficult to beat in the literature. Out-of-sample exchange rate predictability becomes considerably stronger starting in the fall of 2008 during the worst stage of the global financial crisis. We show that exchange rate predictability can substantially improve the performance of conventional carry trade strategies: a smart carry portfolio that incorporates the information in the elastic net forecasts avoids the crash experienced by a conventional carry portfolio in late 2008 and markedly improves portfolio performance thereafter.
Keywords: Carry, Currency excess return, Panel predictive regression, Machine learning, Penalized regression, Short-horizon predictability, Global financial crisis
JEL Classification: F31, G11, G12, G15
Suggested Citation: Suggested Citation