Machine Learning and Currency Carry Strategy
Log in to collectAcademic paper
The Rise and Fall of the Carry Trade: Links to Exchange Rate Predictability
Ilias Filippou; David E. Rapach; Mark P. Taylor; Guofu Zhou
- Washington University in St. Louis
- ?Washington University in St. Louis - John M. Olin Business School
- Federal Reserve Bank of Atlanta
- ?Research Department, Federal Reserve Bank of Atlanta
- Brookings Institution
- Centre for Economic Policy Research
- ?Centre for Economic Policy Research (CEPR)
Strategy in a nutshell
The strategy invests in currencies from 14 countries, switching to the Euro for Germany, Italy, France, and the Netherlands after 1999. It uses country-specific (inflation, unemployment, yields, market metrics) and global (policy uncertainty, geopolitical risk, volatility, illiquidity) predictors. A panel regression with elastic net predicts exchange rate changes, which are combined with interest rate differentials to estimate excess returns. Portfolio weights are determined via mean-variance optimization using an exponentially weighted variance-covariance matrix (decay 0.94), and rebalanced monthly.
Economic rationale
The "smart" carry strategy improves the classic carry trade by incorporating predicted exchange rate changes, mitigating risks from currency appreciation. Traditional carry trades performed well historically but faltered during crises due to large USD gains. By accounting for predictable currency moves and applying machine learning to reduce predictor dimensionality, the strategy avoids overfitting, providing more reliable excess returns and outperforming conventional approaches, especially during turbulent periods.