Machine Learning Models for Trading Dollar Neutral Portfolio of NDF Currencies

This report explores various machine learning based strategies for trading NDFs. With total of 20 currencies under study, the best single currency performance was observed for a hybrid model based on lasso and elastic net, reporting an average correct prediction of 53%. Further, we build an equal weighted as well as an optimized dollar neutral portfolio comprising all the 20 currencies, and investigate the performance of our overall strategy. With various checks for robustness, our final portfolio has an information ratio of 0.8, beating the benchmark momentum portfolio with information ratio of 0.3