PREDICTING EMERGING MARKET RETURNS WITH SIMPLE MACHINE LEARNING TECHNIQUES

This study applies simple machine learning techniques to predict Emerging Market (EM) equity returns. It examines excess returns for major EM indices versus relevant benchmarks. Predictor variables were drawn from the same security’s historical excess returns. The intuition is that past performance can be predictive of future investment interest and that predictive relationship is unique for each security. Therefore, a variety of models need to be considered. A total of five models were applied: traditional logistic regressions, ridge, random forest, boosting and neural nets. The results indicate that newer machine learning-based predictive relationships, such as random forest, boosting, and neural nets, generate valuable out-of-sample results. In nearly all countries, there was a machine learning active strategy that delivered higher returns than a passive long strategy. The paper identifies avenues for further exploration of such investment strategies.