Predicting Equity Returns with Forecast Combinations of Deep Learning and Ensemble Methods

We contribute to the literature by analyzing forecast combination methods in the context of machine learning to predict equity returns. Whilst individual models lack robustness, forecast combinations display stability and are able to produce improved results with Sharpe ratios up to 3.16. We use decision trees in genetic algorithms to analyse the structure of variable influence. The impact of these variables displays inconsistencies and shows variations across different models and data. We propose a new performance measure for risk premium forecasts which leads to more robust evaluation than existing performance measures, whilst providing economic interpretability. This measure can be linked to the advantages models offer for portfolio choice