Machine Learning - The Future of Equity Premium Prediction
Sammanfattning: Predictions of the equity premium have historically been made by using traditional predictive regressions. Despite the great promise of machine learning applications for prediction tasks, it has largely been overlooked in the financial literature. We predict the monthly equity premium, defined as the monthly excess return on the S&P 500, using three linear and five non-linear machine learning models. The models are evaluated against a benchmark consisting of the historical average return. Six of our models outperform the benchmark, with the three most successful being non-linear models. We perform a statistical evaluation of the machine learning forecasts, finding that three of the outperforming models are significantly different from the benchmark. Additionally, the models are evaluated economically by calculating an implied Sharpe ratio using the predictive results, showing meaningful economic gains even for models with only a slight increase in predictability. Successively, we translate our forecasts to their inherent directional prediction, finding that four models beat the benchmark. By formulating a naïve investment strategy, we show that also the directional predictability can be exploited to generate a higher Sharpe ratio than that of the market.
HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)