• Emilie Krogsgård Rom
4. term, Science in Economics, Master (Master Programme)
This thesis investigates whether machine learning techniques such as random forest and gradient boosting improve cross-sectional asset return forecasting statistically and economically compared to traditional OLS regression. The analysis deviates from extant literature in three ways: (1) addressing the U.S. bias by analyzing Scandinavian stocks, (2) including fewer variables to get an eye-level comparison, and (3) excluding micro caps. Surprisingly, this thesis finds OLS exhibiting superior predictive performance that translates into economic profitability relative to machine learning techniques. However, OLS is only marginally better than gradient boosting and all three portfolios generate significant Sharpe ratios. Reasons for the underperformance might include limitations in the data size and potential over-/underfitting. Nevertheless, there is moderate support for the results when looking at the deviations in the methodology (1-3) and when seeing the existing results in the literature through a critical lens.
Publication date2023
ID: 532546585