Machine Learning Methods in Asset Pricing: An Analysis of Cross-sectional Stock Returns with Macroeconomic Factors in Brazil

Abstract

The literature on financial machine learning has growth rapidly with studies encompassing several asset classes, chiefly for the stock market. We apply machine learning methods to Brazilian stocks cross section of monthly excess returns by using Brazilian stock factors while adding a vast set macroeconomic ones as our research primary contribution, for which the literature on Brazilian equities is scarce. We confirm recent results that ML models drive a substantial improvement in out-of-sample R2 predictive power over traditional OLS models. Running an out-of-sample variable importance analysis, we also found macroeconomic factors overweight firm-related ones, with a slight predominance of country risk (EMBI Brazil Index), followed by the expectations of economics conditions, and Brazil’s commodities composite index, and credit-to-GDP ratio. Our findings suggest a high relevance of macroeconomic factors when predicting monthly excess returns for Brazilian stocks