Comparing Earnings Prediction Models to Analysts’ Consensus Forecasts: A Machine Learning Approach

Details

Supervisors

Vande Kerckhove, Corentin

Faculty

Louvain School of Management

Degree label

Master [120] : ingénieur de gestion, à finalité spécialisée

Abstract

Value investing is an investment approach that involves selecting undervalued assets based on fundamental analysis, with the expectation that their true intrinsic value will be recognized by the market over time, leading to abnormal long-term returns. We apply machine learning techniques, specifically histogram-based gradient boosting regression trees, to predict one-year net income growth using a trailing three-year window of historical company fundamentals. The premise is that an accurate earning prediction leads to the selection of better-performing stocks, enabling the development of a long-holding, systematic value investing strategy. We assess the predictive performance of the approach with that of analysts' consensus forecasts alone and a similar approach combining analysts' consensus forecasts and company fundamentals as model input data. With respective R² values of 0.045 and 0.232, results do not show that making use of fundamental data alone can offer comparable results to analysts’ forecasts. Moreover, while the combined approach shows marginal improvements over analysts' consensus forecasts, the potential of combining consensus forecasts with historical fundamentals remains underutilized with the chosen configuration. These findings suggest future research should make use of more sophisticated models incorporating historical fundamentals alongside analysts' consensus.