Title: Forecasting corporate earnings with machine learning
Authors: Christoffer Thimsen - Aarhus University and CREATES (Denmark)
Jorge Wolfgang Hansen - Aarhus University and CREATES and the Danish Finance Institute (Denmark) [presenting]
Abstract: A comparative empirical analysis is performed for a set of machine learning methods for predicting future firm-level earnings. In addition, we show how to incorporate monthly and quarterly data into the forecast models. We consider generalized linear models, dimension reduction, boosted regression trees, random forests, and neural networks. We find that the best performing machine learning models are superior to existing accounting-based predictive models and our findings are consistent across industries and firm characteristics. Utilizing intra-year data enhances also the practicability of our forecasts and makes them a useful alternative to the commonly applied analyst forecast. In particular, we show that the forecasts from the machine learning models are unbiased and outperform the usually over-optimistic financial analysts forecasts. Our findings show that machine learning significantly increases our ability to predict future economic outcomes and gives insight into where machine learning is particularly useful.