Despite evidence of their herding behavior and optimism bias, equity analysts have been acknowledged for providing the best proxy for future corporate earnings for many decades. Previous studies have documented equity analysts’ superior forecast accuracy compared to statistical models throughout the 1980s and 1990s. The past decades have shown signi cant enhancements in CPU e ciency and computational power, enabling the use of big data and arti cial intelligence. While the use of machine learning has revolutionized areas in the nancial industry ranging from algorithmic trading to risk management and fraud detection, the application of machine learning has not been explored in the context of corporate earnings forecasts. This study leverages state-of-the-art machine learning algorithms to explore the potential of machine learning as a standalone estimator for future corporate earnings per share. By implementing a machine learning algorithm based on accounting gures and stock information we compare the forecast accuracy to those of equity analysts at horizons of one, two, and four quarters to evaluate the applicability of machine learning. We nd that the Catboost algorithm produces more accurate earnings per share forecasts than equity analysts on average across all horizons, which is statistically signi cant on horizons of two and four quarters. In addition, we nd that the Catboost algorithm forecasts are more accurate than equity analysts’ when there is a low analyst coverage and a high forecast dispersion among analysts as well as when the market capitalization is small. The results of this study provide preliminary support for the potential to use machine learning as a standalone estimator for future corporate earnings.
|Educations||MSc in Finance and Accounting, (Graduate Programme) Final Thesis|
|Number of pages||100|