Precisely predicting stock prices is a major challenge in the intricate world of financial markets. Enhancing the data from stock market anomalies—factors that affect a stock’s return—is one strategy. Especially in international stock investments, traditional approaches that aggregate data from these anomalies frequently run into restrictions.
Machine Learning (ML) techniques, a subfield of Artificial Intelligence (AI), present a viable remedy, nevertheless. A study titled “Stock market anomalies and machine learning across the globe” by academics from Kaiserslautern and Munich, published in the Journal of Asset Management, demonstrates how these approaches can combine many characteristics to improve stock return projections.
Similar to predicting the weather, predicting market returns also requires a large number of data points. The temperature and humidity at high altitudes, air currents, cloud cover, and length of sunshine are a few examples of these. Extensive financial data and clever ways to combine this information are necessary to assess whether an investment is likely to be lucrative, much as precise weather forecasting depends on thorough meteorological data.
Among this data are what are known as capital market abnormalities. According to Professor Dr. Vitor Azevedo, a co-author of the study from the University Kaiserslautern-Landau, more than 400 of these have been identified by reputable financial journals in recent years as predicting factors for stock returns.
A prominent illustration is the “Price-Earnings Ratio” (PER) of a stock. Using this criterion, so-called value strategies can purchase (apparently) affordable stocks with low price-to-earnings ratios. An additional illustration is the “Short-Term Reversal” effect, which states that stocks that performed poorly the month before typically do well the next month.
Which of these oddities, though, is significant? What effect do they have when combined, and how do they connect to each other? In order to ascertain whether artificial intelligence could provide answers to these queries, Azevedo, Sebastian Kaiser of Roland Berger, and Professor Dr. Sebastian Müller of the Technical University of Munich conducted a study.
According to Azevedo, conventional techniques like regression analysis have inherent limitations in this situation. They employed machine learning techniques that could find intricate correlations in large data sets because of this. Expert circles frequently refer to this strategy as a nonlinear combination.
The economists looked at several ML techniques in order to conduct their analysis. They examined 68 countries’ worth of about 1.9 billion stock-month anomaly observations between 1980 and 2019.
These AI models perform noticeably better than conventional techniques, as they discovered. In contrast to roughly 1% for traditional methods, the machine learning models achieve an average monthly return of up to 2.71% in stock return prediction, demonstrating their amazing accuracy, says Professor Azevedo.
The study’s conclusions show how beneficial such technology can be to the financial sector. It could be used by financial managers to create new stock price models in the future. According to their study, the researchers from Kaiserslautern and Munich recommend, among other things, rigorous data preparation to accurately incorporate outliers and missing values, especially when working with international data. Before using these AI tools, they also advise examining ethical and regulatory considerations.
