Imagine being smarter than the market and generating consistent returns. A new study suggests that this could be possible thanks to machine learning (ML) models that analyze international stock market data.
In the complex world of financial markets, accurately forecasting stock prices is a significant challenge. One approach is based on improving information about stock market anomalies, factors that influence a stock’s performance. Traditional methods that combine information from these anomalies often reach their limits, especially in global stock investments. However, machine learning (ML) methods, a branch of artificial intelligence (AI), offer a promising solution. These methods can aggregate various factors to enhance stock profitability predictions, as demonstrated by a study conducted by researchers from Kaiserslautern and Munich, published in the “Journal of Asset Management.”
Predicting stock returns is akin to forecasting the weather and requires a multitude of data points. These include, for example, temperatures and humidity at high altitude, as well as air currents, cloud cover, and duration of sunlight. Just as detailed weather data are crucial for accurate weather predictions, a large amount of financial data and intelligent methods to combine this information are essential in determining whether an investment is likely to be profitable.
These data include so-called capital market anomalies. “More than 400 of them, identified in recent years by leading financial journals, are considered predictive of stock profitability,” explains Professor Dr. Vitor Azevedo from the University of Kaiserslautern-Landau, co-author of the study. An example is the well-known “price-to-earnings ratio” (P/E ratio) of a stock. Value strategies can use this metric to invest in (seemingly) affordable stocks with low P/E ratios. Another example is the “short-term reversal effect,” where stocks with the lowest returns in the previous month tend to outperform those with the highest returns in the following month.
However, which of these anomalies are relevant? How do they interrelate, and what is their impact when combined? In the study, Azevedo, Professor Dr. Sebastian Müller from the Technical University of Munich, and Sebastian Kaiser from Roland Berger aimed to determine if artificial intelligence could address these questions. “Traditional methods such as regression analyses have their limits in this context,” notes Azevedo. “That’s why we used machine learning methods capable of uncovering complex relationships within large datasets.” This approach is often termed nonlinear combination in expert circles.
For their analysis, the economists examined various ML approaches. They scrutinized nearly 1.9 billion observations of anomalies in stock months from 1980 to 2019 across 68 countries and employed 46 different machine learning models to identify patterns and predict future market movements.
“We found that these AI models significantly outperform traditional methods. Machine learning models can predict stock returns with notable accuracy, achieving an average monthly return of up to 2.71 percent, compared to approximately 1 percent for traditional methods,” adds Professor Azevedo.
Feedback neural networks and composite predictors emerged as the top performers, particularly those employing specific techniques such as elastic net feature reduction and percentage-ranked performance as targets. Even after accounting for transaction costs of up to 300 basis points, the models held strong, demonstrating their robustness and potential profitability.
Importance for Financial Markets
This research sheds light on the power of machine learning in financial markets, which could lead to new investment strategies. Financial managers could potentially use it in the future to develop new stock pricing models.
Additionally, the study’s findings challenge traditional asset pricing theories and suggest that market inefficiencies may exist, exploitable through sophisticated machine learning models.
The researchers from Kaiserslautern and Munich recommend, among other things, careful data preparation to properly account for outliers and missing values, especially when working with international data, as they write in their study. Furthermore, they advise reviewing ethical and regulatory concerns before implementing these AI techniques.
Investors, analysts, and financial professionals should closely monitor the developments in this exciting field. The study was funded by Projekt DEAL.
RPTU Kaiserslautern-Landau, Department of Financial Management
Gottlieb-Daimler-Straße 42, Kaiserslautern, 67663, Germany
Azevedo, V., Kaiser, G.S. & Mueller, S. Stock market anomalies and machine learning across the globe. J Asset Manag 24, 419–441 (2023). https://doi.org/10.1057/s41260-023-00318-z
Note: Article prepared with information from the Rheinland-Pfälzische Technische Universität Kaiserslautern-Landau and scientific study.