The Rise of Artificial Intelligence in Investment Strategies: Evaluating the Performance of Machine Learning in Stock Market Prediction

Details

Supervisors

Candelon, Bertrand

Faculty

Louvain School of Management

Degree label

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

Abstract

enThis thesis investigates the transformative impact of artificial intelligence on investment strategies. The study focuses on the predictive capabilities of machine learning models in stock market forecasting. The research begins by explaining how these contemporary approaches developed through various historical periods. The forecasting capabilities of these modern methods are then analyzed through a wide range of models, including Prophet, Random Forest, Gradient Boosting, Support Vector Regression, Multi-Layer Perceptrons, and Long Short-Term Memory networks. The results are subsequently compared to a traditional statistical tool, namely ARIMA, which serves as a benchmark to evaluate the machine learning models. The study applies these models to predict S&P 500 Index movements through empirical evaluation. The research uses log returns, price levels, and rolling one-year returns as prediction targets and evaluates the performance across short-, mid- and long-term horizons. The accuracy of these forecasts is evaluated using a combination of various statistical metrics (including MAE, RMSE, R², among others) and finance-specific indicators (such as Sharpe Ratio, KS Test, Diebold-Mariano Test). The results of this analysis show that machine learning models indeed show a stronger performance than traditional methods such as ARIMA in short-term and volatile market conditions. Interestingly, no single model consistently dominated across all forecasting targets and timeframes. On the contrary, every model’s performance depended heavily on the market regime, the forecasting objective, and the time horizon. Furthermore, the thesis is complemented by an extensive literature review to put these findings into perspective and compare them. A risk assessment is also included with research about limitations in the use of AI and expert interviews on current applications in investment banks. The major risks components are comprised of overfitting, lack of transparency, and a systemic risk amplification. These challenges underscore the importance of being able to explain an AI’s thinking and strong regulatory framework and oversight. As a conclusion, this study unites historical, empirical, and theoretical dimensions. It offers valuable insights for investors, researchers, and policymakers by contributing to the academic understanding and practical application of AI in financial forecasting.