Stock Price Forecasting and Portfolio Selection Through Machine Learning: An Application on BIST Participation 30 Index

This study aims to forecast stock prices of companies listed in the BIST Participation 30 Index using machine learning techniques and construct optimized portfolios based on these forecasts. Two methods, Linear Regression (LR) and Gated Recurrent Unit (GRU), were applied for price forecasting, and the results were used to create equal-weighted and return-weighted portfolios using the Markowitz mean-variance model. The analysis shows that the GRU model significantly outperforms LR in terms of forecast accuracy, leading to more profitable portfolio strategies. The return-weighted portfolio consistently showed higher performance compared to the equal-weighted portfolio and the benchmark index. These findings highlight the effectiveness of machine learning models, particularly deep learning algorithms like GRU, in enhancing investment strategies and portfolio management within the context of portfolio selection. The study provides a framework for future research to explore other indices and machine learning models.