Stock portfolio construction using LSTM neural networks and fractal analysis
Robert V. Garafutdinov - Cand. Sc. (Econ.), Associate Professor of Information Systems and Mathematical Methods in Economics Dept., Perm State University
Maksim V. Shevchenko - Student, Perm State University
Abstract
The article tests the hypothesis that combining the methods of fractal analysis and prediction of stock returns using LSTM neural network in the construction of investment portfolio allows improving its characteristics (the ratio of return and risk) compared to the classical Markowitz method and stock index. The study includes several stages: selection of assets to be included in the portfolio; obtaining forecasts of asset returns using LSTM networks; forming an optimal portfolio by maximising the Sharpe ratio; reducing the number of assets in the portfolio using fractal analysis; comparing the resulting portfolios with benchmarks. The research reveals that the combined portfolio constructed on the basis of LSTM predictions and fractal analysis has the best characteristics, confirming the hypothesis of the study. The parameters of the LSTM portfolios are better than those of the Markowitz portfolios of the same assets. Scientific novelty of the article consists in the fact that for the first time the authors have applied a combination of neural network forecasting and fractal analysis methods for portfolio construction.
Keywords: Russian stock market; investment portfolio; modern portfolio theory; return prediction; neural networks; LSTM; fractal analysis; Python
For citation: Garafutdinov R. V., Shevchenko M. V. Stock portfolio construction using LSTM neural networks and fractal analysis. Digital models and solutions. 2025. Vol. 4, no. 2. Pp. 5–17. DOI: 10.29141/2949-477X-2025-4-2-1. EDN: YGNYUG.
