The use of generative neural networks to predict market trends and manage asset portfolios

Abstract

This article presents a study of the use of generative neural networks for predicting market trends and managing asset portfolios, performed according to the IMRAD structure. The Introduction section substantiates the relevance of using artificial intelligence in financial modeling. The growing volume of market data and the complexity of its analysis require the use of modern algorithms capable of identifying hidden patterns and predicting trends. The necessity of transition from traditional methods of analysis to models based on generative neural networks, which demonstrate potential in working with unstructured data and can provide high accuracy of forecasts, is substantiated. The Methods section describes the architecture of the developed model based on variant autoencoders and generative adversarial networks (GANS). The stages of data preprocessing, forming training samples, and configuring the hyperparameters of the model are described. Comparative characteristics of algorithms based on forecasting quality metrics such as accuracy, completeness, and the Sharpe coefficient for evaluating the effectiveness of asset portfolio management are presented. The study used data obtained from several stock exchanges over the past ten years, which allowed for large-scale testing of the model. The "Results" section demonstrates that the proposed approach makes it possible to significantly improve the quality of forecasting market trends compared to classical methods. The analysis showed that generative neural networks are able to identify long-term dependencies, which has a positive effect on optimizing the structure of the asset portfolio. The improvement in risk and profitability indicators is confirmed by statistical tests, and the model demonstrates stable performance in a volatile market environment. The Discussion section analyzes the strengths and weaknesses of the method. The prospects for integrating algorithms into automated asset management systems are discussed, and potential limitations related to the interpretability of neural network analysis results are identified. The result of the study is the conclusion about the high efficiency of generative neural networks in forecasting and managing investment portfolios, which opens up new opportunities for the use of artificial intelligence in financial markets.