Mining Latent Patterns in Stock Market Dynamics via Association Rules

Source abstract

This study is dedicated to the development of a methodology for identifying complex non-linear dependencies in financial markets through association rule mining. The relevance of this work is driven by the inherent complexity of forecasting stock index dynamics due to high volatility, the inefficiency of traditional linear models, and the presence of hidden (latent) patterns within massive financial datasets. The article proposes an approach that integrates data mining methods, specifically association rule discovery algorithms (e.g., Apriori, FP-Growth), to identify stable correlations between various technical, fundamental, and macroeconomic indicators. These discovered rules are subsequently integrated into predictive models of non-linear dynamics, such as artificial neural networks or decision tree-based machine learning methods, enabling the accounting of complex factor interactions. Experimental validation of the methodology was conducted using data from stock markets. The results demonstrate an increase in the accuracy of short- and medium-term forecasts, the model's ability to identify early signals of trend formation, and the recognition of non-obvious predictors of market movements. The findings hold significant practical value for the development of automated investment decision-support systems, risk management systems, and algorithmic trading, providing a deeper understanding of the nature of financial markets.