Journal of Operational Research and Its Applications ( Applied Mathematics ) - Lahijan Azad University

Sudden and severe stock price crashes pose a significant challenge to capital markets. The substantial losses incurred from such events underscore the need for more effective risk forecasting tools. This study aims to enhance the predictive power of risk models for stock price declines in the Tehran Stock Exchange and commenced with a comprehensive literature review to identify key financial factors influencing stock price volatility. Given the high dimensionality of the dataset and the extended time period, metaheuristic algorithms were employed for feature selection. 10 algorithms, namely Ant Colony Optimization, Hill Climbing, Las Vegas, Whale Optimization, Simulated Annealing, Genetic Algorithm, Tabu Search, Particle Swarm Optimization (PSO), Honey Bee (HBA) and Firefly were utilized to reduce dimensionality and enhance model performance. Five variables, namely "Return on Equity," "Debt Ratio," "Cash Flow to Income Ratio," "Negative Skewness of Stock Returns," and "Logarithm of Sales," were selected based on the outcomes of metaheuristic algorithms. Attention to these five variables is of great importance for economic actors and investors; these variables serve as key indicators in analyzing the financial status and performance of companies and can assist in identifying potential risks. Subsequently, ANNs were implemented to develop predictive models. The models were trained and evaluated using historical data from the Tehran Stock Exchange spanning from 2001 to 2020. The findings of this research demonstrate that combining metaheuristic algorithms for model reduction and optimization, along with advanced machine learning techniques, yields results that can significantly improve risk management and investment decision-making.