A Hybrid Forecasting Model for Stock Price Prediction: The Case of Iranian Listed Companies

Abstract

This paper introduces advanced computational methods for stock price prediction, integrating Fast Recurrent Neural Networks (FastRNN) with meta-heuristic algorithms such as the Horse Herd Optimization Algorithm (HOA) and the Spotted Hyena Optimizer (SHO). By challenging the Efficient Market Hypothesis (EMH) and Random Walk Hypothesis, our research demonstrates the effectiveness of these hybrid models in semi-strong or weak-form efficient markets. The study leverages data from five listed Iranian companies (2011–2021) and 25 factors encompassing technical, fundamental, and economic considerations. Our findings highlight the superior accuracy of the FastRNN optimised by HOA, SHO, and a Generative Adversarial Network (GAN) in forecasting stock prices compared to conventional FastRNN models. This research contributes to the multidisciplinary field of computational economics, emphasising advanced computing capabilities to address complex economic problems through innovative econometrics, optimisation, and machine learning approaches.