Enhancing Stock Market Predictions Through the Integration of Convolutional and Recursive LSTM Blocks: A Cross-market Analysis

Abstract

This study explores convolutional and recursive LSTM blocks within a singular architecture for forecasting stock prices. We propose a method that integrates convolutional networks, which learn to process signals through filters, with recursive LSTM blocks to account for critical temporal information often overlooked in convolutional approaches. Our investigation primarily revolves around two research questions: (1) Can integrating convolutional and recursive LSTM blocks within a singular architecture enhance prediction accuracy? and (2) What is the impact of training and testing with disparate data distributions? The latter question arises from our experiment of training the model using data drawn from the Indian Stock Market and testing the predictions with New York Stock Market data, thus deviating from the traditional focus on uniform stock market distributions. Our results reveal a notable improvement in prediction accuracy (MAPE reduction of 2.22%), strongly suggesting that pre-processing data via Convolutional Neural Networks (CNN) benefits LSTM blocks and can enhance the performance of stock market prediction methodologies.

Supported by Universidade Federal de Ouro Preto, FAPEMIG (APQ-01518-21), CAPES and CNPq (308400/2022-4).