INTEGRATING SYMBOLIC GENETIC PROGRAMMING WITH LSTM FOR FORECASTING CROSS-SECTIONAL PRICE RETURNS: A COMPARATIVE ANALYSIS OF CHINESE AND JAPANESE STOCK MARKET

This paper introduces an advanced framework integrating Symbolic Genetic Algorithm (SGA) with Long-Short Term Memory Neural Network (LSTM) to forecast cross-sectional price returns using technical and fundamental indicators of 4,700 listed stocks in China and over 4,600 listed stocks in Japan spanning from 2014 to 2022. Leveraging the S&P Alpha Pool Dataset for both countries, the framework incorporates data augmentation and feature selection techniques. The research showcases substantial enhancements in Rank Information coefficient (Rank IC) by 588.03% in China and 194.27% in Japan, respectively. Moreover, a straightforward rule-based strategy rooted in the proposed hybrid SGA-LSTM model outperforms major Chinese stock indexes, delivering average annualized excess returns of 22.35% and 16.26% compared to the CSI 300 index and CSI 500 index in China, respectively, and surpasses the N225 index by 4.56% and the TPX index by 5.10% in Japan. These findings underscore the effectiveness of LSTM with SGA in optimizing the accuracy of cross-sectional stock return predictions and offer valuable insights for fund managers, traders, and financial analyst.