Can Long-short Term Memory Neural Network With Symbolic Genetic Algorithm Predict Stock Price Change Basing on Fundamental Indicators

Abstract

Abstract This paper presents an enhanced framework that combines Symbolic Genetic Algorithm (SGA) with Long-Short Term Memory Neural Network (LSTM) for predicting cross-sectional price returns using fundamental indicators of 4,500 listed stocks in China. The study addresses the challenges posed by fundamental indicators resembling smart beta factors in efficient markets and the low frequency of fundamental indicator updates for deep learning models (DNN). The proposed DNN framework incorporates data augmentation and feature selection techniques, resulting in significant improvements in Rank Information coefficient (Rank IC) and IC information ratio (ICIR) by 1,128% and 5,360% for fundamental driven data. Additionally, a rule-based strategy based on the hybrid SGA-LSTM model outperforms major Chinese stock indexes, generating impressive average annualized excess returns compared to the CSI 300 and CSI 500 indexes. These findings highlight the effectiveness of LSTM with SGA in optimizing cross-sectional stock return predictions based on fundamental indicators, providing valuable insights for financial professionals..