Quantum-inspired neural networks enhance stock prediction accuracy

The inherent complexities and turbulence of the stock market often challenge the predictive capabilities of traditional neural networks like long short-term memory (LSTM) models. However, recent explorations into quantum-inspired computing to deal with the inherent volatility of the stock market offer novel paradigms for handling uncertainty. In this study, we investigated whether incorporating quantum-inspired elements into LSTM neural networks could improve stock price prediction accuracy, especially during periods of high market volatility. We developed a quantum-inspired LSTM (QiLSTM) model featuring a layer designed to mimic quantum concepts like wave-like behavior using sine functions and compared its performance against a traditional LSTM model. We hypothesized that the QiLSTM would outperform the standard LSTM, especially in volatile markets, due to its potential to better capture uncertainty and errata. Using stock data from 2010 to 2024, we tested both models during a recent low volatility period (late 2023-early 2024) and a high volatility period corresponding to the COVID-19 market crash (early 2020). Supporting our hypothesis, the QiLSTM model demonstrated statistically significant superior performance during the high volatility period, achieving lower mean squared error (MSE) and mean absolute percentage error (MAPE) with large effect sizes. During the low volatility period, performance differences were large, with the traditional LSTM significantly outperforming QiLSTM on both MSE and MAPE, indicating that the quantum-inspired layer may be detrimental in low volatility cases.