Optimal trading volume decisions based on LSTM price-volume function and mean-variance utility

Deep learning has made significant progress in predicting financial asset prices, yet converting these predictions into trading decisions remains a challenge. This paper proposes a method that leverages LSTM predictions to determine optimal trading strategies, bridging the gap between deep learning forecasts and practical trading strategies. Our approach has two stages. In the first stage, we use LSTM to build a nonlinear price-volume function that captures how prices respond to trading volume based on historical data. In the second stage, we embed this learned function into a mean-variance utility framework and use Monte Carlo simulation to find the optimal trading volume that maximizes utility under market constraints. We test our method on five blue-chip stocks from different industries in the CSI 300 index. The results show that our method outperforms buy-and-hold and simple moving average strategies in cumulative returns, annualized returns, and Sharpe ratio, while achieving lower volatility and maximum drawdown.