Predicting Stock Prices of Chinese Liquor Companies with the LSTM Network: A Case Study on Shede Spirits

Abstract

This paper presents an investigation into the application of Long Short-Term Memory (LSTM) networks for predicting stock prices of Chinese liquor companies, specifically Shede Spirits. The study utilized a one-year dataset of Shede Spirits' stock prices obtained from Yahoo Finance, covering daily opening, closing, and high prices. The data was preprocessed through sorting, normalization using MinMaxScaler, and creation of input-output pairs with a lookback window of 20 days. The dataset was split into training and testing sets with an 80:20 ratio. A two-layer LSTM model with 32 hidden units per layer was constructed and implemented using PyTorch. Dropout regularization was applied between LSTM layers to prevent overfitting. Experimental results showed the LSTM model's effectiveness in predicting Shede Spirits stock prices, achieving a Root Mean Squared Error (RMSE) of 0.023 and an R^2 score of 0.89 on the test dataset. The LSTM model demonstrated superior performance in capturing non-linear patterns and long-term dependencies compared to linear regression and ARIMA methods. The training loss plot indicated convergence towards an optimal state, and the dropout regularization successfully prevented overfitting. In conclusion, this research highlights the potential of LSTM networks as a powerful tool for stock price prediction in the Chinese liquor industry, offering benefits to investors and businesses. It also emphasizes the importance of employing advanced AI techniques in financial forecasting tasks.