Applying LSTM Model to Predict the Japanese Stock Market with Multivariate Data

Applying LSTM Model to Predict the Japanese Stock Market with Multivariate Data

Using machine learning methods to analyze and predict time series data is a hotspot issue. Because of its potential profitability, it has attracted a lot of research and investment, particularly in the financial field. Compared with other machine learning prediction models, long short-term memory (LSTM) is very effective for processing time series data, due to its special network structure. In this study, we use three models to predict the Japanese stock market movements. These models can be used to learn and predict multivariate data by adjusting the structure and hyperparameters. The original dataset is made up of NIKKEI 225 and some individual stocks. Subsequently, several well-known technical indicators are calculated and added as a new dataset. Two efforts were also made to improve the quality of the dataset. Multiple sets of numerical experiments are established to examine the impact of increasing the number of features on these models and the impact of lengthening the training data on these models. The results show that lengthening the length of training intervals and increasing the number of features can improve the model performance effectively. The LSTM model has better performance than the encoder-decoder LSTM model and CNN-LSTM model in stock market prediction.