A PCA-LSTM Model for Stock Index Forecasting: A Case Study in Shanghai Composite Index

Financial data prediction is greatly facilitated by the advanced machine learning techniques. Thus, financial index prediction based on supervised learning is attracting increasing attention with the surge of quantitative trading. In view of the hysteresis nature of the traditional financial index prediction, this paper proposes a novel PCA-LSTM approach for real-time stock index forecasting, which is both accurate and efficient. Specifically, a large dataset is first built up by analyzing the trading indicators and calculating the technical indicators of the Shanghai composite index in the last 10 years. PCA is then employed for data feature extraction, via redundancy removing and dimensionality reducing. In particular, we further formulate a 4-layer LSTM neural network to model the extracted temporal features, by considering the current states of trading indicators and the information of previous technical indicators. A number of comparative experiments are conducted on the new established dataset, which verify the effectiveness of our method.