Series decomposition Transformer with period-correlation for stock market index prediction

Stock price forecasting has been always a difficult and crucial undertaking in the field of finance. In the last few decades, deep learning models based on RNNs and LSTMs have dominated the research, where the stock price data are modeled as time series data. However, the high volatility of stock prices and the decay of information learned from historical data prevented these models from achieving more accurate predictions in this problem. Recently, Transformer has been gradually applied in time series prediction, but the methods aim to feed the highly-uncertain social media information as the additional auxiliary information into Transformer, rather than improving the ability to extract features from historical series. In this paper, we propose a Series Decomposition Transformer with Period-correlation (SDTP), which uses the period-correlation mechanism and series decomposition layers to further discover relation between historical series and learn the changing trends in the stock market for high forecasting accuracy and generalizability. The extensive experimental results show that the proposed SDTP model generally outperforms the state-of-the-art methods on a collection of datasets.