Dynamic forecasting of the Shanghai Stock Exchange index movement using multiple types of investor sentiment

Shangkun Deng, Chongyi Xiao, Yingke Zhu, Yu Tian, Zonghua Liu*, Tianxiang Yang

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Direction forecasting of stock market movements using the market investor sentiment is a significant subject in the research area of financial market. In this study, a novel integrated machine learning approach LightGBM-NSGA-II-SW is proposed for stock index prediction and simulation trading by combing the LightGBM (Light Gradient Boosting Machine), NSGA-II (Non dominated Sorting Genetic Algorithm-II), and SW (Sliding Window) methods. For the proposed method, first, three types of investor sentiments, including the individual, institutional, and foreign investor sentiment, are utilized as features to forecast the movement direction of the Shanghai Stock Exchange (SSE) index, and simulation trading is executed based on the direction prediction signal. Then, LightGBM is employed as the classifier for stock index direction prediction, while NSGA-II is essential for hyper-parameter optimization by incorporating multiple objectives that include the hit ratio, accumulated return, and maximum drawdown. In addition, the sliding window method is utilized to prepare the training and testing periods for realizing dynamic prediction and simulation trading of the SSE index. Finally, the proposed method generates an average hit ratio of 60.34%, an annual accumulated return of 28.43%, an average maximum drawdown of 8.35%, and a Sharpe ratio value of 3.24. The performances of the proposed method are substantially superior to the results produced by the benchmarks. Furthermore, the relative importance of the sentiment features of three types investors are comprehensively investigated, and the essential investor sentiment features for forecasting the SSE index direction in different periods are explored. Experimental results demonstrate that the proposed methodology could provide beneficial references for investors to make decisions more rationally, and it could be applied as an effective monitor of the market investor sentiment for market regulation and policymaking in the Chinese security market.

Original languageEnglish
Article number109132
JournalApplied Soft Computing
Volume125
DOIs
Publication statusPublished - 2022 Aug

Keywords

  • Dynamic forecasting
  • Hyperparameter optimization
  • Investor sentiment
  • Sliding window
  • Stock index

ASJC Scopus subject areas

  • Software

Fingerprint

Dive into the research topics of 'Dynamic forecasting of the Shanghai Stock Exchange index movement using multiple types of investor sentiment'. Together they form a unique fingerprint.

Cite this