Analysis of quantitative trading investment influence factors based on KPCA-LightGBM algorithm

Chuwei Feng; Jiali Li; Mengliang Liu

doi:10.54097/hbem.v12i.8338

Authors

Chuwei Feng
Jiali Li
Mengliang Liu

DOI:

https://doi.org/10.54097/hbem.v12i.8338

Keywords:

Impact Factors, Quantitative Investment, KPCA, LightGBM, Stock Price Forecast.

Abstract

This study proposes a stock price forecasting model based on Kernel Principal Component Analysis (KPCA) and Light Gradient Boosting Machine (LightGBM) to examine the connection between the impact factors and the CSI 300 index. As a case, this study focuses on the CSI 300 Index. Data gathered from January 2005 to July 2022, including 16 clear indicators such as CPI, IPI, and Dow Jones Industrial Index, were simplified into five dimensions by the KPCA. Subsequently, the LightGBM algorithm was applied through 5-fold cross-validation to adjust the parameters and achieve optimal model performance. The results demonstrate that the combined KPCA-LightGBM model has strong generalizability and stability, with an RMSE and R2 of 0.001 and 0.837, respectively. Furthermore, reducing factors to five dimensions by KPCA can also provide improved decision-making metrics for investors.

Downloads

Download data is not yet available.

References

Li Bin, Lin Yan, Tang Wenxuan. ML-TEA: A set of quantitative investment algorithms based on machine learning and technical analysis [J]. Systems engineering theory and practice, 2017, 37 (05): 1089 - 1100.

Pan Li, Xu Jianguo. Risk and characteristic factors of A-share market [J]. Financial Research, 2011, No.376 (10): 140 - 154.

Li Bin, Shao Xinyue, Li Yueyang. Quantitative investment research on fundamentals driven by machine learning [J]. China Industrial Economy, 2019, No.377 (08): 61 - 79.

Kozak, S., S. Nagel, and S. Santosh. Shrinking the Cross Section [EB / OL]. Journal of Financial Economics,

Kelly, B.T., S. Pruitt, and Y. Su. Characteristics Are Covariances: A Unified Model of Risk and Return [EB / OL]. Journal of Financial Economics,

Ma Xiaojun, Sha Jinglan, Niu Xueqi. Design and Application of P2P Project Credit Rating Model Based on LightGBM Algorithm. Quantitative Economics and Techno-Economic Research, 2018, 35 (05): 144 - 160.

Gao Ni, He Yiyue, Ma Xincheng. Index timing strategy based on low-frequency component EEMD-SVR prediction modeling [J]. Statistics and Decision, 2022, 38 (02): 140 - 145.

Xu Xuechen, Tian Kan. A new method of stock index prediction based on emotional analysis of financial text [J]. Quantitative Economic and Technological Economic Research, 2021, 38 (12): 124 - 145.

Huang Qiuli, Huang Zhuxing, Yang Yan. Empirical study on stock prediction based on recursive feature elimination and stacking integrated learning [J]. Journal of Nanning Normal University (Natural Science Edition), 2021, 38 (03): 37 - 43.

Ye Kang, Deng Xiaowei, Lu Xuebin. Research on stock market index prediction based on CEEMDAN-GRU model [J]. Business Research, 2021, 28 (03): 64 - 72+132

Xie Chengyan, Fang Hongbin, Guo Mengjie, Yang Mengzhuo. Research on the prediction of the CSI 300 index based on the inversion error combination optimization algorithm [J]. Journal of Fujian University of Business, 2020 (06): 14 - 22.