Analysis of Maximum Drawdown Warning for the CSI 300 Index Based on Large Models

Ruixi Fang; Siyi Wang

doi:10.54097/k6471b27

Authors

Ruixi Fang Massey Institute, Nanjing University of Finance and Economics Nanjing,210000, China
Siyi Wang School of Computer Science and Engineering, Southeast University, Nanjing, 210000,China

DOI:

https://doi.org/10.54097/k6471b27

Keywords:

Large Language Model, Sentiment Analysis, Maximum Drawdown, Risk Warning, Quantitative Investment.

Abstract

With the development of behavioral finance market sentiment has been proven to be an important factor influencing asset price fluctuations and the financial text data from the Zhihu platform and uses a large language model to construct a daily negative sentiment index, systematically exploring its leading relationship with the maximum drawdown risk of the CSI 300 Index in the future. The research results show that there is a significant positive correlation between the negative emotion index and the maximum future drawdown. Through in-depth analysis of sentiment groups, it is found that as sentiment index rises, the average market drawdown expands significantly, and the probability of a major drawdown also increases significantly. The timing strategy constructed based on this sentiment index demonstrated outstanding performance in backtesting. It not only achieved higher investment returns but also significantly reduced the volatility and drawdown of the investment portfolio, notably improving risk-adjusted returns. This study confirmed the effectiveness of large language models in the sentiment analysis of financial texts and promoted the innovative application of intelligent technologies in the field of quantitative investment.

Downloads

Download data is not yet available.

References

[1] Agrawal, A., Jain, S., & Ch, S. B. (2023). Large language models in finance: A survey. In Proceedings of the fourth ACM international conference on AI in finance (pp. 374-382).

[2] Alostad, H., & El-Halees, A. (2022). Forecasting stock market movements using social media data: A BERT-based deep learning approach. Applied Soft Computing.

[3] Araci, D. (2020). FinBERT: Financial sentiment analysis with pre-trained language models.

[4] Guo, W., Wang, B. C., & Han, D. (2025). The impact of Sino-U.S. trade event news on Chinese stock index returns—Based on ChatGPT sentiment analysis strategy. South China Journal of Economics, 7, 119–135.

[5] Hashamia, R., & Maldonado, F. (2025). Can news predict the direction of oil price volatility? A language model approach with SHAP explanations. arXiv.

[6] Lopez-Lira, A., & Tang, Y. (2023). Can ChatGPT forecast stock market movements? Return predictability and large language models.

[7] Manh, A. T., Tuan, L. M., & Anh, L. T. (2024). Can GPT models predict the stock market? A comprehensive study on various stock tickers.

[8] Naimoli, A. (2023). The information content of sentiment indices in forecasting Value at Risk and Expected Shortfall: A Complete Realized Exponential GARCH-X approach. International Economics, 176, 100459.

[9] Roeder, J., Palmer, M., & Muntermann, J. (2022). Data-driven decision-making in credit risk management: The information value of analyst reports. Decision Support Systems, 158, 113770.

[10] Silcenco, O., Machad, M. R., Ugulino, W. C., & Braun, D. (2024). A retail-corpus for aspect-based sentiment analysis with large language models.

[11] Zhang, S., & Ding, S. Q. (2025). Analysis of systemic risk spillover effects of financial institutions in China: From the perspective of investor sentiment. Friends of Accounting, 10, 35–42.