基于结构先验引导与残差修正的金融波动率预测

Financial Volatility Forecasting Based on Structural Prior Guidance and Residual Correction

摘要

针对金融波动率序列固有的非线性、低信噪比预测难题,提出一种融合结构先验和深度学习的自适应门控混合模型(SCLA-GARCH).该模型利用广义自回归条件异方差(GARCH)捕捉线性结构,并行构建CNN-LSTM-Attention网络学习其残差中的非线性模式,并由自适应门控机制动态融合.实证分析结果表明,该模型预测误差显著低于多种基准模型,均方误差(MSE)最大降幅达59%.

Abstract

Aiming at the prediction challenges caused by the inherent nonlinearity and low signal-to-noise ratio of financial volatility series, we proposed an adaptive gated hybrid model SCLA-GARCH that fused structural priors with deep learning. The model utilized the generalized autoregressive conditional heteroskedasticity (GARCH) to capture linear structures, and constructed a parallel CNN-LSTM-Attention network to learn nonlinear patterns in its residuals, which were dynamically fused by an adaptive gating mechanism. Empirical analysis results show that the prediction error of this model is significantly lower than that of multiple benchmark models, with a maximum reduction of 59% in mean squared error.

关键词

波动率建模 / GARCH模型 / 神经网络 / 金融波动率 / 预测

Key words

volatility modeling / GARCH model / neural network / financial volatility / forecast

参考文献

[1] 孙伟业. 基于GARCH-LSTM融合模型的上证综合指数波动率预测研究[D]. 南京: 南京林业大学, 2025.

[2] (Sun W Y. Research on the Volatility of the Shanghai Composite Index Based on the GARCH-LSTM Fusion Model[D]. Nanjing: Nanjing Forestry University, 2025.)

[3] 朱宝, 翟世婷. 新常态经济下中国股票市场高质量发展研究---基于波动聚集、风险收益和非对称效应多重视角[J]. 中国证券期货, 2025(3):4-17.

[4] (Zhu B, Zhai S T. Research on the High-Quality Development of China’s Stock Market under the New Normal Economy-Based on Multiple Perspectives of Volatility Aggregation, Risk-Return and Asymmetric Effects[J]. China Securities and Futures, 2025(3): 4-17.)

[5] Bollerslev T. Generalized Autoregressive Conditional Heteroskedasticity[J]. Journal of Econometrics, 1986, 31(3): 307-327.

[6] 孙芹芹. 国际股票市场的波动聚集与溢出效应研究---基于日间与隔夜收益率的视角[D]. 北京: 北京交通大学, 2023.

[7] (Sun Q Q. Research on Volatility Clustering and Spillover Effect of International Stock Market Based on Intraday and Overnight Yields[D]. Beijing: Beijing Jiaotong University, 2023.)

[8] 刘丽萍, 杨天兴. 宏观经济对股市波动的影响---基于GARCH-MIDAS-RTSRV模型的证据[J]. 运筹与管理, 2024, 33(2):184-189.

[9] (Liu L P, Yang T X. Impact of Macroeconomy on Stock Market Volatility: Evidence Based on GARCH-MIDAS-RTSRV Model[J]. Operations Research and Management Science, 2024, 33(2): 184-189.)

[10] 许灏. 重尾和轻尾分布下两种风险模型破产概率的研究[D]. 长沙: 湖南师范大学, 2021.

[11] (Xu H. Research on the Ruin Probability on Two Risk Models Respectively with Heavy-Tailed and Light-Tailed Distribution[D]. Changsha: Hunan Normal University, 2021.)

[12] 朱福敏, 宋佳音, 刘仪榕. 基于Levy-GARCH模型的股票市场尾部风险度量研究[J]. 中央财经大学学报, 2024(1):47-60.

[13] (Zhu F M, Song J Y, Liu Y R. Research on Tail Risk Measurement of Stock Market Based on Levy-GARCH Model[J]. Journal of Central University of Finance & Economics, 2024(1): 47-60.)

[14] 宋锐彪, 刘海鸿. 基于小波分析的滑动GA-BP-GRACH模型对股票的预测研究[J]. 数学的实践与认识, 2023, 53(3):117-126.

[15] (Song R B, Liu H H. Research on Stock Prediction Based on Wavelet Analysis of Sliding GA-BP-GHACE Model[J]. Practice and Understanding of Mathematics, 2023, 53(3): 117-126.)

[16] Mersha M, Lam K, Wood J, et al. Explainable Artificial Intelligence: A Survey of Needs, Techniques, Applications, and Future Direction[J]. Neurocomputing, 2024, 599: 128111-1-128111-25.

[17] 孟繁瑞, 王翔, 俄文娟, 等. 基于GCN-CS-LSTM的车辆多模态行驶轨迹预测[J]. 江苏大学学报(自然科学版), 2024, 45(5):506-512.

[18] (Meng F R, Wang X, E W J, et al. Multi-modal Trajectory Prediction of Vehicles Based on GCN-CS-LSTM[J]. Journal of Jiangsu University (Natural Science Edition), 2024, 45(5): 506-512.)

[19] Hochreiter S, Schmidhuber J. Long Short-Term Memory[J]. Neural Computation, 1997, 9(8): 1735-1780.

[20] 白军成, 孙秉珍, 郭誉齐, 等. 融合三支聚类与分解集成学习的股票价格预测模型[J]. 运筹与管理, 2024, 33(8):213-218.

[21] (Bai J C, Sun B Z, Guo Y Q, et al. Stock Price Forecasting Model Integrating Three-Way Clustering and Decomposition Ensemble Learning[J]. Operations Research and Management Science, 2024, 33(8): 213-218.)

[22] 鲁斌, 霍泽健, 俞敏. 基于LSTNet-Skip的综合能源系统多元负荷超短期预测[J]. 中国电机工程学报, 2023, 43(6):2273-2283.

[23] (Lu B, Huo Z J, Yu M. Multi Factor Ultra Short Term Forecasting of Integrated Energy Systems Based on LSTNet Skip[J]. Chinese Journal of Electrical Engineering, 2023, 43(6): 2273-2283.)

[24] 董小刚, 韩元元, 秦喜文. 基于深度确定性策略梯度算法的股票投资组合策略研究[J]. 东北师大学报(自然科学版), 2025, 57(1):29-34.

[25] (Dong X G, Han Y Y, Qin X W. Research on Stock Portfolio Strategy Based on Deep Deterministic Policy Gradient Algorithm[J]. Journal of Northeast Normal University (Natural Science Edition), 2025, 57(1): 29-34.)

[26] Vaswani A, Shazeer N, Uszkoreit J, et al. Attention Is All You Need[EB/OL]. (2017-06-12)[2025-12-10]. https://arxiv.org/abs/1706.03762.

[27] Pardo F D M, López R C. Mitigating Overfitting on Financial Datasets with Generative Adversarial Networks[J]. The Journal of Financial Data Science, 2019, 201(2): 367-383.

[28] Flavell J H. Metacognition and Cognitive Monitoring: A New Area of Cognitive-Developmental Inquiry[J]. American Psychologist, 1979, 34(10): 906-911.

基金资助

吉林省科技发展计划项目(20250201081GX)