High-frequency financial data exhibit pronounced nonlinear characteristics and high volatility, significantly complicating accurate market forecasting. Traditional predictive models, reliant upon static correlations among features, fail to capture dynamic causal structures, thereby lacking robust generalization across varying market conditions. To address these challenges, this study proposes a novel forecasting model, the meta-causal transformer (MCT), which synergistically integrates meta-learning with dynamic causal inference. Specifically, an incremental causal discovery algorithm, leveraging a sliding-window approach and adaptive Granger causality testing, is introduced to dynamically reconstruct causal relationships among market variables. An adaptive decay mechanism further enhances this capability, allowing real-time tracking of rapidly evolving causal patterns. The causal-constrained Transformer architecture utilizes attention masks explicitly to eliminate spurious correlations driven by market noise, thus strengthening the model's interpretability and predictive robustness. Additionally, a meta-learning framework is employed to ensure rapid adaptation of model parameters to diverse causal scenarios during market regime shifts. Empirical evaluations on Level 2 high-frequency order book datasets from the Chinese A-share and U.S.stock markets demonstrate that the MCT model achieves substantial improvements, yielding a 7%-15% increase in directional prediction accuracy and a reduction in prediction latency by 1-3 time steps compared to state-of-the-art benchmarks. This research provides an interpretable and dynamically adaptive causal inference methodology, offering robust decision-making support for real-time high-frequency trading systems.
为验证MCT模型的有效性,本文基于沪深300和标普500股票的Level-2高频订单簿数据进行实验,并与传统时序预测方法[22],深度学习模型[23]及前沿模型TFT[7],随机欠采样提升算法(Random Under-sampling Boosting, RUS-BOOST)[24],自适应学习策略引擎(Adaptive Learning Policy Engine, ALPE)[25]进行对比。实验采用方向预测准确率Acc,ROC曲线下面积(Area Under the Curve of the Receiver Operating Characteristic, AUC),预测延迟及因果结构稳定性(jaccard指数TLatency)作为评估指标。其计算公式分别为:
GOODELLJ W, KUMARS, LIMW M, et al. Artificial Intelligence and Machine Learning in Finance: Identifying Foundations, Themes, and Research Clusters from Bibliometric Analysis[J]. J Behav Exp Finance, 2021, 32: 100577. DOI: 10.1016/j.jbef.2021.100577 .
[2]
LIUZ H, SHAMC W, MAL Y, et al. ViT-LOB: Efficient Vision Transformer for StockPrice Trend Prediction Using Limit Order Books[C]//2024 10th International Conference on Applied System Innovation (ICASI). New York: IEEE, 2024: 436-438. DOI: 10.1109/ICASI60819.2024.10547868 .
[3]
LEOND, SANDOVALJ, CRUZA, et al. Deep Heterogeneous AutoML Trend Prediction Model for Algorithmic Trading in the USD/COP Colombian FX Market through Limit Order Book (LOB)[J]. SN Comput Sci, 2024, 5(5): 631. DOI: 10.1007/s42979-024-02930-1 .
[4]
LIZ W, LIUF, YANGW J, et al. A Survey of Convolutional Neural Networks: Analysis, Applications, and Prospects[J]. IEEE Trans Neural Netw Learn Syst, 2022, 33(12): 6999-7019. DOI: 10.1109/TNNLS.2021.3084827 .
[5]
YAOY M. Data Analysis on the Computer Intelligent Stock Prediction Model Based on LSTM RNN and Algorithm Optimization[C]//2022 IEEE International Conference on Electrical Engineering, Big Data and Algorithms (EEBDA). New York: IEEE, 2022: 480-485. DOI: 10.1109/EEBDA53927.2022.9744859 .
[6]
SIRISHAU M, BELAVAGIM C, ATTIGERIG. Profit Prediction Using ARIMA, SARIMA and LSTM Models in Time Series Forecasting: A Comparison[J]. IEEE Access, 2022, 10: 124715-124727.
[7]
TSANTEKIDISA, PASSALISN, TEFASA, et al. Using Deep Learning for Price Prediction by Exploiting Stationary Limit Order Book Features[J]. Appl Soft Comput, 2020, 93: 106401. DOI: 10.1016/j.asoc.2020.106401 .
[8]
ZHANGY J, TUMIBAYG M. Stock Market Prediction Model Based on LSTM Deep Learning: The Case of Top Corporate Company in China[C]//2022 4th International Conference on Artificial Intelligence and Advanced Manufacturing (AIAM). New York: IEEE, 2022: 451-455. DOI: 10.1109/AIAM57466.2022.00092 .
[9]
HUX K. Stock Price Prediction Based on Temporal Fusion Transformer[C]//2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI). New York: IEEE, 2021: 60-66. DOI: 10.1109/mlbdbi54094.2021.00019 .
[10]
CHOEU, SON H, RYU B, et al. Predicting Stock Price Index Using the Wavelet Transform and Temporal Fusion Transformer Model[C]//Proceedings of Annual Conference of KIPS.Seoul, Korea: Korean Information Processing Society, 2024: 401-404. DOI:10.3745/PKIPS.Y2024M10A.401 .
[11]
SHABANIM, TRAND T, MAGRISM, et al. Multi-head Temporal Attention-augmented Bilinear Network for Financial Time Series Prediction[C]//2022 30th European Signal Processing Conference (EUSIPCO). New York: IEEE, 2022: 1487-1491.
RENX L, SUNW Y. Asymmetric Risk Spillovers between the Chinese Stock Market and Index Futures Market Based on Granger Non-causality Test in Quantiles[J]. Oper Res Manag Sci, 2021, 30(8): 190-197. DOI:10.12005/orms.2021.0265 .
CHENJ B, CAIR C. Research on Algorithms of Granger Causality Discovery for Non-stationary Hawkes Process[J]. Comput Sci Appl, 2021, 11(4): 821-831. DOI: 10.12677/CSA.2021.114084 .
CUIS Y, WANGX J. Short-term Photovoltaic Power Prediction by Ensemble Learning Based on Feature Engineering[J]. Sci Technol Eng, 2022, 22(2): 532-539. DOI: 10.3969/j.issn.1671-1815.2022.02.013 .
[18]
SHIJ T, YAOH X, WUX, et al. Relation-aware Meta-learning for E-commerce Market Segment Demand Prediction with Limited Records[C]//Proceedings of the 14th ACM International Conference on Web Search and Data Mining. New York: ACM, 2021: 220-228. DOI: 10.1145/3437963.3441750 .
[19]
CHANGS H, HSUC W, LIH Y, et al. Short-term Stock Price-trend Prediction Using Meta-learning[C]//2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC). New York: IEEE, 2021: 2900-2905. DOI: 10.1109/smc52423.2021.9658607 .
[20]
FALATOURIT, DARBANIANF, BRANDTNERP, et al. Predictive Analytics for Demand Forecasting: A Comparison of SARIMA and LSTM in Retail SCM[J]. Procedia Comput Sci, 2022, 200: 993-1003. DOI: 10.1016/j.procs.2022.01.298 .
MAZ G, XUX H, LIUX E. Three Analytical Frameworks of Causal Inference and Their Applications[J]. Chin J Eng, 2022, 44(7): 1231-1243. DOI: 10.13374/j.issn2095-9389.2021.07.04.002 .
LIL T, CUIW. Causal Inference of Iron Ore Price Fluctuations: Analysis of Influencing Factors and Effects[J]. J Technol Econ, 2024, 43(8): 36-45. DOI: 10.12404/j.issn.1002-980X.J24051802 .
[25]
VASWANIA, SHAZEERN, PARMARN, et al. Attention is All You Need[C]//Proc of the 31st International Conference on Neural Information Processing Systems (NIPS'17). Red Hook, NY: Curran Associates Inc, 2017: 6000-6010. DOI:10.48550/arXiv.1706.03762 .
[26]
KISIELD, GORSED. Axial-LOB: High-frequency Trading with Axial Attention[C]//2022 IEEE Symposium Series on Computational Intelligence (SSCI). New York: IEEE, 2022: 1327-1333. DOI: 10.1109/SSCI51031.2022.10022284 .
[27]
SEKHARP C, PADMAJAM, SARANGIB, et al. Prediction of Cryptocurrency Using LSTM and XGBoost[C]//2022 IEEE International Conference on Blockchain and Distributed Systems Security (ICBDS). New York: IEEE, 2022: 1-5. DOI: 10.1109/ICBDS53701.2022.9935871 .
[28]
SONGY P, CAIC C, MAD X, et al. Modelling and Forecasting High-frequency Data with Jumps Based on a Hybrid Nonparametric Regression and LSTM Model[J]. Expert Syst Appl, 2024, 237: 121527. DOI: 10.1016/j.eswa.2023.121527 .
[29]
GOUDARZIM, BAZZANAF. Identification of High-frequency Trading: a Machine Learning Approach[J]. Res Int Bus Finance, 2023, 66: 102078. DOI: 10.1016/j.ribaf.2023.102078 .
[30]
NTAKARISA, IBIKUNLEG. Minimal Batch Adaptive Learning Policy Engine for Real-Time Mid-Price Forecasting in High-Frequency Trading [EB/OL]. (2024-12-26)[2025-05-21].
[31]
SHAMSM Y, ELSHEWEYA M, EL-KENAWYE M, et al. Water Quality Prediction Using Machine Learning Models Based on Grid Search Method[J]. Multimed Tools Appl, 2024, 83(12): 35307-35334. DOI: 10.1007/s11042-023-16737-4 .
[32]
RIMALY, SHARMAN, ALSADOONA. The Accuracy of Machine Learning Models Relies on Hyperparameter Tuning: Student Result Classification Using Random Forest, Randomized Search, Grid Search, Bayesian, Genetic, and Optuna Algorithms[J]. Multimed Tools Appl, 2024, 83(30): 74349-74364. DOI: 10.1007/s11042-024-18426-2 .
[33]
HIDAYATT, MANONGGAD, HENDRY, et al. Performance Prediction Using Cross Validation (GridSearchCV) for Stunting Prevalence[C]//2024 IEEE International Conference on Artificial Intelligence and Mechatronics Systems (AIMS). New York: IEEE, 2024: 1-6. DOI: 10.1109/AIMS61812.2024.10512657 .
[34]
BOGOUSSLAVSKYV, COLLIN-DUFRESNEP. Liquidity, Volume, and Order Imbalance Volatility[J]. J Finance, 2023, 78(4): 2189-2232. DOI: 10.1111/jofi.13248 .
[35]
SUK K, CAOL H, ZHAOB T, et al. N-IoU: Better IoU-based Bounding Box Regression Loss for Object Detection[J]. Neural Comput Appl, 2024, 36(6): 3049-3063. DOI: 10.1007/s00521-023-09133-4 .