Leakage-Controlled Evaluation of Cross-Temporal Multimodal Fusion for Stock Movement Prediction
Main Article Content
Keywords
stock movement prediction, multimodal financial data, cross-temporal attention, gated fusion, walk-forward evaluation
Abstract
Financial news can complement price histories, but irregular timing makes multimodal stock-direction models vulnerable to leakage and overstatement. We evaluate Cross-Temporal Contrastive Alignment Fusion (CTCAF), which combines a five-day causal news memory with a recurrent price representation and a current-day gated cross-attention anchor. On public NIFTY data, three expanding walk-forward folds and five fixed seeds give descriptive Macro-F1 values of 0.467 for CTCAF, 0.495 for gated cross-attention fusion, and 0.510 for XGBoost; their seed-level standard deviations are 0.050, 0.038, and 0.021. The CTCAF-minus-gated fold-mean differences are minus 0.004, zero, and minus 0.080, so no advantage is observed at the independent-fold level. On StockNet-12, CTCAF and the gated baseline differ by only 0.001 Macro-F1. Ablations likewise show no consistent benefit from the contrastive term or longer text memory. Under this protocol, the added temporal-fusion components do not improve direction classification, underscoring the need for chronological controls and fold-level reporting in multimodal financial research.
References
- [1]Peng, Y., and Jiang, H., “Leverage financial news to predict stock price movements using word embeddings and deep neural networks,” arXiv preprint arXiv:1506.07220, 2015.
- [2]Chen, Q., “Stock movement prediction with financial news using contextualized embedding from BERT,” arXiv preprint arXiv:2107.08721, 2021.
- [3]Shah, D., Isah, H., and Zulkernine, F., “Predicting the effects of news sentiments on the stock market,” arXiv preprint arXiv:1812.04199, 2018.
- [4]Zong, C., Wan, J., Cascone, L., and Zhou, H., “Stock movement prediction with multimodal stable fusion via gated cross-attention mechanism,” Complex & Intelligent Systems, 11, 396, 2025, https://doi.org/10.1007/s40747-025-02023-3.
- [5]Hwang, E., and Kim, Y.-H., “Interdependency between the stock market and financial news,” arXiv preprint arXiv:1909.00344, 2019.
- [6]Villamil, L., Bausback, R., Salman, S., Liu, T. L., Horn, C., and Liu, X., “Improved stock price movement classification using news articles based on embeddings and label smoothing,” arXiv preprint arXiv:2301.10458, 2023.
- [7]Chen, Q., and Robert, C.-Y., “Graph-based learning for stock movement prediction with textual and relational data,” The Journal of Financial Data Science, 4, (4), 152–166, 2022, https://doi.org/10.3905/jfds.2022.1.104.
- [8]Chen, Q., “Stock price prediction with LLM-guided market movement signals and transformer model,” FinTech and Sustainable Innovation, 1, A12, 2025, https://doi.org/10.47852/bonviewFSI52025703.
- [9]Elahi, A., and Taghvaei, F., “Combining financial data and news articles for stock price movement prediction using large language models,” arXiv preprint arXiv:2411.01368, 2024.
- [10]Saqur, R., Kato, K., Vinden, N., and Rudzicz, F., “NIFTY financial news headlines dataset,” arXiv preprint arXiv:2405.09747, 2024, https://arxiv.org/abs/2405.09747.
- [11]Xu, Y., and Cohen, S. B., Stock movement prediction from tweets and historical prices, Proceedings of the 56th annual meeting of the association for computational linguistics, 1970–1979, Association for Computational Linguistics, 2018, https://doi.org/10.18653/v1/P18-1183.
- [12]Sahay, S., Okur, E., Kumar, S. H., and Nachman, L., “Low rank fusion based transformers for multimodal sequences,” arXiv preprint arXiv:2007.02038, 2020.
- [13]Jiang, M., and Ji, S., “Cross-modality gated attention fusion for multimodal sentiment analysis,” arXiv preprint arXiv:2208.11893, 2022.
- [14]Khorasani, A., and Demazure, T., “Cross-temporal attention fusion (CTAF) for multimodal physiological signals in self-supervised learning,” arXiv preprint arXiv:2602.02784, 2026.
- [15]Lin, S., Chen, Y., Qi, Y., Ma, C., Cao, B., Zhang, Y., Liu, X., and Guo, J., “CSPO: Cross-market synergistic stock price movement forecasting with pseudo-volatility optimization,” arXiv preprint arXiv:2503.22740, 2025.
Article Sidebar
This work is licensed under a Creative Commons Attribution 4.0 International License.