Keywords
Transfer Learning
Abstract
Stock market prediction is inherently challenging because financial markets differ significantly in their microstructure, regulatory environment, investor composition, and macroeconomic drivers. A predictive model trained on U.S. equities may perform poorly when applied to Chinese A-shares or South Korean equities, and vice versa. Most existing approaches treat each market as an isolated prediction problem, failing to leverage the transferable knowledge that exists across markets—such as the universal momentum anomaly, the value effect, and the low-volatility anomaly. Furthermore, the design of neural network architectures for stock prediction has largely relied on manual engineering, with architectures optimized for one market often suboptimal for others. In this paper, we propose TransNAS-Stock (Transfer Learning and Neural Architecture Search for Cross-Market Stock Prediction), a novel framework that jointly addresses the challenges of cross-market knowledge transfer and architecture optimization through two key innovations. First, we develop a Market-Invariant Representation Learning (MIRL) module that learns stock embeddings that are invariant to market-specific characteristics but sensitive to universal predictive factors, enabling positive knowledge transfer across markets. Second, we introduce a Multi-Task Neural Architecture Search (MT-NAS) component that searches for optimal stock prediction architectures across multiple markets simultaneously, discovering market-general and market-specific architecture modules. By optimizing architectures jointly across markets, MT-NAS identifies architecture patterns that generalize across markets while exploiting market-specific opportunities. Extensive experiments on six cross-market scenarios—including transfers from S&P 500 to CSI 300, from FTSE 100 to KOSPI 200, and from Nikkei 225 to ASX 200—demonstrate that TransNAS-Stock significantly outperforms both market-specific models and naive transfer learning baselines, with an average improvement of 7.2% in directional accuracy over S3G (Lu, Hu, and Zhang, 2026), the state-of-the-art stock state space graph model, and with particularly strong gains in low-data transfer scenarios where the target market has limited training data.
References
1. Araci, D. (2019). FinBERT: Financial Sentiment Analysis with Pre-trained Language Models. *Proceedings of the 1st Workshop on Financial Technology and Natural Language Processing*, pages 38-44.
2. Bai, S., Kolter, J.Z., and Koltun, V. (2018). An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling. *arXiv preprint arXiv:1803.01271*.
3. Baker, M., and Wurgler, J. (2011). Investor Sentiment in the Stock Market. *Journal of Economic Perspectives*, 21(2): 129-151.
4. Ben-David, S., Blitzer, J., Crammer, K., Kulesza, A., Pereira, F., and Vaughan, J.W. (2010). A Theory of Learning from Different Domains. *Machine Learning*, 79(1-2): 151-175.
5. Caruana, R. (1997). Multitask Learning. *Machine Learning*, 28(1): 41-75.
6. Chen, Y., Lu, Y., and Wang, B. (2020). Stock Movement Prediction with Sector Information using Graph Convolutional Networks. *IEEE Transactions on Neural Networks and Learning Systems*, 31(12): 5419-5429.
7. Chen, Z., Li, Y., and Ma, X. (2022). TS-NAS: Neural Architecture Search for Time Series Classification. *Proceedings of the 2022 ACM SIGKDD International Conference on Knowledge Discovery and Data Mining*, pages 2346-2458.
8. Cho, K., Van Merriënboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., and Bengio, Y. (2014). Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation. *Conference on Empirical Methods in Natural Language Processing (EMNLP)*, pages 1724-1734.
9. Deng, Y., Wang, B., and Zhang, Y. (2021). Deep Learning for Cryptocurrency Market Regime Detection. *Proceedings of the 2021 ACM International Conference on Information and Knowledge Management*, pages 3452-3459.
10. Duong, L., Cohn, T., Bird, S., and Cook, P. (2015). Low Resource Speech Recognition Using Multi-Task Learning and Transfer Learning. *Proceedings of Interspeech*, pages 3219-3223.
11. Fama, E.F., and French, K.R. (1993). Common Risk Factors in the Returns on Stocks and Bonds. *Journal of Financial Economics*, 33(1): 3-56.
12. Fama, E.F. (1965). The Behavior of Stock-Market Prices. *Journal of Business*, 38(1): 34-105.
13. Finn, C., Abbeel, P., and Levine, S. (2017). Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks. *International Conference on Machine Learning (ICML)*, pages 1126-1135.
14. Fischer, T., and Krauss, C. (2018). Deep Learning with Long Short-Term Memory Networks for Financial Market Predictions. *European Journal of Operational Research*, 270(2): 654-669.
15. Ganin, Y., and Lempitsky, V. (2015). Unsupervised Domain Adaptation by Backpropagation. *International Conference on Machine Learning (ICML)*, pages 1180-1189.
16. Lu, Y., Hu, K., and Zhang, L. (2026). S3G: Stock State Space Graph for Enhanced Stock Trend Prediction. *ICASSP 2026-2026 IEEE International Conference on Acoustics, Speech and Signal Processing*, pages 4081-4085. IEEE.
17. Yuping Deng , Hanbin Ouyang , Pusheng Xie , Yanfang Wang , Yang Yang , Wenchang Tan , Dongliang Zhao , Shizhen Zhong & Wenhua Huang (2020): Biomechanical assessment of screw safety between far cortical locking and locked plating constructs, Computer Methods in Biomechanics and Biomedical Engineering, DOI: 10.1080/10255842.2020.1844882
18. Bao, W., Xu, K., & Leng, Q. (2024). Research on the Financial Credit Risk Management Model of Real Estate Supply Chain Based on GA-SVM Algorithm: A Comprehensive Evaluation of AI Model and Traditional Model. Procedia Computer Science, 243, 900-909.