Benchmarking Machine Learning Methods for Stock Prediction
Keywords: Stock forecast, Benchmark, Evaluation, AI4Finance
Abstract: Machine learning has been widely applied to stock movement prediction. However, research in this field is often hindered by the lack of high-quality benchmark datasets and comprehensive evaluation methods. To address these challenges, we introduce \textit{BenchStock}, a benchmark that includes standardized datasets from the two largest stock markets (the U.S. and China) along with an evaluation method designed to facilitate a thorough examination of machine learning stock prediction methods. This benchmark covers a range of models, from traditional machine learning techniques to the latest deep learning approaches. Using BenchStock, we conducted large-scale experiments predicting individual stock returns over three decades in both markets to assess both short-term and long-term performance. To evaluate the impact of these predictions in actual market conditions, we constructed a portfolio based on the predictions and used a backtesting program to simulate its performance. The experiments revealed several key findings that have not been reported: 1) Most methods outperformed the S\&P 500 in the U.S. market but experienced significant losses in the Chinese market. 2) Prediction accuracy of a method was not correlated with its portfolio return. 3) Advanced deep learning methods did not outperform traditional approaches. 4) The performance of the models was highly dependent on the testing period. These findings highlight the complexity of stock prediction and call for more in-depth machine learning research in this field.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9856
Loading