Representation learning for financial time series forecasting
Keywords: representation learning, contrastive predictive coding, cpc
TL;DR: Automated feature generation on the basis of contrastive predictive coding (CPC) is used to generate embeddings as input to improve the performance of downstream financial time series forecasting models
Abstract: The accurate forecasting of financial time series remains a significant challenge due to the stochastic nature of the underlying data. To improve prediction accuracy, feature engineering has become a vital aspect of forecasting financial assets. However, engineering features manually often requires domain expertise. We propose to utilise an automated feature generation architecture, Contrastive Predictive Coding (CPC), to generate embeddings as input to improve the performance of downstream financial time series forecasting models. To benchmark the effectiveness of our approach, we evaluate forecasting models on predicting the next day's log return on various foreign exchange markets with and without embeddings. Finally, we assess our CPC architecture by employing the same trained encoder on different currency pairs and calculating the Sharpe ratio of our strategies.
Primary Area: learning on time series and dynamical systems
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.
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: 2975
Loading