ADDRESSING FINANCIAL MARKET UNCERTAINTY: EXTREMA PREDICTION WITH SEQUENCE MODELS

ICLR 2025 Workshop ICBINB Submission3 Authors

Track: tiny / short paper (up to 2 pages)

Keywords: Extrema Prediction, Sequence Models, Data Shift

TL;DR: Our paper highlights the challenges of deep learning models in predicting local extrema types, revealing their limited generalization under market data shifts.

Abstract: Deep learning models, particularly sequence-based architectures, are widely used for trend prediction and time series analysis in financial markets. This paper investigates a fundamental aspect of chart pattern formations, the prediction of local extrema types. By classifying extrema into four distinct categories using historical extrema data, we aim to provide a novel perspective on chart pattern identification. However, our findings reveal that these models struggle to generalize under data distribution shifts, achieving significantly lower prediction accuracy on out-of-training data. These results underscore the limitations of deep learning-based strategies in dynamic financial environments and highlight the need for robust methods to address market variability.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Submission Number: 3