Test-time loss improves financial forecast accuracy by tracking price trends
DPAL: A Test-Time Adaptation Loss for Financial Time Series Forecasting
Machine Learning
Summary
Forecasting financial prices is hard because it’s important not only to predict the exact next value but also to get the direction and overall price path right. The authors created a new way to update forecasting models after they are deployed that looks at prediction errors, direction of price changes, and entire price paths. They call this new approach DPAL, which helps the models adjust better to new data. Testing on several financial datasets showed DPAL often makes the forecasts closer to actual prices.
What this means in practice
- •For quantitative trading teams: Update deployed forecasting models to better track price trends and reduce pricing errors in financial time series.
- •For energy market analysts: Improve forecasting accuracy for price trajectories of volatile commodity markets using enhanced test-time adaptation loss.
Authors
Wuzhou Lian, Haochun Wang, Yuefan Deng
Abstract
Test-time adaptation (TTA) enables time series forecasters to update after deployment as new observations become available. However, adaptation objectives that focus primarily on prediction errors at individual time steps do not fully address the needs of financial time series forecasting. In addition to pointwise return fitting, return direction and the price trajectories reconstructed from sequences of returns are also important. To better account for these properties, we introduce Direction- and Price-Aware Adaptation Loss (DPAL), which combines pointwise return fitting with directional and price-calibration terms. We apply DPAL within existing TSF-TTA frameworks without modifying the source forecaster or the adapter architecture. Across a diverse collection of financial datasets, DPAL reduces reconstructed-price mean squared error in most evaluated settings.