The Label Horizon Paradox: Rethinking Supervision Targets in Financial Forecasting

TL;DR

This paper reveals that in financial forecasting, the best supervision signals often differ from the actual prediction targets, and introduces a method to automatically find optimal proxy labels, improving forecasting accuracy.

Contribution

It uncovers the Label Horizon Paradox and proposes a bi-level optimization framework to identify optimal proxy labels during training.

Findings

Consistent performance improvements over baseline models.

Theoretical grounding in signal-noise trade-off.

Effective automatic proxy label identification.

Abstract

While deep learning has revolutionized financial forecasting through sophisticated architectures, the design of the supervision signal itself is rarely scrutinized. We challenge the canonical assumption that training labels must strictly mirror inference targets, uncovering the Label Horizon Paradox: the optimal supervision signal often deviates from the prediction goal, shifting across intermediate horizons governed by market dynamics. We theoretically ground this phenomenon in a dynamic signal-noise trade-off, demonstrating that generalization hinges on the competition between marginal signal realization and noise accumulation. To operationalize this insight, we propose a bi-level optimization framework that autonomously identifies the optimal proxy label within a single training run. Extensive experiments on large-scale financial datasets demonstrate consistent improvements over conventional baselines, thereby opening new avenues for label-centric research in financial forecasting.