TimeRFT: Stimulating Generalizable Time Series Forecasting for TSFMs via Reinforcement Finetuning

2026-04-18

arXiv.org

AI Summary

This work addresses the limitations of time series foundation models in downstream forecasting tasks, where distribution shifts and data scarcity often impair the generalization of supervised fine-tuning. To overcome these challenges, the authors propose TimeRFT, a reinforcement learning–based fine-tuning paradigm that introduces a quality-aware, fine-grained temporal reward mechanism and a difficulty-aware data selection strategy. These innovations effectively reduce reliance on stable data distributions and large-scale datasets. Extensive experiments demonstrate that TimeRFT consistently outperforms existing methods across multiple real-world forecasting benchmarks, achieving superior accuracy and generalization under varying data volumes and distribution shift scenarios.

Abstract

Time Series Foundation Models (TSFMs) advance generalization and data efficiency in time series forecasting by unified large-scale pretraining. But TSFMs remain lacking when adapting to specific downstream forecasting tasks for two reasons. First, the non-stationary and uncertain nature of time series data lead to inevitable temporal distribution shifts between historical training and future testing data, while current Supervised FineTuning (SFT)-based methods are prone to overfitting and may degrade generalization. Second, training data availability varies across forecasting tasks, requiring TSFMs to generalize well under diverse data regimes. To address these challenges, we introduce the Time series Reinforcement Finetuning (TimeRFT) paradigm for TSFM downstream adaptation, which consists of two task-specific training recipes: i) A forecasting quality-based temporal reward mechanism that conducts a multi-faceted evaluation of the contribution of each prediction step to overall forecasting accuracy. ii) A forecasting difficulty-based data selection strategy to identify time series samples with generalizable predictive patterns and informative training signals. Extensive experiments demonstrate TimeRFT can consistently outperform SFT-based adaptation methods across various real-world forecasting tasks and training data regimes, enhancing prediction accuracy and generalization against unforeseen distribution shifts.

Problem

• time series forecasting
• distribution shift
• data scarcity
• generalization
• foundation models

Innovation

• Reinforcement Learning
• Time Series Forecasting
• Foundation Models
• Distribution Shift
• Fine-tuning