Reasoning-Aware Training for Time Series Forecasting

2026-05-08

AI Summary

This work addresses the lack of interpretable qualitative reasoning in existing time series foundation models, while avoiding the numerical discontinuity and high computational cost incurred by directly integrating large language models (LLMs). To this end, the authors propose STRIDE, a novel framework that unifies semantic reasoning and numerical forecasting within a continuous embedding space for the first time. STRIDE distills lightweight reasoning traces from an LLM and dynamically injects them into the time series encoder as cross-modal priors, jointly optimizing cross-entropy and quantile losses. This approach achieves a principled balance between interpretability and predictive accuracy, offering plug-and-play enhancement for diverse time series models. STRIDE attains state-of-the-art performance on GIFT-Eval with 0.674 MASE and 0.454 CRPS, and demonstrates strong in-domain and out-of-domain reasoning and forecasting capabilities on TFRBench.

Abstract

Time Series Foundation Models (TSFMs) excel at numerical forecasting but operate as black boxes lacking qualitative reasoning. Conversely, applying LLMs directly to temporal data introduces a modality gap: text tokenizers fragment continuous numerical values, degrading mathematical relationships and exploding sequence lengths, leading to computational overhead. To resolve this, we introduce STRIDE (Strategic Time-series Reasoning Injected via Distilled Embeddings), a novel framework natively integrating LLM reasoning into the continuous embedding space of TSFMs. Instead of discrete tokens, STRIDE distills reasoning traces into a lightweight LLM, dynamically projecting its mean-pooled hidden states as a cross-modal prior into the target numerical encoder. The architecture is jointly optimized using cross-entropy and quantile losses. Evaluations demonstrate STRIDE establishes state-of-the-art numerical forecasting on GIFT-Eval (0.674 MASE, 0.454 CRPS) compared to TSFMs and exhibits superior in-domain and out-of-domain numerical as well as reasoning performance on TFRBench. Specifically, STRIDE acts as a plug-and-play enhancement, consistently improving diverse TSFMs (e.g., Chronos-2, Timer-S1) across various LLM configurations. Thus, injecting semantic reasoning as a continuous prior equips TSFMs with human-interpretable reasoning while fundamentally improving predictive accuracy.

Problem

• Time Series Forecasting
• Foundation Models
• Modality Gap
• Qualitative Reasoning
• Numerical Embeddings

Innovation

• Time Series Foundation Models
• LLM Reasoning Integration
• Continuous Embedding Prior
• Cross-Modal Distillation
• Interpretable Forecasting