Efficient Test-Time Scaling for LLM-based Time Series Forecasting

2026-08-09

AI Summary

This work addresses the high computational cost of test-time scaling and the tendency toward global shape mismatch in long-horizon forecasting observed in existing large language model (LLM)-based time series prediction methods. To overcome these limitations, the authors propose SCALER, a coarse-to-fine framework that first employs a lightweight Transformer to predict a coarse-grained outline of the future sequence and then explicitly guides the LLM to iteratively refine residual details. By circumventing lengthy prompts and reward-model-based selection mechanisms, SCALER substantially reduces the number of tokens processed at each step. Experimental results demonstrate that SCALER consistently outperforms strong baselines across long-term, short-term, and zero-shot time series forecasting tasks while significantly lowering inference costs.

Abstract

Long-term time series forecasting benefits from preserving global structure such as trends and seasonality. Recent LLM-based forecasters often improve accuracy through test-time scaling (e.g., iterative refinement), but these methods are computationally expensive and increasingly prone to global-shape mismatch as the prediction horizon extends. We propose SCALER, a coarse-to-fine forecasting framework that first employs a lightweight Transformer tailored to long-term shape modeling to predict a coarse representation of future dynamics. This predicted shape then serves as a compact guide for an LLM to perform test-time scaling via iterative coarse-to-fine residual token refinement, while processing substantially fewer tokens at each step. By guiding refinement with an explicit future-shape prediction, SCALER reduces reliance on long description prompts, and its fixed-step refinement avoids costly reward-model-based selection, further lowering computational overhead. Experimental results demonstrate that SCALER outperforms strong forecasting baselines in long-term, short-term and zero-shot forecasting while significantly reducing the inference cost associated with scaled LLM for time series forecasting. Code: https://github.com/xuanmay2701/SCALER.

Problem

• test-time scaling
• time series forecasting
• global structure
• computational efficiency
• long-term forecasting

Innovation

• test-time scaling
• coarse-to-fine forecasting
• LLM-based time series forecasting
• global structure modeling
• efficient inference