# Conversational Time Series Foundation Models: Towards Explainable and Effective Forecasting

arXiv: https://arxiv.org/abs/2512.16022 · HTML: https://mlsanity.com/paper/2512.16022/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 17 Dec 2025.

## Built on by (most cited first, 8 of 8)

- [Adaptive Collaboration with Humans: Metacognitive Policy Optimization for Multi-Agent LLMs with Continual Learning](https://mlsanity.com/paper/2603.07972/) (2026)
- [KairosAgent: Agentic Time Series Forecasting with Fused Semantic Reasoning](https://mlsanity.com/paper/2605.30002/) (2026)
- [TimeRouter: Efficient and Adaptive Routing of Time-Series Foundation Models](https://mlsanity.com/paper/2606.11625/) (2026)
- [AME-TS: Anchored Mixture-of-Experts for Time Series Forecasting](https://mlsanity.com/paper/2605.25166/) (2026)
- [Interpretability in Deep Time Series Models Demands Semantic Alignment](https://mlsanity.com/paper/2602.02239/) (2026)
- [Speaking Numbers to LLMs: Multi-Wavelet Number Embeddings for Time Series Forecasting](https://mlsanity.com/paper/2606.26487/) (2026)
- [Coarse-to-Fine Multi-Resolution Diffusion Models for Trajectory Generation in Urban Systems](https://mlsanity.com/paper/2608.14570/) (2026)
- [Latent Inference-Time Guidance of Time Series Foundation Models](https://mlsanity.com/paper/2609.38058/) (2026)

## Abstract (arXiv)

The proliferation of time series foundation models has created a landscape where no single method achieves consistent superiority, framing the central challenge not as finding the best model, but as orchestrating an optimal ensemble with interpretability. While Large Language Models (LLMs) offer powerful reasoning capabilities, their direct application to time series forecasting has proven ineffective. We address this gap by repositioning the LLM as an intelligent judge that evaluates, explains, and strategically coordinates an ensemble of foundation models. To overcome the LLM's inherent lack of domain-specific knowledge on time series, we introduce an R1-style finetuning process, guided by SHAP-based faithfulness scores, which teaches the model to interpret ensemble weights as meaningful causal statements about temporal dynamics. The trained agent then engages in iterative, multi-turn conversations to perform forward-looking assessments, provide causally-grounded explanations for its weighting decisions, and adaptively refine the optimization strategy. Validated on the GIFT-Eval benchmark on 23 datasets across 97 settings, our approach significantly outperforms leading time series foundation models on both CRPS and MASE metrics, establishing new state-of-the-art results.