KReF: Training-Free Retrieval for Long-Term Time-Series Forecasting and Predictive Uncertainty

arXiv cs.LG Papers

Summary

KReF introduces a training-free retrieval framework for long-term time-series forecasting that constructs empirical predictive distributions from similar historical lookback-future pairs, achieving strong CRPS performance across multiple benchmarks.

arXiv:2608.06748v1 Announce Type: new Abstract: Probabilistic long-term time-series forecasting commonly relies on trained models. Training-free conformal methods typically construct intervals around a pre-existing point forecaster and do not natively represent a complete predictive distribution; sequential variants additionally suffer from increasingly delayed feedback at long horizons. We propose KReF, a training-free retrieval framework that treats retrieved historical futures as a querylocal empirical predictive distribution. After robust preprocessing, KReF embeds each lookback using handcrafted statistics or frozen random Fourier features and retrieves similar historical lookback-future pairs. Their similarity weights directly define predictive masses, quantiles, CRPS, and a weighted-mean point forecast. KReF further uses the observed query lookback to construct a probability-integral-transform map and applies validation-selected expansion and shrinkage rates to adapt interval boundaries. Across six LTSF benchmarks and four horizons, KReF obtains the lowest CRPS in all 12 dataset-embedding settings and the lowest IS90 in 9 settings. Without gradient-based fitting, its point forecasts also match or surpass trained baselines on two of six datasets. An archive-oracle analysis further reveals substantial headroom under finer horizon- and channel-wise routing. These results establish retrieval as a useful and underexplored inductive bias for LTSF.
Original Article
View Cached Full Text

Cached at: 08/10/26, 08:03 AM

# KReF: Training-Free Retrieval for Long-Term Time-Series Forecasting and Predictive Uncertainty
Source: [https://arxiv.org/abs/2608.06748](https://arxiv.org/abs/2608.06748)
[View PDF](https://arxiv.org/pdf/2608.06748)

> Abstract:Probabilistic long\-term time\-series forecasting commonly relies on trained models\. Training\-free conformal methods typically construct intervals around a pre\-existing point forecaster and do not natively represent a complete predictive distribution; sequential variants additionally suffer from increasingly delayed feedback at long horizons\. We propose KReF, a training\-free retrieval framework that treats retrieved historical futures as a querylocal empirical predictive distribution\. After robust preprocessing, KReF embeds each lookback using handcrafted statistics or frozen random Fourier features and retrieves similar historical lookback\-future pairs\. Their similarity weights directly define predictive masses, quantiles, CRPS, and a weighted\-mean point forecast\. KReF further uses the observed query lookback to construct a probability\-integral\-transform map and applies validation\-selected expansion and shrinkage rates to adapt interval boundaries\. Across six LTSF benchmarks and four horizons, KReF obtains the lowest CRPS in all 12 dataset\-embedding settings and the lowest IS90 in 9 settings\. Without gradient\-based fitting, its point forecasts also match or surpass trained baselines on two of six datasets\. An archive\-oracle analysis further reveals substantial headroom under finer horizon\- and channel\-wise routing\. These results establish retrieval as a useful and underexplored inductive bias for LTSF\.

## Submission history

From: Yang Zhang \[[view email](https://arxiv.org/show-email/c16b3fe2/2608.06748)\] **\[v1\]**Fri, 7 Aug 2026 03:16:25 UTC \(1,554 KB\)

Similar Articles

Stationarity-Aware Retrieval-Augmented Time Series Forecasting

arXiv cs.LG

SARAF is a Stationarity-Aware Retrieval-Augmented Forecasting framework that adaptively balances relevance and diversity in retrieval for time series forecasting, modulating diversification strength based on dataset-level stationarity to handle non-stationary regime shifts. Accepted to KDD 2026, it demonstrates competitive performance over strong baselines on eight real-world datasets.

Retrieval-Corrected Conformal Prediction for Time Series

arXiv cs.LG

Introduces Retrieval-Corrected Conformal Prediction (RCCP), a retrieval-augmented calibration method for time series prediction intervals that selects similar past residuals and applies a scalar conformal correction to achieve target coverage with low overhead.