Time series forecasts combined adaptively for more reliable results
Latent Inference-Time Guidance of Time Series Foundation Models
Machine Learning
Summary
Predicting future events from past data can be tricky, and different methods often work better at different times. The authors show a way to combine multiple prediction models by letting the combination change over time based on hidden patterns. This makes use of advanced time series models without needing to retrain them, improving forecast accuracy in many scenarios. Their method also comes with guarantees that the combined predictions can be understood and reliably reconstructed.
What this means in practice
- •For financial analysts: Improve stock or asset price forecasts by combining multiple model predictions depending on market conditions.
- •For energy demand planners: Create more reliable energy consumption forecasts by switching weights on various prediction models as usage patterns shift throughout the day.
Authors
Chloé Hashimoto-Cullen, Amaury Durand, Laurent Bozzi, Benjamin Guedj, Yannig Goude, Sylvain Le Corff
Abstract
Time Series Foundation Models (TSFMs) currently provide state-of-the-art results in forecasting tasks. They are available out-of-the-box and rely on in-context learning to make their predictions, which makes the quality of their performance highly sensitive to the user-selected lookback, covariates, horizon and training data distributions. In practise, the quality of the forecasts are variable but complementary, which highlights the need for a principled ensembling approach, rather than selecting the best context. This paper introduces Latent Inference-Time Guidance for TSFMs, which adaptively combines a pool of TSFM forecasts through a time-dependent latent space with independent components. The framework comes equipped with identifiability and reconstruction guarantees, whilst maintaining the off-the-shelf aspect of foundation models. We provide experiments on datasets at various frequencies and from multiple domains: these show that the approach is competitive with traditional ensembling approaches.