Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Regime-Switching Forecasts with Distribution-Free Uncertainty: Deep Switching State-Space Models Meet Conformal Prediction

Echo Diyun LU, Charles Findling, Marianne Clausel, Alessandro Leite, Wei Gong, Pierric Kersaudy

TL;DR

The paper tackles calibrated uncertainty under regime-switching nonstationarity by marrying a regime-aware Deep Switching State-Space Model (DS3M) with Adaptive Conformal Inference (ACI) and Aggregated ACI (AgACI). A unified, residual-based conformal wrapper is applied atop strong baselines (e.g., S4, MC-Dropout GRU, Sparse GP) to produce online, distribution-free predictive intervals with finite-sample guarantees. Empirical results on synthetic and real data show near-nominal coverage and competitive point accuracy, with improved interval efficiency, especially in regime-rich settings. The approach offers a practical calibration layer that remains robust to nonstationarity without retraining, enhancing reliability for time-series forecasting in changing environments.

Abstract

Regime transitions routinely break stationarity in time series, making calibrated uncertainty as important as point accuracy. We study distribution-free uncertainty for regime-switching forecasting by coupling Deep Switching State Space Models with Adaptive Conformal Inference (ACI) and its aggregated variant (AgACI). We also introduce a unified conformal wrapper that sits atop strong sequence baselines including S4, MC-Dropout GRU, sparse Gaussian processes, and a change-point local model to produce online predictive bands with finite-sample marginal guarantees under nonstationarity and model misspecification. Across synthetic and real datasets, conformalized forecasters achieve near-nominal coverage with competitive accuracy and generally improved band efficiency.

Adaptive Regime-Switching Forecasts with Distribution-Free Uncertainty: Deep Switching State-Space Models Meet Conformal Prediction

TL;DR

The paper tackles calibrated uncertainty under regime-switching nonstationarity by marrying a regime-aware Deep Switching State-Space Model (DS3M) with Adaptive Conformal Inference (ACI) and Aggregated ACI (AgACI). A unified, residual-based conformal wrapper is applied atop strong baselines (e.g., S4, MC-Dropout GRU, Sparse GP) to produce online, distribution-free predictive intervals with finite-sample guarantees. Empirical results on synthetic and real data show near-nominal coverage and competitive point accuracy, with improved interval efficiency, especially in regime-rich settings. The approach offers a practical calibration layer that remains robust to nonstationarity without retraining, enhancing reliability for time-series forecasting in changing environments.

Abstract

Regime transitions routinely break stationarity in time series, making calibrated uncertainty as important as point accuracy. We study distribution-free uncertainty for regime-switching forecasting by coupling Deep Switching State Space Models with Adaptive Conformal Inference (ACI) and its aggregated variant (AgACI). We also introduce a unified conformal wrapper that sits atop strong sequence baselines including S4, MC-Dropout GRU, sparse Gaussian processes, and a change-point local model to produce online predictive bands with finite-sample marginal guarantees under nonstationarity and model misspecification. Across synthetic and real datasets, conformalized forecasters achieve near-nominal coverage with competitive accuracy and generally improved band efficiency.

Paper Structure

This paper contains 14 sections, 4 equations, 1 figure, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. DS3M forecaster.
  5. Adaptive Conformal Inference (ACI).
  6. Aggregated ACI (AgACI).
  7. Unified CP wrapper for baselines.
  8. Experimental Setup
  9. Results
  10. Discussion and Future Direction
  11. Conformal Prediction Primer
  12. What We Mean by Regimes and Switching
  13. Datasets and Summary
  14. ACI Pseudo-code

Figures (1)

  • Figure 1: Prediction vs. truth with ACI intervals. Black: observed series; Blue: DS$^3$M mean; Grey: DS$^3$M native MC interval; Orange: ACI band. Coverage near $0.90$ indicates successful calibration; narrower orange bands at similar coverage indicate more efficient uncertainty than the native MC band.