Adaptive Conformal Predictions for Time Series (2022-02-15T00:00:00.000000Z)

TL;DR

This work argues that Adaptive Conformal Inference (ACI), developed for distribution-shift time series, is a good procedure for time series with general dependency, and proposes a parameter-free method, AgACI, that adaptively builds upon ACI based on online expert aggregation.

Abstract

Uncertainty quantification of predictive models is crucial in decision-making problems. Conformal prediction is a general and theoretically sound answer. However, it requires exchangeable data, excluding time series. While recent works tackled this issue, we argue that Adaptive Conformal Inference (ACI, Gibbs and Cand{\`e}s, 2021), developed for distribution-shift time series, is a good procedure for time series with general dependency. We theoretically analyse the impact of the learning rate on its efficiency in the exchangeable and auto-regressive case. We propose a parameter-free method, AgACI, that adaptively builds upon ACI based on online expert aggregation. We lead extensive fair simulations against competing methods that advocate for ACI's use in time series. We conduct a real case study: electricity price forecasting. The proposed aggregation algorithm provides efficient prediction intervals for day-ahead forecasting. All the code and data to reproduce the experiments is made available.

Authors

Margaux Zaffran

1 papers

Aymeric Dieuleveut

2 papers

Olivier F'eron

1 papers

TL;DR

Abstract

Authors

References45 items

Expert Aggregation for Financial Forecasting

Random Forests for Spatially Dependent Data

A Gentle Introduction to Conformal Prediction and Distribution-Free Uncertainty Quantification

Adaptive Conformal Inference Under Distribution Shift

Regularized quantile regression averaging for probabilistic electricity price forecasting

CRPS Learning

Conformal prediction interval for dynamic time-series

Forecasting day-ahead electricity prices: A review of state-of-the-art algorithms, best practices and an open-access benchmark

Robust Validation: Confident Predictions Even When Distributions Shift

Conformal prediction interval estimation and applications to day-ahead and intraday power markets

Conformalized Quantile Regression

Conformal Prediction Under Covariate Shift

Exact and Robust Conformal Inference Methods for Predictive Machine Learning With Dependent Data

Introduction to Online Convex Optimization

Additive models and robust aggregation for GEFCom2014 probabilistic electric load and electricity price forecasting

Probabilistic forecasting of electricity spot prices using Factor Quantile Regression Averaging

Distribution-Free Predictive Inference for Regression

Electricity price forecasting: A review of the state-of-the-art with a look into the future

Optimal learning with Bernstein online aggregation

A second-order bound with excess losses

The Hybrid Wild Bootstrap for Time Series

A Simple Bootstrap Method for Time Series

Time series prediction with performance guarantee

Network Traffic Demand Prediction with Confidence

A primal-dual perspective of online learning algorithms

A tutorial on conformal prediction

Prediction, learning, and games

Bootstrap Methods for Time Series

Inductive Confidence Machines for Regression

Machine-Learning Applications of Algorithmic Randomness

Aggregating strategies

Multidimensional Additive Spline Approximation

Opera

Conformal prediction for dynamic time-series

Random forests for time-dependent processes

Application of conformal prediction interval estimations to market makers' net positions

EUPHEMIA Public Description Single Price Coupling Algorithm

Random Forests for Time Series

Forecasting spot electricity prices Deep learning approaches and empirical comparison of traditional algorithms

2018) proves the following Theorem A.2 about SCP quasi-exact validity. Theorem A.2

Recent advances in electricity price forecasting: A review of probabilistic forecasting

Markov Chains and Stochastic Stability

The Book Review Column 1 Coverage Untyped Systems Simple Types Recursive Types Higher-order Systems General Impression 3 Organization, and Contents of the Book

. Achieves valid coverage for every simulation settings with a well chosen γ , or for dependence such that ϕ < 0 . 95 . It is robust to the strength of the dependence

Its validity strongly depends on the data distribution (it is valid on a MA(1) noise, not in AR(1) and ARMA(1,1) noise). When the method is valid , it produces the smallest intervals

Field of Study

Venue Information

Name

Type

URL

Alternate Names