A1088
Title: Sequential conformal prediction for time series
Authors: Chen Xu - Georgia Institute of Technology (United States) [presenting]
Yao Xie - Georgia Institute of Technology (United States)
Abstract: A new distribution-free conformal prediction algorithm is presented for sequential data (e.g., time series), called the sequential predictive conformal inference (SPCI). The nature that time series data are non-exchangeable is specifically accounted for, and thus many existing conformal prediction algorithms are not applicable. The main idea is to adaptively re-estimate the conditional quantile of non-conformity scores (e.g., prediction residuals) upon exploiting the temporal dependence among them. More precisely, the problem of conformal prediction interval is cast as predicting the quantile of a future residual, given a user-specified point prediction algorithm. Theoretically, asymptotic valid conditional coverage is established upon extending consistency analyses in quantile regression. Using simulation and real-data experiments, a significant reduction is demonstrated in the interval width of SPCI compared to other existing methods under the desired empirical coverage. Extensions to multivariate time series are also discussed.
