Extensions to complex serial correlation and high-dimensional model averaging

Develop conformal-inference-based prediction intervals for model averaging that handle dependent data with more complex forms of serial correlation and high-dimensional candidate model sets, and establish corresponding coverage guarantees.

Background

The paper delivers finite-sample and asymptotic coverage results for model averaging under exchangeability and under stationarity with weak dependence, along with algorithms and demonstrations in linear settings.

In closing, the authors explicitly flag open questions regarding extending conformal methods for model averaging to more complex serial dependence structures and to high-dimensional settings, which would require new theoretical developments and algorithmic design.