Practical use of random conformal predictive distribution functions in insurance

Determine how the random conformal predictive distribution functions introduced by Vovk, Shen, Manokhin, and Xie (2019) can be utilized in practice for predictive inference in insurance, including clear procedures for interpretation and deployment in actuarial decision-making.

Background

The paper proposes a deterministic conformal predictive distribution function G_n based on a specific non-conformity measure, providing an alternative to the random conformal distributions previously studied in the literature.

The authors note that prior work by Vovk et al. (2019) presents a general framework yielding random distribution functions from conformal prediction, but they explicitly state uncertainty regarding the practical utility of such random functions for insurance prediction tasks.

References

It bears remaking that non-parametric distribution functions based on conformal prediction have been considered by Vovk et al. (2019). However, what Vovk et al. (2019) propose is a general framework for generating a random distribution function, and it is unclear how a random distribution function can be practically useful for predictive inference in insurance.

— Conformal prediction of future insurance claims in the regression problem  (2503.03659 - Hong, 5 Mar 2025) in Section 3 (Proposed method), paragraph following Theorem \ref{thm:converge}