Sample Risk Premium
An additional tail-risk component in data-driven policy evaluation that is overlooked by averaging criteria and that can become a dominant factor under the CVaR criterion.
Timeline (last 90 days)
The paper decomposes the policy's regret relative to an oracle that knows the true distribution into three components: expected post-sample regret (PS), the outcome-only CVaR gap, and the sample risk premium (SR).
Under the expected loss criterion, the sample risk premium vanishes; but under the CVaR criterion, it can become the dominant factor.
For loss distributions with continuous density near the tail cutoff point, even if the two components decay at the same rate, the limiting ratio of the sample risk premium to post-sample regret can still reach thousands of times.
Finite-sample numerical experiments show that under both settings, the sample risk premium accounts for about 74% to 97% of the total regret.
In fields sensitive to tail risk, such as finance, healthcare, and autonomous driving, the sample risk premium should become a routine metric for policy evaluation.
The paper's decomposition framework can be used to diagnose sources of risk: if the sample risk premium accounts for too high a proportion, it may be necessary to increase the sample size or improve the sampling strategy; if post-sample regret dominates, then generalization ability needs attention.
All Facts (6)
The paper decomposes the policy's regret relative to an oracle that knows the true distribution into three components: expected post-sample regret (PS), the outcome-only CVaR gap, and the sample risk premium (SR).
50%UnverifiedUnder the expected loss criterion, the sample risk premium vanishes; but under the CVaR criterion, it can become the dominant factor.
50%UnverifiedFor loss distributions with continuous density near the tail cutoff point, even if the two components decay at the same rate, the limiting ratio of the sample risk premium to post-sample regret can still reach thousands of times.
50%UnverifiedFinite-sample numerical experiments show that under both settings, the sample risk premium accounts for about 74% to 97% of the total regret.
50%UnverifiedIn fields sensitive to tail risk, such as finance, healthcare, and autonomous driving, the sample risk premium should become a routine metric for policy evaluation.
50%UnverifiedThe paper's decomposition framework can be used to diagnose sources of risk: if the sample risk premium accounts for too high a proportion, it may be necessary to increase the sample size or improve the sampling strategy; if post-sample regret dominates, then generalization ability needs attention.
50%