Skip to content

Unified Inference in High Quantile Regression

Home / Open Questions / Unified Inference in High Quantile Regression

Background: In high quantile regression for dependent time series, the determination of the appropriate limit theorem is often ambiguous because different regimes of quantile extremeness require distinct inferential frameworks.

Question / Future Work: There is a need to develop more robust, unified inferential procedures that do not rely on the practitioner’s ability to correctly categorize the rate at which quantile levels approach the boundary, particularly when the sample size or tail data is limited. This is especially challenging in practice, where the underlying tail behavior and the relationship between multiple high quantile levels are unknown, leading to potential model selection ambiguity and inconsistent performance of statistical tests.

Why It Matters: This is a fundamental bottleneck in high quantile statistics where asymptotic validity often depends on specific, often unverifiable, assumptions about the rate of extremeness, hindering the reliability of automated or general-purpose statistical software for tail analysis.

Evidence: It can be seen from our developed asymptotic theory that the two situations are guided by different limit theorems for inference, while in practice it can be ambiguous to determine which limit theorem to use in a given application.

Metadata & Links