Saved in:
Bibliographic Details
Main Author: Petrosyan, Armen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.04490
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914239892946944
author Petrosyan, Armen
author_facet Petrosyan, Armen
contents Standard risk metrics used in model validation, such as the Kolmogorov-Smirnov distance, fail to converge at practical rates when applied to high-frequency financial data characterized by heavy tails (infinite skewness). This creates a "noise barrier" where valid risk models are rejected due to tail events irrelevant to central tendency accuracy. In this paper, we introduce a Weighted Kolmogorov Metric tailored for financial time series with sub-cubic moments ($\mathbb{E}|X|^{2+δ}<\infty$). By incorporating an exhaustion function $h(x)$ that mechanically downweights extreme tail noise, we prove that we can restore the optimal Gaussian convergence rate of $O(n^{-1/2})$ even for Pareto and Student-t distributions common in Crypto and FX markets. We provide a complete proof using a core/tail truncation scheme and establish the optimal tuning of the weight parameter $q$.
format Preprint
id arxiv_https___arxiv_org_abs_2601_04490
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Restoring Convergence in Heavy-Tailed Risk Models: A Weighted Kolmogorov Approach for Robust Backtesting
Petrosyan, Armen
Probability
Standard risk metrics used in model validation, such as the Kolmogorov-Smirnov distance, fail to converge at practical rates when applied to high-frequency financial data characterized by heavy tails (infinite skewness). This creates a "noise barrier" where valid risk models are rejected due to tail events irrelevant to central tendency accuracy. In this paper, we introduce a Weighted Kolmogorov Metric tailored for financial time series with sub-cubic moments ($\mathbb{E}|X|^{2+δ}<\infty$). By incorporating an exhaustion function $h(x)$ that mechanically downweights extreme tail noise, we prove that we can restore the optimal Gaussian convergence rate of $O(n^{-1/2})$ even for Pareto and Student-t distributions common in Crypto and FX markets. We provide a complete proof using a core/tail truncation scheme and establish the optimal tuning of the weight parameter $q$.
title Restoring Convergence in Heavy-Tailed Risk Models: A Weighted Kolmogorov Approach for Robust Backtesting
topic Probability
url https://arxiv.org/abs/2601.04490