Revue d'économie et de statistique appliquée
Volume 23, Numéro 1, Pages 102-116
2026-06-28

Comparative Analysis Of Extreme Value Theory Approaches For Value At Risk (var) Estimation

Authors : Cheraitia Hassen .

Abstract

This study evaluates statistical techniques for Value at Risk (VaR) estimation, focusing on Extreme Value Theory (EVT) approaches compared to the Historical VaR benchmark. Using 249 daily log returns of AAPL.INC stock (NASDAQ) from May 9, 2022, to May 5, 2023, we applied GEV, GPD, Hill, Pickands, and Moments estimators. The GPD approach showed the closest alignment with Historical VaR, yielding a Mean Squared Error (MSE) of nearly zero (4.2×〖10〗^(-6)) and consistent results across confidence levels (e.g., VaR at 99%: -0.053 for Historical vs. -0.051 for GPD). Pickands estimator provided reasonable accuracy (MSE=2.2×〖10〗^(-5)), while Hill, Moments and GEV exhibited higher MSE values (1.9×〖10〗^(-4),2.1×〖10〗^(-4) and 6.5×〖10〗^(-4), respectively). These findings highlight the potential of EVT for accurate tail risk modeling, with GPD emerging as the most robust approach.

Keywords

Value-at-Risk(VaR ; Extreme value Theory (EVT) ; semi parametric estimators ; risk management ; financial modeling