Extremal Dependence-Based Specification Testing of Time Series

TL;DR

This paper introduces a new specification test for time series models based on extremal dependence of residuals, offering an easier-to-implement alternative to existing methods and demonstrating its effectiveness through simulations and real data application.

Contribution

It develops a nuisance parameter-free Portmanteau-type test based on tail copulas for serial independence in residuals, improving ease of use and detection power.

Findings

Test performs well in simulations

Detects violations missed by standard tests

Applicable to risk forecasting models

Abstract

We propose a specification test for conditional location--scale models based on extremal dependence properties of the standardized residuals. We do so comparing the left-over serial extremal dependence -- as measured by the pre-asymptotic tail copula -- with that arising under serial independence at different lags. Our main theoretical results show that the proposed Portmanteau-type test statistics have nuisance parameter-free asymptotic limits. The test statistics are easy to compute, as they only depend on the standardized residuals, and critical values are likewise easily obtained from the limiting distributions. This contrasts with extant tests (based, e.g., on autocorrelations of squared residuals), where test statistics depend on the parameter estimator of the model and critical values may need to be bootstrapped. We show that our test performs well in simulations. An empirical application to S&P 500 constituents illustrates that our test can uncover violations of residual serial independence that are not picked up by standard autocorrelation-based specification tests, yet are relevant when the model is used for, e.g., risk forecasting.