Unconditional density vs conditional density functions in estimating value-at-risk

ABSTRACT

This research compares the performance of 10 Value-at-Risk (VaR) models, including pure density models (student’s t, mixture of normal, double gamma, kernel, and GPD) and conditional density models (GARCH-student’s t, GARCH-mixture of normal, GARCH-double gamma, GARCH-kernel, and GARCH-GPD). We employ these models and test their performance in Asian markets, including over the 2008 financial crisis period. Findings show that the density functions used in conjunction with the GARCH models are capable of capturing the high peak, fat-tails, and volatility clustering of returns and can improve the accuracy of VaR forecasts. Furthermore, the proposed GARCH-double gamma provides the best fit for all tested Asian markets, both developed and emerging markets, and for foreign exchange rates markets.

KEYWORDS:

• Value-at-Risk
• density Function
• extreme Value Theory
• double Gamma

JEL CLASSIFICATION:

• B23
• C14
• C22

Disclosure statement

No potential conflict of interest was reported by the authors.