Forecasting Value at Risk and Expected Shortfall based on serial pair-copula constructions

Abstract

In this paper, we use a serial dependence structure of financial assets based on pair-copula construction (PCC) to estimate risk measures in a very flexible way. This structure considers dependence with past observations isolating the effect for other lags, in a way that strengths the capacity for correct modeling. We present an algorithm to compute VaR, CAViaR, ES and CARES from this serial PCC structure. The results indicate that a considerable difference in dependence estimation is influenced by other lags. Such pattern is only observed through the proposed approach, and not by others usually considered both in academic and practical work. This pattern is consistent for returns and volatilities. Regarding the risk measures, the results indicate an absolute superiority of the approach used over two concurring, parametric ARMA–GARCH and non-parametric historical simulations.