Fitting a Pareto-Normal-Pareto distribution to the residuals of financial data (Q1424662)

A model for Value-at-Risk (VaR) and expected shortfall estimation is proposed. In this model the log-returns of a financial series are described by an AR-GARCH process with Pareto-Normal-Pareto innovations (PNP, i.e., the PDF \(f(x)\) of the innovations is Gaussian if \(x\in[u_1,u_2]\) and generalized Pareto if \(x\not\in[u_1,u_2]\), were \(u_i\) are some fixed thresholds). The parameters of AR-GARCH processes are fitted by a pseudo-maximum likelihood estimator (PMLE), and the residuals are used to fit PNP by MLE. Results of simulations and application to South African stock exchange data are presented.