Nonparametric Estimate for Conditional Quantiles of Time Series

Abstract

This paper investigates a nonparametric approach for estimating conditional quantiles of time series for dependent data. The considered estimate is obtained by inverting a kernel estimate of the conditional distribution function. We implement the technique on four simulated samples with light and heavy-tailed distributions and on real financial data, by calculating VaR using the nonparametric procedure. The good performance of the estimator is illustrated with backtesting.