Tail expectile process and risk assessment (Q2278671): Difference between revisions

The authors consider expectiles as a least squares analogue of quantiles. The \(p\)-th quantile \(q_p\) of the distribution of a random variable \(Y\) may be defined as the minimizer \(q_p=\operatorname{arg}\min_{\theta \in R}E\{\rho_p(Y-\theta)-\rho_p(Y)\}\), with equality if the distribution function is increasing where \(\rho_p(y)=|p -1_{(y\leq 0)}||y|\) and \(1_{(\cdot)}\) is the indicator function. This successfully extends the conventional definition of quantiles as left-continuous inverse functions. It is possible to replace the absolute deviations in the asymmetric loss function \(q_p\) with squared deviations to obtain the \(\xi_p\)-th expectile of a random variable \(Y\) with finite first moment as \[ \xi_p=\operatorname{arg}\min_{\theta \in R} E\{\eta_p(Y-\theta)-\eta_p(Y)\}, \] where \(\eta_p(y)=|p -1_{(y\leq 0)}|y^2\). The expectiles are determined by tail expectations rather than tail probabilities. Under the model of heavy-tailed distributions, the authors derive joint weighted Gaussian approximations of the tail empirical expectile and quantile processes. The obtained result is used to introduce and study new estimators of extreme expectiles and the standard quantile-based expected shortfall, as well as a novel expectile-based form of expected shortfall. The estimators are built on general weighted combinations of both top order statistics and asymmetric least squares estimates. Some numerical simulations and applications to actuarial and financial data are provided.