主题：Testing for Neglected Nonlinearity in the Conditional Quantile Using Neural Networks
主讲：Georg Keilbar,School of Business and Economics,Humboldt-Universität zu Berlin
摘要：We propose a specifification test for the conditional quantile by comparing a linearmodel to an alternative model represented by a neural network. The test is formulatedas a testing problem with nuisance parameters under the alternative. We considerthree test statistics, theaverage, theexponential-averageand thesupremum Waldstatistic as functions of a Wald process and derive their asymptotic distribution. Tosimulate critical values, we propose a consistent bootstrap procedure. Further, weshow the asymptotic optimality of theaverageandexponential-average Waldtestagainst local alternatives in a correctly specifified maximum likelihood setting. In anempirical application, we examine the nonlinearity of systemic risk in the frameworkof conditional value-at-risk (CoVaR).