Title: Jackknife, small bandwidth and high-dimensional asymptotics
Authors: Yukitoshi Matsushita - Tokyo Institute of Technology (Japan) [presenting]
Taisuke Otsu - London School of Economics (United Kingdom)
Abstract: A jackknife based inference procedure for semiparametric models is proposed. It is shown that a jackknife empirical likelihood statistic is asymptotically pivotal under standard asymptotics, while it is not under non-standard asymptotics allowing for small bandwidths or many covariates. A modified jackknife empirical likelihood is proposed that is fully automatic and works for under both standard and non-standard asymptotics. Our findings are applied to three non-standard settings: small bandwidth asymptotics for semiparametric density-weighted average derivatives, many/weak IV asymptotics for IV models, and many covariates asymptotics for linear regression models.