Asymptotics of Quantiles and Rank Scores in Nonlinear Time Series (Q4956027)

The author deals with a problem of parameter estimation in a rather general nonlinear time series model which covers both stationary and nonstationary cases and includes, for instance, self-exciting threshold autoregressive (SETAR) models, exponential autoregressive models, random coefficient autoregressive models and SETAR models. The notion of regression and autoregression quantiles and rank scores is extended to these time series models. The asymptotic realization of these nonlinear quantiles gives the possibility to obtain the limiting distributions of a class of L-estimators of model parameters. In particular, the limiting distributions of the least absolute deviation estimator and trimmed estimators are derived. Their asymptotic efficiency is compared with conditional least squares estimators (CLSE). The practical performance of \(L\)-estimators is studied within SETAR models considering the following error distributions: standard normal, Tukey's contaminated Gaussian and Cauchy distribution. The \(L\)-estimators turn out to be asymptotically more efficient than CLSE for heavy-tailed distributions. This effect becomes obvious even for small contamination and sample size \(n=5\).