Vienlaicīgās ticamības joslas neperametriskajā regresijā

Abstract

Darbâ tika apskatîta vienlaicîgo ticamîbu joslu konstruçðana neparametriskajai regresijai, izmantojot San un Loader un rezidiju butstrapa metodes. Ðî darba mçríis ir pârâdît, ka heteroskedastiskâ gadîjumâ (ja kïûdu dispersija ir atkarîga no regresora), meþonîgâ rezi- diju butstrapa metode strâdâ labâk nekâ klasiskâ metode, kas balstâs uz parametriskiem nosacîjumiem par kïûdu sadalîjuma veidu.

In this work was investigated a constructing of the simultaneous confidence bands in nonparametric regression, using Sun and Loader and residual bootstrap methods. The purpose of this work is to show, that in case of heteroskedasticity (then errors depend on exogenous variables) wild residual bootstrap method works better than classical method, which is based on the type of error distribution.