Asymptotically Sufficient Statistics in Nonparametric Regression Experiments with Correlated Noise英文文献资料.docVIP

HindawiPublishingCorporation JournalofProbabilityandStatistics Volume2009,ArticleID275308,19pages doi:10.1155/2009/275308 ResearchArticle Asymptotically Suf?cientStatisticsin Nonparametric Regression Experiments with Correlated Noise AndrewV.Carter Department ofAppliedProbabilityandStatistics,University ofCalifornia, SantaBarbara,SantaBarbara,CA93106-3110, USA Correspondence shouldbeaddressedtoAndrewV.Carter,carter@ Received30July2009;Accepted 16December 2009 Recommended byA.T.A.Wood We ?nd asymptotically su?cient statistics that could help simplify inference in nonparametric regression problems with correlated errors. These statistics are derived from a wavelet decomposition thatisusedtowhitenthenoiseprocessandtoe?ectivelyseparate high-resolution andlow-resolution components. Thelower-resolution components containnearlyalltheavailable information about the mean function, and the higher-resolution components can be used to estimate the error covariances. The strength of the correlation among the errors isrelated tothe speed atwhich thevariance ofthehigher-resolution components shrinks, and this isconsidered anadditionalnuisanceparameterinthemodel.WeshowthattheNPRexperiment withcorrelated noise is asymptotically equivalent to an experiment that observes the mean function in the presence of a continuous Gaussian process that is similar to a fractional Brownian motion. These results provide atheoretical motivation forsome commonly proposed wavelet estimation techniques. Copyright q2009AndrewV.Carter. Thisisanopenaccess article distributed under theCreative Commons Attribution License, which permits unrestricted use,distribution, andreproduction in anymedium,providedtheoriginalworkisproperlycited. 1.Introduction A nonparametric regression NPR problem consists of estimating an unknown mean functionthatsmoothlychangesbetweenobservationsatdi?erentdesignpoints.Thereare nobservationsYioftheform   i Yiμ ξi fori1,...,n, 1.1 n whereμistheunkno