拟蒙特卡罗方法用于估计多元回归函数.pdfVIP

维普资讯 哥 丑曹 喜 事 霉 ； 应用概率统计 第十四卷 ChitleseJour,udofAppliedPrd~abi6ty ■ ． ● ． ． 第四期 199@,年 ¨月 alldStatisticsVn1．14No．4Nov．19q8 Quasi-】ⅥonteCtrioM ethodsforEstimating aM ultivariateRegressionFunction* WANOXIAOQUN (Tsin#tu U*tbcrtity,Beljiq，tooo~4) Abstract Quasi-MonteCarlo(QmC)methodsaxewidelylimedforsolvingdifferentproblemsofnumer． icedandeysisand statistics．suchas~ntegTation．optintization．experht~entde design，sil．ulationof stochasticproc~ses，etc．Inthispaper-thesemethodsaJesnggestedofr~timatingamultivariate regremsionfIlnction．It shownthat．inarathergeneralcase．the uniformdesign (respec- tively,tile “represeutativepointsdesign)ttogetherwithstaudardQMCestimates(respetively． ntodifiedQMCestimatesu【singquasi-?alldolnimportancesampling))ofFouriercoelficients0fa regressionfnnctionprovideallasymptoticallyoptimalprocedureofprojectionestimationofthe regression fimction． Keywords：QItui·MonteCarlomethods．discrepancy,experimentaldesign．multivariate nonparanletticregression． AMSSnbj+ctCla~lflcatlon：62(；07，62K99．65(：05． §1． Introduction Considerasimpleregressionmode1．Letamtdtivariateregre~ioaflmction，(￡)maybe observed： (￡)=f(4+f()，zED，where(()isazeromeanrandomvar；,ab