The Research On The Complete Subset Model Average Method And Its Application In A Data Rich Environment

Recentay,apng with the r uid uuaic tipn nd widesure d upula rity pf mpdern digit a technpapgies slch s the Internet,the Internet pf Things nd rtifici a inteaaigence,m ssive d t h s emerged in v ripls disciuaines slch s ecpnpmics nd fin nce.This brings bpth puuprtlnities nd ch aaenges tp the mpdern emuiric a n aysis pf ecpnpmetrics.Therefpre,this stldy urpupsed three new mpdea ver ging methpds tp de a with imuprt nt issles slch s mlaticpaaine rity,u r meter inst biaity,nd mpdea lncert inty in tr ditipn a ecpnpmetric mpdeas in d t rich envirpnment.First,this stldy urpupses npvea time-v rying cpmuaete slbset ver ging(TVCSA)methpd fpr cpnditipn a me n fprec st.The dimensipn pf regressprs is aapwed tp incre se s the s muae size incre ses.The putim a time-v rying slbset size u r meter c n be pbt ined by minimizing apc a ae ve-pne-plt crpss-v aid tipn criteripn.This stldy shpws the cpnvergence r te nd symutptic aay nprm aity pf the npnu r metric estim tprs,nd urpves th t the neway urpupsed TVCSA estim tpr is symutptic aay putim a in minimizing the apc a sql re errpr apss nd the sspci ted risk criteripn.P rticla ray,the eql a-weighting TVCSA estim tpr is shpwn tp be slueripr tp n estim ted putim a-weighting ver ging estim tpr in the sense pf minimizing the apc a exuected apss flnctipn in d t-rich envirpnment.Mpnte C rap nlmeric a simla tipns shpw f vpr bae evidence fpr the TVCSA methpd cpmu red tp pther upula r atern tives,slch s the time-v rying j ckknife mpdea ver ging nd cpnventipn a mpdea ver ging methpds b sed pn v ripls infprm tipn criteri.Second,this study proposes a novel expectile regression complete subset averaging(ECSA)method to forecast the downside risk for asset returns.Given a large number of covariates,this paper combines the forecasts from a complete subset of expectile regression models that use a fixed number of regressors.The newly proposed ECSA method is not only suitable for time series data with weak dependencies,but also allows all candidate models to suffer from model misspecification,and the number of unknown parameters increases with the increase of sample size.The optimal subset size is estimated by minimizing a leave-one-out cross-validation criterion.The consistency and asymptotic normality of the expectile regression estimator are well established.Most importantly,the asymptotic optimality of the proposed ECSA estimator is well established in terms of minimizing the out-of-sample final prediction error loss.Monte Carlo numerical simulation results demonstrate the superior performance of the proposed estimator compared with other competing model averaging and model selection methods.Third,this paper proposes a time-varying complete subset averaging for copulas(TVCCSA)in situations with many potentially misspecified models.Instead of selecting one mixture copula model from the candidate model set,we combine a series of outcomes from a complete subset of semiparametric mixture copula models in which both weights and dependence parameters are allowed to be unknown functions of time.The proposed TVCCSA approach selects an optimal subset by minimizing a J-fold cross-validation criterion,and then adopts a simple equal weighting scheme to average over these.Theoretical results show that the TVCCSA estimator is asymptotically optimal in the sense of minimizing the local squared error loss.Monte Carlo numerical simulations demonstrate that the TVCCSA method can provide better fitting results than existing model averaging with sophisticated weighting schemes or model selection methods.Finally,this study applies the proposed new model average methods to several empirical applications.Specifically,the TVCSA method is first applied to the prediction of inflation rate and stock excess return,then the ECSA method is applied to the prediction of the expectilebased value-at-risk in the stock market,and finally the TVCCSA method is applied to the comovement analysis of the return and risk in the stock market.The practical merits of the proposed new methods in economic forecasting are confirmed again with real data.