Average of Variable Coefficient Models Based on Compound Quantile Regression
To address the problem of averaging variable coefficient models,a new semi-para-metric modeling strategy is used,i.e.,a series of non-nested candidate models are construc-ted by rotating continuous variables as indicator variables.The candidate models are esti-mated based on composite quantile regression,and the discard-one cross-validation criterion is applied to select weights for model averaging prediction.A simulation study demonstrates the better finite-sample nature of the method,and finally the method is applied to Boston housing data to further illustrate its accuracy and validity.
composite quantile regressionmodel averagevarying coefficient model
NETL
NSTL
万方数据
