|
Abstract:
|
Estimation of the covariance structure for longitudinal data poses significant challenges, as they are usually collected at irregular time points. Fan, Huang, and Li (07) proposed using a varying-coefficient partially linear model coupled with a semiparametric cov structure. After estimating other components, they estimated the parametric correlation structure via quasi-maximum likelihood method and showed its good finite-sample performance based on extensive simulation studies. In this work, we study the semiparametric varying-coefficient partially linear model under weaker smoothness conditions on the varying regression coefficient function, propose to estimate regression coefficients using a newly developed difference-based technique, and provide large-sample properties of our estimators. The particular focus is given to asymptotic properties of the quasi-maximum likelihood estimator.
|
