Semiparametric Predictive Regression: Theory and Practice

This thesis first studies a semiparametric single-index predictive regression model with multiple nonstationary predictors that exhibit co-movement behaviour. Then, this thesis extends the proposed model to a partially linear single-index predictive regression with stationary predictors in the linear part. Orthogonal series expansion is employed to approximate the unknown link function in the model and nonlinear least squares estimators for the unknown parameters are derived from an optimization under the constraint of an identification condition for the single-index parameter. In the empirical studies, we provide ample evidence in favour of nonlinear predictability of stock return using our proposed model.