In financial time series nonlinear effects and time-varying effects are observed. In this dissertation we propose a predictive regression model with time varying coefficients and functional coefficients. It allows for nonstationary predictors. We establish asymptotics for the coefficient estimation and show oracle properties of the resulting estimators under stationary and nonstationary settings. Simulations demonstrate good finite sample performance of our estimators. A real example illustrates the use of our methodology.