Predictive Regression is an important tool in exploring return predictability. We introduce an efficient procedure to select and estimate the active predictors and change points in a high-dimensional structural break predictive regression, where the number of change points is allowed to vary with the sample size, and the predictors are allowed to be stationary, cointegrated and nonstationary with sparse active variables. We first select the active predictors initially by a Sure Independence Canonical Screening (SICS) procedure. Then, we estimate the change points by a Ratio-controlled Regression Screening (RRS) procedure. Finally, we eliminate the redundant break points and predictors by information criteria (IC). It is shown that the true break points and active predictors could be estimated and selected consistently. Simulations and Empirical Studies show that the proposed procedure performs quite well.