| Summary: | 碩士 === 逢甲大學 === 統計與精算所 === 97 === This paper investigates structural change problems in regression and return prediction models, it is difficult to fit those models when there exists structural changes. Here specific targets can be separated into the two parts. First, for segmented regression models, we discuss how to classify the groups in mixture data, and apply the Bayesian method and the data augmentation to estimate unknown parameters include threshold values, intercepts, piecewise slopes and latent variables. Second, for return prediction models, the stock return is affected by some state variables in certain periods to arise the structural changes problem. We consider a structural break in instability of return prediction model with heteroskedasticity and employ a Bayesian approach to identify a structural break at an unknown breakpoint in the model. We simultaneously make inference unknown parameters and a breakpoint. Finally, we carried out simulation studies and empirical examples for both models. The illustrations of simulation studies and empirical examples present the well performance of the proposed
algorithms. The results display the Bayesian method is proposed which can properly simultaneously identify the changepoint and estimate unknown parameters above two models.
|
|---|
