A1275
Title: Forecasting under structural breaks in nonparametric time series regression
Authors: Sze Him Isaac Leung - The Chinese University of Hong Kong (Hong Kong) [presenting]
Chun Yip Yau - Chinese University of Hong Kong (Hong Kong)
Yuanbo Li - University of International Business and Economics (China)
Abstract: In this presentation, the challenge of forecasting in non-parametric time series regression models that experience a structural break is addressed. A novel weighted kernel estimator designed to estimate the post-break regression function and predict future observations is introduced. Unlike traditional methods, this approach assigns weights that are both time and location dependent, enabling the model to make effective use of pre-break data. The findings reveal that incorporating pre-break observations can significantly enhance the estimation of the post-break regression function by reducing the mean squared forecast error. This improvement arises from managing the bias-variance trade-off associated with using pre-break data. The optimal weight allocation for both conditional and unconditional forecasting scenarios is also explored, providing theoretical insights into the asymptotic properties that guide this process. Simulation studies demonstrate the superior performance of the weighted kernel estimator compared to standard post-break methods, showing notable reductions in MSFE. To showcase the practical applicability of the method, a case study on forecasting the volatility of the Nasdaq 100 index is provided.
