Humboldt-Universität zu Berlin - High Dimensional Nonstationary Time Series

SFB649DP2017 023

Penalized Adaptive Method in Forecasting with Large Information Set and
Structure Change

Xinjue Li
Lenka Zbonakova
Wolfgang Karl Härdle

In the present paper we propose a new method, the Penalized Adaptive Method (PAM), for a data driven detection of structure changes in sparse linear models. The method is able to allocate the longest homogeneous intervals over the data sample and simultaneously choose the most proper variables with help of penalized regression models. The method is simple yet flexible and can be
safely applied in high-dimensional cases with different sources of parameter changes. Comparing with the adaptive method in linear models, its combination with dimension reduction yields a method which selects proper significant variables and detects structure breaks while steadily reduces the forecast error in high-dimensional data. When applying PAM to bond risk premia modelling, the locally selected variables and their estimated coefficient loadings identified in the longest stable subsamples over time align with the true structure changes observed throughout the market.

SCAD penalty, propagation-separation, adaptive window choice, multiplier bootstrap,
bond risk premia

JEL Classification:
C13, C20, E37