A study on structural change detection scheme in linear dynamic models

Abstract

When we estimate a model for a time series or forecast its future level based on the estimate model, frequent structural changes cause persistent and large estimation or forecast errors. If we solve such problems to reduce errors by considering structural changes, it will be great help to establish economic policies, corporate and business strategies. In particular, as Korea that has undergone various economic shocks including oil shock, foreign exchange crisis, and unstable credit market, it is necessary to identify the existence of structural change, to understand the significance of causal events and to perform a right forecast. In order to solve such problems, statistics, distribution and adapting algorithm have been developed. However, statistics that have been developed so far in view of forecasting have only been on the measurement of the degree of errors caused by structural change and they are not supported by meaningful statistical distribution, there could be hypothesis test errors. Also, statistics that have been developed so far in view of estimation the focus has been only on a single aspect of unit root and structural change, therefore limiting a full assess to the given problems. In this paper, various linear dynamic models are assumed, and based on a null hypothesis, statistics and distributions are derived through a simulation method. The simulation method has an advantage in solving complex problems considering the attributes of parameters, regressors, and type and numbers of structural changes. We apply these methodologies to a default rate, real interest rate, foreign exchange rate, and business cycle indexes, and confirm its practical application.