Monte Carlo Evaluation of the Methods Estimating Structural Change Point in Panel Data

Abstract

In this study, we investigate the existence of structural break in a panel data consisting of N time series of T unit length, and the estimation performance of Simple Mean Shift Model, Fluctuation Test, Wald Statistic Test, Kim Test which are based on common break assumption are examined to determine the break date. In this context, 108 Monte Carlo simulations are performed, each of which consisted of 3000 repetitions for the factors number of cross-sections, time dimension, break size and break rate factors, which are considered to influence the performance of the tests. As a result of the Monte Carlo simulations, the Simple Mean Shift Model approach predicts the break point with a higher performance than the other methods. In addition, if the breakpoints are at the midpoint of the series, the Wald Statistic and Kim Tests show the highest performances, while the Fluctuation Test shows the highest breakpoint predictive performance if break occur in the third quarter of the series. Generally, as the number of cross-sections increases, the estimation performance of the tests increases, whereas as the time dimension increases, the performance of methods other than the Simple Mean Shift Model decreases. As a final point, it has been observed that there is no significant effect of the break size on the predictive performance of the methods.