A0691
Title: A modified sequential procedure to estimate the number of breaks in trend
Authors: Daisuke Yamazaki - Kyushu University (Japan) [presenting]
Abstract: For univariate time series data with structural breaks in the deterministic trend, the number of breaks can be estimated by sequential testing. However, the existing method has two drawbacks. First, the procedure requires a consistent breakpoint estimator in order to apply the sequential tests, but the trend point estimator is inconsistent when the number of breaks is under-specified. This inconsistency results in the low power of the sequential tests. Second, the tests suffer from the non-monotonic power problem. This is because the long-run variance estimator of the error term is inconsistent when the number of breaks is under-specified. To solve these problems, we propose a modified procedure to estimate the number of breaks in trend. First, we develop a new breakpoint estimator that is consistent even when the number of breaks is under-specified. Second, in order to avoid the non-monotonic power problem, we construct a modified long-run variance estimator. Simulation results show that the power of the modified test is much higher than that of the existing test so that the proposed method has good finite sample properties.
