U-statistic Type Tests for Structural Breaks in Linear Regression Models

This article introduces a U-statistic type process that is based on a kernel function which can depend on nuisance parameters. It is shown here that this process can accommodate very easily anti-symmetric kernels very useful for detecting changing patterns in the dynamics of time series. This theory is applied to structural break hypothesis tests in linear regression models. In particular, the flexibility of these processes will be exploited to introduce a simultaneous and joint test that exhibit statistical power against changes in either intercept or slope. In contrast to the literature, these tests are able to distinguish between rejections due to changes in intercept from rejections due to changes in slope; allow control of global errors rate; and are explicitly designed to have power when the distribution error is asymmetric. These tests can also incorporate different weight functions devised to detect changes early as well as later on in the sample, and show very good performance in small samples. These tests, therefore, outperform CUSUM type tests widely employed in this literature.