A test for changing trends with monotonic power

摘要

We propose a powerful test for changing trends in which no nuisance parameters are needed to be nonparametrically estimated, and the only inputs required are the regression residuals under the null hypothesis. The new test allows for serial dependence, conditional heteroskedasticity and time-varying unconditional variance in error terms. Monte Carlo experiments show the test has monotonic power against abrupt or smooth structural changes in trends. (C) 2016 Elsevier B.V. All rights reserved.