Improved confidence sets for the date of a structural break

Prior studies have proposed constructing confidence sets for the break date by inverting a sequence of tests for the date of a structural break. In this study, we improve these confidence sets by taking the direction of the break into account. Even when the break direction is unknown, we can consistently estimate it, enabling us to use the proposed method. Simulation results show that the proposed method effectively reduces the length of the confidence sets, while maintaining a good coverage rate. An empirical example illustrates the usefulness of the proposed method.