Testing for multiple level shifts with an integrated or stationary noise component

Abstract

The paper analyzes the detection and estimation of multiple level shifts regardless of the order of integration of the time series. We show that it is possible to extend the Bai-Perron methodology (1998) to the I(1) and NI(1) nonstationary cases so that a unified framework to test for the presence of multiple level shifts in a robust way is designed. The finite sample performance of the proposed statistics is carried out, establishing a comparison with other existing approaches in the literature. The paper illustrates the implementation of the statistics focusing on the real exchange rate with time series that either cover a long time period or provide a worldwide analysis. Robust detection of multiple level shifts is of great importance to define the statistical approach that is used to test the purchasing power parity hypothesis.