Testing for Unit Roots with Breaks: Evidence on the Great Crash and the Unit Root Hypothesis Reconsidered
| Author | Chung‐Ming Kuan,Luis C. Nunes,Paul Newbold |
| DOI | http://doi.org/10.1111/1468-0084.00076 |
| Date | 01 November 1997 |
| Published date | 01 November 1997 |
OXFORD BULLETIN OF ECONOMICS AND STATISTICS, 59, 4 (1997)
0305-9049
TESTING FOR UNIT ROOTS WITH BREAKS:
EVIDENCE ON THE GREAT CRASH AND THE
UNIT ROOT HYPOTHESIS RECONSIDERED*
Luis C. Nunes, Paul Newbold and Chung-Ming Kuan
I. INTRODUCTION
In a frequently cited study, Nelson and Plosser (1982) applied Dickey–
Fuller tests to annual data on fourteen U.S. macroeconomic time series.
The null hypothesis of a unit autoregressive root was not rejected against
the alternative of trend stationarity in thirteen cases, the exception being
the unemployment rate. However, Perron (1989) noted that these tests
lack power when the generating process is stationary about a broken
trend. Perron re-analysed the thirteen series for which Nelson and Plosser
failed to reject the null hypothesis, allowing for a break in trend in 1929.
The null hypothesis was then rejected for ten of these series. This
research provoked considerable interest in both the detection of struc-
tural breaks, and inference about the order of integration in the possible
presence of such breaks. A number of authors, including Christiano
(1992) and Zivot and Andrews (1992), urged the importance of endogen-
ous rather than exogenous selection of a break date. Analysing the data
from this perspective, the latter found less stong evidence against the null
hypothesis than did Perron for ten series, but stronger evidence for the
other three: industrial production, nominal GNP, and real GNP.
The results reported in this paper extend the analysis of Zivot and
Andrews in two ways. First, we permit a structural break under the null
hypothesis, as well as under the alternative. Second, we take account of
the sequential nature of the testing process.
The value of allowing a break under the null has often been stressed,
from the original study of Perron (1989). This point is discussed for
example by Perron (1994) and Vogelsgang and Perron (1994). Banerjee et
al. (1992) also consider unit root models with a shift in mean. When this
allowance is not made of course, rejection of the null hypothesis would
not imply strong evidence of trend stationarity, merely evidence against
the hypothesis of a unit root process with no break.
*We are extremely grateful to an associate editor, whose comments greatly improved the
presentation of our research. 435
© Blackwell Publishers Ltd, 1997. Published by Blackwell Publishers, 108 Cowley Road, Oxford
OX4 1JF, UK & 350 Main Street, Malden, MA 02148, USA.
To continue reading
Request your trial