In the past decade, we have seen the development of a new set of tests for
structural change of unknown timing in regression models, most notably the
SupF statistic of Andrews (1993, Econometrica 61, 825-856), the ExpF and Av
eF statistics of Andrews-Ploberger (1994, Econometrica 62, 1383-1414), and
the L statistic of Nyblom (1989, Journal of American Statistical Associatio
n 84, 223-230). The distribution theory used for these tests is primarily a
symptotic, and has been derived under the maintained assumption that the re
gressors are stationary. This excludes structural change in the marginal di
stribution of the regressors. As a result, these tests technically cannot d
iscriminate between structural change in the conditional and marginal distr
ibutions. This paper attempts to remedy this deficiency by deriving the lar
ge sample distributions of the test statistics allowing for structural chan
ge in the marginal distribution of the regressors. We find that the asympto
tic distributions of the SupF, ExpF, AveF and L statistics are not invarian
t to structural change in the regressors. To solve the size problem, we int
roduce a 'fixed regressor bootstrap' which achieves the first-order asympto
tic distribution, and appears to possess reasonable size properties in smal
l samples. Our bootstrap theory allows for arbitrary structural change in t
he regressors, including structural shifts, polynomial trends, and exogenou
s stochastic trends. It allows for lagged dependent variables and heteroske
dastic error processes. (C) 2000 Elsevier Science S.A. All rights reserved.
JEL classification: C22.