UNIT ROOTS IN THE PRESENCE OF ABRUPT GOVERNMENTAL INTERVENTIONS WITH AN APPLICATION TO BRAZILIAN DATA REGINA CELIA CATI, a MARCIO G. P. GARCIA b AND PIERRE PERRON a,c * a Department of Economics, Boston University, 270 Bay State Road, Boston, MA 02215, USA b PUC, Rio de Janeiro c CRDE, Universite  de Montre Âal SUMMARY This paper considers econometric issues related to time-series data that have been subject to abrupt governmental interventions. The motivating example for this study is the Brazilian monthly in¯ation rate (1974 : 1±1993 : 6) which we use throughout for illustration. This series has been heavily in¯uenced by the eect of so-called shock plans implemented by various governments starting in the mid-1980s. The plans act as `inliers' in the sense that the series is temporarily brought down to low levels before returning to its previous trend path. We analyse the eects on standard unit root tests and measures of persistence caused by the presence of these `inliers'. We show a substantial bias in favour of concluding that the series is stationary and that shocks have temporary eects. We then construct appropriately corrected statistics which take into account the presence of the plans. These show, unlike the standard tests, that the stochastic behaviour of the in¯ation rate was indeed unstable over this period. Simulation results are presented to support the adequacy of our corrected statistics. Copyright # 1999 John Wiley & Sons, Ltd. 1. INTRODUCTION Non-stationarity in economic data can take various forms; for example, an autoregressive unit root or the presence of structural changes in a functional relation among a set of variables. In this paper, we discuss an alternative form of non-stationarity related to the eects of abrupt governmental interventions also referred to as `shock plans'. Our analysis is directly motivated by the time-series properties of the Brazilian in¯ation rate. This series is characterized by important increases starting in the early 1980s, turning into hyperin¯ation by the end of the 1980s. Yet this period of very high in¯ation has been marked by a few (®ve that are important until the early 1990s) `shock plans' which have brought in¯ation to a low level for a short period of time. Intuition suggests that, in this highly volatile period with an ever-increasing trend path for in¯ation, standard statistical measures related to the issue of non-stationarity and the persistence of shocks would show the series to be highly persistent and non-stationary. Yet exactly the opposite occurs. Standard unit root tests suggest that in¯ation was stationary in that period and that shocks aected its level in a temporary manner. Indeed, standard measures suggest that in¯ation was `more stationary' and less persistent in this hyperin¯ation period than in the 1970s when in¯ation was moderate. CCC 0883±7252/99/010027±30$17 . 50 Received 31 January 1996 Copyright # 1999 John Wiley & Sons, Ltd. Revised 6 October 1997 JOURNAL OF APPLIED ECONOMETRICS J. Appl. Econ. 14: 27±56 (1999) *Correspondence to: Pierre Perron, Department of Economics, Boston University, 270 Bay State Road, Boston, MA 02215, USA. E-mail: perron@bu.edu Contract grant sponsors: ACDI; Universidade de SaÄo Paulo; Universite de MontreÂal; SSHRC; NSERC; FCAR.