Business cycle asymmetry and duration dependence: an international perspective
preprintposted on 29.03.2006, 10:19 by Terence Mills
The business cycle behaviour of macroeconomic variables has long been of interest to economists, and some attention has recently focused on two aspects of this behaviour – the ‘stylised facts’ of cyclical asymmetry and duration dependence. Cyclical asymmetry is where the economy behaves differently over the expansion and recession phases of the business cycle and, consequently, cannot be captured by linear models. There has thus been much interest in non-linear specifications that can distinguish between these phases and which are sufficiently flexible to allow different relationships to apply over them. Examples of such specifications are the threshold and smooth transition models (see, for example, Teräsvirta and Anderson, 1992) and the various Markov-switching regime models, all of which stem from the original model proposed by Hamilton (1989). Duration dependence, on the other hand, concerns the question of whether, for example, the probability of a cyclical expansion is dependent on how long the expansion has been running, or whether business cycle lengths tend to cluster around a particular duration (see, for example, Diebold and Rudebusch, 1990, and Diebold, Rudebusch and Sichel, 1993). Duration dependence and switching regime models are related: Hamilton (1989) assumed that the state transition probabilities in his regime switching model were duration independent so that, for example, after a long expansion the economy was no more likely to switch to the recession regime than after a short expansion. Models such as Kim and Nelson (1998) attempt to marry these two features. Much of the empirical work using these models is either country specific or restricts analysis to just post-World War II data. Recently, however, there have become available much longer and wider macroeconomic data sets. Consequently, it is now possible to “turn business cycle theories loose on perhaps the greatest macroeconomic laboratory available: the extant record of macroeconomic historical statistics for a broad cross-section of countries since the late 19th century” (Basu and Taylor, 1999, pp. 45-6). Focusing attention on long runs of macroeconomic data has its problems, of course, for it becomes difficult to maintain the assumption of a stable model structure in the presence of the impacts of two world and various civil wars, three monetary regimes, oil price shocks, etc. This is important not just for examining the stylised facts of business cycles but also for deciding upon which methods to use to obtain the cyclical components of the time series under investigation, as such components are the necessary data on which most analyses are based. Consequently, we focus attention in this paper on nonparametric techniques for extracting cyclical components and for modelling and testing asymmetry and duration dependence. Section 2 thus introduces the data set used for the exercises and discusses the technique employed to obtain cyclical components. Sections 3 and 4 set out the hypotheses concerning cyclical asymmetry and duration dependence that are of interest and the statistics that are used to test them. Results are presented in Section 5 with conclusions following in the final section.
This paper forms part of the ESRC funded project (Award No. L1382511013) “Business Cycle Volatility and Economic Growth: A Comparative Time Series Study”, which itself is part of the Understanding the Evolving Macroeconomy Research programme.
- Business and Economics