Reliable leading indicators of the business cycle are of great importance for policy makers, firms, and investors. It is therefore not surprising that economists set out on an intensive quest for such leading indicators, ever since the initial attempts of Mitchell and Burns (1938) for the US economy. This research has provided much insight into the construction, use, and evaluation of leading indicators, see Marcellino (2006) for a recent survey.
Reliability of a leading indicator variable includes aspects such as consistency and timeliness. By consistency we refer to the property that a leading indicator should systematically give an accurate indication of the future course of the economy and should not produce false turning point signals too frequently, for example. Timeliness means that in order to be useful, a leading indicator variable should have a considerable lead time with respect to business cycle turning points. Most of the currently popular leading indicator variables are believed to have a lead time between six and eighteen months. At the same time, it appears to be the case that many of these variables have a considerably longer lead time at business cycle peaks than at troughs. For example, the Composite Index of Leading Indicators (CLI) currently published by The Conference Board has led cyclical downturns in the economy by eight to twenty months, and upturns by one to ten months during the post-World War II period (The Conference Board, 2001).
In this paper, we develop a formal approach to investigate whether leading indicator variables have different lead times at peaks and troughs. For this purpose, we propose a novel Markov switching vector autoregressive model, where economic growth and leading indicator variables share a common cycle determined by a single Markov process, but such that their regime-switching is not exactly synchronous with the length of the displacement, or lead/lag time varying across the different regimes. We follow a Bayesian approach for estimation of the model parameters, with posterior results being obtained through flexible Markov Chain Monte Carlo techniques. The advantage of Bayesian analysis of the model is that it allows us to treat the lead/lag times as unknown parameters. We can then use their posterior distributions to conduct statistical inference on the asymmetry of the lead/lag structure at peaks and troughs.
We provide an empirical application involving monthly US industrial production (IP) and The Conference Board’s CLI over the period 1959-2004. We find that on average the CLI leads IP by more than seven months at peaks, but only by three and a half months at troughs. This suggests that, in terms of timeliness, the CLI is most useful for signalling oncoming recessions. The posterior results provide convincing evidence in favor of the presence of a non-synchronous common cycle with asymmetric lead times. The Bayes’ factor relative to an alternative specification with equal lead times at cyclical downturns and upturns is very large. The same applies to models with synchronous cycles and with independent cycles in the different variables. In addition, the CLI is more consistent and more timely in terms of signaling oncoming recessions when embedded in the general model specification. In order to examine the practical usefulness of allowing for asymmetric lead times we conduct a business cycle dating and forecasting exercise for the period from October 1987 to July 2004, using real-time data for both the CLI and IP. We find that allowing for asymmetric lead times leads to more timely and precise identification of peaks and troughs for the 1990-1991 and 2001 recessions, as well as more accurate out-of-sample forecasts of turning points as well as IP growth rates.
The paper is organized as follows. In Section 2, we introduce our novel Markov switching vector autoregressive model. In addition, we describe (nested) alternative specifications, which allow for a non-synchronous common cycle but with identical lead times at all possible regime switches, for a synchronous common cycle, and for independent cycles. To facilitate interpretation of the models, we focus on the bivariate case, where both economic growth (or the coincident indicators) and the leading indicators are represented by a single variable. We provide details of the Bayesian approach for parameter estimation and inference in Section 3. In Section 4 we discuss the empirical results based on estimating the different model specifications over the complete sample period. In Section 5, we consider the real-time performance of the alternative cycle representations in terms of identifying peaks and troughs, and forecasting turning points and IP growth. Finally, we conclude in Section 6.
Download
PDF Ebook Do Leading Indicators Lead Peaks More Than Troughs?
