The determination of business cycle turning points is a classic problem in economic statistics. Many of our basic notions of the lead-lag relations among macroeconomic time series are informed by traditional methods of dating turning points for individual series and comparing them to turning points of the overall economy. Chronologies of business cycle turning points (in the jargon, reference cycle chronologies) are currently maintained in the United States by the NBER Business Cycle Dating Committee, in Europe by the CEPR, and by similar organizations in other countries.
This paper compares two approaches to dating business cycles. The dominant current approach, both in the academic literature and in the real-time practice of dating committees, is to date reference cycles by focusing on one, or a few, highly aggregated time series. Hamilton (2010) surveys the academic literature on identifying peaks (dating and predicting recessions). All the methods he discusses define recessions or turning points in terms of single highly aggregated series such as GDP or a monthly index of coincident indicators.
Press releases of the NBER Business Cycle Dating Committee indicate that its current practice is to focus on a few highly aggregated series; for example, the press release announcing the 2007:12 peak (NBER (2008)) gives greatest weight to three aggregates (establishment employment, GDP, and GDI), gives secondary weight to five more aggregates (industrial production, household employment, real manufacturing and trade sales, real personal income less transfers, and monthly consumption), and mentions no other series. We will use the term “average then date” to describe the dating of reference cycles using a single highly aggregated series, such as GDP.
As Harding and Pagan (2006) point out, this average then-date approach contrasts with the approach of the pioneers of business cycle dating, who considered a large number of disparate disaggregated series, identified turning points in those disaggregated series, then determined reference cycle turning points based on the distribution of the turning points of the disaggregated series; see Burns and Mitchell (1946, p. 13 and pp. 77-80). We refer to this latter approach as “date then average.”
