About the Recession Probabilities
On this website we publish our recession probability estimates for the US economy obtained using the methodology in Paap et al. (2009), as described below. The website is updated every month, shortly after the release of The Conference Board's Coincident Economic Index (CEI) and Leading Economic Index (LEI).
Model Explanation
We use a bivariate Markov-Switching vector autoregressive model for the monthly growth rates of The Conference Board's Coincident and Leading Economic Indexes. The model has three key features, which aim to capture the most salient business cycle characteristics of these variables:
Stylized fact I: The CEI and LEI are pro-cyclical, that is, they increase during expansions and decline during expansions. The two variables show similar cyclical patterns.
Model feature: The business cycle behavior of the CEI and LEI is characterized by different mean growth rates in recession and expansion regimes. The regimes are not imposed exogenously, but are incorporated endogenously by means of a latent two-state Markov process. The CEI and LEI are assumed to share the same business cycle, which is achieved by having only a single Markov process governing the regime switches of both variables.
Stylized fact II: The cycle of the LEI leads the cycle of the CEI, while the LEI has a considerably longer lead time when entering a recession than when entering an expansion.
Model feature: The cycles in the CEI and LEI are non-synchronous with the non-synchronicity at peaks and troughs allowed to be different. The current estimates based on the sample period January 1959 - September 2009 indicate that, on average, the LEI leads the CEI by 12 months at peaks and by 4 months at troughs.
Stylized fact III: Many US macro-economic variables such as output have shown a large and persistent decline in volatility since the mid-1980s.
Model feature: The model account for this `Great Moderation' by incorporating a single structural change in the variances and covariance of the shocks. The current estimates indicate that the variances of the CEI and LEI declined by about 50% since the beginning of 1984.
A more elaborate description of the model can be downloaded using the link below. For full details we refer to Paap et al. (2009).
March 22, 2008: Model Description [Adobe PDF]
