Leading Probability of a US Recession

This leading probability index of US recessions has been developed in 2005, concluding a two year research on leading indicators, the US business cycle and links with financial markets. This is a 'quasi real-time ' model which detected the last six US recessions with an average (resp. median) real-time lead of 8 (resp. 7) months. The leading recession index is based upon US non-financial time series, non subject to significant revisions. The model is built on a markov-switching detection principle ''à la Hamilton''. Estimates are produced with MSVARLIB on a monthly sample starting in february 1960.


Recession probability update - March 2008 - "The shallowest recession ever ?"

In March 2008, the leading probability of a recession was down to 1%. Since the beginning of the year, odds of a full-blown recession beyond the next 3 quarters have continuously declined.
Preliminary estimates indicate that the coincident filtered probability of a recession was up from 17% in February to 44% in March. Our coincident diffusion index also based on very preliminary March data has been in the red zone for the third month in a row. For now, coincident indicators - that may however considerably be revised in the coming months - show that activity has stalled [...]
Since January 2007, the behaviour of the leading probability has been rather hectic and unprecedented [...].

Latest full report:

March 2008 - The shallowest recession ever ? , (Release, April 2008)

Note : new updates may be released on this website with a varying timing, but real-time estimates are reserved for proprietary use. A signal needs to be persistent to launch a recession call. Historical results based on quasi real-time vintages can be found below.



An average lead of 8 months on the last six US recessions

How should the filtered probability be interpreted?

Following Hamilton, a probability rising above 50% is likely to announce a recession. Here are the leads based upon this conservative criterion related to the following recessions: 1970 (6), 1974 (4), 1980 (11), 1982 (7), 1990 (14), 2000 (7). The model offers an average lead on the NBER datation of 8 months, and a median lead of 7 months. In practice, a rising and persistent filtered probability (at least 3 consecutive hikes above a 30% threshold) has been robust enough to detect past incoming recessions. Note that we may suspect that the model also detects the 1960 recession far in advance (at least 2 months, but the lead may be downward biased because of the lack of available data).

Click here, for a 48 year long review of the US business cycle with this index.

The model has produced only one false signal in 1966-1967, corresponding to the credit crunch, that some economists consider close to a recession, see for instance Chauvet M. (1998). "An Econometric Characterization of Business Cycle Dynamics with Factor Structure and Regime Switching", International Economic Review, vol. 39. Interestingly, a rising probability persisting above 5% is a short-leading indicator of more minor slowdowns such as (1962, 1968, 1973, 1978, 1984, 1995 and 1998). The probability has been a reliable measure how persistent, pervasive and pronounced is a slowdown to degenerate into recessions. For further information on research and references about recession and business cycles modelling, you may refer to some personal working papers and programs and to ressources on Business Cycle Economics and Econometrics.


Historical records

Here are archived public reports. Releases may be made public, two or three weeks after the month of reference, when probabilities suggest activity may significantly slow.


Estimations are conducted every year and probabilities are calculated out-of samples. (Present sample estimate: 1960/2 - 2007/1). Because series are non subject to major revision, there is no significant "vintage effect". Some minor differences may arise because models are estimated every year and probabilities are calculated out-of samples. See here, for comparisons on "vintage effect", and "estimation effect". An enhanced version of the model has been introduced in early 2008, using a two-month averaged signal rather thant a one-month signal to increase its readability.


Disclaimer

This model is proprietary, copyrighted Ó2005 by Benoît BELLONE, all rights reserved. No further information will be provided about its design and its updates, except those presented on this website. Note that Benoît Bellone takes no responsibility for any use of the information delivered on his website, which remains entirely at the reader's risk.

 

 

| Home / Contact info | Working papers and Publications | MSVARlib Gauss open source Library |Links |Business Cycle Economics |Leading US recession index |