Probability of Recession and the TMN Model

Since The National Bureau of Economic Research (NBER) is extremely late in announcing peaks and troughs in the US Economy, TMN thinks it would be useful to have its own model in predicting recessions.

Possible reasons for the late declarations of peaks and troughs could vary and include many, but two possible ones are the following: 1) Political reasons, as the ruling party may benefit from delaying a declaration that a recession had started or accelerating a declaration that a recession had ended; 2) Conflict of interest between the committee members of NBER who decide the peaks and troughs.

TMN is using a simple approach as it believes that the more simple the approach the better chances exist for the model to be robust as economic data change and may get revised every month. For that reason the model employs only two variables that may get revised when new data arrive, and the methodology used is the logit binary choice model. Furthermore, the model is using monthly data and calculates the probability of the economy being in a recession in the current month by using the latest available data.

The model tries to capture changes in the following areas by using the appropriate variables/indexes according to TMN's judgement: 1) Shipping Market Conditions, i.e. demand and supply for seaborne transportation of goods; 2) Stock Market Conditions; 3) Credit Market Conditions; 4) Employment Conditions and 5) and Manufacturing Conditions (note that the only variables subject to revisions are in 4 and 5). 

In the attempt to decide the cut-off probability value that will define the start and end of a recession, TMN uses the optimal cut-off probability as described by Palepu (1986). The cut-off probability - of the model - that defines when a recession has started or ended is 26%; i.e. probabilities that are higher (lower) than 26% declare that the economy is in a recession (expansion) the current month.  The results of the model can be found below, where the estimated probability of recession is plotted against the SP500 and the recessions as defined by NBER. 

The current probability that the US Economy was in recession during September 2010 is now standing at very low levels, 0.40%. It is now interesting to look at how far ahead the TMN model identifies peaks and troughs in respect to the NBER's declaration dates and actual peaks and troughs. 

As it can be noted from the table above, the TMN model signals peaks and troughs well ahead of the NBER. Looking at actual dates of peaks and troughs the TMN does not perform bad either. During the 1990s recession, the TMN model was late one month in identifying both the start and end of recession. In the case of the early 2000 recession, the TMN model identified the start of recession 4 months ahead of the actual date. On the other hand, there was a lag of 2 months behind in identifying the end of recession. Finally, the TMN model did not perform bad for the recent recession either. It recognised the start of recession 3 months after the actual date and it coincided exactly on the same month as the actual date for the end of the recession.

Overall, TMN believes that the model is very useful in identifying the overall economic situation in the US and certainly faster than NBER's declaration dates of peaks and troughs. Finally, two good indicators for the business climate in the US are the Aruoba-Diebold-Scotti Conditions Index provided by the Federal Reserve Bank of Philadelphia, and the Russell Investments Business Cycle Index.