Another Interpretation

Monday Greg Mankiw blogged about a new model for predicting recession:

Tim Kane of the Joint Economic Committee staff has a new paper on Employment Numbers as Recession Indicators. The abstract:

This paper investigates the value of employment data as real-time recession indicators. Among popular monthly labor measures, the unemployment rate is the most useful as an indicator of recession, whereas two top measures of employment growth–payroll jobs and civilian employment–have little value. Two other series, the labor force participation rate and the employment-population ratio, also provide little or no value in anticipating a recession. The best pre-recession employment indicator is actually weekly claims for unemployment insurance (UI). The paper reviews a new technique for predicting recessions, and develops an employment recession probability index. The index indicates a 35.5 percent chance that the U.S. economy is in recession, sharply up from 10 percent last month.

Today, Kane used the model to interpret the latest data:

In this morning's BLS Employment Situation report for Jan 2008, the unemployment rate is 4.9. Therefore the new employment-based recession probability index (RPI) is 0.060 (or 6.0%).

The RPI is a combination of the two most valuable employment indicators of a recession's early stages: weekly initial unemployment insurance (UI) claims and the unemployment rate.

The 4-week moving average of initial UI claims was reported yesterday at 325,750, which is 17,000 lower than 4 weeks ago and essentially unchanged from the October average. Alone, trends in UI claims suggest a 4 percent recession probability. The unemployment rate is 0.1 points lower than December, but 0.1 higher than three months ago, suggesting an 8 percent recession probability. Combined, this yields an overall recession probability of 6 percent.

I sure hope his model is right, but that claims report on Thursday was pretty lousy. If we get confirmation of a trend up in claims the odds of recession, even using this model, will rise significantly. I've been using claims as my primary indicator on employment for years. It always seemed logically to be the most accurate information because of how its produced. The weekly numbers come from actual unemployment insurance claims data provided by each state. It would seem to be the most up to date information available. Too bad I didn't build a model to prove it.