Thursday, December 29, 2011

An examination of the model for ECRI's black box

- by New Deal democrat

Two weeks ago, an article I wrote, A Peek inside ECRI's black box gained considerable traction. There I examined a number of data series which likely generated ECRI's recession warning in September, using as my starting point the work of ECRI's founder, Prof. Geoffrey Moore, who in a 1961 manuscript set forth a number of leading indicators that predicted recessions and recoveries both before and after World War 2. A commenter at Seeking Alpha, dlamba, suggested reading Prof. Moore's later work, "Leading Indicators for the 1990's" (1991) in which he laid out in considerable detail his 1980's research into both Long and Short leading indicators, and also suggested a high-frequency Weekly Leading Index which, while slightly less reliable, could be updated in a much more timely and frequent fashion.

The book itself describes the research and the results thereof, as well as setting forth the methods by which values for each indicator were computed. A description of all of that, and the present values of each indicator, are beyond the scope of this post. Instead, let me simply describe the data that as of 1990, Moore recommended for each index.

Moore proposed 4 Long Leading Indicators:

Real M2
Dow Jones Bond Average
Housing Permits
The ratio of price to unit labor cost in manufacturing

These turn negative at least 12 months before the onset of recession, and on average 14 months before, and turn positive on average about 11 months before the end of a recession.

He also proposed 11 Short Leading Indicators:

S&P 500 stock price index
Average workweek in manufacturing
Layoff rate under 5 weeks
Initial claims for unemployment insurance
ISM manufacturing vendor performance
ISM manufacturing inventory change
Journal of Commerce change in commodity prices
Change in deflated nonfinancial debt
New orders for consumer goods and materials
Dun and Bradstreet change in business population
Contracts and orders for plant and equipment

On average these turn negative about 8 months before the onset of recession and turn positive 2 months before the end of a recession

Finally, he proposed a Weekly Leading Index made up of indicators whose values could be computed weekly, or if monthly were reported at the beginning of the next month. While this index would be a little less reliable than the two others, it would have the advantage of being more timely. It's components are:

Real M2
Dow Jones Bond Average
S&P 500 stock price index
Average workweek in manufacturing
Layoff rate under 5 weeks
Initial claims for unemployment insurance
ISM manufacturing vendor performance
ISM manufacturing inventory change
Journal of Commerce change in commodity prices
Dun and Bradstreet new business formation and large business failure
Real estate loans, deflated, growth rate

Many of these are the same as on the Long and Short Leading Indicator list, but are calculated weekly for purpose of the weekly index. On average these turn 12 months before the onset of a recession and turn positive three months before the end of a recession

Prof. Moore also set forth a series of signals that could be used to predict recessions and recoveries (p. 76):

"The first signal of recession occurs when the six-month smoothed growth rate in the leading index first declines below 2.3%. The second signal requires the leading index growth rate to fall below -1.0% and the coincident index to fall below 2.3%. The third signal is set off when the leading rate is below zero and the coincident below -1.0%. ... [T]he first signal of recovery ... is set off when the six month smoothed rate of change in the leading composite first goes above +1.0%."

In case it isn't clear from the above quote, in a recent interview Lakshman Achuthan explicitly stated that it is a decline in the coincident indicators -- nonfarm payrolls, industrial production, real income, and real retail sales -- rather than a decline in GDP, that defines a recession for them.

At its website, ECRI maintains that it "refined" the above indicators throughout the 1990's both before and after its founding in 1995. My examination of the Long Leading Indicators in the above list, for example, strongly suggests that at least one of those has subsequently been replaced.

I will examine the forecasting record, up to the present, made by each of the 3 above indexes, in future posts.