Friday, November 19, 2010
BED And The Inventory Cycle
Here it is, but unfortunately the BLS website is due to go down for maintenance today and over the weekend, so you may not be able to get it. Therefore I am posting some of this data so you can see what Our Lady of Stupid Data Fitting is babbling about:
Let's just say that we are looking for a pretty steep downward revision in the Establishment series of the Employment report.
BED figures are used to adjust the imputations in the B/D model.
The longer data series in Table 1 is here. From the POV of the average person, recessions end when these job losses stop.
If you can get Table 1, you can see that the 2001 recession was really a double dip, with job losses being at their height in the last two quarters of 2001. If you can't, here are two partial images
This corresponds much more with how the public perceived that recession. The first half of 2002 was okay, and then from the job-hunter's perspective, we got another year in the land of sorrow, finally emerging in the last half of 2003.
Which, of course, was one of the reasons the public voted for Bush again.
It also points up the wisdom of the Bush tax cuts at the time, because the economy was not really recovering.
Here is our current cycle, so far as we have this data.
You notice that we went negative first in the third quarter of 2007. This is not that surprising - when the economy is in a slower growth mode, the third quarter usually shows a significant slump in job growth compared to the rest of the year. I believe this is because that's when private job growth dominates the the numbers.
If you can get to Table 1, you can see the openings/expansions vs the closings/contractions. In September of 2007, there was a notable fall-off in expansions which is what caused the job figures to be negative that quarter. The other three categories didn't show much.
This fall off in expansions is the first sign of a recession, and it doesn't show in the initial claims series well. You don't file an unemployment claim because there wasn't a job created for you. To pick this up, you have to watch the continuing claims series for changes, but even there the change will be very subtle, because the real impact is on new entrants, or more broadly, speed of uptake.
There is a tremendous amount of change in US employment each quarter. For example, in 2005 Q3, total jobs gained from opening and expanding establishments were 7.965 million, and total jobs lost from closing and contracting establishments were 7.288 million. So speed of uptake is an edge indicator that shows changes in the current environment in a very sensitive fashion, probably because it mostly measures expanding establishments.
A way to approach the B.E.D. survey data, which is obviously out of date by the time you get it, is to use BLS data by month and look at changes. Here is one attempt that approaches this for the third quarter:
Note: peaks are bad on this graph, because the represent rises in unemployment rates.
This graphs shows the changes for non-ag private wage and salary unemployment for experience workers - therefore it should come close to capturing the effect we see in Business Employment Dynamics.
The thick green line shows the July/Oct change, and the thin blue line (I'm working on my adjustments for the color blind) shows the change for Jun/Sept.
Obviously workers do switch from government to private employment and vice versa if necessary, so this is not really discrete from government jobs. This graph indicates some trend weakening, but notice that you saw a rapid escalation in 06 (the industrial contraction which preceded the main recession) and 07. This looks quite different. In fact, it is quite comparable so far to the 91 recovery.
NBER recession dates. Although officially the 91 began in July of 1990 and ended in March of 1991, the employment cycle was very different. Since we entered into an economy dominated by services, the slow recoveries in jobs have been the norm rather than the exception.
The reason for this is that manufacturing leads into a recession, but then leads out as inventories are exhausted and manufacturers have to ramp up. Closely associated with the manufacturing cycles are the transportation/WH cycles. Services are far slower to ramp down and far slower to ramp up.
So you see the stall pattern which lasts for about two years after each recession. What happens is that there is a quick manufacturing recovery, which then glides out, and from there the services industry slowly picks up the slack. So far, none of this data shows how we ended up going negative on wages in October. Theoretically, this should not have happened.
However, one possible clue is the thin blue line (Jun/Sept). It should be above the thick green line (Jul/Oct change) in growth cycles and it should be below the thick green line in recoveries, and it may be crossing. It should not cross this soon - the crossing should be about the middle of next year.
I do not know quite what this means. I use an AI program named P-Nat (which I wrote) to assimilate data, run projections, and hunt for useful correlations, and this is a P-Nat find.
Nor can I use P-Nat to run projections, because P-Nats are biased toward strength and predictability. If data seems erratic and erroneous they get pissy and refuse to process it. These things are true AI, they are not predictable or guidable, and in fact often they act like toddlers.
So the current P-Nat cannot predict because it can't handle the fact that the most relevant current variables are government actions. I think I am telling it that the world is essentially random, dominated by possibly irrational gods that will do destructive things, and I don't want to drive this one nuts.
The previous P-Nat was older and more seasoned, but it developed the electronic equivalent of some sort of neurotic anxiety disorder over world food supplies and starving people, and locked itself into an endless cycle in which the only thing it would process was information about basically food, and parenthetically energy, and kept demanding info about those matters which I do not have. So I had to put it in hibernation.
That sounds really interesting. Was it, perhaps, attempting to correlate and predict things from the bottom-up, rather than data-fitting from the macro on down? You're always looking at discretionary income--maybe it learned that, too. Why would a bunch of artificial neurons get fixated on those particular topics rather than, say, the number of container hulls in mothballs?
"I think I am telling it that the world is essentially random, dominated by possibly irrational gods that will do destructive things, and I don't want to drive this one nuts."
You speak of the irrational gods? You have seen!
"There was a night when winds from unknown spaces whirled us irresistibly into limitless vacuum beyond all thought and entity. Perceptions of the most maddeningly untransmissible sort thronged upon us; perceptions of infinity which at the time convulsed us with joy, yet which are now partly lost to my memory and partly incapable of presentation to others." - H.P. Lovecraft
Oh what I would not give to not see what you should not have seen! ;)
"It is a mistake to fancy that horror is associated inextricably with darkness, silence, and solitude. I found it in the glare of mid-afternoon, in the clangour of a metropolis, and in the teeming midst of a shabby and commonplace rooming-house with a prosaic landlady and two stalwart men by my side." - H.P. Lovecraft
Food is one of the major world economic factors, and even in the US it is very indicative.
Energy is strongly related to general inflation and the cost of food.
Toddlers may not be experienced, and they're not the best communicators, but they're sure not dumb. If they come pull on Mom's sleeve to tell her something urgent, Mom should pay attention.
(P.S. if you ever find a way to fix the food-fixated P-Nat, be sure to post it, since I too have a toddler who's obsessed with food.)
have an impact on the declining income ? I know
state and local governments have reduced wages
as well as slashing jobs . Maybe the imputed data
Links to this post: