What does recalculation look like? How about a big NGDP shock?
Here is a very interesting Slate.com graph showing the progression of the recession by color-coded jobs losses and gains. Red shows jobs lost, and the bigger the circle the more jobs lost. Blue is job gains. They are 12-month changes, so if jobs rose strongly for 8 months, then fell at the same rate for 4 months, the 12 month change would still be positive, despite the negative trend over the most recent 4 months.
I am going to ask you to either print out my post, or take a few notes before looking at the graph, as I’d like to point out a few interesting patterns. Regional economics is not my specialty, so I will undoubtedly get a few details wrong.
I’d like you to look at the following three months; August 2007, 2008, and 2009. Get used to toggling back and forth. (Hit “start” to get rid of the box.) Here’s what those dates indicate to me:
August 2007: The first signs of the sub-prime bubble bursting appear. Normally growing regions like California’s OC/Inland Empire area show losses. So does Florida’s Gulf Coast. Those are the first signs of recession (which hasn’t yet arrived.) What about the group of cities near Lake Erie, the Detroit/Cleveland/Buffalo area? Well who wants to live there! It’s cold, cloudy, and taxes are high. Of course they are losing jobs. They’ve been losing jobs for decades. And then there are spotty job losses in rural counties where factories have closed down.
Now toggle up to August 2008. Ouch! Compared to 2007 it looks like red ink all over the place. But I think that is misleading. Rather, I would argue that the changes over those 12 months mostly show the effects of “recalculation,” or shifts away from two key industries, along with a very mild recession due to NGDP growing at less than 3%.
The job losses have increased dramatically in California and Florida, but Vegas and Phoenix are still hanging in there (those results may be biased by the 12 month change discussed earlier.) But now there is a second structural problem. Oil had peaked the month before at $147, and the 12 months of relentlessly higher gasoline prices devastated the auto belt from Michigan down to northern Alabama. I’m sure some other energy intensive midwestern industries were also affected. And perhaps furniture making in North Carolina was hurt by housing. New England and upstate NY are a bit misleading, as with near-zero population growth they are always on a knife edge between gaining and losing jobs.
Am I sugar-coating a bad situation? Am I trying to deny that these structural problems could cause a recession? My response is that if you want to see a real demand-side recession, if you want to see what happens if NGDP falls significantly, toggle back and forth between August 2008 and August 2009. Look at what 12 months of falling NGDP does. And then tell me that our problems are all caused by a few nutty housing markets in SoCal/Vegas/Phoenix and Florida. Keep toggling until the 2009 mass of red is permanently impressed in your retina, and connected to the part of your brain that stores the phrase “falling NGDP.”
PS. I spent Christmas in the county with America’s biggest jobs gain between August 2008 and August 2009. This means that it’s even misleading to say “Arizona” was a sub-prime fiasco, as Tucson is still doing OK.
HT: Mark Lewis
Tags:
13. January 2010 at 16:06
August 08 to Jan 09 is sufficient.
13. January 2010 at 18:49
Statsguy, Yes, but August 2009 was even worse.
But that’s a good point, as it shows the close correlation between the onset of NGDP declines and the collapse of the job market.
And no one can claim it was just the delayed reaction to the first 8 months of recession, as no economist’s model showed anything like that job loss coming.
13. January 2010 at 19:05
BTW, you will like this:
“A few Fed officials have suggested continuing or expanding mortgage-backed securities purchases past their scheduled completion at the end of March to nurture a weak recovery and prevent potential disruptions to housing markets.
Dudley, however, said it would “seem prudent” for the Fed to stick to its plan to end the Fed’s mortgage-backed securities purchase program in March as the economy is starting to improve.
Dudley said the impact on mortgage rates from the ending of the purchase program would likely only lead to “a little bit of upward pressure” on rates. But if mortgage rates were to rise a lot, leading to a significant deterioration in the economy, the Fed could always decide to step back in and buy more, he said.”
http://www.reuters.com/article/idUSTRE60D03G20100114
Reread that last sentence. Not “if markets EXPECTED mortgage rates to rise a lot”, but rather, “after we have definitive proof that the ship has in fact struck the iceberg, we will attempt to change course”.
Also, separately, regarding Bullard –
“Manufacturing activity was weak in the other Districts. Richmond reported widespread weakness across shipments, new orders, and employment within its manufacturing sector and Atlanta saw orders and production drop back after an increase in November. The St. Louis District reported a continued decline in activity, persistent weakness in employment, and plant closings, on net.”
From the Fed’s Beige Book. It strikes me that St Louis region took a hit after the rest of the country (judging by that interactive graph), but now that the economy has struck home Bullard’s had a little dose of reality.
You really should give a lot more credence to “long and variable lags”. The problem isn’t the economy lagging policy, it’s policy lagging the economy. 🙁
14. January 2010 at 09:36
This is somewhat related at least to the post’s topic, but David Beckworth posted an interesting paper on the US as an optimal currency area. His study indicated that it is not, especially the Rust Belt and Energy area diverge in their reaction to monetary policy and ends up supporting some regional autonomy in monetary policy.
http://macromarketmusings.blogspot.com/2010/01/optimal-eurozone-and-optimal-dollar.html
Thoughts?
14. January 2010 at 11:18
Hmmm. About that map…the grey counties don’t seem to have any data at all. Why is that?
Also, some counties will be gray one month and not gray another. This is even more confusing. If a county has data for February, then it should also have data for March. Why is there no data for some months and then there is data for other months?
And most importantly…where are they coming up with the county data? The county 12 month net change data (for those few counties that I choose to check) do not match the official BLS data. The only number that “matches” (although this can be off slightly) is the nationwide number on the top of the map.
Finally, I will note that the 2009 data is in production while the 2008 data has been benchmarked and therefore they are not really comparable. But hey, this map seems to have more problems then just comparing the benchmarked to production data!!! Other than that it’s a great map.
P.S. Scott, those questions are not really for you since you had nothing to do with the creation of the map.
15. January 2010 at 09:26
statsguy, Great link. I’ll try to do a post when I have time.
OGT, I think the efficiency cost outweights any macro advantages. But in this crisis a beggar thy neighbor race would have been desirable.
Tom, I think the grey areas are counties where the change that month was too small to trigger a color (say less tha 100 jobs.) The changes are absolute numbers, not percentages. So the rural counties often have little change. Not sure, but I think that is the reason.
15. January 2010 at 13:00
Scott,
Looking at Sussex Co. Delaware LAUS Employment 12 month change for 2007 January to December we have:
BLS: -294, -913, -385, -561, -360, -155, -196, -660, 194, -662, -569, -1108
Slate: X, -645, X, -369, X, X, X, -379, 466, X, X, X.
The less than one hundred theory is not right. Numbers that are greater than one hundred are not in the graph. Where there are actual numbers, they do not match. There really should be no gray areas, because each county has a LAUS estimate. There really is no salvation for this graph.
15. January 2010 at 14:39
Scott,
A few other counties I’ve check out:
El Paso County, Texas.
Jan. 2008, BLS 7660, Slate 3563
Feb 2008, BLS 7914, Slate 3487
Mar 2008, BLS 7986, Slate 5277
Palm Beach County, FL
Mar 2008, BLS -8294, Slate -7138
Apr 2008, BLS -3453, Slate -1734
May 2008, BLS -7537, Slate -1675
Clark County, NV
May 2008, BLS 26817, Slate 27973
June 2008, BLS 24933, Slate 25410
July 2008, BLS 22023, Slate 27686
Aug 2008, BLS 22457, Slate 28642
Sep 2008, BLS 19960, Slate 22800
Los Angeles County, CA
Oct 2008, BLS -110887, Slate -110887 BINGO!
Nov 2008, BLS -161567, Slate -161567 BINGO!
Dec 2008, BLS -187361, Slate -187361 BINGO! Three right in a row. Lord have mercy.
Denver County/city, CO
May 2009, BLS -10504, Slate -6533
June 2009, BLS -12195, Slate -7235
July 2009, BLS -12812. Slate -8742
Sacramento County, CA
July 2009, BLS -33142, Slate -42891
Aug 2009, BLS -32986, Slate -43029
Sept 2009, BLS -34689, Slate -45776
OK, I think I have taken enough of a sample to show that the Slate map is coo coo for cocoa puffs. The only correct data found in my sampling of the map was in LA County.
17. January 2010 at 11:22
Tom, Thanks for all that work. I won’t try to defend them, as I haven’t studied the issue. You should send them the data and ask for a clarification.
Did Slate say they used absolute changes in employment from BLS numbers?
24. January 2010 at 15:26
Scott,
Slate reports that they are using BLS’s LAUS data and “the numbers you see show the change in the number of people employed compared with the same month in the previous year.” So it is the absolute change in employment from the BLS’s LAUS data. It is interesting that they actually got LA County, CA right! But that also confirms they are using LAUS over the year absolute change in employment data from BLS.
Why could there be so much bad data? LAUS has a preliminary estimate for the current month and a revised estimate for the previous month. Perhaps they are neglecting to revise the previous months data. In addition, the 2008 and 2007 data have been further revised through the benchmarking process. I don’t know how long that map has been in existence but if it is an old map they would need to update all the old data every year to reflect the benchmarking process. Finally, there should be no gray data for any county for any month. The existence of gray counties on the map is completely unexplainable.
I see lots of misinformation in the press. It will be interesting experiment to see how long the map will survive with bad data!
25. January 2010 at 07:41
Tom, I agree there are problems, but a couple comments:
1. You are right about the grey areas. I assume they must be requiring a minimum change to show up red or blue. The grey areas are rural countries with small populations. No other answer makes sense, regardless of what they claim.
2. I still think the overall pattern is roughly correct even if they are using flawed preliminary data. Even with more precise data, the massive increase in red areas between mid 2008 and mid 2009 would probably look similar. If not, I’ll make a retraction.
Thanks for investigating that. Normally I would have, but the blog is sort of overwhelming me and I just don’t have time to be as careful as I’d like to be.
25. January 2010 at 10:50
Scott,
Curiosity has gotten the better of me and I have informed the author of the Slate map with the problems I have found and have asked for his comment on the questions I have raised on your blog.
26. January 2010 at 07:51
Tom, Good, let me know what you find.
27. January 2010 at 13:52
[…] Tyler Cowen links to a graph showing the effect of “recalculation.” Oddly, a fews weeks back I linked to a similar graph arguing that it showed recalculation was a minor factor in the current […]
3. February 2010 at 17:07
Scott,
I emailed the author of the Slate map, Chris Wilson, regarding the problems with the employment numbers I found on his map. He had a short response say he would look in to the discrepancy and then asked me if I was getting my numbers from BLS. That was the extent of his reply. I replied to him that I was indeed using BLS data and outlined possible reasons why his calculations were incorrect. That was over a week ago and I have not heard back nor has his map been updated with corrected data. Many questions were asked but little was given in the way of a reply. Perhaps a more substantial reply is forthcoming.
P.S. I am glad to hear that you are not going to stop posting but just cut down a bit. Whew, that’s good news!
4. February 2010 at 06:27
Tom, Thanks. My fear is that he won’t reply. I hope I am wrong.