All hail Mark Sadowski
Here’s Mark from yesterday’s comment section:
I’ve developed my own relatively simple nowcast model based on the subcomponents of GDP using currently available monthly FRED data. (For example the last available new data was the Value of Manufacturers’ Shipments for Capital Goods: Nondefense Capital Goods Excluding Aircraft Industries for June on 7/25, which I used to project private fixed nonresidential investment.) My estimates for 2013Q2 have been consistently more optimistic than those of the major modelers. My current estimate is actually a frightenly high 2.0% RGDP growth for 2013Q2.
I’m sure that the major modelers have access to better data than me, and their models are far more sophisticated, but it will be very interesting to see the final results and where who has been wrong and why.
The reported figure was 1.7%, which suggests that Mark was closer than the experts, who predicted 1%. Not to rain on his parade, but these early numbers are often revised. And the revisions done by the BEA have in no way resolved the puzzle that I discussed last month. In the 6 months between 2012:3 and 2013:1, NGDP rose at a rate of 2.2% and NGDI rose at a rate of 5.1%. The first number is borderline recession and the second is boom. So which is right? Probably neither, but the 2.2% seems especially wrong. RGDP growth averaged only 0.6% over those 2 quarters. Yet industrial production rose at a 3.4% annual rate, housing starts were up at a 50% annual rate, there was relatively rapid job growth. Sorry, but that’s not an economy skirting recession. I also suspect the 5.1% figure is too high. We could average the two (3.65%), but my hunch is 4% is about right.
The Federal government really needs to fix this problem, it’s inexcusable. And set up an NGDP futures market while you are at it. They should probably base the payoffs on the average of NGDP and NGDI, at least until the problem is sorted out.
I wonder what Mark’s model tells us about GDP growth in the prior two quarters.
PS. 2013:2 NGDI numbers are released next month.
Tags:
31. July 2013 at 06:55
Maybe the NSA should be given the task of estimating GDP? They could have their machines comb through emails and banking records, use Bayesian methods to match their aggregate data to past yearly GDP revisions. We could get a daily update of NGDP to-date. If they are going to collect the data might as well do some good with it!
31. July 2013 at 07:28
Always kind of interesting how little the markets care about GDP. You might think that a number that was 0.7% off from the surveys would cause the markets to jump. But, barely a reaction.
31. July 2013 at 07:35
Yeah, well, I went to college with Mark for a couple years, so…
#smallworld
#reflectedglory
31. July 2013 at 07:48
Just to make sure I understand the culture on this blog: If Peter Schiff says the government numbers are wildly optimistic, he’s an idiot. If a Market Monetarist says they’re wildly pessimistic, it shows what a great economist he is. Right?
31. July 2013 at 07:56
(BTW my snarky comment isn’t really about Sadowski, it’s more about the previous post where Scott is challenging the official GDP numbers as not reflecting the true economic reality.)
31. July 2013 at 08:02
Doug, The markets know enough to pay attention to employment, not RGDP.
Bob, I’ve never claimed any systematic bias in the GDP numbers. That’s Schiff’s claim. It’s all a big conspiracy in his view. The GDP numbers have random errors in my view. No comparison.
In addition, his specific claims don’t pass the laugh test. I’d never criticize someone who said actual inflation was higher than reported. That’s not what I objected to. But if they start suggesting 8% or whatever, all you can do is laugh.
31. July 2013 at 08:17
Prof. Sumner,
Krugman deserves praise for these last two posts:
http://krugman.blogs.nytimes.com/2013/07/31/yellensummers-and-the-twilight-of-the-vsps
http://krugman.blogs.nytimes.com/2013/07/31/for-the-birds
In particular, it’s nice to see him criticize the Rubinites. Sadowski helpfully pointed out that DeLong is under the spell of the Rubinites. Good for Krugman that he is not.
31. July 2013 at 08:35
Ezra Klein just wrote a long post repeating off-the-record arguments Obama Administration members are making for Larry Summers:
http://www.washingtonpost.com/blogs/wonkblog/wp/2013/07/31/the-case-for-larry-summers
You will roll your eyes multiple times.
31. July 2013 at 09:11
Yglesias just wrote the best potential case for Larry Summers that I’ve seen here:
http://www.slate.com/blogs/moneybox/2013/07/31/brilliance_is_overrated.html
31. July 2013 at 09:39
I’m actually still kind of stunned at how well my model performed.
First, a minor correction. When I said I used the Value of Manufacturers’ Shipments for Capital Goods: Nondefense Capital Goods Excluding Aircraft Industries to project private fixed nonresidential investment I should have said I used it to project the equipment portion of private nonresidential investment.
To put this in perspective here were the projections of the major modelers as of yesterday:
1. Credit Suisse: +1.1%
2. Bank of America -Merrill: 0.9%
3. Goldman: +0.8%
4. Macroeconomic Advisers: 0.6%
5. Barclays and Royal Bank Scotland: +0.5%
6. JP Morgan: +0.5%
7. Morgan Stanley: 0.3%
And I have compared my component by component projections and they were all so close that I’m still pinching myself.
My biggest deviation on a rate of change basis was residential investment. I nowcasted +22.2% and it was actually +13.4%. My biggest deviation on a dollar basis was on net exports. My projection was -$434.6 billion converted to 2009 dollars and it was actually -$451.3 in 2009 dollars. So I was off by $16.7 billion or about 0.1% of RGDP.
The eight components I projected were:
1) PCE
2) equipment
3) private nonresidential structures
4) inventories
5) residential investment
6) net exports
7) government consumption
8) government investment
I projected nominal expenditures by fitting the components of GDP to the following dataset:
http://research.stlouisfed.org/fred2/graph/?graph_id=131093&category_id=0
I used June CPI to project June’s PCEPI. And I used the projection of PCEPI to project the GDP deflator.
All components were projected in nominal terms first. To project June’s PCE I used June’s retail sales. I deflated PCE by the projection of PCEPI and the rest of the components by the projected GDP deflator.
Now, the thing about this is I actually hold forecasting and nowcasting in great contempt. In my opinion a chimpanzee with slightly above average math aptitude and a copy of MS Excel can do it.
The amount of time and effort I put into this has to be a tiny fraction of what the major modelers did and yet my model generated an error only a fraction as large.
P.S. It would take a lot of work to figure out what my model would have told us about GDP growth in the prior two quarters because I only developed the model a couple of weeks ago. I would have to find the prerevised data and that would be an enourmous task.
31. July 2013 at 09:52
And a quick observation about the sequestration. The big cuts to federal spending on consumption and investment clearly occured in the first quarter:
http://www.bea.gov/iTable/iTableHtml.cfm?reqid=9&step=3&isuri=1&910=X&911=0&903=94&904=2012&905=2013&906=Q
So all this fuss about lags and delays in the impact of the sequestration is pure unadulturated grade A horse manure. The sequestration is well behind us, not ahead of us and this is something my model was detecting through monthly government wage and salary accruals and public construction expenditures.
31. July 2013 at 10:36
Correction:
The big cuts to federal *consumption* spending occured in the fourth quarter of 2012.
31. July 2013 at 10:37
Interesting stuff on Wren-Lewis’s blog:
http://mainlymacro.blogspot.co.uk/2013/07/japan-and-consumption-tax.html#comment-form
Interesting in part because he uses the monetary-offset point viz. the UK 2009 VAT cut (if it had any effects, it’s only because the BoE let it) and much more interestingly because of the comment from a Japanese macroeconomist that suggests that NGDP targeting is being discussed there.
31. July 2013 at 11:38
Thinking about the model’s performance has forced me to confront a conceptual inconsistancy. Deflating PCE and non-PCE differently is likely to make the model biased. The simplest solution is to deflate everything by the projected GDP implicit price deflator. If I had done that the projected RGDP growth would have been 1.7%. (Wow!)
(Incidentally my projection of the change in PCEPI was +0.0% and the actual change was +0.0%. My projection of the change in the GDP implicit price deflator was +0.5% and it actually was +0.7%.)
Now the point of all this is not just to think out loud how to refine the model (which actually is something of a waste of time), but to venture a guess as to why it may have outperformed the major models. I bet that the major models all try and deflate their components before summing them. I focused on projecting the nominal quantities and deflated them afterwards.
In short, the key to accurate nowcasts may be to focus on projecting NGDP first and then converting it to RGDP later.
So it all comes down to NGDP.
31. July 2013 at 15:17
I believe you get 1.8% RGDP growth if you just do a linear extrapolation staring from Jan 2012 (or 2.1% for linear extrapolation from Jan 2010).
http://research.stlouisfed.org/fredgraph.png?g=l3t
I think the takeaway is that the sequester hasn’t had much impact and monetary policy has been delivering roughly the same expectations since at least 2010.
http://informationtransfereconomics.blogspot.com/2013/07/all-hail-linear-extrapolation.html
31. July 2013 at 16:10
I certainly agree that the federal government should do better.
Regarding Yellen-Summers if this were a boxing match they would have to stop it. It’s getting ugly and while I thought Summers was fine before, it’s working so well than even I am now convinced that Summers is unacceptable.
Yellen is creaming Summers in public battle
http://diaryofarepublicanhater.blogspot.com/2013/07/yellen-trouncing-summers-by-mile-in.html
31. July 2013 at 16:11
Oh and good job Mark! Very impressive.
31. July 2013 at 17:34
“I’ve never claimed any systematic bias in the GDP numbers. That’s Schiff’s claim. It’s all a big conspiracy in his view.”
Public choice theory, hello?
Incentives, psychological/career pressure, changing models to give more favorable output, etc.
Why deny the obvious and lump all the doubts and disagreements with what the government claims (NSA, hello?) into some straw man “conspiracy theory”?
Try to at least make it appear that you’re not a bootlicking apologist for government statistics.
31. July 2013 at 18:39
Travis, I think Yglesias’s post was actually against Summers.
Mark, Great job, Keep me updated on improvements.
W. Peden, Very interesting. I agree with his comment, even the part about the tax increase slowing growth under inflation targeting.
31. July 2013 at 20:39
Given his accuracy on 2q13, Mark Sadowski must have been spot-on in predicting the 1q13 revision.
1. August 2013 at 04:58
Anonymous,
It would take a lot of work to figure out what my model would have told us about GDP growth in the prior two quarters because I only developed the model a couple of weeks ago. I would have to find the prerevised data and that would be an enourmous task.
However since there seems to be some interest I could do a “backcast” based on the revised data.
1. August 2013 at 05:01
Anonymous,
If you’re talking about nowcasting data revisions there is no such thing (i.e. it’s impossible.)
1. August 2013 at 06:37
Come on Scott, you sound like a rookie.
“The reported figure was 1.7%, which suggests that Mark was closer than the experts, who predicted 1%.”
The month before it was 1.8%, which was then revised down (after the media hyped the 1.8%) to 1.1% (or 1% I forget). Let’s wait until the revised number of 1% comes out before we give Mark credit.
1. August 2013 at 08:46
StimuLOL, If I sound like a “rookie” you sound like someone lacking in basic reading comprehension skills. Didn’t I say exactly the same thing as you?