Scott Sumner Slips
Long-time readers know that I am not a fan of Scott Sumner’s signature idea, namely that the Great Recession was caused by Ben Bernanke’s tight monetary policy. However, if you’re really an expert on my writings, you’ll know that I’ve said Sumner would be a very formidable debating opponent–much more than Krugman. Indeed, when I said Sumner was “insane,” I meant it as a (backhanded) compliment: Sumner can back up his “outlandish” (to me) views with all sorts of internally-consistent facts and arguments. It’s like you run into a guy who claims to be Napoleon, and you realize five minutes into the conversation that you can’t prove he’s wrong.
In that context, then, it’s refreshing to see Sumner slip. (This is just a re-hash of Sumner’s same slip back in 2011. I don’t know if he saw my response at the time.) It reassures me that he’s a man, and can be beaten…
Anyway, in a recent post, Scott is kinda-sorta taunting the people who warned of the housing bubble, last time around. Since a popular US housing price index is above its peak (in nominal terms) from the prior boom, Sumner wants to know if they think another crash is coming? (My answer is “yes,” and Scott at least concedes I am being consistent–he probably thinks I’m insane!)
Then to drive home just how goofy these bubble-theorists were, Scott asks them a bunch of provocative questions:
Is it possible that the housing boom was not a bubble? Is it possible that fundamentals (such as building restrictions and lower real interest rates) support much higher real housing prices during the 21st century than during the 20th century? Is it possible that the real problem was nominal, a fall in NGDP engineered by a monetary policy that (during 2008) held the Fed’s target interest rate far above the equilibrium interest rates? Is that why unemployment stayed low as housing construction fell in half between January 2006 and April 2008, and then soared when tight money pushed NGDP down in late 2008? [Bold added.]
Scott thinks he’s got a real zinger here. In fact, superficially it’s so good that Arnold Kling admitted defeat back in 2011 when Sumner made the same point. But as I pointed out at the time, Kling threw in the towel unnecessarily. Scott’s point blows up in his face once we pick better data. As with two of Krugman’s examples (here and here), the attempt to destroy a coordination-of-resources story (and replace it with a shortfall-in-demand story) actually turns against them. And since it was Sumner who picked this example, that gives it extra significance when it actually supports the Austrian (and Klingian) view.
First, let’s make sure we get what Sumner is doing. He’s saying that the casual association of the financial crisis of 2008, and more generally the Great Recession of 2007-2009, with the collapse of the housing boom, doesn’t actually work when you look carefully at the numbers. Specifically, between early 2006 and mid-2008, new home starts fell in half, while the national unemployment rate didn’t move up very much. See for yourself:
So, Sumner is arguing that *clearly* the slowdown in house construction has little to do with the onset of the Great Recession or the financial crisis that struck in September 2008.
Yet as I pointed out back in 2011, “new housing starts” isn’t the right metric. Clearly a much better test of the Kling/Austrian story–about workers needed to move out of construction and into other sectors, and this reallocation (or “recalculation” in Kling’s terminology) takes time–would be to look at *employment* in construction, and relate that the to the national unemployment rate. If you doubt me, here’s what Sumner himself said when he thought he blew up Kling back in 2011: “So housing starts fall by 1.3 million over 27 months, and unemployment hardly changes. Looks like those construction workers found other jobs, which is what is supposed to happen if the Fed keeps NGDP growing at a slow but steady rate.” So clearly, Sumner thought that the collapse in housing starts was a good proxy for construction employment.
But we don’t need to use a proxy for construction employment. We can use total construction employment itself. And when you compare *that* to the national unemployment rate…
…the fit is gorgeous. That’s exactly what Kling (or Murphy) require for their story. Not only does the story work for the crash, but the prior boom works too: The national unemployment rate falls, as more and more workers are sucked by the real estate bubble into construction.
If you’re trying to put your finger on the problem, it’s this: Sumner just assumed that a large drop in new housing starts went hand-in-hand with a large fall in construction employment. But as the data show, that’s not what happened. So no, there weren’t a bunch of “construction workers [who] found other jobs” because the Fed kept NGDP growth up.
If you want to offer various theories about why that should be, go ahead. It’s an interesting puzzle, presumably having to do with rates of growth, the fact that you build shopping centers etc. around new housing developments, that there is a lag for add-on work to newly constructed houses, etc. But, it’s not my job to explain *why* the collapse in new housing starts didn’t translate into a collapse in construction employment. Once we realize that, apparently, the one didn’t cause the other, then Sumner’s whole point falls apart. We are back to the original “common sense” view that it’s not a coincidence that the housing bubble collapsed, and then the financial crisis / Great Recession happened.
If you’re curious, the following sheds some light on it:
Awesome. Thanks. Do you think you could time the next pop using Hal Snaar’s accordeon and some common sense about the cantillon effect, charting cap goods orders and unemployment in these sectors ?
The only reason your second graph looks like a coordination problem is because the fall in construction employment happened at the same time as the fall in non-construction employment.
https://fred.stlouisfed.org/graph/?g=iS1d
Growth in both started falling at the same time in mid 2006 and then they both started dropping faster around the end of 2007/beginning of 2008. It wasn’t a coordination problem. It was that there were no jobs to coordinate into.
A single force that can affect the national economy was affecting them together.
*”Not only does the story work for the crash, but the prior boom works too: The national unemployment rate falls, as more and more workers are sucked by the real estate bubble into construction.”*
It wasn’t a bubble, though. They were building homes that were needed. The funny thing is, if you look at Phoenix – the quintessential bubble city – construction and non-construction employment also both start to drop at the same time in mid 2006. And, in Phoenix, where you might think the drop in construction should have led the collapse, the drop in non-construction employment was actually a little sharper than the drop in construction employment.
https://fred.stlouisfed.org/graph/?g=iS1L
Also, job growth in Phoenix was running at a little over 100,000 per year in 2004 and 2005. About half of that growth was from in-migration, mostly from California, where certainly they didn’t have a bloated construction sector. That’s why so many were moving to Phoenix, after all. At that time, at the peak of homebuilding, Phoenix was permitting about 60,000 units per year. There weren’t even too many homes in Phoenix.
Yet, something caused all of that job growth and migration to suddenly break down in mid 2006. Notice in the Fred graph that the Phoenix unemployment rate didn’t rise until the end of 2007 even though total employment growth had fallen by half. That’s because the first thing that happened in 2006 was that all those families stopped moving away from LA all of a sudden. At the same time that employment growth began to trend down. All these things happened in unison.
Eventually, when you get to “Galactic Brain” level on this, you will see that the recession actually caused the housing bust. Not the other way around.
Here are delinquencies, unemployment, and GDP growth (inverted right axis).
https://fred.stlouisfed.org/graph/?g=iS2q
It all happened together. And the foreclosures that really caused the actual cash flows to mortgage securities to be impaired were from all those delinquencies in 2009 and 2010. If not for the recession, 80% of the foreclosures wouldn’t have happened.
Thanks Kevin. Believe it or not, I was thinking, “Ah, Sumner and his fans could come back with…” and so I was going to look at the kind of stuff you were pointing out.
You are going to be shocked to hear this, but I actually think your graphs support *my* story better than Sumner’s! I am sure that you (and Sumner if he saw your graphs) would be thinking, “Jeez Bob, don’t you know when you’re beat?”
I’ll try to post on this tomorrow.
Kevin wrote: “Eventually, when you get to “Galactic Brain” level on this, you will see that the recession actually caused the housing bust. Not the other way around.”
Does it involve tachyons?
Looking forward to your post.
And, now that you mention it, yes it does!
@ Kevin Erdman,
If these houses were needed then why was the vacancy rate climbing during the same period? The rental vacancy rate rose from 8% in 2000 to 10.2% in 2004, and the homeowner vacancy rate rose from 1.5% to 1.8% in 2005. A 27.5% increase in vacant rentals and a 20% increase in homeowner vacancy is not what you would expect during a housing shortage.
There is a lot of regional variance. I don’t think you’ll find rental vacancies that high many places where quantities or prices were rising. The housing bust didn’t happen because there were too many houses in Detroit.
The rust belt population decline was under way well before 2000. Metropolitan Detroit lost more residents from 1970 to 1990 in absolute numbers and even more in terms of % of national population than it did from 1990-2010, or 2000-present day or almost any equal length starting and end point. This is true for almost every rust belt city. By the time of the rising vacancy rates in the early 2000s the total population losses of these areas can’t account for even half of the increase. Probably less than a quarter at best (and that is if none of the buildings were torn down during that span). The numbers of people moving out during that time frame were high in percent but low in impact nationally because the sum total of the major cities that it was occurring in was only a couple % of the total US population by then.
“Also, job growth in Phoenix was running at a little over 100,000 per year in 2004 and 2005. About half of that growth was from in-migration, mostly from California, where certainly they didn’t have a bloated construction sector. That’s why so many were moving to Phoenix, after all. At that time, at the peak of homebuilding, Phoenix was permitting about 60,000 units per year. There weren’t even too many homes in Phoenix.”
Phoenix might have been a casualty of the bust, but you can’t extrapolate from one city to the entire country. Using full country FRED
Opps, continued.
Using full country FRED data US population rose ~19 million from 2000 to 2007 and total housing units rose almost 14 million in that same span, and the median sq foot of new housing starts rose from 2000 to 2005 by about 13%.
The US was adding more housing units per new person than it had previously done and those new units were larger than the historical average and growing.
When did I lose you, baconbacon? This is all stuff you should know from reading my stuff.
In a regime defined by regional housing shortages, growth periods will be characterized by new homes that are larger and less expensive than existing homes because new homes can’t be built where homes are small and expensive.
BTW, housing starts compared to population growth has been quite moderate by historical standards for at least 20 years. It had been very low in the mid-1990s, and was probably due for some catch up, but the rate of new units didn’t ever really have a sustained period of catch up.
North Korea just needs to give Kevin internet access, and he can plan their economy!
We went through this a few years ago. Housing starts don’t equal net housing gains and straight population changes are one of several demographic shifts that occurred during this time period. Doing this from memory but
1. Building in places like Phoenix tends towards a much higher net per housing start than in already established areas as you are rarely tearing down to build the new units.
2. Population growth from ~1950 through 1990 was also accompanied by smaller household size, dropping from >4 (iirc) to under 2.5. Since then (unless it has changed very recently) this trend has flattened and slightly reversed.
3. Population grown isn’t equal. An adult immigrant needs an adult’s amount of housing when he is added to the population, a child born does not need an additional unit built until they move out into the world as an adult (not strictly true but a much better approximation). The 2000-2005 boom was coming 18-25 years after some of the lowest fertility rates in US history, when it would be reasonable to expect lower than typical demand.
4. US vacancy rates rose significantly during the period of heavy building and rising prices, which fits the bubble theory far better than a shortage theory.
PS any grumpyness/aggressiveness in my tone that is above my typical in the near future please chalk up to the recent root canal + stomach virus I have had. I have been told I get excessively combative during times of stress.
Become the master of your domain, baconbacon!
It’s not just that the decline in construction employment coincides with the spike in unemployment. It’s that in percentage terms relative to the (employment) size of the sector, the decline in construction employment is disproportionately large.
Indeed you could look at all post recessions (I would personally exclude 1981 as double counting the 1980 recession and 1945 as a spurious recession) and percentage departures of employment from trend by sector. Durable manufacturing is pretty consistently hit the hardest and services the least hard. In general, the further employment is from final consumption, the bigger the cycical swings in employment. Seems consistent with an Austrian type story where roundaboutness and capital are important to the business cycle.
How does Monetarism explain a chart like this? I think that it can’t.
Good points Andrew. That was one of my Krugman pushbacks (which I said “here” in the post above) too.
@Andrew,
Couldn’t that chart be explained by the fact that some industries are just more volatile than others and more sensitive to changes in AD ? I don’t think that would mean they were suffering from “unsustainable booms” every time a downturn comes along.
Musical Chairs Model doesn’t explain *why* certain sectors are more sensitive to fluctuations in AD.
But this is potentially a fruitful direction for thought: Why does your comment implicitly assume that sectors assymetrically respond to *downward* fluctuations in AD, but not upwards fluctuations? The assymetric response to upwards fluctuations in AD, that’s the unsustainable boom.
“If you want to offer various theories about why that should be, go ahead. It’s an interesting puzzle, presumably having to do with rates of growth, the fact that you build shopping centers etc. around new housing developments, that there is a lag for add-on work to newly constructed houses, etc”
I thought for a moment you were kidding with this as there is a specific Austrian explanation for this.
There’s an Austrian explanation for why construction employment stays constant even when new home starts falls off a cliff?
ABCT posits that credit expansion will pull capital into projects where the expected return exceeds the actual return. The bust starts when it is realized that the actual returns will be lower and capital stops flowing into these sectors. However the capital that has been invested is no longer (as) fungible leaving us with a mis-allocation of capital. The majority of capital is expected to remain in these sectors and people adjust their consumption and investment patterns to reflect this lower level of return on their investments.
This fits your graphs very well, first you have a large run up in employment in construction in concert with where you would expect an increase in capital, you have a flat line of hiring when capital stops flowing in and a decline in employment as projects get finished and no new ones are started.
baconbacon wrote: “you have a flat line of hiring when capital stops flowing in and a decline in employment as projects get finished and no new ones are started.”
Oh sure, we’re on the same page. What I meant in my line (that you thought was a joke) was that I don’t know exactly what the projects are that need to be finished. (It doesn’t take 2.5 years to build a house.)
re: 2.5 years to build a house. No, however looking at the FRED graphs new housing starts in 2006 were at 2004 levels through early 2006 and in June/July they were still starting 1.7-1.8 million units (annualized I assume) which is 2001-2002 levels. Further new housing units are only a portion of the construction market.
The big question, which I don’t see an obvious answer from an Austrian or Monetarist perspective is how the dollar value all commercial and industrial loans held by US banks could have almost doubled from 2004 through late 2008.
Hey baconbacon,
Sure, I’m not disagreeing with anything you’re saying. And look at what you’re saying now: You’re talking about the trends, rates of growth, etc. That’s what I meant in my OP, that I could get into the specifics and try to explain why Sumner’s proxy wasn’t a good proxy after all, but that it wasn’t really my job to explain it.
Residential Building represents less than 15% of all construction employment-even back at the peak.
I have seen housing construction projects take >2.5 years but generally because something goes wrong like the builder going bankrupt, and they stopped building for most of that time.
There’s a common sense explanation which involves the word “start”.
You see if you are just starting a new building construction project, that’s probably a bad time to lay off workers. Consider maybe laying them off approx 1½ years after the project is started (hint: that’s known as the “FINISH” of the project, also fits with chart #3 above).