I've been writing my forthcoming book — trying to dot all the i's and cross all the t's with the data analysis — and I came across something that was a bit shocking. There seemed to be some trouble in the Case-Shiller housing price index model which had some ambiguity with the shocks. While there was a decent fit to the CS index level, the growth rate was not well-described. This is not a good qualitative fit to this growth rate data:
But if you squint at that data, you see a pattern that looks a bit more like the pattern in wage growth — see the dynamic information equilibrium model (DIEM) described here. So I went to Shiller's (continuously updated) data directly to get a longer term time series, and sure enough a DIEM describes CS index (i.e. housing price) growth pretty well:
The dashed line shows a fit to the small fluctuations in the aftermath of the Great Recession (possible step response to the recession shock) that aren't that big in the growth rate, but create a noticeable effect in the level.
Here, you can see the large effect on the level of those two post-recession fluctuations in the growth rate (difference between the dashed and solid green curves). Overall, this is a great description of both the level and the growth rate of housing prices. But it raises a big question:
In the aftermath of the S&L crisis of the late 80s, the growth rate for housing prices in equilibrium increases steadily — much like wages. But there's no positive shock sending prices higher, no event between 1990 and 2006, just steady acceleration. Sure, the growth rate becomes insane at the end of that trajectory (> 10%), but that's the result of the equilibrium process. There's no "bubble" unless there's always a bubble.
Did new ways of financing allow this acceleration to increase for a longer time? There's some evidence that wage growth is cut off by recessions (i.e. when wage growth reaches NGDP growth, it triggers a recession). What is the corresponding housing price growth limit? The pattern of debt growth roughly matches the housing price growth starting in the late 80s, but the shock to debt growth in the Great Recession comes well after the shock to housing prices.
However, this improved interpretation of the housing price data puts the negative shock right when the immigration freak-out was happening and the shock to construction hires (not construction openings, but hires). This would raise a question as to why a shock to the supply of new housing would cause prices to fall. I've previously discussed how increasing the supply of housing should actually increase prices because it should be considered more as general equilibrium (supply doesn't increase much faster than demand). But also the dynamic information equilibrium model that produces this kind of accelerating growth rate is (d/dt) log CS ⇄ CS (just like the wage growth model). This has the interpretation that the demand for housing is related to the growth rate of housing prices (since the left side is demand).
It's important to note that the shock to housing price growth rate comes before any decline in housing prices. The decline in growth rate would only have become noticeable in early 2006, but whatever the shock was that triggered it could have come as early as mid-2005 — which points to another possibility: hurricane Katrina. Or maybe simply housing prices themselves were the cause. Eventually, regardless of creative financing options, people become unable to afford prices that grow faster than income. It's not so much of a bubble, but supply and demand for a scarce resource. Maybe demand for housing really was that great. Judging by the increase in homelessness in the aftermath (e.g. Seattle), that is a plausible explanation. If it was truly a bubble, then there'd be a glut of housing, but prices seem to have continued on their previous path .
Update 5 December 2018
See Kenneth Duda's comment below and my response for a bit more on possible causes, including the OTS decision to loosen requirements set by the CRA. I would like to point out that the difference between the dashed green curve and the solid one in the CS-level diagram above could be interpreted as a result of mis-management by the Fed and the government in 2008. However, it could also be interpreted as a measure of the general level of panic and overreaction (the downturn appears to overshoot the price level path resulting in a deeper dive than indicated by later prices).
Update 9 December 2018
Per Todd Zorick's comment below and expanding on footnote , here are the housing price indices for Seattle:
And for Dallas:
As you can tell, the shocks are quite different. Dallas doesn't have much of a housing collapse, but seems to be undergoing a fairly large shock over the past couple years. Do note that the time periods are different (since the 90s for Seattle, since 2000 for Dallas) because that was the data on FRED.
 I will note that recent data from Seattle shows a decline (a similar turnaround might be visible in the Shiller data as a downward deviation at the end of the graph above):
You can also see the tail end of the drop in 1990 associated with the birth of grunge.
It is possible this is part of the early signs of the hypothetical upcoming recession.
Jason, I am convinced by Kevin Erdmann's arguments that there was never a housing bubble. He is nearly complete with a book on this topic. Please see:ReplyDelete
One thing I interpret his work is that the drop in house prices was due to poor policy decisions at the Fed and at Fannie/Freddie in 2008-2009, which resulted in (a) lots of unemployment/foreclosure, and (b) large-scale denial of financing to potential buyers. These factors together led to a temporary drop in housing prices, enabling every pundit on the plant to incorrectly characterize housing prices before 2008 as a "bubble".
Thanks Kenneth! Yes, I'm aware of Kevin's view (forgot to include it here, but referenced it in a link in this earlier post:Delete
The issue I see with any particular policy being a causal factor is that this crash in housing prices was set in motion in late 2005 or early 2006 -- well before anything that happens in 2008 and at a time when unemployment was under 5% and heading downward.
I'm actually still open to other explanations for the cause, though. There is a case to be made that your b) is a great explanation: in Feb of 2005, the OTS said that large banks no longer had follow some CRA rules which cut the availability of subprime mortgages. This has the right timing (before the end of 2005).
There are some other events in that time period as well:
It could even be the financial pressure from the shorts -- a self-fulfilling prophecy -- where if a lot of people believe it's a bubble then it becomes a bubble.
Jason, here are a couple of posts on the timeline of events in 2005-2007 that cover some issues that I think have been underappreciated. In sum, the bubble in places like Arizona was a bubble in an inferior good (substituting housing in AZ for housing in CA). It was already a bad sign. And, so the collapse in housing starts in early 2006 was the collapse in a market for inferior goods. A terrible sign. One reason unemployment didn't spike until late 2007 is because the economy in 2005 produced a mass migration away from cities with high incomes, and the first result of the crash in the bubble in inferior housing was a contraction in that migration. Look at a place like Phoenix and compare its employment growth patterns and its unemployment rate in 2005-2007.Delete
I think you are correct about the self fulfilling prophecy. While loose lending terms were still allowing mortgage lending to grow in 2006-2007, there was a significant outflow from home equity. Some of that was due to home equity borrowing, but much of it was due to wealthy owners selling out of housing altogether or moving to less expensive cities, and that former home equity was re-invested in other asset classes (including money markets that were buying CDOs and MBSs). The rise of the CDO market in 2006 was really an early sign of financial disequilibrium. It was treated as one more sign of excess, and that sort of massive conceptual error in policymaking was the primary cause of the crisis.
Thank you for these links and great visualization!Delete
The one piece of data I'm trying to reconcile with all of these metrics is that housing prices start to reverse their growth rate almost coincident with a drop in JOLTS Construction Hires (JTS2300HIR) in 2006, but before the drop in construction openings in 2007. This results in an increase in openings per hire (shown here) indicating a tight construction labor market (difficult matching).
That's also coincident with the beginning of the drop in housing starts. It seems that if we could have flooded the construction labor market, it might have prevented the chain of events leading to the fall.
The Fed seems almost completely reactive in this scenario, and it's hard for me to believe that a 3% and rising fed funds rate at the end of 2005 would have such an enormous effect on housing when interest rates had not been that low except in the immediate aftermath of a recession since the 1960s. Pulling up a handy mortgage calculator, the difference in monthly payments between a 3% rate and 4% rate for a 500k house is only about $200 per month. In an otherwise booming economy, I can't imagine such a change having that large of an effect when, say, the comparable rise in the mid-90s from 3% to 6% had almost no effect. Sometimes tighter money causes housing markets to crash, sometimes it doesn't?
The following is an interest report, but especially if one is open to the story that the seeds of the crisis were planted in the 2003-06 period, with too many new nonprime borrowers dependent on an inherent contradiction: to wit, the only way the large pool of new nonprime borrowers were going to survive in the event rates started to rise from when they entered their loans (e.g., in the reported 75% of nonprime loans that were "short-term hybrids," which would mean repricing to new, higher rates after 2-3 years, and which were entered into at ~90-100% loan-to-value, leaving virtually no equity cushion should prices fall) would be if housing prices continued to rise and a distressed owner could sell or access new equity. And yet higher mortgage rates and increased selling pressure in the market are precisely the conditions that would be likely to inhibit or even stall growth in housing prices.Delete
I understand that this is probably the conventional narrative, but it does appear to describe powerful economic forces. This is also the way that the Fed might have been culpable, even in a delayed way, in that any rise in interest rates was bound to hit a large group of vulnerable borrowers disproportionately hard. Next, add to that the fact that the financial system had, along the way, incurred leveraged exposure to these borrowers. As for the broader recession, I'm satisfied that a sufficiently disrupted financial system and set of credit markets would be enough to cause investment and consumption to fall across the board.
Whether we might have been able to build and grow out of the problem with a large enough supply of labor and credit is an interesting counterfactual, but I still think probably contingent on the Fed keeping rates low throughout the period, as higher rates appear to have coincided with household mortgage debt and mortgage debt service payments rising to historically high levels just when the weakest nonprime borrowers would have needed ample willingness and ability to lend, borrow, and buy in the housing market to save them from their inability to service their debt.
Worth adding that the "conventional narrative" presented above might be perfectly consistent with the interesting insights that have been shared into how it's possible there never really was a "housing bubble" during the same period.Delete
Finally, these are some great videos from 2005 and 2006 for anyone with the time and inclination and assuming one is not already familiar with them. Certainly not meant to be exhaustive, but thought-provoking slides and quotes throughout for someone still considering new directions for research (e.g., comments on "impact of immigration on housing" circa Dec. 2005; observations that housing busts had historically occurred again broader economic distress, and that, as of Dec. 2006, the weakest markets were in IN, OH, and MI, which may or may not have some relevance to broader trends in offshoring and/or the auto industry that were hitting around the same time). I apologize if this is "more heat than light," but there really is so much that can be learned about the topic and the time period.Delete
https://www.c-span.org/video/?190445-1/housing-market-forecast (Dec. 2005)
https://www.c-span.org/video/?195768-1/national-housing-outlook (Dec. 2006)
https://www.c-span.org/video/?195768-3/home-mortgage-business (Dec. 2006)
Could it be that openings increased because employers had less urgency about filling outstanding openings? I have heard speculation about that before. Is that a thing that ever happens in JOLTS data, or just idle speculation from people looking for explanations?
The other comment I would make about that is that since a key factor in the turning market was the whipsaw shift up then down in migration away from the Closed Access cities, the first signs of contraction are odd. Unemployment in Phoenix in late 2007 was under 4%, even though over the previous year or two employment growth had cratered there. But, the contraction was the result of the migration collapse, so it was a simultaneous collapse of supply and demand growth both.
The shifts in the data you mention seem to correlate more or less with the shifts in the Phoenix economy. First, the migration inflows began to drop in 2006. Then, by late 2007, those shifts were deep enough that it started to show up in broader signs of dislocation - rising unemployment, tenant vacancies, etc.
Unfortunately the JOLTS data doesn't really tell us why something is happening. All we really know is that construction hires per vacancy fell. That could be due to less urgency. That could be due to more scrutiny of job candidates. That could be due to fewer people applying -- either because the supply is gone or because they could find jobs elsewhere.Delete
There are really only 1 and 1/2 business cycles in the detailed jolts data (there is some additional data from the Fed, but it's aggregate, and Barnichon 2009 has different methodology). However for all of those other sectors, hires and openings (vacancies) follow each other fairly closely. This long period between construction hires falling and openings falling is different.
It could even be a measure of the irrational exuberance in the housing market -- construction firms thinking the good times were going to keep going. Nobody paying attention to their HR departments saying they can't find people!
As a side note (despite the shock coming in 2008, much later than these things happening in 2006) the unemployment rate doesn't show a whole lot in terms of regional timing differences in the recession (the plains were hit last, Florida was hit first). Oddly, Minnesota is also among the first to feel the recession.Delete
Is there some intuition, either endogenous or exogenous to your model, for why we should see this particular relationship in housing prices and/or wages in a smoothly functioning economy (i.e., acceleration or growth in the growth rate) and not a different relationship or none at all?*ReplyDelete
These basic, perhaps even fundamental, economic relationships that your research could be uncovering, and the relatively simple models that correspond to these relationships, seem to be an important part of your work. In some of your earlier posts, perhaps because they included some of your original derivations, I want to say that I felt like I had a better intuition for why certain relationships might be the way they appeared to be (i.e., given a specific information equilibrium). That might have been a false impression, though, caused by having something like a series of equations to follow. Have I lost sight of a theoretical insight along the way, or has your work been primarily empirical all along, such that I was mistaken to read too much into any given DIEM?
* I apologize if you've answered this question in one of your posts already and I have neglected to read carefully enough or simply forgotten.
Usually, there is a sense of matching states having equivalent amounts of information on either side (like hires and vacancies) in any DIEM.Delete
Bringing that logic to this model says that there's equivalent information in price states and price growth states (or in the identical wage model, wage states and wage growth states). I have yet to understand the 'story' behind this relationship other than demand for X is stronger when the growth rate of the price of X is higher (i.e. housing with faster growing prices increases the demand for it). An asset with a rapidly growing value is of course going to draw additional demand.
My best guess is that these dynamics arise in scenarios where scarcity of the item (workers in tight labor market, houses in tight housing market) doesn't just increase the price, but results in a kind of "bidding war" (multiple job offers, or literal bidding in the case of housing).
Thank you. That feels like a satisfying answer to me.Delete
Also, I appreciate the relevance to your comment to Todd below, in that these relationships might be dependent on the "regime," so to speak. The more housing or labor acts like a scarce resource, the more we might expect this particular effect of price states reflexively informing price growth states, but in cases where, for example, housing is not artificially constrained (I'm thinking along the lines of zoning restrictions) and where land is not capturing a significant share of new economic value (I'm thinking along the lines of Henry George's theories), then you'd see the effect less and maybe even have an entirely different relationship/model. I hope that's a reasonable interpretation, anyway.Delete
Interesting- the implication is that we are consigned to repeated unsustainable housing price surges followed by crashes. This is entirely consistent with our experience in California real estate going back to the 1960s. I recall seeing a time series of housing prices in Amsterdam that goes back to the 1600s, it would be interesting to get those data and see how they match up with your model. Also, different places must have different information transfer coefficients for housing prices, as places such as Texas experienced a much less pronounced price rise than the SF Bay area or Seattle, for instance.ReplyDelete
This dynamic does seem to appear only since the 70s. Earlier Shiller data has less of a pattern. Massive building in the post-WWII era seems to have made housing affordable and non-bubbly.
Yes, different places experienced different shocks. Going to put together an update with Seattle and Dallas for a start ...