Monday, October 22, 2018

Housing: Part 326 - Another example of how much priors determine conclusions

Adam Ozimek has a recent post up at Moody's about the causes of the slow housing recovery.  Adam generally does great work, and the funny thing about this post is that he approaches the topic just as I would like to think I would.  He tries to find neutral references to measure new data against.  He uses reasonable logic to take inferences from those measures.  He looks at individual metro areas instead of the national numbers.  But, his conclusions are upside-down wrong.

This is because, as is common on this topic, he (reasonably) builds on the conventional presumptions about the bubble and bust: A credit bubble led to a housing price shock, which led to overbuilding, which led to an inevitable bust.

But, I have found that, to the contrary, there was a housing supply shortage in the dynamic coastal urban centers which led to a housing price shock in those cities (accelerated by more flexible lending to young households with high incomes) , which led to a migration event as renters were forced away from and existing owners tactically sold out of those cities.  That migration event stressed the housing supply of the main destinations of those migrants which led to a secondary housing price shock in those cities (Phoenix, etc. - "Contagion cities"), which led to a moral panic about lending and building, which caused the migration event to suddenly stop, which left those cities with large inventories of unsold homes.  And, when the entire episode was blamed on excess lending, post-crisis lending policy was tightened to the extreme, creating a post-crisis housing bust from 2009-2012 in credit constrained areas while wealthier areas stabilized.

Put more simply, prices can rise because of high demand or low supply, and where one cause is presumed vs. another, it will tend to lead to diametrically opposite conclusions.  One clue is quantity.  If quantity is rising, that suggests a demand cause.  If quantity is declining, that suggests a supply cause.  This is especially tricky with an inferior good.  Rice prices and quantities could both rise if rice is an inferior good to meats and other foods in the midst of a famine.  Effectively, there was a supply-caused price spike in the Closed Access cities from the late 1990s to 2005, and eventually there was a price spike for houses in less expensive "Contagion" cities where houses were purchased as inferior substitutions for Closed Access homes.  The supply-constrained market eventually led to a bubble market in an inferior good.

In the national data, both types of cities are mashed together, and rising quantities in the cities serving as inferior substitutes combined with the rising prices of the Closed Access cities made it look like a national housing bubble.  The financial crisis developed because these extreme supply-constraint problems were treated as if they had been caused by excess demand, so a credit bust was engineered in an attempt to solve a supply bust.  Rather than a bubble-bust, we had a bust-bust.

All of that is a prologue to reviewing Ozimek's post.

First, he compares the current rate of permitting in each MSA to the rate of permitting before the crisis.  He finds that the cities where supply has recovered the most are the cities where price growth is the highest now and has been strongest since the bubble peak.  He concludes, reasonably, that supply constraints aren't the primary driver of rising prices, because building appears to be strongest where prices are rising the most.

He does several regressions to test the effect of home prices on the recovery in permits, and finds in several models that rising prices either since the bubble peak, since the bust trough, or just in the last year, all correlate with more supply recovery.  And deep price drops from the bubble peak to the trough correlate with lower supply recovery.  This also convincingly suggests that supply constraints aren't the reason for current price appreciation.  And, cities where permits haven't recovered seem to have overbuilt in the bubble (which is why their prices had declined so much in the bust).

He does one final regression where he adds job growth to the mix, and when he does that, price increases since the trough become less significant, and the recovery of permits is explained by job growth and by the depth of the previous price declines.  In other words, even in that regression, we might conclude that permits are strongest where job growth is strong and weakest where cities had overbuilt during the boom.

All of these conclusions are reasonable - obvious, even - if we operate from the conventional presumptions about the boom and bust.  But, the presumptions are doing all the work here.

Ozimek even takes a reasonable precaution when measuring the recovery of permits.  He is trying to measure the recovery to a reasonable level, not to bubble levels.  So, he doesn't compare current permit rates to the peak rates of 2005.  He compares them to the average from 2000 to 2004.  But, this only seems like a reasonable precaution because of the conventional presumptions.

This wasn't a building bubble.  This was a migration event.  The places that were building more homes weren't places that suddenly had spontaneous speculative bubbles.  The peak building years of 2004-2005 mostly reflected an acceleration of long-standing migration patterns.  So, cities that built more in 2004-2005 were generally cities that had always built more.  The credit bust after 2007 didn't undo a bubble.  It undermined longstanding migration patterns.

I have reviewed construction employment numbers at the state level.  After controlling for existing rates of construction, states that built more in 2004-2005 have done well.  They still have construction employment that is above average.  States that had low levels of construction employment before 2004 still have about the same level of construction employment today that they did then.  States that had high levels of construction employment before 2004 have suffered deep cuts in construction employment, and most of that drop in construction employment happened after the credit bust was imposed.

So, one reason that job growth and the depth of the bust are the most significant variables in Ozimek's last regression is that, after 2007, the credit bust killed off job growth and mortgage lending in the cities where growth had previously been the highest before the bubble.  The drop in home prices wasn't due to overbuilding.  It was due to the sharp collapse in mortgage lending.  That is why home price appreciation in most cities before the bust was uniform across the market, but home price collapses after 2007 were most severe at the low end in every city.

We can see this by looking at housing permits as a proportion of population.  I also have included scatterplots showing MSA incomes, home prices, and permits/capita.  And, a couple of graphs comparing home prices in Seattle and Atlanta.

Several notable items are clear here:
  • The reason some cities have high home prices is because they have high incomes and low rates of homebuilding.
  • That relationship has been strengthening over time.
  • One reason cities with expensive homes have shown the most recovery to pre-crisis permit rates is because they never had high permit rates.  Their permit rates during expansions are highly politically constrained.  Ironically, it is cities like LA and San Francisco (the red lines at the bottom) that Ozimek would identify as having recovered the most, even though they still have much lower building rates than the other cities.  And, in fact, the reason their job growth is strong is because families there have gone back to stuffing ever more uncomfortably into a stagnant housing stock in those cities because the credit bust has obstructed the avenue for building homes in the less expensive cities.
  • Another reason cities with expensive homes have recovered the most is that the credit bust was highly correlated with incomes.  Cities with higher incomes are less credit constrained.  This is clear in the graphs comparing Seattle and Atlanta.  (One at market prices, and one with prices indexed to 2000 to compare relative changes.)  This is what most cities look like.  Top and bottom moved together during the boom, then after the credit bust, the bottom in every city dropped significantly compared to high tier markets.  So, the low end in Seattle has performed similarly to the high end in Atlanta.  So, the credit bust has caused low priced cities to decline more than high priced cities and the low tier within each city to decline more than the high tier.  That is because it was a credit bust that caused the collapse.
  • In the graph of housing permits, notice that there aren't typically surges in permits.  Permits rise up to a typical expansion level for each city and then remain fairly level for the remainder of the expansion.  The difference between cities is much larger than fluctuations over the course of a building expansion than changes within a city.  That is because permitting rates are largely a product of migration patterns.  The reason rates of building were high in Atlanta and Phoenix was because a lot of people move there.  There were a few cities that had building spikes.  Phoenix did have one from 2001 to 2005, but that spike in building was matched household by household by a spike in in-migration because of the Closed Access migration event.  Builders in Phoenix weren't building tens of thousands of spec homes in 2005.  In fact, they were holding lotteries among buyers because they couldn't get lots permitted fast enough.
  • If you think about the dominant effect of migration on these building patterns, it would be very difficult for the cities that build the most to overbuild, because for every home's worth of natural local growth, there are one or two households moving into the city.  If there isn't a negative shock to migration, then inventories in the cities that build the most will naturally be worked off the quickest.  The reason these cities might end up with excess inventory is because of local income shocks or migration shocks.  Here, there was a migration shock followed by an income shock.
  • Oddly, permits declined in every city at roughly the same time even though clearly there wasn't oversupply in LA or San Francisco.  In fact, the end of the migration event meant that population started to rise again in LA and San Francisco just as housing starts collapsed.  The reason that cities that saw the largest drop in home prices and the weakest labor growth have had the weakest housing permit recovery is that the cities that had the largest collapses were cities that welcomed in-migrants.  The credit bust killed off longstanding migration patterns, which created a housing collapse in those cities and limited job growth because job growth had previously been correlated with rising population.
In a later post, Ozimek noted that the job recovery has been uneven and that places with lower home prices have seen lower job growth since the crisis.  This also can be explained by the credit bust and the decline of migration.  Before the crisis, households were moving away from places where incomes are higher.  This is a perverse pattern, but it did lead to lower job creation in prosperous places and more job creation in less prosperous places.  Instead of fixing that perversion by building more homes in prosperous places, we added a new perversion to it.  We made it more difficult for households to build homes in low cost places.  This has tilted job growth more toward prosperous places, but instead of creating more prosperity, it just intensifies the cost pressure.

This is the problem with having a canonized set of presumptions that are all wrong.  Ozimek did everything right in his analysis.  The presumptions are the issue.  Until the public comes around to the correct presumptions, good analysts will be drawn to all the wrong conclusions about what has happened and what we should do about it.  For starters, for those households who haven't been drummed out of low tier homeownership by foreclosure, we would do wonders for working class balance sheets if we opened up the mortgage window to FICO scores under 760 again so that prices in those markets could recover by 20% or more to where they should be.  And, the borrowing and building that would be triggered by that shift would create several positive developments.  It would be disinflationary, because rent inflation would finally be tamed.  It would also likely boost interest rates.  It's quite amazing how easy it is to get a headache to go away when you stop hitting your head with a mallet.  But it is the presumptions about the cause of this whole mess that will have to change for us to do that.

1 comment:

  1. Exactly. Ozimek was getting so close and then hopped off the rails.

    But at least he is thinking of the housing market in terms of supply constraints and geographies.

    I like your example of inferior goods.

    In percentage terms, some neighborhoods in Los Angeles exploded in value even though, or rather because, they were previously inferior goods. The Hollywood area was a grunge pit for decades, and now made a comeback in the last generation or so. Bakersfield and Palmdale are filling up with low income people. Older residents of inland areas, who actually fled there for perceived safety and refuge a generation ago, are complaining about dangers.

    For some reason, macroeconomists have a hard time thinking about property zoning and supply and demand.