Monday, August 28, 2017

Housing: Part 252 - The Deceptive Limits of Knowledge

One of the lessons that has really hit home for me in this housing research is how much our assumptions and priors affect our interpretations, and how much our conclusions really are simply a regurgitation of our priors after they have been run through a Rube Goldberg logic machine.  We make so many claims based on a mixture of facts, constructed information that may or may not be accurate, and assumptions that may seem perfectly reasonable but are not.

I can't say that I've applied much of this learning.  I regularly make bold claims based on my own versions of constructed information.  Very early posts I wrote about housing markets contain what I still consider to be clever and well-argued justifications for housing trends, which today I would say generally missed the most important points.  I don't know if anything I wrote was flatly incorrect, but in practice, coming up with the wrong answer to the right question isn't much worse than coming up with the right answer to the wrong question.  Yet, I find it very easy to be confident about my current conclusions, some of which are almost certainly wrong in some way that I will eventually realize.

One example of an assumption that doesn't explain as much as it seems like it should is the idea that cities where people want to live would naturally be much more expensive.  Another is that when mortgages outstanding grow at about the same rate as total real estate values, the growth in mortgages seems like it must have caused the rise in values.  In both of these cases, there really isn't a reason to believe that is the case without corroborating evidence, but it seems so reasonable, that these assumptions can become placeholders in a conclusion that end up doing a lot of the work.  They seem so reasonable, it doesn't seem necessary to vet them.

Yet, in so many cases, just a slight error in how information is constructed can turn our conclusions 180 degrees backwards.

In any case, here is a good example of the problem.  The authors post a chart of high and low tier housing markets, aggregated from 16 major markets.  They note that low tier homes rose higher in the boom years, fell lower in the bust, and now have overtaken high tier homes again.  They tell investors:
If you focus on lower-priced homes, beware that you are investing in a more volatile section of the market from a pricing perspective and beware that lower-priced homes have appreciated the most.
First, there is an assumption problem here.  It just makes sense that lower tier markets are more vulnerable, in the "World to end tomorrow.  Women and children to be hurt the worst." sense.  It makes sense that marginal markets would be most affected by economic contractions.  Defaults would be higher in low tier markets.  So, this conclusion rests easy.  It doesn't seem like it needs to be vetted.

Then, there is a constructed information problem.  Low tier homes were only more volatile than high tier homes during the boom in a handful of markets.  It shows up in their chart, because the 16 markets they aggregate contains all of the cities where that happened.  It didn't happen anywhere outside the 16 markets they reference, and it only happened in about half the markets they do reference.  There is a specific pattern that causes this to happen that is probably mostly tied to maxing out tax benefits on homeownership when homes rise above about $500,000.  Actually, this should provide a little bit of a positive skew to potential gains for investors in low tier homes and a negative skew for investors in high tier homes, because tax subsidies create an asymmetrical set of outcomes for investors that don't claim owner-occupier subsidies.

Source
The first graph here shows YOY high and low tier home prices in LA and Portland going back to 1987.  Notice that there is no systematic excess volatility among low tier homes anywhere until the mid 2000s.  In LA, low tier homes rose higher than high tier homes during the boom, and then dropped farther.  But in Portland there was only a short period, well after the bust had begun, where low tier homes dropped somewhat farther than high tier homes did.

So, there is basically one episode where low tier prices collapsed more than high tier prices, and that happened to coincide with a sharp public policy shift where we essentially made it illegal to make a mortgage to low tier owner-occupiers.  This coincided with a decline in working class homeownership rates in general and a decline of more than 10% in homeownership rates among families that typically used to be first time homebuyers - young families with average or above average incomes.

Recently, it appears that the collapse in homeownership may have finally stopped.  But, this was a one time shock.  You can't collapse a bridge twice.

Source: Zillow Data
So, if we deconstruct their constructed information, we really have two types of cities.  First, we have Closed Access cities, where low tier housing is more volatile than high tier housing, because of these owner-occupier tax issues.  And we have the other 80% of the country, where there never was a difference in volatility, except for that one-time shock.  In a few cities, like Phoenix and Dallas, it appears that population inflows have made up for the lack of low tier demand created by the mortgage shock, and low tier house prices may have recovered back to parity with high tier homes.  But in most cities, from Seattle to Detroit and everywhere in between, there is a very typical pattern.  Prices moved together until the end of 2008, then low tier prices take an extra step down.

(Admittedly, this shows up more clearly in Zillow data.  Low tier markets, in general, seem to have an upward bias and high tier have a downward bias in the Case-Shiller data, compared to Zillow, over time.  This seems to be the case across cities.  Case-Shiller follows individual properties while Zillow takes a market snapshot of all properties at a point in time.  But, I'm not sure how that would create this difference.  So, the volatility is basically the same in Case-Shiller vs. Zillow, but Case-Shiller makes low tier properties look like better long term investments.  The authors use Case-Shiller data, which should make low tier properties look better, but since there has been this z-shaped boom, bust, and recovery, that only makes it look like an unsustainable boom in a volatile market segment.  Another example where a simple difference in a data set of constructed information leads to totally opposite conclusions.)

So, basically the exact same data with a couple of assumptions or facts switched out, and you get two completely different conclusions.  If you account for the mortgage market shock and you use the long term drift in Zillow values, then you're buying up low tier homes in cities across the country with great risk/reward profiles.  If you don't account for the mortgage market shock, you make some reasonable assumptions about market behavior, you aggregate the data among various cities, and you use the long term drift in Case-Shiller values, then you're bracing for a contraction in low tier housing markets that look especially perilous.

These are not massive errors of judgment.  These are tiny little shifts in data with some very reasonable assumptions filling in some harmless looking blanks in the narrative.  And you've got one investor long and the other short - one investor looking at markets always swinging to extreme equilibriums that must correct and another looking at some stable and boring market equilibriums with occasional policy shocks.

No comments:

Post a Comment