It’s common knowledge that Seattle housing prices are increasing. I frequently hear that Seattle prices are currently in a bubble, and buyers should hold off. I also hear that houses are selling very quickly and for cash. On the other hand, it’s folk wisdom that demand is being driven by the tech sector. So if that’s true, then maybe it’s not a bubble. Or, maybe house prices are going up for good reason, but it’s also a bubble, but no one knows when the bubble will pop.
Any good model needs to have criteria for acceptance and falsifiability. A bubble is a nebulous term in finance, which appears to be a real phenomena, but is also extremely hard to measure. Robert Shiller’s careful econometric study of mean reversion in equity prices and valuations made a strong claim that there is some predictability in asset markets (when detrending growth and inflation). He attributes this in large part to ‘irrational exuberance,’ and the psychology of investors. This is hard to measure. When a new home buyer in Seattle is going out today to buy a home, is it based on rational expectations? Or is it because everyone else is buying homes, and he wants to get in before it’s too late? Shiller measured this as well, and found that people’s expectations of housing price increases (last in 2003) were much more optimistic than reality.
What is extremely interesting, is when controlling for inflation Shiller finds that based on his housing prices index, prices don’t increase that much. Since 1990 housing prices have only increased about 28% in real terms, which is orders of magnitude less than equity markets. Real estate though has a duality, as it generates wealth over time, either in the form of rental or consumption. Seattle is also often compared to San Francisco, and it’s true that the real-estate market has increased massively in San Francisco due to the tech boom. However, San Francisco is also surrounded by water and has extremely restrictive construction regulation both in the city and down near the Menlo Park region. In the short-run all cities are static, but Seattle can continue to expand, build, and create new neighborhoods.
While housing prices on average don’t seem to consistently increase, not dramatically at least, over the long-run, they do vary. Based on Zillow Seattle data to 2006, a one standard-deviation movement would be on average $40,000.
From here we know a few things:
1.) Seattle prices have been increasing for nearly six years straight, even when controlling for inflation.
2.) There is no strong evidence indicating that real-estate assets always increase in real value, as unlike equity there is no cash-flow or (measured) risk-premium
3.) Real-estate markets have high transaction costs, are very illiquid, and often rely on cyclical factors such as cost of credit.
Above is a graph of the Seattle Case-Shiller house price index, as well as the same index that I adjusted using the Seattle-CPI measure of inflation. Just eyeballing the chart there are two obvious hypothesis. The first is that Seattle housing prices are consistently increasing, probably due to the tech boom and it being an awesome city, with an exception for the housing-crisis. The second is that housing prices generally don’t increase, but do follow a mean-reverting random walk. We see the same thing using Zillow data, which only goes back to 2006, but has a better methodology using real micro-data.
To start I will just let the time-series speak for itself. I chose an AR(1) to model this data as a simple Gaussian distribution. Some other specification might be better, but the general point here is to show how the inflation-adjusted series is a stationary process. The question these models raise is whether the time-series process captures the dynamics of Seattle real-estate, or if there is some additional structural knowledge about the economy that means the recent increase isn’t just cyclical.
One explanation could be the leverage cycle. John Geanakoplos argues that there is a cycle of leverage driving asset price variation in addition to interest rates. If this were true, increases and variation in the Seattle housing market could be explained in part due to cyclical variation in credit. It’s hard to pin down the mechanism of action here. In retrospect the housing crisis came down to sub-prime mortgages. Asset prices due to a leverage cycle could be due to very small changes in the preferences of investors and homebuyers. As a starting point, I have plotted the R2 values of regressions of Seattle housing prices on the Bank of America US High Yield Spread. Using a high yield spread might show if something is going on as it proxies for credit-market risk, but won’t help in pinpointing causality. On this chart a positive lag represents a lag. So a lag of 6 means we are using a 6 month old Credit Spread to explain the current house price. A negative lag means using data from the ‘future’ to explain current house prices. Using the Zillow data, we see that the future of the high yield data explains around 50% of the current price. We know this was the case during the housing-crisis, so the core question is whether it is also cyclical on a lower scale, or only happens rarely during a crisis. If it only happens during a crisis, these charts are just picking up a structural issue, and have no use in understanding the dynamics of housing or the future. But otherwise, it could mean increasing house prices are able to predict increased risk in credit markets. This data keeps the question alive, but isn’t sufficient to answer it.
To finish, I decided to throw the high yield data into a vector autoregression of Seattle housing prices. These aren’t much different from the auto-regressive models, but I was interested in seeing if there were any joint-dynamics. A shock to the high yield series does lower average house prices by a few thousand dollars, but nothing massive. It would be interesting though to see what the dynamics are like if there were a better measure of leverage, ease-of-credit, and liquidity in the Seattle market. If Shiller’s research is correct, these variables could explain most of the movement in the house prices, which would be essentially mean-reverting.
What would be really interesting is if Seattle house prices do have an upwards trend, and by explaining the factors that are mean-reverting it would be easier to search for variables that predict price increases. This could be related to more traditional supply and demand factors. For example, if there are areas that are exceptionally nice in Seattle, living there could be a signal to others as to your value. In this case social-signalling and agglomeration could result in a few areas that are explained poorly by a model that captures the mean-reverting pricing factors. This would require a more granular view of subsets of the housing-market. My ultimate prediction is that Seattle house prices, in real terms, will decrease over the coming decade as more houses are built and new firms open offices in cheaper locations creeping the city away from the center. Bellevue has become a beautiful area, but before Microsoft it wasn’t particularly noteworthy. In addition, certain subsets of the market will increase massively due to specific economic factors. Whether those subsets are predictable, I am not sure yet.
Code I used to generate charts: Github
Shiller’s paper: https://www.nber.org/papers/w13553
Geanakopolos’s paper: https://www.nber.org/chapters/c11786
Washington Employment Data: https://fortress.wa.gov/esd/employmentdata/reports-publications/economic-reports/monthly-employment-report
FRED CPI Series: https://research.stlouisfed.org/fred2/series/CUURA423SAH
Zillow Seattle Data: zillow.com/seattle-wa/home-values/
FRED Case-Shiller series: https://research.stlouisfed.org/fred2/series/SEXRNSA/downloaddata
FRED Bank of America High Yield Debt Series: https://research.stlouisfed.org/fred2/series/BAMLH0A0HYM2