When I was thinking about the previous housing post, I decided to revisit some basic data on debt outstanding to get a sense of the relationship between debt and home prices. It appears that, as with many measures pertaining to this subject, the story the numbers tell flips upside down, depending on if you look at it from a national level or from a more local level.
Here I am using the New York Fed Quarterly Report on Household Debt and Credit (Total Debt Balance Per Capita By State, Chart 20) for the debt measure, and the All Transactions House Price Index from the FHFA for the home price measure for various states.
In this graph of national measures, I have also added the national S&P/Case-Shiller home price index. Since the housing boom was largely facilitating the movement of Americans away from expensive cities, the S&P/Case-Shiller index of all existing homes rose more than indexes based on sales of homes, especially during the boom. The S&P/Case-Shiller measure moves more in line with rising per-capita debt levels until 2006. It is probably the case that the all-transactions measure understates changing home values, because Federal Reserve Flow of Funds data during the boom don't point to rising leverage, suggesting that prices and debt were rising in parallel.
In any case, using the FHFA average price measure, it appears that, at the national level, before 2004, debt was rising faster than the prices of homes for sale. Then, after 2005, debt continued to rise, even though prices leveled off. This appears to support the idea that rising debt fueled rising prices and also that rising price, then, led to more rising debt, in the classic positive feedback of a bubble.
The New York Fed provides debt data on several states, and comparing per capita debt among states seems to confirm this story. States where per capita debt was the highest in 2008 were "bubble" states - California, Nevada, Arizona, New Jersey and Florida.
But, what happens if we compare debt levels to home prices?
For the following graphs, I have indexed both home prices and debt levels to 1 in January 1999 to compare relative changes over time. In this next graph, in each state, I index the ratio of per capita debt/average home price to 1 in January 1999. On this measure, the relative order of the states is flipped upside down from the basic measure of debt-per-capita. Here, which is a broad estimate of changing leverage, it is the non-bubble states where leverage increased during the boom - Illinois, Michigan and Ohio. It's only after prices fall in the bubble states that debt/price levels rise. In fact, in the bubble states, debt/price levels were slightly declining during the boom.
Research has shown that, in the aggregate, homeowners harvest about a quarter of new home equity gains. Comparing the change in home prices over time to the change in debt, it appears that this data reflects that tendency. The next graph is a scatterplot comparing the change in debt to the change in home prices in each of these states over various periods of time. In each case, for each percentage point increase in home prices in a given state, debt rises by between 0.2% and 0.3%.
Over time, we should expect the relationship between debt and price to be close to 1:1. This is not because of equity extraction, but simply that if leverage levels remain fairly stable over time, then if, over a long period of time, property values double, we should generally expect debt outstanding to double too.
What's interesting is that leverage over the 1999-2005 time period was relatively stable. But, what we can see here is that, apparently, the stable level of national leverage was really a mixture of places where prices were relatively stable but leverage was rising; and places where prices were rising and leverage was declining.
A hypothetical state with no change in home prices would have expected debt per capita to rise by about 50% from 1999 to 2005 and about 20% from 2005 to 2011, for a total of about 70% over the total period.
(An aside: Remember back to the first graph. It could be that the all-transactions measure of home prices was understated by about 20% in 2005, and reconverged with other price measures by 2011. In that case, if we could use other price measures, the 1999-2005 plot (blue) would move right by about 20% and the 2005-2011 plot (red) would move left about 20%. This would pull their y-axis intercepts closer together.)
So, according to this measure, if there was a debt bubble, it was concentrated in the places with the least price appreciation. The subtle issue here is that there never would have been a moral panic in favor of watching home prices drop by 20% or 30% based on higher leverage in states with stable home prices. The early rise in foreclosures in 2007 did emanate from Michigan and Ohio. But, this was related to local economic problems, and, in fact, debt growth in those states from 2005-2007 was quite a bit lower than the US average. It was speculation in places like Phoenix that led to complacency about "disciplining" the housing market. In fact, if the focus had been more on working class households losing their homes in the rust belt in 2007, maybe public sentiment would have been more counter-cyclical. Maybe that would have fed a more typical populist response in favor of inflation.
This lines up with a separate analysis I have done regarding price and rent. Across metro areas, changing prices from the 1990s to 2005 correlate pretty strongly with changing rents. And, if you assume that home prices have a moderate sensitivity to real long-term interest rates, then interest rates basically explain the change in home prices across the country and rent inflation explains price changes in local hot spots. Using such a model, prices in 2005 are not out of line. They reflect interest rates and rent inflation. In order to make prices in 2005 look high, in the aggregate, you must assume that home prices are not sensitive to real long-term interest rates. The relationship between rent inflation and price remains, regardless of the sensitivity to interest rates. And, since rent inflation has little effect on home prices in Texas and Ohio, then interest rate sensitivity is more important to prices in low priced places than it is in high priced places.
In other words, given the undeniable relationship between rent inflation and price inflation, in order to believe prices were too high in 2005, you must believe that it was places like Texas and Ohio where prices were especially too high, not California and Massachusetts. And, similarly, as shown above, leverage rose more where prices were more moderate. It appears that if one is to believe that debt was the driving force in the housing market, that would need to be squared with the fact that prices didn't seem to be highly sensitive to changing debt levels during the boom, and, as researchers like Mian and Sufi have pointed out, some of the effect is from causation in the other direction, where rising prices lead to cash out refinancing.
Here, it may be worthwhile to look at the 2005-2011 period more closely. There was still an expansion of debt from 2005-2007. Here, we can see that there was a large difference between states during that period that was unrelated to the concurrent change in home prices (the blue dots in this graph). But, in a plot of the 2005-2007 change in debt against the 1999-2005 change in price (red dots), the correlation is quite strong, and similar to other periods.
During that time, there was a combination of two factors - the harvesting of equity, as noted by Mian and Sufi; and the slow tendency toward an equilibrium leverage level that I mentioned above. Surely, over long periods of time, we should expect debt levels to rise at a similar rate as price levels as new mortgaged buyers replace older owners. In other words, even after prices stopped rising, we should expect debt to continue to rise in the places where prices are the highest as new leveraged owners enter the market and as existing owners harvest home equity. In any natural long-term scenario, this should continue over time until the regression line approaches a slope of 1.
Then, from 2007-2011, prices and debt both sharply moved into negative territory and the relationship steepened. This was the period dominated by foreclosures, short sales, etc.
Looking back at the previous graph, for the entire period from 1999-2017, the relationship remains fairly stable. If prices have gone up an additional percentage point over that long period of time in a given state, per capita debt has only risen by about 0.2%. This shouldn't be the case. That should tend toward 1:1.
The lack of a strong relationship is clear if we look at each individual state, over time. In each of the following graphs, the black line is the US aggregate number, and the colored lines are individual states. I have categorized them, roughly, by the type of market, although the "Closed Access" states are really a mixture of Closed Access metros, Contagion, and Rust Belt areas. In each graph, I have clearly marked the 4th Quarter of 2003 and 2005, to get a sense of where each state was at those points in time.
Remember back to the first graph, also. These graphs are based on the all-transactions home price measure, which may be somewhat understated from 2003-2007, so the aggregate US measure is probably the useful comparison to use as an estimate of how leveraged households were becoming in each state, and with a more comprehensive measure, it may have moved more in line with the 45 degree line until 2006 (stable leverage).
By 2003, leverage (relative to 1999) was high in Michigan and Ohio (where home prices were especially low) and was low in Texas (where prices were near the national average). During that period, it appears that price and debt were inversely related, if anything, and where debt was high, it was likely the result of households in economically challenged locations using home equity as a financial safety net. Note also that Nevada and Arizona had roughly moved with the national average in terms of both debt and price over that time.
From 2003 to 2005, prices accelerated in all states. In Ohio and Michigan, debt growth since 1999 retracted back toward the national average and home prices didn't rise as sharply as in other places. In Texas, both prices and debt levels remained relatively low.
During this period, debt levels did rise more quickly than average in Nevada and California. In other states, debt rose at a rate similar to the national average. There is no systematic difference here, regarding debt, between the Closed Access and Contagion states and the other states. In general, the "bubble" states didn't move up during this period, or even move diagonally along the 45 degree line. They moved horizontally to the right. Valuations changed in those states, but there is little sign of systematic differences in debt levels.
After that, leveling off and declining prices dominate the behavior, so where prices had previously risen more, debt continued to rise more as prices stabilized, then where prices declined more, debt declined more along with them. This creates a pattern of concentric circles around the national average, especially in the Contagion states. It is the lagging nature of debt growth that creates that counterclockwise shape. And, after all of that, generally, states that had higher appreciation from 1999 to 2005 have had prices rebound so that they have more price appreciation today than the US average (with the exception of Nevada) and they have less debt than the US average (with the exception of New York).
The only other states with per capita debt growth higher than the national average from 1999-2017, besides New York, are Texas and Pennsylvania. Those are also the two states who had the smallest price shocks after 2005.
There is little evidence here that debt was an important causal factor in systematic differences between states. There is some evidence of price as a causal factor in the differences in debt growth between states. I would suggest one other factor that seems important here, and that is property taxes. There seems to be a relationship between higher property taxes and less volatile housing markets, mostly because higher property taxes moderate prices when they are rising.