The vexing question of the U.S. housing bubble origins
Back in 2005, a massive increase in mortgage credit and U.S. housing prices resulted in household debt almost doubling. Today, while the consequences of that bubble are still very evident, the dynamics of the housing bubble are not fully understood. A debate about these dynamics continues courtesy of a new National Bureau of Economic Research working paper released yesterday, featuring arguments between two teams of economists.
One side of the debate, Atif Mian of Princeton University and Amir Sufi from the University of Chicago, have long argued that credit growth in the 2000s flowed to low-income borrowers who used the credit to replace wage growth and keep their consumption up. Mian and Sufi’s extensive data work shows that zip codes with low-income growth and low credit scores saw high levels of credit growth during the housing bubble.
Earlier this year, another research team published a working paper that challenged some of the findings of Mian and Sufi’s 2009 paper. These economists, Manuel Adelino of Duke University, Antoinette Schoar of the Massachusetts Institute of Technology, and Felipe Severino of Dartmouth College, argue that Mian and Sufi’s focus on zip codes does not consider that the composition of buyers changed over the time within zip codes. The three researchers show that, within zip codes, the growth of credit accrued to middle- and higher-earning households whose income was growing.
Mian and Sufi replied with two papers. The first paper argues that Adelino, Schoar, and Severino’s choice of income source– reported income from mortgage applications–was prone to fraud, and therefore overstated the positive relationship between income and credit growth. The second paper notes that the three economists sort their mortgage by credit scores in 2006, and that these scores were themselves influenced by the run up in housing prices. Therefore, they are not an unbiased measure of borrowers’ status when the bubble started.
The new working paper released yesterday by Adelino, Schoar, and Severino responds to Mian and Sufi’s criticisms. On the first point, the three economists point out that their original paper also used tax data from the Internal Revenue Service, and that the positive relationship between individual borrowers and credit growth holds up. But, they maintain, even when they do use income data from mortgage applications, the relationship still holds up if they look at areas where misreporting wasn’t a major problem. The three economists contend that if misreporting were driving all of their results, then zip codes where fewer mortgages purchased by mortgage finance giants Freddie Mac and Fannie Mae—and therefore more likely to be fraudulent—would have stronger correlations between income growth and credit growth. But that doesn’t show up in the data.
On the second point, Adelino, Schoar and Severino concede that using 2006 credit scores does mean that they are biased up. But the level of bias, according to their analysis, is so small (at the most 18 points) as to not have an effect on their analysis. This means that Mian and Sufi’s argument that households that were really low-credit ones but got sorted upward arbitrarily wouldn’t hold up. In fact, the three economists claim that the data analysis in Mian and Sufi’s second paper actually supports their original analysis.
Ultimately, what is at stake in this debate? Adelino, Schoar, and Severino agree with Mian and Sufi that low-credit score zip codes saw large increases in credit. The resulting debate seems to be about which households took on the debt—a question not addressed by Mian and Sufi in their 2009 paper. But a 2011 paper by the pair does find that increased home equity loans by existing homeowners explained a significant part of the increase in household debt during the bubble years. So perhaps there is some agreement there. We are left with the question: How much of the defaulting debt was in the hands of low-credit score individuals (those in the bottom 20 percent)? The dates for the two different estimates don’t overlap. In 2006, Mian and Sufi have the share at a bit over 40 percent, whereas Adelino, Schoar, and Severino have it at 36 percent. What seems still up for debate is the importance of the level by credit score versus the trend.
A response from Mian and Sufi may be forthcoming, but the recent paper seem to be converging to a spot closer to Mian and Sufi’s work. These debates that take place in gated working papers may seem far removed from everyday life, but they actually have important implications for our understanding of the link between credit and its allocation, wealth inequality, and economic stability then and now.