Predictive foreclosure measure
The predictive foreclosure measure estimates what the foreclosure rate may have been absent federal policy implementing institutional forbearance and the foreclosure moratorium. In 42 states and Washington, DC, the 90-or-more-day mortgage delinquency rate exceeded the foreclosure rate at the trough of the Great Recession. In these states, we regress the actual foreclosure rate on the 90-or-more-day delinquency rate in the following manner:
For% = α + 90%(x1) + Pol(x2) + Ɛ
In this model, 90% represents the 90-or-more-day mortgage delinquency rate, and Pol is a dummy variable for federal public policy that is 0 in all months before March 2020 and 1 in the months since March 2020. The Coronavirus Aid, Relief, and Economic Security (CARES) Act extended financial assistance as well, but we hypothesize that this support would have largely reduced the 90-or-more-day mortgage delinquency rate. The data cover the period between January 2005 and May 2021.
To estimate county-level foreclosure rates absent these public policy actions, we first apply the coefficients x1 and x2 to county-level data, and we set Pol equal to 0 for all months. This model presumes the CARES Act would have kept the financial assistance of $1,200 but not implemented forbearance or the foreclosure moratorium. If the CARES Act generally were not included, the 90-or-more-day delinquency rate may have been higher than reported.
Across the remaining eight states—Florida, Hawaii, Iowa, Maine, New Jersey, New York, Vermont, and Wisconsin—the 90-or-more-day mortgage delinquency rate was less than the foreclosure rate at its Great Recession low point. This suggests that other factors, such as tax foreclosures, may have contributed to the foreclosure rate.
To account for the possibility that delinquencies other than missed mortgage payments may have played a larger role in historical foreclosure rates during times of economic stress, we regress the state-level foreclosure rate on the unemployment rate and five-year home price changes:
For% = α + UR(x1) + 5HPI(x2) + Pol(x3) + Ɛ
In this model, UR represents the state unemployment rate, 5HPI represents the five-year home price change, and Pol represents the public policy actions of institutional forbearance and the foreclosure moratorium. We then take the statewide coefficients x1, x2, and x3 and apply them to county-level data to estimate a county-level foreclosure rate. In our county-level calculation, we set Pol equal to 0 for all months. Similar to the previous model, this model assumes that public policy had no effect on labor and housing market conditions. But the absence of these policy actions might have reduced five-year home price changes by introducing more housing supply and might have raised the unemployment rate even higher than reported. As a result, these results might be underestimates.
We calculated state and county foreclosure rates and 90-or-more-day mortgage delinquency rates using CoreLogic’s Market Trends data. We used unemployment rate data from the Bureau of Labor Statistics and calculated home price index trends using Products and Services provided by Black Knight Data & Analytics, LLC.
This is only one method of estimating foreclosure rates absent forbearance and the foreclosure moratorium. There are other and potentially more sophisticated methods for estimating a foreclosure rate. And as these policies sunset, future methods will likely provide more accurate results.
Demographic and household statistics
We calculated the household and demographic statistics using 2019 American Community Survey data. The data are calculated using both the Public Use Microdata Area (PUMA) and pretabulated summary-level data when available.
Not all counties met the 100,000-person minimum population threshold to be identifiable in the public use data. For these smaller counties, we used a PUMA-to-county crosswalk to create county-level weighted estimates. To check that our estimates were reasonable, we compared a series of calculations from the crosswalked data with census summary data and found marginal differences. We indicate which variables were created this way under the PUMA tab with a “Y.”