Disability and mortgage delinquency

Abstract

In this study, we examine the effect of disability on mortgage delinquency. Using 2007–2017 data from the Panel Study of Income Dynamics (PSID), we study the impact of disability on ex-post delinquency rate and ex-ante self-assessed delinquency risk. We find that disability substantially increases a household’s likelihood of falling behind on mortgage payments. Our point estimates suggest that on average, households with disabilities are about 41.65 percent more likely to be delinquent. This effect is highly cyclical and is much stronger during economic downturns. Households with disabilities are 56.38 percent more likely to be delinquent during the 2007–2009 U.S. housing crisis and its aftermath, and whereas the effect is rather muted during the subsequent economic recovery. Additionally, we discover that households with disabilities are generally conscious of their greater vulnerability to financial troubles. Households with disabilities are 27.28 percent more likely to report possible future delinquencies. This finding suggests that in addition to increasing ex-post delinquency rates, disability can also be psychologically burdensome for many homeowners, and this effect spreads beyond households who actually fall into delinquency.

Keywords:

• Credit risk
• disability
• mortgage delinquency
• JEL Classification D14
• I10
• R31

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 The 2017 American Community Survey reports that among 76,684,018 homeowners, 48,168,243 have a mortgage. See https://factfinder.census.gov/faces/tableservices/jsf/pages/productview.xhtml?pid=ACS_17_SPL_K202508&prodType=table (accessed on July 1, 2019).

2 See https://www.federalreserve.gov/releases/z1/dataviz/z1/balance_sheet/chart/#series:liabilities;units:usd (accessed on July 1, 2019)

3 Under the Dodd-Frank Act, a loan servicer can start a foreclosure proceeding after a borrower is more than 120 days behind mortgage payment.

4 A 2014 U.S. Census report shows only 20.4% of adults with a disability has a college degree, and that is substantially lower than the percentage of non-disabled adults who completed college (42.8%). Fewer than half of people with a disability between the ages of 18 and 64 were employed as compared to 77.5% for people without disabilities (Taylor, 2018).

5 See Deng et al. (2000), Foote et al. (2008), Elul et al. (2010), Quercia, Pennington-Cross, and Tian (2012), and Gerardi et al. (2013).

6 See Schwartz and Torous (1993), Quigley and Van Order (1995), Capozza et al. (1998), Yang et al. (1998), Ambrose et al. (2001), and Kau et al. (2011).

7 The positive correlation between financial leverage and mortgage delinquency has been consistently documented in most empirical studies examining mortgage defaults and delinquencies. This literature is vast and dates back several decades, and we do not attempt to list all of them here. Quercia and Stegman (1992) and Jones and Sirmans (2015) provide excellent reviews of this body of literature.

8 The PSID was conducted annually from 1968 to 1997 and biennially after 1997.

9 Disability includes both physical and mental conditions that limit major life activities. We limit the scope of our study to work-limiting physical disability because PSID family data do not contain information about mental health and non-work-limiting disabilities. Henceforth, we use the word “disability” throughout the paper to mean work-limiting physical disability.

10 Information on mortgages more than the first two are not collected during our sample period. Households having more than two mortgages are sparse. For example, the PSID did survey households about whether or not they have a third mortgage in 1999 and 2001. No household reported having a third mortgage in both years.

11 While self-assessed LTV ratios may differ from the household’s LTV ratios estimated using market-based house value measures, it is a relevant variable that homeowners use to make housing-related decisions. Previous studies show that self-assessed LTV ratios are good predictors of mortgage delinquencies and defaults (Gerardi et al. 2018), home improvements (Bian 2017, Melzer 2017), mobility (Coulson and Grieco 2013), and downsizing (Bian 2016).

12 In addition to house values and mortgages, the PSID also collects information on other assets and liabilities and uses it to calculate for each household their net worth excluding housing equity. Specifically, this imputed variable is constructed as the sum of seven asset categories net of total debt (sum of eight debt categories). The seven asset categories are farm or business, checking and saving accounts, other real estate assets, stocks, vehicles, annuity and IRA, and other assets. The eight debt categories are farm or business debt, other real estate debt, credit card debt, student loan, medical debt, legal debt, loan from relatives, and other debt.

13 We realize controlling for regional fixed effects at a finer level is preferable as housing market conditions may vary widely across a state. Unfortunately, we are unable to do so because more detailed locational variables are not available in our data.

14 Correlation between high default risk and low family income and limited education are documented by Deng et al. (2000), Foote et al. (2008), Elul et al. (2010), Quercia, Pennington-Cross, and Tian (2012), and Gerardi et al. (2013). See Quercia and Stegman (1992) and Jones and Sirmans (2015) for reviews of the large body of literature documenting the link between high LTV ratios and mortgage defaults.

15 $1.746/2.450=71.27\%$ and $1.747/2.450=71.31\%.$

16 See Kau et al. (1992) and Kau, Keenan, and Kim (1994).

17 Family health care expenditure include expenses for hospital and nursing home, doctor, prescription drugs and insurance.

18 Our results are robust to a wide range of the threshold values used for ${\mathit{income} \mathit{reduction}}_{t+2}$ and ${\mathit{health} \mathit{care} \mathit{expense} \mathit{increase}}_{t+2}.$ In addition to using $5,000, we re-estimated our models with many different threshold values between $3,000 and $15,000, and results are qualitatively similar.

19 An alternative to estimating a multinomial logit model is to use an ordered logit model. However, the validity of applying an ordered logit model hinges critically on the proportional odds assumption – the relationship between each pair of choices must be identical. In other words, the mechanism that governs the choice between “not at all likely” and “somewhat likely” must be the same as the one affecting the choice between “somewhat likely” and “very likely”. We test the proportional odds assumption using the Brant (1990) test and find this assumption is violated with our data. Consequently, we choose to estimate a multinomial logit model, which does not need the proportional odds assumption to hold true.

20 See Cohen et al. (2016) for a review of studies documenting the spillover effects of foreclosures on local house prices. Bian et al. (2019) find that the presence of nearby foreclosures substantially increases a home’s time-on-market.

Additional information

Notes on contributors

Xun Bian

Dr. Xun Bian is an Associate Professor of Finance and Real Estate in the College of Business and Economics at Longwood University.