Government Stimulus and Mortgage Payments during COVID-19: Evidence from the US Census Household Pulse Survey

Abstract

We investigated the impact of governmental stimulus payments and how they were employed by individuals—whether saved, spent, or used to pay down debt—on mortgage repayment. We determined that there was a positive effect for individuals who were eligible for the Economic Impact Payment (EIP) Stimulus and used it to increase their ability to make their next mortgage payment. However, this did not affect their overall likelihood of having paid off their mortgage. These findings held after various demographic controls were employed, as well as after controlling for alternative measures of spending meant to disentangle the EIP from other long-term patterns of saving and spending. Differences by Race and Socioeconomic status or age were also explored. Our results provide preliminary evidence that the EIP had a positive effect on mortgage payments during the COVID-19 pandemic, and show that future government stimulus payments should take into account patterns in saving affecting repayment.

Keywords:

• Mortgage
• COVID-19
• economic impact payment
• government stimulus

Statement of Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Statement of Competing Interests

The authors report there are no competing interests to declare.

Human Subjects Research

The data employed for this study was conducted and made available by the U.S. Census. All human subject protocols in the initial data collection were performed by the U.S. Census and its agents. In keeping with IRB standards for our university, since data was provided in a public-use, deidentified fashion, it is considered exempt from separate approvals.

Data Availability Statement

Data are publicly available online through Census.gov. Code is available from the authors upon request.

Notes

1 As seen in Elmer and Seelig (2020) consumers do appear to be affected by insolvency in making their mortgage payments

2 On a slightly different note, our conclusions are in keeping with Hubbard and Mayer (2009), who advocated for additional governmental support in the midst of crisis to help cushion against underwriting losses. While the situation is quite different today, we also find a positive role for government intervention via stimulus payments relating to a reduced likelihood of missed mortgage payments.

3 Prior versions of this work with no weighting did not show substantially altered results. Furthermore, regressions used “probability” weighting, while summary statistics employed “importance” weighting. This was due to restrictions in how weights can be employed in calculating means and in the marginal Probit analysis.

4 Differences in response rates and participant characteristics between the Household Pulse Survey and other datasets—such as the American Community Survey (ACS) in particular—are provided in detail in Peterson et al. (2021).

5 Results from the OLS regressions were similar and are suppressed due to space constraints.

6 We further hypothesized a smaller effect on paying off mortgages, rather than on making the next mortgage payment. This represented an additional justification for employing these two outcome variables - to further test our empirical structure.

7 Regressions were repeated, but categorizing individuals who were able to make a payment on the mortgage because it had been “deferred” as also having high confidence in their ability to pay. The results were virtually unchanged compared with employing our main variable in the analysis.

8 In keeping with the previous literature, we considered employing mental health characteristics as an instrumental variable for the choice of how the stimulus money was used (spend/save/pay debt) due to an individual’s current state, but we believed that there was still insufficient evidence from the literature to rely solely upon this choice. We felt more confident instead employing this as an additional control factor. Furthermore, each of the variables was initially coded using a Likert Scale of reporting a particular problem (anxiety/feeling down, etc.) during the last seven days: 1. not at all, 2. several days, 3. more than half of the days, 4. nearly all the time. In order to create a binary structure, individuals who experienced the issue either more than half the time or nearly all the time were coded as having the problem, and those who had it not at all or only for several days were coded as it being absent.

9 Income was coded by category: <$25K, $25-$35K, $35-$50K, $50-$75K, $75-$100K, $100-$150K, $150-$200K, over 200K; Marital status was recoded as either 1. single, 2. married, or 3. widowed/separated/divorced; Race was coded in a hierarchy of Hispanic, followed by the other racial categories of White alone, Black alone, Asian Alone, Other.

10 Some of the more recent Census estimates show a lower fraction of the population having at least a college degree. See: https://www.census.gov/newsroom/press-releases/2020/educational-attainment.html

11 In a more typical sample we might expect the “always single” group to be closer to one-third of the sample. See: https://www.unmarried.org/statistics/

12 The increased age in our sample relative to the national average of 31-45 in the United States (https://worldpopulationreview.com/state-rankings/median-age-by-state) was anticipated since surveys are generally filled out by individuals at home and that tends to be disproportionately represented by older individuals. This may also explain why our average number of children in the home(0.61-0.86) is somewhat smaller than the national average of around 0.94 and why the restricted non-elderly sample has more.

13 We experimented with using even lower income cutoffs and less education, but, as apparent from the summary statistics, there was a substantial fraction of the sample having exactly a High School degree and also with income at the higher breakpoint. We chose this set of criteria to retain sufficient observations in the sample to make reasonable conclusions.

14 These two categories were either included, with the separate individual variable for each of spending and psychology categories, or else by employing the score and predict functions in STATA to retain the principal factor score of the combined group of each of psychology or spending patterns. As can be seen, the results are generally similar after each of these methods were employed.

15 While we controlled for alternate measures of spending patterns, and a fairly complete set of income and demographics characteristics, it is still possible that some omitted features of income and SES were creating artifacts in this analysis and being captured by stimulus payments.

16 See www.fmerr.org for additional program details.