The views expressed in this blog post and the associated paper are solely those of the author and do not necessarily reflect the views of the Federal Reserve Bank of Philadelphia, the Federal Reserve Board, or the Federal Reserve System. Any errors or omissions are the responsibility of the author. No statements here should be treated as legal or investment advice.
Does age matter for credit access? Due to lack of data, this question remains largely unanswered. The question is potentially important for policymakers because the US population is aging quickly. To the extent that credit access is important for people’s quality of life, knowing the answer to this question may be important for policies that government agencies wish to make in the future.
In a new paper, I use the 2018 to 2020 vintages of the anonymized Confidential Home Mortgage Disclosure Act (CHMDA) data to explore the empirical relationship between age and mortgage access. The CHMDA data set is a large data set of mortgage applications and outcomes that covers a large part of the US mortgage market. Starting in 2018, CHMDA began reporting borrower’s age, which gives me the opportunity to answer the open question posed above. Specifically, I use the data set to estimate the conditional correlation between borrower’s age and: (1) the probability that his or her mortgage application gets rejected and (2) the coupon rate charged on his or her originated mortgage.
In the first part of the paper, I use a sample of 5 million single-borrower refinance mortgage applications to study the relationship between applicant age and rejection probability. I focus on single-borrower applications because for applications that have two borrowers, it is unclear whose age should be the economically meaningful one. I focus on refinance mortgage applications because the selection bias that stems from homeownership status is likely to be less severe than in a sample of home purchase mortgage applications.
To estimate the conditional correlation between borrower’s age and rejection probability, I run a linear regression where the outcome variable is an indicator variable that equals one if the application was rejected and the main explanatory variable of interest is a vector of age group indicator variables. The age groups are: 18 to 24, 25 to 29, 30 to 39, 40 to 49, 50 to 59, 60 to 69, and 70+. The first group is used as the reference group. The regression controls for applicant (e.g., income, credit score, debt-to-income ratio, loan-to-value ratio, and more), loan, and property characteristics. I also include time and location fixed effects to absorb variation in mortgage application outcomes across time and geographies.
The baseline regression results show that there is a generally positive relationship between applicant age and reject probability. Applicants in the 30 to 39, 40 to 49, 50 to 59, 60 to 69, and 70+ age groups are 0.5%, 1.3%, 2.4%, 3.5%, and 5.5% more likely to be rejected, respectively, than applicants who are in the 18 to 24 age group. These economic magnitudes are large when compared to the sample’s unconditional rejection probability of 17.5%. The baseline pattern holds within lenders, which suggests that the result is not driven by differential sorting across lenders; that is, it is not the case that older borrowers are more likely to apply for mortgages at stricter lenders. The pattern holds when I drop observations from 2020 due to concerns related to COVID-19 and the pattern also shows up across loan types: conforming, non-conforming, and government guaranteed loans.
Since the literature on unequal mortgage access has largely been focused on differences in access across racial and ethnic groups, it is interesting to compare estimates of the age effect to those of the race and ethnicity effects. Previous works find that black and Hispanic applicants are 1-2% more likely to be rejected than white applicants. The age effects discussed above are hence comparable to the race and ethnicity effects documented in the literature. Therefore, taken at face value, the baseline results suggest that age is a comparably important determinant of mortgage access as race and ethnicity.
The new vintages of the CHMDA also gives the reasons why each application was rejected. Using this new information, I find that similarly to the baseline rejection pattern, older applicants are more likely to be rejected because of “insufficient collateral.” In the context of a mortgage refinance, an applicant is likely to be rejected due to insufficient collateral when the appraised value of the property comes in too low relative to the requested loan amount. This situation could occur if the property value decreased substantially between the time of initial purchase and the time of refinance request. The result suggests that older applicants, for whatever reason, are more likely to be associated with lower quality collateral. This result is in-line with the work by Campbell and co-authors, which found that houses that were sold because their elderly owners unexpectedly pass away received substantially larger discounts and conjectured that inability to maintain one’s home in old age could be a potential explanation for the discount.
In the second part of the paper, I use a sample of 1.7 million home purchase mortgages and 1.1 refinance mortgages that were sold to government sponsored entities (GSEs) to study the relationship between borrower age and coupon rate. I follow the identification strategy used by Bartlett and co-authors, which relies on the GSEs’ loan-level pricing adjustment (LLPA) schedule. The logic behind the strategy is that the coupon rates on loans that were originated to be sold to the GSEs should be solely determined by the LLPA schedule. Therefore, conditional on the schedule, any statistically significant differences in coupon rates across demographic groups can be interpreted as being caused by factors other than unobservable credit risk.
In practice, the aforementioned identification strategy is implemented by estimating a linear regression where coupon rate is regressed onto the demographic variable of interest, along with LLPA schedule fixed effects. Effectively, the regression is comparing coupon rates across loans that belong to applicants of different demographic groups but fall in the same LLPA schedule cell. Using this identification strategy, I find that there is a generally positive relationship between borrower age and coupon rate. This pattern holds for both home purchase and refinance mortgages within lenders in non-COVID-19 years, and once mortgage points have been accounted for. Like the rejection results, the age effect is comparably large to the race and ethnicity effect documented by previous studies.
The main findings discussed thus far are that older mortgage applicants seem to face higher barriers to access because for refinance mortgage applications, they face higher rejection rates, and for home purchase and refinance mortgages that were sold to the GSEs, they pay higher coupon rates. The rejection results are conditional correlations and not causal statements. Therefore, they should be interpreted as potentially being driven by selection bias and any combination of supply and demand-side mechanisms.
Selection bias is a significant concern and could potentially play a large role. It is rare for young individuals to buy a house, let alone refinance a mortgage. Therefore, the younger applicants who appear in the sample are likely to be of very high credit quality, beyond the observable characteristics that I can control for. On the other hand, the older applicants who appear in the sample are likely to be of low credit quality because anecdotally, would-be retirees prefer not to carry debt into retirement. The combined effect of these two selection issues could give rise the baseline rejection result.
Age-related mortality is another potential explanation. In the US, when a borrower dies, his or her mortgage is either (1) prepaid through asset sale, (2) assumed on to heirs, (3) or put into foreclosure. In this light, death introduces uncertainty in loan performance to the lender. Therefore, a rational and risk-averse lender should price age-related mortality risk properly. Age-related mortality risk fits several of the findings in the paper. In particular, rejection rates increase with age and the increase accelerates in old age. This pattern lines up well with the way in which probability of death within one year evolves with age. Other potential explanations include unintended consequences from demographic-blind statistical models and taste-based age discrimination.
The coupon rate results are also conditional correlations and not causal relationships. However, if the LLPA schedule identification strategy holds, then selection bias can be eliminated from the list of potential mechanisms. Therefore, the results should be interpreted as being driven by any combination of the following mechanisms: differences in shopping behavior across age group, market segmentation and unequal degrees of competition, differences in coupon rates and points menu offerings, and taste-based age discrimination. Other mechanisms not listed can also contribute.
Two important caveats need to be discussed. First, the results do not necessarily show that lenders use age to make lending decisions because the correlations discussed above are not informative about the underwriting models that lenders use. To be able to make such statements, a fair lending analysis of an individual lender’s activities, which is not an accurate description of the analyses presented in the paper, needs to be performed. Second, since the correlations presented above are not necessarily informative about the variables that are considered in lenders’ underwriting models, it follows that the results do not indicate whether or not the lenders included in the study are legally or illegally using borrower age to make lending decisions.
To conclude, I show that, in large parts of the US mortgage market, age is positively correlated with rejection probability and coupon rate. The results suggest that, for many important mortgage products, older individuals face higher access barriers. Furthermore, the findings suggest that age is a comparably important determinant of mortgage access as race and ethnicity. I hope that the current paper will be a starting point for a rich literature that explores why age matters for credit access and, as the US population continues to age, an informative piece of research for credit-related policies that may be made in the future.
*Natee Amornsiripanitch is a Senior Financial Economist at the Federal Reserve Bank of Philadelphia.
