Bowling alone, buying alone: The decline of co-borrowers in the US mortgage market☆

What is the role of co-borrowers in determining the stability of the mortgage market? While a large literature has documented the relationship between borrower financial and demographic characteristics and the stability of the mortgage market, the role of co-borrowers is largely overlooked.² This paper provides stylized and empirical facts associated with co-borrowers in the US mortgage market since the early 1990s. We document that the change in the co-borrower share is heterogeneous across local mortgage markets. Further, this heterogeneity has implications for the performance of local mortgage markets — default rates, house price growth and credit growth.

The paper unfolds in three steps. First, we document changes in the regional distribution of mortgages with a co-borrower. It is known that the co-borrower share – the share of mortgage originations with a co-borrower relative to all mortgage originations – declined in the US since the 1990s (Tzioumis, 2017). However, we show that there is a spatial heterogeneity associated with this decline. Further, this heterogeneity is time-varying. For instance, at the county level, the coefficient of variation of the share increases three-fold since 1992, reaching its peak in 2006 and then drops sharply with the onset of the financial crisis.

In the second step we focus on the role of co-borrowers for mortgage market stability. The presence of a co-borrower significantly decreases the default probability, when compared to mortgage originations that have similar observable characteristics but do not include a co-borrower. Using Fannie Mae and Freddie Mac mortgage credit performance data, we show that the co-borrower’s presence reduces the mortgage default probability by more than 60 percent, relative to the average mortgage default probability.

Intuition for the finding – the co-borrower’s presence reduces the mortgage default probability – comes from the double-trigger hypothesis literature. Recent studies have established that households do not default ruthlessly on their mortgages as predicted by models based on option theory (Kau et al., 1992, Kau et al., 1994), but when the event of negative home equity coincides with adverse life events, e.g. an unemployment shock (Schelkle, 2018, Bhutta et al., 2017, Campbell and Cocco, 2015). Under the assumption that most mortgages with a co-borrower are taken by individuals who are bound together by some relation, for example they are married or co-habiting couples, the incidence of double-trigger defaults will be less likely due to within-household risk sharing.³

We also show that the performance differential for mortgages with and without a co-borrower varies over time. The discovered time pattern supports the double-trigger hypothesis. The hypothesis predicts that, conditional on home equity, the default rate for mortgages without a co-borrower, relative to mortgages with a co-borrower, will increase during economic downturns: a higher incidence of unemployment shocks during economic downturns would make mortgages without a co-borrower more susceptible to default. We analyze the default differential in different years by focusing on mortgages from the same origination cohort and controlling for loan-level and local mortgage market characteristics, so that levels of home equity would be similar. We find that the difference in default probabilities was the highest for mortgages issued just before the onset of the crisis. Mortgages issued in the early 2000s were 1.5 percentage points more likely to default, if they did not have a co-borrower. However, mortgages issued without a co-borrower just before the crisis were 3.5 percentage points more likely to default than mortgages with a co-borrower. Since more than 50 percent of issued mortgages in 2005 and 2006 did not have a co-borrower, the default probability difference of 3.5 percentage points has significant implications for the mortgage market during the crisis.

After establishing the relationship between the presence of a co-borrower and the mortgage default probability, we show that local mortgage markets with a low co-borrower share have higher default rates. We construct a measure of mortgage credit performance at the local level and find that a one-standard deviation increase in the co-borrower share reduces default measures at the county level by 60 percent of their standard deviation. Hence, the heterogeneous decline in the co-borrower share across the US has led to regional disparities in mortgage default rates.

After documenting stylized facts and the relevance of co-borrowers for the mortgage default probability, in the third step we establish the implications of a low co-borrower share for local mortgage market outcomes. We focus on growth in house prices and growth in purchase and refinance mortgages. Using the difference-in-differences framework, we show that counties with a low co-borrower share prior to the financial crisis experienced lower house price growth, lower refinancing growth and lower growth of credit extended to purchase mortgages. House price growth was lower by one-third of the standard deviation. Other outcomes differed by less than one-third standard deviation, but the difference in all outcomes persisted for three years. We also check the robustness of this result by using the difference-in-differences model in conjunction with propensity score matching.

Higher default rates in regions with a lower co-borrower share suggest a potential explanation for why house price growth and credit growth would be lower too. A large literature shows that foreclosures depress neighborhood prices, e.g. Harding et al., 2009, Campbell et al., 2011, Anenberg and Kung, 2014, Gerardi et al., 2015. Mian et al. (2015) establish that foreclosures accelerate the fall in house prices, consumption and investment. Further, the fall in house prices erodes home equity and limits refinancing opportunities (Beraja et al., 2019).⁴ Thus, higher default rates in regions with a low co-borrower share could have created spillovers on house prices and other economic outcomes through fire sales. Unlike these studies, we do not prove that higher foreclosures drive the decline in local outcomes. However, our results are robust to controlling for local demand effects, aggregate trends in lending standards and state-level reforms.

Related literature: Besides the double-trigger literature, we contribute at the junction of three related fields: regional divergence in the US mortgage market, innovation in the mortgage market and changes in mortgage credit allocation before the financial crisis. A growing literature documents regional divergence within the US mortgage market, especially after the financial crisis, and its implications for the redistribution of resources as well as the effectiveness of monetary policy. Mian et al., 2013, Mian and Sufi, 2014 uncover how the geographic distribution of home equity creates significant wealth effects and tightens credit constraints, leading to changes in local demand, employment and access to credit. Kaplan et al. (2020) obtain similar estimates for the non-durable consumption response and argue that the wealth channel alone is sufficient. Beraja et al. (2019) provide evidence that regions with more deteriorated household balance sheets refinance less even in the environment of low interest rates. Our results point to the co-borrower share as yet another factor that can explain regional divergence within the US mortgage market.

Foote et al. (2018) document technological changes in the US mortgage market. Since the early 1990s, the advancement in information technology, the widespread adoption of credit scoring models and automated underwriting exogenously affected loan processing time and credit allocation. Due to these changes, financial intermediaries relied more on credit scores and less on debt-to-income ratios over time. This paper shows that during the same period, while technological changes were altering the role of loan characteristics in underwriting, the share of mortgages with a co-borrower were also declining. Further, we show that this decline had additional implications for the stability of the mortgage market.

Our paper also relates to research that has analyzed changes in credit allocation in the US mortgage market prior to the crisis. Mian and Sufi (2009) note that mortgage lending standards relaxed and credit was inefficiently allocated to subprime borrowers.⁵ On the other hand, Adelino et al. (2016) argue that years before the crisis were marked with an increase in mortgage availability across the income distribution. While these papers provide evidence about efficiency, or lack thereof, in allocation of mortgage credit before the crisis, we show that there was another dimension of credit allocation which has been overlooked until now. While mortgages without a co-borrower are more likely to default, a larger section of the market was increasingly captured by these mortgages and lending institutions failed to price the default differential (Tzioumis, 2017).

The remainder of this paper is organized as follows. In the next section we focus on data description and provide summary statistics. In Section 2 we present stylized facts related to the change in the share of mortgages without a co-borrower. Section 3 establishes the importance of co-borrowers for the stability of the mortgage market and Section 4 focuses on the regional implications of the co-borrower share. Section 5 discusses and concludes.