Boom and bust cycles are recurring features in housing markets. These episodes are often associated with significant expansions in the amount of mortgage debt households incur. The mortgage crisis of 2008 was no exception, with a near doubling of outstanding household mortgage debt between 2000 and 2008 in the United States. In this paper, we ask what role relaxations in credit constraints play in amplifying or even causing house price bubbles.
The Role of Credit Standards
Access to credit is typically constrained across two dimensions. First, lenders impose a debt-to-income (DTI) constraint to account for the borrower’s ability to make his or her mortgage payment. Second, since borrowers cannot guarantee repayment, lenders require collateral. How much they are willing to lend against that collateral gives the loan-to-value (LTV) constraint.
LTV constraints and DTI constraints are distinct objects and need to be considered separately because whether one, the other, or both relax has different implications for house price cycles, monetary policy transmission, and foreclosure waves.
Finding 1: Stability of the Aggregate CLTV Distribution
The housing boom of the early 2000s saw mortgage credit to income rise significantly across all income groups in the United States. It is often assumed that this relaxation of DTI constraints went hand in hand with a relaxation in LTV constraints.
To investigate the validity of this assumption we analyze deeds records from across the United States. This data source has three important advantages. For starters, second and third loans—also called piggyback loans—are included. This allows us to calculate a combined-loan-to-value (CLTV) ratio that reflects all the money not in the down payment. For example, if a buyer puts 5% down and gets two loans, one for 80% and a second for 15% (the piggyback loan), we say that their CLTV ratio is 95%. Second, we observe a purchase price for all sales transactions. In fact, we use data only from sales, and omit refinances, precisely because we know the true value of the home when properties are sold. Lastly, we observe loan and sale amounts for the near universe of mortgages, not just those in private-label securities or originated by one lender.
We document that the distribution of CLTVs for purchase mortgages, which includes all the liens on a property and avoids potential bias in appraisals that would affect measurement of CLTVs for refinances, remained unchanged across home buyers between 1996 and 2015. In other words, the distribution of dollars of debt per dollars of housing collateral at purchase has not moved.
Of course, loan sizes at origination increased significantly with the housing cycle (driven by changes in home values), but the CLTV (loan size divided by home value) distribution did not become more skewed towards very high-CLTV loans.
In short, the narrative that a flood of low-down payment buyers entered the market during the housing boom does not fit the data. Rather, CLTV ratios remained constant and DTI constraints relaxed.
Finding 2: A Shift in the Role of Government Programs
Next, we show that the steadiness of CLTV ratios masks a dramatic shift in the source of high-CLTV loans.
Government programs designed and operated by the Federal Housing Administration (FHA) and the Veteran’s Affairs (VA) seek to encourage homeownership among their target populations: low-income, middle-income, and first-time homeowner households for FHA and military families for VA. The underlying argument for the existence of the FHA and VA programs is that they grant qualifying borrowers access to mortgage credit that they would not normally have access to in a world where mortgages are provided only by private markets.
We show that, consistent with the common narrative, the share of privately securitized high-CLTV loans for home purchases increased substantially during the housing boom. In 2000, 8% of all purchase loans were private-sector loans with down payments less than 5% and this share increased to 30% in 2006. However, highly leveraged loans had always been available. Indeed, once FHA and VA loans are included, approximately 30% of all purchase loans were high-CLTV even back in 2000. The big change between 2000 and 2006 was simply a switch in whether high-CLTV mortgages were explicitly backed by the government or privately securitized.
After the bust, high-CLTV loans remained available and highly utilized when government programs with a mission to provide access to low down payment mortgages stepped back up to the plate.
Finding 3: Stability of the Users of High-CLTV Loans
The natural next question is whether the steady aggregate usage of high-CLTV loans in the United States over the time series masks changes in which borrowers were utilizing these products.
It could be, for example, that government-guaranteed loans went to the Midwest before the boom and then, during the boom, the private-sector’s high-CLTV loans went to borrowers buying homes in Arizona, California, Florida, and Nevada. However, we show this was not the case. ZIP code-by-ZIP code utilization of high-CLTV loans stays very steady, just the source of these loans changes.
It could be that government-guaranteed, high-CLTV loans went to middle-income borrowers pre- and post-boom, but during the boom, the private sector made high-CLTV loans to low-income borrowers. But this was not the case either.
We use a repeat-sales methodology to show that the same specific properties are purchased with high-CLTV loans over the time series. Voter records in North Carolina allow us to follow movers and document that the same specific borrowers use high-CLTV, government-guaranteed loans pre- and post-boom and private-sector, high-CLTV loans during the boom. Furthermore, we see no reallocation of high-CLTV loans from low house price-growth ZIP codes to high house price growth ones, from recourse states to non-recourse states, from small homes to large homes, from old homes to new homes, from low-DTI borrowers to high-DTI borrowers, or from pessimists to optimists.
This last finding is particularly informative as it speaks directly to the leverage cycle model’s prediction that greater access to high-CLTV loans allows more optimistic buyers to bid up asset prices. We do show, reasonably, that more optimistic borrowers use high-CLTV loans in all periods. But the fraction of optimists with high-CLTV loans did not increase significantly in the boom.
We next look for evidence of a smoking gun that private-sector, high-CLTV loans went to different borrowers than their government-guaranteed counterparts. If the private-sector loans were different in unobservable ways, we would expect to see differences in performance, especially in the crisis, but this is not what we find. Instead, we find that their delinquency rates are very similar, even after controlling for geography and credit scores.
In short, the same people and places used high-CLTV loans between 1996 and 2015, and only the source of these loans—private-sector or government—changed.
Finally, we show that the shift to privately supplied high-CLTV loans follows the growth in house prices rather than leading it. This is inconsistent with models that predict house prices rose once credit access expanded.
Our results raise important questions about how to model credit cycles. We show that highly levered loans were available throughout the last two decades, suggesting that the housing boom and subsequent financial crisis cannot be explained by changes in aggregate purchase CLTVs. Instead, lenders expanded credit in proportion to rising house prices and allowed DTI ratios to go up across the population.
Our results suggest changes in collateral values (credit cycle models) or broad changes in house price expectations were at the heart of the last housing bubble. We do not find that the financial system loosened LTV constraints beyond the pre-boom period. But one might argue that these LTV ratios already started at very high levels pre-boom due to the role government programs like FHA and VA play in the US housing market.
The exact nature in which loans are guaranteed may be important in understanding how declining house prices affect financial markets and ultimately the economy.
Double trigger models of default explicitly consider separately the roles of debt overhang (corresponding to CLTV) and ability to pay (corresponding to DTI) as triggers of mortgage default. In light of our results, macroprudential regulations that seek to cap LTV ratios may prove less impactful than those limiting DTI ratios.
In addition, the fact that high-CLTV loans went from being explicitly guaranteed by the government during some periods (through the FHA/VA programs) to being privately securitized, and therefore no longer explicitly guaranteed, may have implications for overall financial stability. Private securitization might create misaligned incentives to underwrite risky mortgages, especially if participants expect implicit government guarantees through too-big-to-fail or other government backstops.
Manuel Adelino is an Associate Professor of Finance at the Fuqua School on Business, Duke University and NBER Research Associate. Email: email@example.com
Ben McCartney is in Assistant Professor of Finance at the Krannert School of Management, Purdue University. Email: firstname.lastname@example.org
Antoinette Schoar is the Stewart C. Myers-Horn Family Professor of Finance at MIT, Sloan School of Management. Email: email@example.com
This post is adapted from their paper, “The Role of Government and Private Institutions in Credit Cycles in the U.S. Mortgage Market,” available on SSRN (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3608150)