J Real Estate Finan Econ https://doi.org/10.1007/s11146-018-9659-y
A Farewell to ARMs or Ever Changing Market Segments? Bing Chen 1 & Frank P. Stafford 2
# Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract The mortgage market has given rise to a changing and diverse set of borrowers actively using ARMs. Data from the Panel Study of Income Dynamics (PSID) for 2007 and 2013–2015 are used to study borrowing decisions. One view is that an ARM should be offered and taken by those better able to respond to an upward reset. Yet favorable economic conditions induce a demand for mortgages, including by higher risk borrowers, and for these households transactions occur at higher rates, often as ARMs, especially as of 2007. Panel analysis confirms some response to the spread but also to changing demand for mortgages in the shorter run. During the boom, the use of ARMs as a tool for ‘affordability’ led to actual transaction rates exceeding those for fixed rate mortgages. Analysis confirms substantial payment difficulties. Yet, analysis of mortgage transitions, 2007–13, establishes that the ‘affordability’ component to ARM, though less significant, still was present. ARMs were often taken by minorities and those with less education and with family income under $60,000 per year. Keywords Consumption . Liquidity Constraints . Housing Demand . Mortgages JEL Classification G1 . E21 . R21
* Frank P. Stafford
[email protected] Bing Chen
[email protected]
1
Shenzhen Audencia Business School - Shenzhen University, Guangdong 518060, China
2
Department of Economics and Institute for Social Research, University of Michigan, Ann Arbor, MI 48109, USA
B. Chen, F. P. Stafford
Introduction The reliance on adjustable rate mortgages varies dramatically across economies, ranging from about 90% in Australia1 to under 10% in recent years for the United States (Badarinza et al. 2017). Within U.S. mortgage markets we analyze the distinct components. The mortgage choice process is both changing through time and is diverse at a point in time. Matching of diverse borrowers and lenders is a central feature. Our analysis confirms the choice of adjustable rate mortgages (ARMs) versus fixed rate mortgages (FRMs) by those better able to tolerate risk as measured by net worth excluding home equity. Yet there is a shift to more limited net worth ARM borrowers in the short run as the economy is expanding and home prices are rising. Lending differs substantially over time, creating distinct regimes and market segments. Notably, during the period just prior to the 2008–09 recession, the pattern of ARMs held by those with substantial nonhousing net worth was reversed. Lower net worth families seeking ‘affordability’ or anticipating additional price appreciation were then more likely to match to a lender offering an ARM. As the recession set in, these families often experienced repayment difficulties. Our analysis supports the role of experience and learning, as more educated families were much less likely to transition to an ARM, 2007–2013. History matters. Subsequent to the formation of the Home Owner’s Loan Corporation in the Great Depression (Fishback et al. 2013), the primary form of home mortgages in the U.S. was a fixed rate mortgage with a repayment period of 15 or 30 years. Mortgages with 20% equity at the time of initiation have been commonly exempt from a rate premium or mortgage insurance. With the onset of double digit inflation 1979–1981, the value of outstanding fixed rate mortgages fell, and the concern for future inflation risk on new fixed rate mortgages rose sharply.2 In this setting of surprisingly high inflation and primary reliance on fixed rate mortgages, Congress passed the Garn - St Germain Depository Institutions Reform Act of 1982. This act deregulated the savings and loan associations and allowed banks to provide adjustable rate mortgages. Immediately following passage of the act, a rebalancing occurred, and a high share of new mortgages was adjustable in the period, 1985–1989.3 The ARM share followed the spread between fixed and adjustable rate mortgages quite closely.4 From 1990 to 1995, the share did follow the spread but less so, and the share of ARM originations fell to a low level with some recovery in 1995. Based on data from the Mortgage Bankers Association, Bthe highest share recorded in the Weekly Application Survey, which goes back to 1990, was 36.6 percent in March of 2005… the historical average for the ARM share is 14.4 percent.^ (Fisher and Kan 2015). During 2004–2006, loans were often subprime and offered by Shadow Banks. In this period, mortgages were often securitized, limiting the risk to the initiating lender.5 In Australia the ARM rate is set and reset directly by the ‘cash rate’ of the Reserve Bank. The expected change in prices during the next five years as reported in the University of Michigan Survey of Consumer Attitudes fell from 10% in 1980 and remained between 4 and 5% all the way until 1991. Since 2009 the expected rate of inflation has been 3% or lower. 3 One analysis suggests that the shift to ARMs in 1980’s can be explained primarily by the rate differential rather than compositional responses on the supply side. See Brueckner and Follain 1988. 4 Michelle Clark Neely, BHomebuyers Bear ARMs in the Mortgage Market,^ Federal Reserve Bank of St. Louis, January, 1995, The Regional Economist, January, 1995 p. 12–13. 5 In the securitization a central theme is the level of integrity with which the underlying loans are characterized. 1 2
A Farewell to ARMs or Ever Changing Market Segments?
Many of the mortgages in this time window were issued to families with a temporarily good labor market and income position, and in markets experiencing the most rapid appreciation, namely urban markets with inelastic housing supplies in California, Arizona, Florida and Nevada (Mayer et al. 2009, p. 47; Mian and Sufi 2011).6 Subsequently, many of these families were unable to meet their consumption commitments, and holding a distressed mortgage was a strong predictor of subsequent liquidation of other assets, notably selling off stocks at sharply reduced prices (Chen and Stafford 2016). In the period immediately after 2009, the share of new originations which were ARMs fell sharply. The explanation seems to rest on the idea of the added risk of an expected ARM reset. An expected reset cost was not seen to be worth the small spread given the low nominal rates for both forms, implying a lower cash flow benefit for a qualified borrower. Also at work were the publicized difficulties with risky sub-prime ARMs. The extent of those seeking and finding risky mortgages declined notably as the recession set in. Later, during the period, 2011–2015, some recovery of ARMs occurred. One segment of the ARM market was the growth of borrowers with good credit taking out very large but ‘safe’ loans – certainly not to surmount the problem of ‘affordability’ which was often the claimed motivation for the subprime ARMs in the period of 2004– 2007.7 Rather, the interest rate savings from even a small spread are considerable if applied to a large balance8 and wealthy families, boosted by rising home and equity prices, became the dominant holders of ARMs in 2013 and even more so as of 2015. In this period, there was still a continued higher risk borrowing on smaller loans, notably by minority families, providing continued evidence of the bifurcated nature of the ARMs market. The organization of the paper is as follows: Section IA presents the variability and long term decline in ARMs based on the declining spread over the years studied with the micro data. The distinction between the long-run, elastic supply of mortgage funds versus the short-run, less elastic supply by market segment is developed briefly. The demand for ARMs appears to have shifted outward during favorable economic times and with a less elastic supply by segment, and the average ARM rate paid is then higher. In addition, the composition of demand shifts to borrowers with higher repayment risk, adding to the average transaction rate.9 Section IB sets out the matching aspects of the mortgage market, and Section IC discusses the central elements in mortgage choice. These diverse components of the mortgage market can be considered as the basis for a divergence between the long run supply of mortgage funds and the less elastic match-specific short run supply. Section II offers a discussion of the main features of the mortgage market as of 2007, prior to the housing market bust in 2008– 6
It has been argued that the precipitating factor for the housing downdraft was the bankruptcy reforms of 2005, which made foreclosure relatively attractive (Morgan et al. 2011) 7 According to a study for the Wall Street Journal, 31% of mortgages in the range of $417,001 - $1000,000 originated in the fourth quarter of 2013 were ARMs. In contrast, the average fixed rate mortgage was about $200,000. 8 This raises the question of how the funds distributed from these large mortgages at low rates end up being allocated. Is it consumption or the acquisition of public or private equity? 9 As argued by Bhardwaj and Sengupta (2012), an implicit feature of a high rate ARM can be that of a bridge over a short horizon during which the home price is assumed to increase. While this was an interpretation of the rise in subprime loans up to 2007, this can be in effect to a lesser degree during other periods.
B. Chen, F. P. Stafford
09, and then as of the recovery in 2013–2015. Section IIIA offers empirical analysis of choice in 2007 versus 2013, and includes a matching model to gain some insight on the supply side. The difficulties in repaying the mortgages is assessed in Section IIIB. Section IIIC offers analysis of the transitions across mortgage types, 2007–2013. Section IV concludes.
Pricing and Matching Concepts Responding to and Creating the Spread Over Time The U.S. mortgage market is best thought of as shaped by a diverse matching process. This is essential for understanding the market at a point in time and is needed for an understanding of the changing market composition through time. Though there are compositional shifts and a differentiated matching processes, one can still see the longer run demand for ARM versus fixed rate mortgages. On the assumption that mortgages are part of a wider market and one with international flows (Keller and Mody 2010),10 the relevant longer run supply curve to mortgages by term structure can be considered to be highly elastic,11 allowing the estimation of longer run demand side responses to the ARM-FIXED spread. Moreover, with securitization there is less local supply side concern about holding a particular mix of mortgage types. The ARM and FIXED rates and change in GDP are set out in Figure 1. These two reference rates are in some sense the ‘list prices’ for a well-qualified borrower. In the mortgage matching process, the transaction rates will have a mark-up for borrowers with a higher risk of default, and this was especially true during the 2004–2006 housing boom. Then many first time buyers financed with a high rate subprime ARM even as the ‘listed’ ARM rate for a safe mortgage was at its post-1980 low. Of note is that the ARM rate has generally risen after some time during economic expansions – with the strong growth in the mid 1980’s, after the Gulf War in the mid 1990’s, during the latter part of the .com boom and during the subsequent expansion up to 2008. This rise can be interpreted as shaped by both short run supply and sorting. Even with securitization, the mortgages presumably need some more local supply of origination services, especially for riskier ARM loans which are less often securitized, and this will induce some inelasticity. As the expansion continues, the rate rises more rapidly for an ARM, inducing a smaller spread along with greater loan volume. This can explain why, in studying the role of the spread on the choice of an ARM, the predictive power is greater when the ‘mortgage rule of thumb’ is used (Moench, Vickery and Aragon, 2010). That measure is the spread between the current thirty-year fixed rate and the average of the one-year rate over the prior three years. This is premised on the idea that the three year experience on ARM rates is what affects the choice. Yet it may be more related to a rising demand pushing up the ARM rate faster than the FRM rate, narrowing the spread. 10
This is presumably enhanced by the strong developments in securitization from 1990 forward. This permits the diversification of local market risk and leads to potential indifference of mortgage originators the ARMFRM balance of their lending. 11 One estimate of the global debt market is $100 trillion as of mid-2013, while the approximate value of U.S. mortgage debt on 1–4 family residences is approximately $10 trillion.
A Farewell to ARMs or Ever Changing Market Segments?
Fig. 1 Mortgage rates and the rate of real GDP growth, 1984–2014
Cross-Sectional Mortgage Matching Mortgage decisions have several elements and at the family level or demand side are shaped by diverse factors. Setting out a plausible model of the FRM-ARM choice quickly leads to reliance on simulation of the basic patterns (Campbell and Cocco, 2003). Here we characterize the process as one of matching borrowers and lenders in terms of the transaction rate and the expected default cost. From observed choices we can see the likelihood of subsequent mortgage problems in terms of repayment or even foreclosure. These repayment risk costs and the rate of interest are central, underlying characteristics (Kelvin Lancaster 1966) for both sides of the market - though of opposite sign. In a matching and sorting perspective (Sherwin Rosen 1974; Stephen Salant 1977) the attributes are the base rate and expected repayment cost to the lender. The lender repayment cost can be thought of as including any subsequent unexpected rate increase absorbed in a fixed rate contract.12 The ability of the borrower to repay is a function of risk of default for those who are unable to meet their mortgage and other consumption commitments. Anticipated ability to repay constitutes another central component of cost. While there is great heterogeneity in the features of a mortgage – and presumed matching and sorting of diverse borrowers (B1, B2) and lenders (L1, L2), here we illustrate two central elements or characteristics in the matching – that of the base rate (r/$) and the expected lender share of the cost of the repayment not being recovered (P/$). An ARM can have a lower base rate paid to the lender but also has a lower expected lender cost of lost loan recovery if rates subsequently rise.13 On the other hand, an ARM may be sought by some borrowers who are viewed as having a higher repayment risk14 and while this gets built into the transaction rate, it can instead be considered a component of repayment cost (P/$).15 The overall loan market appears to 12
With a rate movement downward, the borrower has the option to refinance at a lower rate, but with a fixed rate the lender experiences a loss of bond capital if rates rise beyond those anticipated. 13 On the other hand, a recovery cost to a lender in a FRM is the lost return when, as rates decline, the borrower exercises the option to refinance to the new lower rate. 14 This applies to ARMS characterized as ‘sub-prime’ and ‘non-conforming’. 15 The choice of an ARM, while lowering the lender’s reset risk from an unanticipated rise in rates, adds to the lender’s recovery cost if the borrower default probability is thereby increased.
B. Chen, F. P. Stafford
be subject to shifts over time in terms of lending leniency. In addition, the choices set out in Figure 1 can be thought of as the difference between a safe FRM (L1-B1) and an ARM with a higher default risk (L2-B2). For the lender there is a profit function increasing in both the rate per se and lower loan recovery cost. Low equity and a low credit rating of the borrower can add to the expected recovery cost. The two elements of the rate and the cost of recovery may be thought of as concern over both the ‘return on my money and return of my money’ as the two arguments in the lender iso-profit function. For the borrower, the gain function is decreasing in both the interest rate and the share of any loss recovery borne by the borrower. The borrower wants cheap money and lender absorption of loss in event of various downside events. As set out in Figure 2, a lender (or lender with companion securitization entities) (L1) in exchange for bearing more repayment cost (P/$) requires a relatively higher rate (r/$). Other lenders (L2) are willing to take on a given increased repayment cost in exchange for a smaller incremental rate. One can say that L1 is repayment biased and that L2 is rate biased. These repayment biased lenders are those who have less ability to assess risky borrowers and hold safe but low interest loans. Others, such as the rate biased, are better at expeditiously processing borrowers with high repayment risk costs – such as those loans to families with limited non-housing wealth choosing an ARM. The presumption is that there are diverse borrowers and lenders, so the borrower’s relative willingness to accept a lower rate in exchange for bearing more repayment recovery (as with an ARM to a family with more non-housing wealth) will attract and match to those lenders that expect low recovery costs in exchange for a lower rate. Consider those seeking a large mortgage. For them, even as the spread narrows the interest differential can lead to substantial savings via an ARM, a pattern which is observed in both aggregate and in our micro-data, especially for the most recent period. For those having high and stable income and substantial non-housing wealth which can be used to maintain their repayment commitment, these ARM borrowers will have expected repayment costs to the lender (P/$) which are low. These ARMs will have a mortgage match with a low rate in compensation for a low cost of recovery to the lender. Borrowers can be thought of as having a reservation utility. For some this may pre-empt them from participating in the mortgage market and they instead rent or own their home outright. They can be thought of as having a reservation utility, URB, to the right and below B1 or B2. On the lender side there are reservation expected iso-profit contours to the left and above L1 and L2, URΠ. As rates fall in credit markets there will be new entrants and a new competition for borrowers across the diverse lenders. At the same time borrowers at both the intensive and extensive margins of URB will reconsider their possible choices in matching to a lender or not. The matching process includes a wide range of variables that borrowers care about and is differentially offered on the lender side. There is not a market clearing price but a market rate and mortgage attribute surface.16 The repayment risk premium paid by the less reliable borrower would be still higher were there not an underlying matching and This includes lenders such as the Bank of Internet, which ‘lends boldly’ using ARMs and accepts higher risk borrowers, paying relatively high deposit rates, and is insured by the Federal Deposit Insurance Corporation. See BAn Internet Mortgage Provider Reaps the Rewards of Lending Boldly,^ New York Times, August 23, 2015.
16
A Farewell to ARMs or Ever Changing Market Segments? r/$
L2 L1 B2 B1
P/$
Fig. 2 Matching lenders and borrowers in the mortgage market
sorting process. Depending on the relative density of lenders and borrowers in the characteristics space, the locus can have a differential upward slope. The nature of the market, having numerous market segments, suggests that the supply of origination services to each segment is less than fully elastic in the short run. One group of borrowers is those early career buffer stock savers (Deaton 1992). While they may be willing to take on a mortgage with a high rate and a high repayment risk, no lender will match up to them. As they mature and improve their financial position, they may then find a lender offering a high rate ARM (an L2B2 match). Even here this can represent a substantial consumption commitment by the borrower and an elevated default risk. In the context of our multiple regime perspective, it is possible to consider some aberrant departures from one time period to another. Some of the decisions people make, such as those with substantial non-housing wealth taking ARMs, align with predictions from basic microeconomic choice models. Other transactions appear be shaped by naïve and contagious expectations (Fisher 1933; Iacobini 2009) about future home prices, as seems to clearly apply to some U.S. markets, 2004–2006. Further, some of the ex post outcomes may be well outside the range of ex ante expectations. In good times lenders and borrowers may become less sensitive to repayment risk and the market locus in Figure 1 can be hypothesized to flatten in terms of perceived ex ante repayment risk. Lenders will loosen standards, allowing otherwise risky borrowers to participate at lower rates. Yet, overall the transaction rates may rise as a larger share of the high repayment risk mortgages are taken. Policies to limit mortgages choices are primarily based on the assumption of inability of borrowers to understand fully the implications of diverse mortgage choices. Research shows that regulatory integrity and a range of housing market factors, including government policies and interest rates, can predict the onset and severity of housing market volatility (Agnello and Schuknecht 2011). However, some of the outcomes may be realizations of less or greater repayment risk than anticipated and be correlated across a substantial share of the pre-existing mortgages. In the empirical analysis below, we show the mortgage relations developed in the 2004–2006 period that became costly with repayment risk as the 2008–09 recession set in.
B. Chen, F. P. Stafford
Choosing a Mortgage: Central Factors Besides whether adjustable or fixed, the mortgage features include the loan to value (LTV) at initiation (TLTV for transaction LTV), how high the monthly payments are relative to permanent income, whether a second borrowing has been taken on – such as a home equity loan or a second mortgage, or whether there is a substantial line of credit loan. Many of the demand side choices group into the common element of seeking a loan at a low rate for the borrower and at the same time leading the potential lenders to seek a rate premium as a component of P/$ (a ‘recovery mark-up’ on the rate) as a result of repayment risk, as stylized in our simple matching model. There are factors that imply a mortgage with more or less repayment risks and corresponding rate variations. 1. On the supply side lenders will try to cover an expected default cost by a higher rate. In addition, during a period of strong housing demand they may expect to have a better chance of recovering the mortgage balance from a foreclosure sale. For this reason, the supply side caution may be reduced, leading to a greater share of risky mortgages from the perspective of both sides of the market. On the demand there are borrowing constraints as in the buffer stock models (Flavin and Yamashita 2002; Carroll 1994). Among these risky borrowers are those who anticipate notably higher income in future years as they progress through a labor market career (Deaton 1992). In this setting the dynamic value of a dollar today is very high, and they operate with a limited liquidity buffer. Their mortgage choice will be shaped by the desire to limit housing-related cash flow and to gain access to greater current consumption services. They have incentives to take on a mortgage with all of the risky elements. In the context of the matching perspective they may be unable to find lenders willing to take on such an agreement except at a prohibitive rate. The buffer stock perspective can also connect to issues of mortgage market discrimination. Is separation of the mortgage market as set out in Figure 2 based on objective risk and return criteria, or is there persistent discrimination which contributes to what may be called a dual mortgage market (Apgar and Calder 2005)? Notably, African-American families have held little in the way of non-pension wealth and often have limited holdings of liquid assets (Hurst et al. 1998, Table 14). Yet there is empirical analysis at the family and neighborhood level consistent with housing and mortgage discrimination. 2. Closely related to limited buffer stock savings is the role of consumption commitments. This can depend substantially on labor market earnings level and control (Dau-Schmidt 1992) or relate substantially to the extent and composition of financial asset balances (Chetty and Szeidl 2007). As part of protecting consumption commitments, families should be less interested in an ARM or other borrowing with high repayment risk. However, they may match to an ARM if they have substantial and stable income and balance sheet resources or are otherwise risk tolerant.17 Most mortgages in the United States have the option of full prepayment with no penalty. As a result, when mortgage rates fall below the rate on previously issued mortgages, many homeowners act to refinance. Refinancing options can be seen as either a consumption option (Hurst and Stafford 2004; Greenspan and Kennedy 2008) or a financial option. 17 As noted by Badarringza, Campbell, and Ramadoria (2017), a loss to a fixed rate mortgage holder occurs when rates rise. They have protection from the mortgage payment increase with and ARM but are not realizing the higher market returns if invested at the higher rates.
A Farewell to ARMs or Ever Changing Market Segments?
3. Those families financing or refinancing to a lower equity position are often seeking funds for current expenditures and taking a riskier repayment position. They may be willing to seek a mortgage with a repayment risk rate premium above their current rate. The range of current outlays can include possibly pre-existing housing consumption commitments (Wood et al. 2013). Such a cash-on-hand or consumption option often leaves the home owner more vulnerable to future difficulties with the added consumption commitments. If the family is ‘borrowing up’ to finance current or expected expenditure needs they may select an ARM to reduce monthly payments, but of the lender’s side there can be concerns over repayment risk (P/$). Such concerns are embodied in a rate premium – or in no loan transaction. 4. In contrast to the consumption option and borrowing up, those refinancing as a financial option are not seeking liquidity that may be obtained via a risky ARM. These financial option families simply act to realize a capital gain – the discounted stream of reductions in mortgage payments. 5. Speculative motivations can motivate a risky mortgage on the demand side, either from adverse selection of borrowers or unfounded positive price expectations along the lines of (Fisher 1933). These high leverage mortgages can occur in conjunction with the non-recourse mortgage regime in the U.S. 6. Experienced mortgage conditions can shape choice. A range of studies has shown that past experience may have a strong and persistent effect on expectations and choices (Malmendier and Nagel 2011; Ehrmann and Tzamourani 2012). Here we consider inflation expectations of those who were the young cohorts active in the economy in 1979–82. Then the 30 year fixed rate reached a peak of 18%. In contrast, throughout the period 2007 forward inflation has been low and the expected risk of reset to a notably higher rate has been correspondingly low. A simple model indicates a lower probability for those experiencing the period of double digit inflation.18 The factors are set out in Table 1 and apply differentially depending on the nature of the local housing market. In markets with limited and inelastic housing supply, the potential for price volatility is higher (Saiz 2008). On one hand, this should lead both borrowers and lenders to be more cautious and make choices which protect against sudden changes. On the other hand, as set out by Fisher, variability of prices and observed past increases can induce some borrowers and lenders to place highly leveraged bets. Lending may be at a high rate based on an ARM contract covering substantial default risk such as point B2-L2 in Figure 1. Without unfounded optimism on both sides of the market loans would not be made to cover such risky ventures. In this setting there can be home price bubbles. The different motivations suggested by theory can combine in their effects as they play out in the matching process. Depending on regime, the ARM mortgages have often been subprime for other reasons and carried a higher default risk. While the reported market or base rate for a given repayment risk group of borrowers is lower for an ARM, the repayment risk premium embodied in the mortgage transaction rate is higher, leading to a higher contract rate for such ARM mortgages. In terms of the matching framework set out 18
The model was limited by the absence of important covariates, such as net worth, as of 1996. Those age 35– 64 as of 1996 would have been age 20–50 as of 1980–81 and likely had to adjust to the inflationary circumstances. As found in other micro data studies (Coulibaly and Li 2009), families reporting moderately greater risk tolerance were more likely to have chosen an ARM.
B. Chen, F. P. Stafford Table 1 Mortgage demand side ARM choices and supply side offers and informing theory Motivating Theory
Demand Side
Supply Side
Net Notes
Borrowing Constraints
+
–
?
Early life course, future high income
Consumption Commitments Capacity– Education, non-housing wealth, stable income
+
+
+
Having sufficient post-mortgage resources
Financing and Refinancing Financial
?
+
?
Exercising the refinance option
Refinancing Consumption
+
? or -
?
Refinancing to produce cash and cash flow for expenditures
Fisher Price Boom
+
+
+
Suspension of awareness of ARM or other risk
Recall of History Effect
–
0
–
In the period after 1979–81 inflation memory
above, these ARM contracts should have a low base rate. Because of a preponderance of new, higher risk borrowers, especially during the strong economy of 2003–2006, and potential low short run supply elasticity of funds to the different market segments, the transaction rate on ARMs was higher (B2-L2 in Figure 1). Even so, it may have been simply unfounded home price growth expectations on both sides of the market which supported such matching at all, with such borrowers sorting into lenders prepared to make these higher risk loans. Another high risk group includes young consumption constrained families that may opt for a low down payment, and a low initial ‘teaser rate’ adjustable rate mortgage in line with optimization. Yet these choices are also the ones of a speculator who may have no liquidity concerns. Even for a home in the lower price range there is an important role for early life course borrowing constraints. A bifurcation of the market is based on an ARM as both a way to control interest costs by a family with a large mortgage and ample financial and income resources as well as by a family that is liquidity constrained.
The Changing Composition of Mortgages As can be seen in Table 2, the percent owning was greatest in 2007 before the financial crisis (64.1%) and by 2013 was noticeably lower (59.5%).19 The percent of owner families with a mortgage in 2013 was about where it was back in 1996.20 Of note is the strong decline in the share of adjustable rate mortgages, from 20.4% in 1996 down to 8.7 by 2013 and 7.4 by 2015.21 This is consistent with a basic time series assessment of 19
The percent owner in PSID for 2013 is lower than as reported from Census data such as the Current Population Survey. This is because those living with another family member will be recorded as ‘other’ in PSID while in CPS they will report on the tenure arrangement in the dwelling unit even if they are neither the owner nor renter per se. 20 The 1996 estimates are from weighted PSID data. 21 The full time series indicates substantial interim up and down movements of the spread and not a steady downward path.
A Farewell to ARMs or Ever Changing Market Segments? Table 2 Descriptive statistics of regimes, 2007 versus 2013–2015 Year
2007
2013
2015
Owner
64.1%
59.5%
58.8%
Renter
30.8%
34.8%
35.2%
Other
5.1%
5.7%
6.0%
67.4%
63.8%
63.4%
Conditional on owners Have mortgage
Conditional on having mortgage (either fixed rate or adjustable) % fixed rate mortgage holders
88.9%
91.3%
92.6%
% adjustable rate mortgage holders
11.1%
8.7%
7.4%
Fixed rate
153,870
144,518
142,989
Adjustable rate
190,839
158,195
175,397
Average first mortgage balance ($)
Average first mortgage rate Fixed rate
6.24
4.72
4.47
Adjustable rate
6.92
4.37
4.19
Fixed rate
19.7
19.2
18.4
Adjustable rate
21.2
15.8
17.2
Years to pay first mortgage
Net wealth without home equity ($) Fixed rate
327,363
225,785
258,999
Adjustable rate
311,472
385,082
617,890
Fixed rate
339,423
254,373
270,931
Adjustable rate
350,025
316,820
399,992
% with second mortgage
20.5%
14.0%
11.4%
Average house value ($)
Conditional on African Americans having mortgage (either fixed rate or adjustable rate) % fixed rate mortgage holders
85.2%
87.7%
89.7%
% adjustable rate mortgage holders
14.8%
12.3%
10.3%
Conditional on having mortgage (either fixed rate or adjustable rate) and took the mortgage after 1983 Fix-adj. Mortgage rate spread
1.6%
1.2%
1.3%
All the dollar values are 2013 values
a response to the spread. The list spread prevailing in the year the mortgages were taken out is reported as the last row of Table 2. With the general downward movement in interest rates, the zero lower bound and increased securitization have reduced the reported market spread at the time of taking their primary current mortgage. For families in our sample the ‘list’ spread had fallen from 1.6% in 2007 to 1.2% in 2013 and 1.3% in 2015.22
22
The market spread or base for our samples is calculated from the year in which the current first mortgage was taken out. That is, the reported spread in that year (the same source as used for the basic time series recessions – Figure 1).
B. Chen, F. P. Stafford
In contrast to the substantial spread as reported for a standard well-qualified borrower, the observed or transaction spread, comparing those with a FRM v ARM, is very small in both 1996 (based on a separate tabulation not in Table 2) and 2013. Specifically, while the market standard or base spread was 2.45% as of 1996, the transaction spread for borrowers was only 0.16% (8.18–8.02). For 2007, the average paid on an ARM rate was noticeably higher than on a fixed rate (6.92% versus 6.24%). We interpret this reversal as reflecting our matching perspective. A group of demanders for ARMs was part of the housing market boom, concentrated in selected urban markets. The desire to hold an asset that was expected to likely appreciate was matched to lenders better able to accommodate such mortgages. In the context of seeking greater access to mortgage funds, there was also an evident rise in percent of home owners with a second mortgage,23 which commonly has an interest rate premium. Elements of the matching theme and its changing composition can be seen in the wealth position of ARM mortgage holders. As the spread has declined, the ARM savings are substantial only if the mortgage balance is large and the family has sufficient wealth to tolerate a risk of reset. Notably, the average constant 2013 dollar net worth excluding home equity is much greater for ARM borrowers in 2013 ($385,082) when compared to fixed rate borrowers ($225,785). By 2015 the share of ARMS was 7.4%, but of those with an ARM mortgage, the non-housing wealth rose dramatically to $617,890. One reason is that those with an ARM as of 2013 had greater wealth gains by 2015 because they were more likely to be buyers of equities at the low point in 2009 (Chen and Stafford 2016). In addition, the perceived risk of resets and labor income variability discouraged contracts for ARMs for most families with lower wealth. In contrast, during the housing boom, the ARM was somewhat more attractive to those with less net worth exclusive of home equity. This attraction is consistent with matching these borrowers with lenders more adept at managing high risk loans. African-American families have held a fairly stable percent of ARM mortgages, though as will be shown in Section IIIC, there has been surprising mobility in which form of mortgage was held over the period 2007–2013.
Choosing an ARM, 2007–2013 Comparing the Two Years as Cross -Sections From Table 3, the factors predicting a choice of ARM over FRM differ substantially over time. For the ARM rate and the list spread (fix_rate-adj rate of mortgages), in the year the current primary mortgage was taken, the level of the ARM rate matters and the spread is statistically significant once allowance is made for whether the mortgage was taken in 2004–2006. The ARM rate (level) at the time the current mortgage was taken, generally corresponds to stronger growth, a high fixed rate as well, a modest spread, and larger ARM volume. The ARM rate has a substantial positive relationship to choosing an ARM. This we interpret as the result of increased short run demand for 23
Here the term second mortgage applies to any loan with the house as collateral, such as a home equity line of credit or a traditional second mortgage.
A Farewell to ARMs or Ever Changing Market Segments? Table 3 Whether took an ARM mortgage
Intercept
2007
2007
2013
−115.6
−3.7897***
−46.2457
(99.5011)
(0.6046)
(92.4353)
Mortgage Variables year_took_mrtg1_07 (13)
0.0558
0.0213
(0.0491) took_mrtg1_btw (2004–2006)
(0.0456) 0.6914*** (0.1366)
fix_rate-adj_rate of mortgages ARM rate
0.4407
0.3495***
0.4319
(0.3407)
(0.1233)
(0.2984)
0.3005*
0.2693***
0.32**
(0.1575)
(0.0597)
(0.1515)
−0.1174
−0.1192
−0.4149**
(0.1399)
(0.1407)
(0.1719)
−0.3251*
−0.3177
−0.4479**
(0.1945)
(0.1948)
(0.2247)
−0.3576***
−0.3457**
−0.2312
(0.1341)
(0.1347)
(0.1607)
−0.0365
−0.0198
0.2478
(0.1623)
(0.1626)
(0.1976)
−0.2299
−0.286
−12.201
(0.5949)
(0.5959)
(358.7)
0.1642
0.1513
−0.271
(0.2134)
(0.2142)
(0.2589)
0.2319
0.2541
−0.0652
(0.1861)
(0.1869)
(0.215)
−0.0738
−0.0796
−0.1188
(0.1791)
(0.1798)
(0.2028)
Income and Wealth 60 K < total_fam_income_07 (13) < =120 K total_fam_income_07 (13) > 120 K 10 K < wealth_without_home_equity_07(13) < =130 K wealth_without_home_equity_07 (13) >130 K constrain_07 (13) Region and Urban Status Northeast_07 (13) North Central_07 (13) South_07 (13) state_07 (13) in (AZ, CA, FL, NV) and urban_07 (13) = 1 state_07 (13) in (AZ, CA, FL, NV) and urban_07 (13) = 2 plan_move_07 (13)
0.5511**
0.5007**
0.3078
(0.2227)
(0.2237)
(0.2703)
0.8277**
0.7395**
0.5619
(0.3668)
(0.3688)
(0.4137)
0.4132***
0.3929***
0.3239*
(0.1271)
(0.1277)
(0.1689)
−0.2366
−0.2594
−0.1203
(0.2308)
(0.232)
(0.2742)
0.3767
0.3458
−0.246
(0.267)
(0.2681)
(0.2989)
0.3178
0.3188
0.1378
(0.2434)
(0.2452)
(0.2512)
Demographic and Education male_head_ 07 (13) age_head_07 (13) < =34 34 < age_head_07 (13) < =49
B. Chen, F. P. Stafford Table 3 (continued)
49 < age_head_07 (13) < =64 edu_head_07 (13) = 12 12 < edu_head_07 (13) < =16 edu_head_07 (13) > 16 African Americans_07 (13) employed_head_07 (13) employed_wife_07 (13) no_wife_07 (13)
2007
2007
2013
0.2782
0.2723
−0.0133
(0.2359)
(0.2375)
(0.2299)
0.1092
0.1054
0.0416
(0.1948)
(0.196)
(0.2415)
−0.0181
−0.0196
−0.1096
(0.1947)
(0.1955)
(0.2381)
−0.0341
−0.0371
−0.4719
(0.262)
(0.2627)
(0.3023)
0.3985***
0.3761***
0.0029
(0.1408)
(0.1412)
(0.1726)
−0.4208***
−0.403**
−0.0126
(0.1621)
(0.1633)
(0.1876)
−0.2833*
−0.2803*
−0.1689
(0.1511)
(0.1521)
(0.1722)
−0.1644
−0.1652
−0.0346
(0.2336)
(0.235)
(0.2685)
Number of observations
3098
3098
2983
AIC (intercept only)
2374.135
2374.135
1804.765
*, **, and *** denote the significant level of 10%, 5%, and 1% respectively Cases limited to first mortgages from 1984 forward when ARM rate data in the market became available. (Also for Table 7)
mortgages at both the intensive and extensive margins.24 This suggests that the demand for ARMs rises and particularly on the part of higher risk borrowers who match with a lender with willingness to offer riskier loans as represented in Figure 2. This corresponds notably to the rise in subprime lending to borrowers with greater repayment risk, especially up to 2007. It can be argued (Bhardwaj and Sengupta 2012) that the increased rate of ARM activity is premised on the idea of a bridge loan for property anticipated to increase in price by both borrower and lender. Income and wealth variables are in two groups of categoric variables. Families in the lowest annual income range (under $60,000 in 2007 and 2013) were more likely to choose an ARM. In 2007, the wealth pattern is one where both high and very low nonhousing wealth families were more likely to hold an ARM, a pattern consistent with our discussion of diverse matching concepts. By 2013 there was some relative shift away from ARMs being held by low wealth families and toward ARMs being held by high wealth families, as suggested by the data in Table 2. The variable which indicates a family headed by an individual with more than 16 years of education, under age 35, and holding no more than $10,000 in net worth (constrain_07(13)) shows no relationship to 24
Other studies of the spread in shaping mortgage choice (Campbell and Cocco 2015; Miles 2004) find a related pattern and offer the explanation as refinancing inertia rather than shifting demand for ARMs narrowing the spread.
A Farewell to ARMs or Ever Changing Market Segments?
holding an ARM. Our interpretation, as suggested in Table 2, is that while these families may have preferred a risky ARM, lenders are reluctant to offer an ARM at an attractive rate - if at all. In line with the subprime mortgage crisis and the issuance of ARMs in the volatile markets, we can see a strong impact of being in the urban markets (and surrounding suburbs) of cities in California, Arizona, Florida and Nevada as of 2007. Here the effects of a housing boom are at work. This is of interest as a distinct event which we believe has potentially shaped subsequent outcomes. The effect of being in the urban centers of Arizona, California, Florida and Nevada (selected urban markets - SUMs) are shown as very strong predictors on having an ARM mortgage as of 2007 but showing only a weak relationship as of 2013. This was part of a wider loosening of loan standards in 2004–2006. Data from the Federal Reserve of St. Louis (Federal Reserve Economic Data) indicate that as of 2003 Q1 22.0% of banks tightened standards for commercial and industrial loans to large and middle market firms. By 2004 Q2 it was −22.8% and in 2006 Q2 it was −12.3%. By 2007 Q1 it was 0.0% and by 2008 Q4 it was 83.6%. We would expect competition in the commercial loan markets would have carried over into mortgage lending, leading to a more general ‘loose credit window’ of 2004–2006. There is agreement that the FHA relied on lax underwriting standards in the run up to the mortgage crisis. In addition, there was emergence of securitization pools including a rising share of risky mortgages sponsored by private label securitization (Simkovic 2013). Based on data from the Mortgage Bankers Association, Bthe highest share recorded in the Weekly Application Survey, which goes back to 1990, was 36.6 percent in March of 2005… the historical average for the ARM share is 14.4 percent.^ (Fisher and Kan 2015). On this basis, we have added a 1–0 dummy indicator of taking mortgage in 2004–2006 to the 2007 model in Table 3. The relationship is significantly positive. Mortgages issued in this time window are those documented to have a high share of risky features, especially ARM mortgages. One unusual geographic variable is that of plans to move. This is asked in the PSID as part of an effort to follow families in subsequent waves. For a given borrowing and risk position, the ARM rate advantage should induce the choice of an ARM for those planning on a shorter contract duration mortgage. In both 2007 and 2013 those expecting to move in the ‘next few years’ are more likely to have had an ARM mortgage. To gain some insight on the demand side and changing credit standards, we have developed a paired match analysis. For those taking an ARM mortgage as of 2004–2006, was this simply a change in the composition of borrowers or was this mostly from an increase in the use of ARMs conditional on demand side characteristics? The sample of ARM borrowers was matched to the larger sample of FRM borrowers. The result is set out in Table 4 and indicates a conditional shift to greater ARM borrowing, 2004–2006. While education has little bearing on the choice of an ARM in 2007, for 2013 those with a college education or more are somewhat less likely to hold an ARM, though the result is not statistically significant in this cross-sectional analysis. As will be shown in Section IIIC, the role of education is consistent with several recent papers indicating the role of past economic events on subsequent choices and expectations (Malmendier and Nagel 2011) and research showing the role of education in financial decision making. (Hudomiet, Kedzi and Willis, 2011). As part of the sorting process, there can be matching based on both perceived repayment risk and on race. For 2007 we can see
B. Chen, F. P. Stafford Table 4 Matched pair model of mortgage choice ARM
FRM
Took mrtg1 btw (2004–2006) (%)
59.3
49.3
Wealth without home equity (1000)
199.6
212.4
Plan to move (%)
27.9
20.1
African Americans (%)
31.2
22.7
Number of observations
391
1924
In Table 4 the matching variables were income categories, state, and age categories. The set of matching variables was limited based on the sample sizes
there was an especially high probability of African-American mortgage holders to have an ARM. Potentially this could be connected to market segmentation or to the fact that African Americans are more likely to have housing transitions and as of 2007 were more likely to have become recent owners and to seek the greater liquidity of an ARM. A concern is that, as new owners, access to the subprime ARM lenders was greater than to traditional lenders. It is also consistent with efforts to provide wider access to mortgage funds for minorities. The period 2004–2006 was characterized by a greater reliance on ARMs. Was this because the composition of the borrowers shifted with more high risk borrowers or, in addition, was there a wider move to mortgage leniency with ARMs becoming more likely given borrower characteristics? This is indicated in the Table 3. As supported by the Federal Reserve survey of lending standards, the period was one in which more lenient standards were quite widely spread. As a lending form which became more popular in the riskier segment of the mortgage market, we can see the greater use of ARMs, 2004–2006 – not only from a compositional change of characteristics of lenders and mortgagees. This can also be explored with the use of a matched-pair model. Such a model has the advantage of avoiding the functional form in a regression mode. A matching model can be seen as a robustness check on the Table 3 result. Namely, there was a higher probability of an ARM for mortgages being taken out in 2004–2006. As can be seen in Table 4, mortgages were more likely to be ARMs if taken in the lending window of 2004–2006, and were more likely for those families with less wealth net of home equity, those more likely to move and more common among African-American families. Mortgage Difficulties, 2009 By of 2007, many of the lending modes in the recent past, especially within ARMs, were in selected regional markets. As shown in Tables 3 and 4, notably within the time window of 2004–2006. Going forward to the crisis, how did this play out? Our assumption is that many of those transaction choices between borrowers and lenders, though recognizing repayment risks, did not allow for the extent of the subsequent disruptions. To this extent, the family level economic variables can be considered as exogenous shocks.25 25
In the extreme, they were Black Swan outcomes (Taleb 2007; Hendry and Mizon 2014). In a partial equilibrium context, they can considered as unanticipated and exogenous.
A Farewell to ARMs or Ever Changing Market Segments? Table 5 Falling behind in mortgage payments
Intercept
1
2
−3.1251***
−3.2768***
(0.1399)
(0.2817)
Interview Time and Mortgage (excluded April/May) Interviewed before April 2009 Interviewed between June and August 2009 Interviewed after August 2009 took_mrtg1_btw (2004–2006)
−0.9643*
−1.171**
(0.5168)
(0.535)
0.4584***
0.4274**
(0.1757)
(0.1845)
0.7834***
0.4997
(0.3042)
(0.3262)
0.3434**
0.3822**
(0.1636)
(0.1726)
Income and Asset Income head 2007 ($1000)
4.3947 (3.0094)
Income head 2008 ($1000)
−10.4456*** (3.5083)
Income wife 2007 ($1000)
7.7416 (6.5617)
Income wife 2008 ($1000)
−17.8995** (7.2003)
value of checking and saving 2007 < =$2000
0.6777*** (0.2082)
10,000 < value of checking and saving 2007 < =$40,000
−0.4801 (0.343)
value of checking and saving 2007 > $40,000
−0.631 (0.4954)
Labor Market and Demographic Laid off head 2009
1.6611** (0.7222)
Unemployed head 2009
1.3393*** (0.2419)
Disabled head 2009
0.8838***
Laid off wife 2009
1.2074
Unemployed wife 2009
0.6624
(0.3358) (0.8647) (0.428) Disabled wife 2009
1.3903*** (0.4364)
No wife 2009
−0.2174 (0.2227)
B. Chen, F. P. Stafford Table 5 (continued) 1
2
African American
0.5483*** (0.1879)
Number of observations
2876
2876
AIC (intercept only)
1265.557
1265.557
A simple model predicting difficulty making mortgage payment is set out in Table 5. The 2009 data were collected from April to the end of 2009. While the timing is based on the logistics of data collection, it is of note that as the recession set in more fully and short run liquidity of households was diminished through time, the likelihood of falling behind in mortgage payments rose throughout the year. Mortgages taken in the 2004– 2006 time window were far more likely to be those in arrears (Model 1). In the context of liquidity and buffer stock balances, there are strong effects of limited income, liquid assets, and the labor market difficulties – such as unemployment and disability (Model 2). African American families have had limited financial assets and have been subject to mortgage lending discrimination (Apgar and Calder 2005). This research suggests the heterogeneous matching in Figure 2 may be thought of as a form of discrimination in which minorities are primarily in the high rate segment, holding a risky ARM (or possibly risky high rate FRM).
Transitions 2007–2013 In this section, we can explore aspects of mortgage choice dynamics, 2007–2013. This allows us to see the migration to and from an ARM mortgage as well as group specific transitions. As of 2007–2013 the nature of the market was changing. While private lenders became very cautious, the FHA Bplayed a critical role in buttressing the single family mortgage and housing market.^ (Immergluck 2011). Moreover, as implied by our discussion of heterogeneity in repayment risk of borrowers, a fixed-effect panel estimate can be thought of as controlling for various repayment risk factors which may be constant over time but which we cannot observe as covariates. This balanced panel Table 6 Ownership transitions and transitions to and from ARM, 2007–2013 2013
2007
Homeowner without Mortgage
Homeowner without Mortgage
Fixed Rate Mortgage
Variable Rate Mortgage
Not Homeowner
Total
16.54
1.56
0.21
2.28
20.59 40.69
Fixed Rate Mortgage
5.33
29.43
1.68
4.25
Variable Rate Mortgage
0.79
2.3
1.69
0.74
5.52
Not Homeowner
1.98
4.78
0.36
26.07
33.19
Total
24.64
38.07
3.94
33.34
A Farewell to ARMs or Ever Changing Market Segments? Table 7 Mortgage transitions 2007–2013 FRM to ARM
ARM to FRM
Mortgage Variables year_took_mrtg1_07 fix_rate-adj_rate of mortgages ARM rate
0.2875**
0.0093
(0.1236)
(0.1203)
1.9239**
−0.3767
(0.8403)
(0.8163)
1.0046**
−0.1616
(0.3944)
(0.3938)
−0.5906*
−0.0855
(0.3282)
(0.4981)
Income and Wealth 60 K < total_fam_income_07 < =120 K total_fam_income_07 > 120 K 10 K < wealth_without_home_equity_07 < =130 K wealth_without_home_equity_07 > 130 K constrain_07
−0.2615
0.6102
(0.4513)
(0.7269)
0.3633
−0.0093
(0.3443)
(0.4569)
0.9194**
0.1761
(0.4252)
(0.605)
2.3053*
−15.418
(1.2251)
(1260.4)
Demographic and Education edu_head_07 = 12 12 < edu_head_07 < =16 edu_head_07 > 16 African Americans_07 plan_move_07
−0.2506
0.5139
(0.3862)
(0.6099)
−0.8226**
0.344
(0.3992)
(0.5867)
−1.5084**
−0.1856
(0.5953)
(0.7052)
0.9364***
0.3047
(0.3069)
(0.4449)
0.6109**
−0.1474
(0.2606)
(0.4272)
Number of observations
1664
207
AIC (intercept only)
719.197
280.788
Added covariates include urban market indicators and additional labor market indicators for the head and wife as of 2009
is necessarily somewhat more representative of an older population than a cross section. For this reason, even though Table 2 shows a declining home ownership rate, the balanced panel (Table 6) has ownership rates of about 67% in both years. As an overview, Table 6 shows the general migration away from ARMs in a balanced panel. The percent of homeowners without a mortgage rises from 20.6% to 24.6%, and while 5.5% of all families were ARM holders in 2007, by 2013 this had declined to 3.9%. Of interest is the question of which families made transitions to and from FRMs and ARMs. As can be seen in Table 7, some of the basic patterns observed in the cross-
B. Chen, F. P. Stafford
section appear in the panel based models. Notably, African Americans are more likely to have migrated to an ARM from a FRM. By 2007 the overall share of ARMS was about 11% (Table 2), so there is a much larger sample for the Table 7 models where movement from a FRM to an ARM is considered.26 The pattern is one of lower rates of movement to an ARM by those with greater education. More educated borrowers were better able to realize the implicit risks of an ARM. Table 7 also includes models with a set of interim variables for 2009. These covariates are included primarily to serve as controls and are available upon request but are not reported here. Having low income is moderately predictive of a move to an ARM. As suggested by the theme of matching and heterogeneity, there still persists a market for lower income families who, lenders willing, take the risk of an ARM. This is a pattern consistent with the basic descriptive statistics in Table 2 as of 2007 and the cross sectional models in Table 3. At the same time, high wealth is also strongly predictive of moving to an ARM, 2007–2013.27 This further highlights the bifurcated nature of the market for ARMs and changes through time. After the less stringent lending standards of 2004– 2006, the new transactions for ARM mortgages were for those with greater nonhousing wealth – a result suggested by the descriptive statistics in Table 2. Of interest is the role of having been constrained as of 2007 on transitions, 2007–2013. As of 2007 (or 2013), being constrained did not predict a cross-sectional matching with an ARM provider (Table 3). Yet, going forward from 2007 to 2013, the families, constrained in 2007, presumably had acquired more financial market experience but were still likely to be looking ahead to increasing income over their early to middle career. Now they looked more assessable and attractive to the providers on the supply side. We posit that as a result they were then more likely to move to an ARM. Consistent with crosssectional models, those planning to move as of 2007 were more likely to migrate to an ARM, 2007–2013. The 2007–2013 panel offers an assessment of mortgage choices over a 6-year period that included both a cyclical peak, a varying ARM rate, and a varying spread. Here we see that ARMs were more likely to be chosen during the peak and early in the period when rates were high. Mortgage choices did respond to the spread just as they have over longer time periods. Given low inflation, home owners, especially the younger ones, may become habituated to a small spread. Should inflation expectations increase, they will have a more substantial spread to consider. In that context, market heterogeneity will play a role. Some higher wealth families will seek an ARM to control their interest costs, while others will choose the lower risk of an FRM.
Conclusion Are ARMs heading to a farewell and are such mortgages not essential (Fugitte 1984)? The ARM share of mortgages has varied through time, both falling and rising in other 26
Note that the sample of those with an ARM as of 2007 is small, yet there is some indication of more movement to a FRM by African-Americans. The suggestion being that there may be more transitions in both directions. For a study of ownership transitions see Charles and Hurst (2002). 27 Another sub-market for ARMs is for ‘jumbo’ but high risk loans at premium rates, such as from Bank of Internet.
A Farewell to ARMs or Ever Changing Market Segments?
countries and for different market segments (Badarinza et al. 2017). The factors shaping ARM choice are substantially known. The U.S. mortgage market has been characterized by both a period of rapid growth in ARM mortgages up to the late 1980’s and a notable decline after 2006. Over the longer time period the share has varied with the level of economic activity, being higher in periods of greater aggregate demand as measured by the rate of growth of real GDP. Over the last three decades, the rate of inflation has generally moved downward and securitization has increased. As a result, there has been a longer term narrowing of the spread between fixed and adjustable rates, and the share of ARMs has declined. The shift downward in the share of ARMs was accelerated during the Global Financial Crisis of 2008–2010. Yet, going forward, there has been some recovery and continued ARM activity by those with substantial net worth, on one hand, and by those with limited assets, on the other. Choice of ARMs also seems to be more common in expansions and to the new or refinancing owners. As demand increases in the short run, there will be a willingness of borrowers to take on premium rates, often expecting further price appreciation. Given that ARM originators need to assess both the return on their money and return of their money, this can lead to the supply of ARMs funds as being less elastic in the short run. In the context of sorting, higher risk borrowers seek to enter the housing market in expansions and more transactions occur at higher rates. The housing boom in selected markets, 2003–2007, is consistent with a margin of extensive home ownership and choice of ARMs. As shown here and in the wider literature, the use of ARMs was particularly prevalent in selected urban markets, 2004–2007. This extensive use of high repayment risk ARMs both in that setting and otherwise leads to the result of a higher share of ARMs when both the ARM an FRM rates have risen. An adjustable rate mortgage should also be attractive to a family with a strong capacity to manage consumption commitments, and as of 2013 ARMs seem to have attracted these families as measured by the level of non-housing net worth. There is still a substantial share of higher risk ARM borrowers, partly from earlier contracts, even as of 2013. This is indicated by the small spread as reported by homeowners – only 35 basis points as shown in Table 2. This bifurcated pattern, particularly as of 2013, is consistent with a matching market with diverse segments. One group of homeowners likely to seek an ARM is those in the early life course who expect greater future income. These are considered to be constrained families in our analysis. Lenders may regard these younger families as representing more repayment risk as suggested in Figure 2. The analysis as of 2007 indicates that such families were not more likely to have an ARM. As they moved forward, 2007–2013, we assume they had a more established financial history and became more attractive as ARM borrowers. The panel analysis indicates that for 2007–2013 these families were more likely to have exited an FRM and to have acquired an ARM. There was some indication of a learning process about ARMs. From 2007 to 2013, those with more education were clearly less likely to have transitioned from a FRM to an ARM. Those planning to move were generally more willing to choose an ARM since a rate penalty for a longer maturity FRM mortgage would not likely be compensated by reduced reset risk. During the financial crisis, the share of ARM mortgages fell and may have left some lasting aversion to ARM mortgages. This may have been
B. Chen, F. P. Stafford
impacted by ability to repay standards from recent financial market regulations. On the other hand, if the economy moves toward higher and more variable expected longer term inflation, we presume that the mortgage market would respond to such heterogeneity with more diverse offerings. As time passes, memories fade and new cohorts may discover ARMs as attractive. In this context, the ARM share has recovered some in recent years. In such a setting, there could easily be a wider rebirth of interest in ARMs. Acknowledgements We thank Paula Fomby, Charles Brown and other seminar participants at the Michigan Survey Research Center’s research seminar as well as Gavin Wood, Trevor Kollmann and other seminar participants at the Royal Melbourne Institute of Technology. The research reported herein was in part performed pursuant to a grant from the U. S. Social Security Administration (SSA) funded as part of the Retirement Research Consortium. The opinions and conclusions expressed are solely those of the authors and do not represent the opinions or policy of SSA or any agency of the federal government. The collection of data used in this study was partly supported by the National Institutes of Health under grant number R01 HD069609 and the National Science Foundation under award number 1157698.
Glossary of Terms Demographic and Education male_head_ 07 (13): 1–0 if head of family is male age_head_07 (13) < =34: 1–0 if head of family is under age 35 34 < age_head_07 (13) < =49: 1–0 if head of family is age 35–49 49 < age_head_07 (13) < =64: 1–0 if head of family is age 50–64 (age 65 and older excluded) edu_head_07 (13) = 12: 1–0 if head is high school graduate 12 < edu_head_07 (13) < =16: 1–0 if head has some college or a college graduate edu_head_07 (13) > 16: 1–0 if head has graduate education (less than high school excluded) African Americans_07 (13): 1–0 if head is African American (Census definition) employed_head_07 (13): 1–0 if head employed as of date of the survey employed_wife_07 (13): 1–0 if wife employed as of date of the survey no_wife_07 (13): 1–0 if not married or cohabiting Income and Asset Income head 2007 ($1000,000) Income head 2008 ($1000,000) Income wife 2007 ($1000,000) Income wife 2008 ($1000,000) value of checking and saving 2007 > $40,000: 1–0 if liquidity value is greater than $40,000 Labor Market and Demographic 10,000 < value of checking and saving 2007 < =$40,000: 1–0 if liquidity value is above $10,000 and less than or equal to $40,000 value of checking and saving 2007 > $40,000: 1–0 if liquidity value is greater than $40,000 Income and Wealth 60 K < total_fam_income_07 (13) < =120 K: 1–0 if family income of 2006 (2012) is above $60,000 and less than or equals to $120,000 total_fam_income_07 (13) > 120 K: 1–0 if family income of 2006 (2012) is above $120,000 10 K < wealth_without_home_equity_07 (13) < =130 K: 1–0 if wealth less home equity is above $10,000 and less than or equal to $130,000 wealth_without_home_equity_07 (13) >130 K: 1–0 if wealth less home equity is above $130,000 constrain_07 (13): 1–0 if head age is less than 35 and head education is more than college and wealth less home equity is less than or equal to $10,000
A Farewell to ARMs or Ever Changing Market Segments? Interview Date 2009 Interviewed before April 2009: 1–0 indicator Interviewed between June and August 2009: 1–0 indicator Interviewed after August 2009: 1–0 indicator (April and May excluded) Labor Market and Demographic Laid off head 2009: 1–0 if head is laid off at date of the survey Unemployed head 2009: 1–0 if head is unemployed at date of the survey Disabled head 2009: 1–0 if head is disabled at date of the survey Laid off wife 2009: 1–0 if wife is laid off at date of the survey Unemployed wife 2009: 1–0 if wife is unemployed at date of the survey Disabled wife 2009: 1–0 if wife is disabled at date of the survey Mortgage Variables year_took_mrtg1_07 (13): the year the first mortgage was taken took_mrtg1_btw (2004–2006): 1–0 if took in the period 2004–2006 fix_rate-adj_rate of mortgages: prevailing market rate differences of 30 year fixed rate less 1-year Mortgage Bankers adjustable rate in year mortgage taken ARM rate: 1-year ARM rate Mortgage Bankers data by year Region and Urban Status Northeast_07 (13): 1–0 if Census Region is Northeast North Central_07 (13): 1–0 if Census Region is North Central South_07 (13): 1–0 if Census Region is South state_07 (13) in (AZ, CA, FL, NV) and urban_07 (13) = 1: 1–0 if in one of these states and in central metropolitan counties of population of 1000,000 or more state_07 (13) in (AZ, CA, FL, NV) and urban_07 (13) = 2: 1–0 if in one of these states and in fringe metropolitan counties of population of 1000,000 or more plan_move_07 (13): 1–0 if plans to change residence in the next few years
References Agnello, L., & Schuknecht, L. (2011). Booms and busts in housing markets: determinants and implications. Journal of Housing Economics, 20(3), 171–190. Apgar, W. C., & Calder, A. (2005). The dual mortgage market: The persistence of discrimination in mortgage lending. Joint Center for Housing Studies, Working Paper W05–11, December. Badarinza, C., Campbell, J. Y., & Ramadorai, T. (2017). What calls to ARMs? International evidence on interest rates and the choice of adjustable-rate mortgages. Management Science, February, p. 1–14. Bhardwaj, G., & Sengupta, R. (2012). Subprime mortgage design. Journal of Banking & Finance, 36(5), 1503–1519. Brueckner, J. K., & Follain, J. R. (1988). The rise and fall of the ARM: An econometric analysis of mortgage choice. The Review of Economics and Statistics, 93–102. Campbell, J. Y., & Cocco, J. F. (2003). Household risk management and optimal mortgage choice. The Quarterly Journal of Economics, 118(4), 1449–1494. Campbell, J. Y., & Cocco, J. F. (2015). A model of mortgage default. The Journal of Finance, 70(4), 1495– 1554. Carroll, C. D. (1994). How does future income affect current consumption? The Quarterly Journal of Economics, 109(1), 111–147. Charles, K. K., & Hurst, E. (2002). The transition to home ownership and the black-white wealth gap. The Review of Economics and Statistics, 84(2), 281–297.
B. Chen, F. P. Stafford Chen, B., & Stafford, F. P. (2016). Stock market participation: Family responses to housing consumption commitments. Journal of Money, Credit and Banking, 48(4), 635–659. Chetty, R., & Szeidl, A. (2007). Consumption commitments and risk preferences. The Quarterly Journal of Economics, 122(2), 831–877. Coulibaly, B., & Li, G. (2009). Choice of mortgage contracts: evidence from the survey of consumer finances. Real Estate Economics, 37(4), 659–673. Dau-Schmidt, K. (1992). The Effect of Consumption Commitments and the Liquidity Constraint on Labor Supply. The Journal of Economics, 18, 49–56. Deaton, A. (1992). Understanding consumption. Oxford University Press. Ehrmann, M., & Tzamourani, P. (2012). Memories of high inflation. European Journal of Political Economy, 28(2), 174–191. Fishback, P. V., Rose, J., & Snowden, K. (2013). Well worth saving: how the New Deal safeguarded home ownership. University of Chicago Press. Fisher, I. (1933). The debt-deflation theory of great depressions. Econometrica, 1(4), 337–357. Fisher, L., & Kan J. (2015). Chart of the Week – July 2, 2015. Research and Economics, Mortgage Bankers Association. Flavin, M., & Yamashita, T. (2002). Owner-occupied housing and the composition of the household portfolio. American Economic Review, 92(1), 345–362. Fugitte, J. (1984). Borrowers Do Not Need or Want Variable-Rate Loans. American Banker, 149, 22. Greenspan, A., & Kennedy, J. (2008). Sources and uses of equity extracted from homes. Oxford Review of Economic Policy, 24(1), 120–144. Hendry, D. F., & Mizon, G. E. (2014). Unpredictability in economic analysis, econometric modeling and forecasting. Journal of Econometrics, 182(1), 186–195. Hudomiet, P., Kézdi, G., & Willis, R. J. (2011). Stock market crash and expectations of American households. Journal of Applied Econometrics, 26(3), 393–415. Hurst, E., & Stafford, F. P. (2004). Home is where the equity is: Mortgage refinancing and household consumption. Journal of Money, Credit, and Banking, 36(6), 985–1014. Hurst, E., Luoh, M. C., Stafford, F. P., & Gale, W. G. (1998). The wealth dynamics of American families, 1984-94. Brookings Papers on Economic Activity, 1998(1), 267–337. Iacobini, M. (2009). Mirroring people: The science of empathy and how we connect with others. New York: Picador, Farrar, Straus and Giroux. Immergluck, D. (2011). From minor to major player: the geography of FHA lending during the US mortgage crisis. Journal of Urban Affairs, 33(1), 1–20. Keller, S., & Mody, M. A. (2010). International Pricing of Emerging Market Corporate Debt: Does the Corporate Matter?. International Monetary Fund, 10–26. Lancaster, K. J. (1966). A new approach to consumer theory. Journal of Political Economy, 74(2), 132–157. Malmendier, U., & Nagel, S. (2011). Depression babies: do macroeconomic experiences affect risk taking? The Quarterly Journal of Economics, 126(1), 373–416. Mayer, C., Pence, K., & Sherlund, S. M. (2009). The rise in mortgage defaults. Journal of Economic Perspectives, 23(1), 27–50. Mian, A., & Sufi, A. (2011). House prices, home equity-based borrowing, and the US household leverage crisis. American Economic Review, 101(5), 2132–2156. Miles, D. (2004). The UK mortgage market: Taking a longer-term view. Final Report and Recommendations. Moench, E., Vickery, J., & Aragon, D. (2010). Why is the market share of adjustable-rate mortgages so low?. Current Issues in Economics and Finance, 16(8). https://doi.org/10.2139/ssrn.1736208. Morgan, D. P., Iverson B., & Botsch M. (2011). Did the 2005 Bankruptcy Reform Increase Subprime Foreclosures?. Federal Reserve Bank of New York Economic Policy Review. Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure competition. Journal of Political Economy, 82(1), 34–55. Saiz, A. (2008). On local housing supply elasticity. University of Pennsylvania, Wharton School, July. Salant, S. W. (1977). Search theory and duration data: a theory of sorts. The Quarterly Journal of Economics, 91(1), 39–57. Simkovic, M. (2013). Competition and crisis in mortgage securitization. Ind. LJ, 88, 213. Taleb, N. N. (2007). Black swans and the domains of statistics. The American Statistician, 61(3), 198–200. Wood, G., Parkinson, S., Searle, B., & Smith, S. J. (2013). Motivations for equity borrowing: A welfareswitching effect. Urban Studies, 50(12), 2588–2607.