Documente Academic
Documente Profesional
Documente Cultură
Product Innovation &
Mortgage Selection in the
Subprime Era
Souphala Chomsisengphet
U.S. Department of the Treasury, The Office of the Comptroller of the Currency
Timothy Murphy
Marquette University, Economics Department
&
Anthony Pennington‐Cross*
Marquette University, Finance Department
October 2008
*Contact author – Marquette University, College of Business Administration, Associate Professor,
Department of Finance, Straz Hall room 328, PO Box 1881, Milwaukee, WI, 53201‐1881,
Anthony.Pennington‐Cross@mu.edu, 414‐288‐1452
Since the use of exotic or alternative mortgage products played a role in the subprime meltdown,
it is important to understand when these products make sense and how far the market deviated
from rationality. The research presented here begins this analysis using a county level empirical
examination of the mechanisms used to select different types of subprime mortgages from 2000
through July 2007. In particular we examine the use of adjustable rate loans, hybrid rate loans,
interest only loans, non-amortizing loans, and loans with balloon payments.
The empirical results indicate that the economic and financial incentives play a large part in
determining what types of loans people used to finance their home. We also examine how the
use of these types of mortgages changed over time and how much of those changes were driven
by economic and financial fundamentals. We find substantial evidence of the misallocation of
mortgage types over the 2004-2007 time period. This may help us to project what a sensible
subprime mortgage market should look like in the future and how far the mortgage market
deviated from that path.
In the last 15 years, the subprime mortgage market has increased exponentially. At the
beginning of the 1990’s, few people had even heard of a subprime mortgage. Gramlich (2007)
indicates that in 1993 there were very few subprime mortgage originations but by 2005, $625
billion dollars of subprime mortgages were originated. By 2006 the subprime financial crisis
was just beginning as default rates began to increase rapidly, causing a series of subprime
originators and aggregators for Wall Street to fail. In 2007 and 2008 subprime lending was only
a fraction of its heyday (2004-2006). In an almost surreal and slow procession through the entire
financial system, subprime lending is now credited with destroying venerable financial
institutions such as Lehman Brothers (bankrupt), Merrill Lynch (purchased by Bank of America
or Wachovia), Washington Mutual (bankrupt), Fannie Mae & Freddie Mac (government
It is clear that the loans originated more recently performed much worse than loans
originated in the early 2000s. However, the cause of the increased default rates is of
considerable debate. Some stylized facts cannot be denied. For example, considerable fraud
aided the downfall of subprime lending while lenient and likely unethical businesses looked the
other way. Many homebuyers did not understand or did not care to understand the mortgage
products they were using. Very sophisticated financial institutions did not understand the
Our data indicate that the volume of subprime lending (this includes securitized, Alt A,
near prime, A-, Low Documentation, and deep into subprime market segments) increased by
over more than six times from 2000 through 2005, declined slightly in 2006, and retrenched
considerably in 2007. During the rapid growth of subprime mortgage lending, the introduction
of new or exotic mortgage products made it possible for many households to access the mortgage
credit market for both home purchases and mortgage refinances. Compared to traditional fixed-
rate mortgages these alternative mortgages enabled households to borrow more, amortize the
loans more slowly or not at all, and take on more interest rate risk (LaCour-Little and Yang
2008).
This paper examines the use of adjustable rate loans (loans that allow the interest rate
and payments to change over time), hybrid loans (loans that have a fixed interest rate for a short
time period then become adjustable), interest only loans (loans that allow only the interest to be
paid over a pre-specified time period and then usually become amortizing), loans with balloon
payments (loans that do not fully amortize and therefore have a lump sum payment due at the
end of the loan), and non-amortizing loans (loans that allow no or negative amortization) using a
county level data set representing subprime (or non-prime) lending. A reduced form model is
estimated that relates mortgage selection to borrower and local demographic characteristics, and
economic and financial conditions as suggested in the mortgage selection literature. One
question this approach can answer is whether the allocation of product types in subprime lending
parameter stability and out of sample forecasts can reveal how much the allocation of mortgage
Most of the extensive literature on mortgage selection examines the choice between
adjustable and fixed rate mortgages (for example, Dunn and Spatt, 1985; LeRoy, 1996; and
2
Stanton and Wallace, 1999). In particular, Stanton and Wallace (1999) suggest that transaction
costs influence the choice of how many points to pay on a fixed rate mortgage. Brueckner
(1994) suggests that the greater the expected return on other investments the higher the optimal
Campbell and Cocco (2003) explore the role of mortgage choice through the viewpoint of
understanding household risk. They find that an adjustable rate mortgage is more preferable in
an environment when the purchaser is less income risk-averse, has a low default cost, or a high
probability of moving. Szerb (1996) echoes these findings, indicating that larger shocks make it
more likely to use a fixed rate mortgage. Posey & Yavas (2001) also show that a separating
equilibrium exists where the more risky borrowers use adjustable rate loans.
that the mortgage market has been segmented into discrete risk classifications that limit the
mortgage supply of particular products. Empirical results indicate that credit scores, income,
non-real estate debt, and down payments are very important determinants of FHA (Federal
LaCour-Little and Yang (2008) and Chambers, Garriga, and Schlagenhauf (2007)
developed mortgage choice models that provide some of the first research focusing on mortgages
more typically used in subprime lending. Lacour-Little and Yang (2008) model which factors
influence the decision to defer costs to the future in exchange for lower initial payments that are
often offered through alternative mortgage products. Empirical evidence indicates that future
house prices, income, and risk preference play an important role in determining if the loan will
3
Schlagenhauf (2007) indicate that younger and households with fewer assets can benefit from
Piskorski and Tchistyi (2007) attempt to discover an optimal mortgage for both the lender
and borrower. Pareto efficiencies for mortgage choices can be gained through dependence on
expected income realization of the borrower and market interest rates. They make evident that
an ideal alternative mortgage contract between a lender and borrower can be either a
Due to the recent growth and subsequent decline of the subprime mortgage market,
research on the subprime market segment is relatively limited. We extend the literature on
several fronts. First, instead of limiting the analysis to simply adjustable rate versus fixed rate
mortgages or alternative versus traditional mortgage products, this paper examines multiple types
of mortgage products. These include adjustable rate mortgages, hybrid mortgages, balloon
Second, the stability of the results can be tested as the subprime market moved through a boom
and bust to test how far mortgage allocation deviated from normal practices during the boom
period. This may provide insight into one mechanism that contributed to the financial crisis.
The data used to explore the use of different mortgage types is collected from the Loan
Performance (LP) Asset Backed Securities (ABS) data series covering January 2000 through
July 2007. The data includes loans that are securitized in the private label ABS market. It does
not include loans that were held in portfolio. Therefore, the results may only apply to the
4
securities part of the market. LP reports that their data covers over 95% of the securities market.
As the subprime market has evolved, different market participants have attempted to re-label the
segment of the market they are involved in as non-prime, near prime, A-, or Alt-A. In this
research all these types of subprime loans are included under the label of subprime.
Appendix – County Level Distribution of Mortgage Types in 2007 provides maps at the
county level of the market share for various mortgage types over the time period January through
July 2007. They show that there is substantial geographic variation in the use of different types
of mortgages. For example, adjustable rate mortgages or ARMs and hybrid mortgages tend to be
used more often in Florida, Los Angeles, San Francisco, and Chicago areas. Interest only loans
are most popular along the California coast all the way from the San Francisco bay area down to
the border with Mexico. They are also moderately popular around the east coast cities of
Washington, D.C., New York and Boston. However, loans that have balloon payments are less
geographically concentrated. Non-amortizing loans are used most frequently along the east coast,
and Washington, DC and Florida in particular. In total, these maps indicate that many of the
product innovations have been concentrated in high cost locations that have seen house prices
increase substantially over the last 10 to 15 years. Therefore, affordability may be an important
driver of alternative mortgage product use. These geographic concentrations are also consistent
with the Posey & Yavas (2001) separating equilibrium where the higher risk loans/borrowers use
Table 1 provides a description of the variables and Table 2 provides summary statistics
for the variables used in the county level analysis. The statistics are reported on an un-weighted
basis. The average county has 45 percent of its loans with adjustable rates and the vast majority
of these adjustable rate loans are hybrids (98 percent). The average county has over 16 percent
5
of loans with non or negative amortization features, 10 percent with balloon features, and 7
The literature indicates that house prices can play an important role in mortgage
selection. Therefore, we include measures of house price appreciation using the Office of Federal
Housing Enterprise Oversight (OFHEO) repeat sales price index. The change in house prices
over the last year and the last five years are included as proxies for anticipated increases in house
prices. It is anticipated that rising prices will be associated with an increase in all types of
alternative mortgages. Since shocks to the housing market can also play a role in mortgage
selection, two measures of price volatility are included. The first measure is the standard
deviation in the house price index growth rate over the prior year and the second measure is the
volatile of individual house prices around the index itself (dispersion). In general, volatility in
the underlying asset’s value should lead to compensation in the underwriting process, such as by
removing interest rate risk, requiring larger down payments, or requiring larger borrower income
relative to debt.
If riskier households and riskier loans self select into loan types with lower initial
payments, then households living in locations with lower credit scores (FICO scores), smaller
down payments or higher loan to value ratios (LTVs), higher debt burdens (debt to income ratios
or D/I), and higher unemployment rates will all use alternative mortgages more frequently.
If affordability is a driving force behind the use of alternative mortgage products then
locations with higher house price to income ratios (HP/I) will be more likely to use alternative
compensate for taking on the additional risks. Wealth is proxied by the county per capita income
6
times the state average household size. Households that are more likely to move should also
prefer lower initial payments available in alternative mortgage types. To proxy for mobility the
state median age is multiplied times the state average household size.
The spread between 10-year and 1-year treasury yields is included to measure the
steepness of the yield curve. Typically, steep yield curves are associated with rising short term
interest rates and a growing economy in the future due to an accommodative monetary policy. A
steep yield curve should increase the use of fixed rate loans to avoid large payment shocks but
improving economic conditions may encourage households to stretch payments in the short run
Our data set is organized as a panel data set where time is measured in years from 2000
through July of 2007 and the cross section is at the county level and includes only counties in
metropolitan areas. Therefore, the estimation must control for the lack of independence of
repeated observations of counties over a short time horizon. A robust Huber/White Sanwich
estimator of variance is used instead of the traditional approach. It adjusts the variance
covariance matrix to allow for correlation of error terms within each county. This procedure
does not affects only standard error estimates, not coefficient estimates.
Five separate models will be estimated for each mortgage selection or market share
analysis: 1. ARM share of all loans, 2. Hybrid share of ARM loans, 3. Non-Amortizing share of
all loans, 4. Balloon share of all loans, and 5, Interest Only share of all loans. Each separate
regression will be estimated using the same explanatory variables discussed above.
7
Market shares are bounded at zero and one. Therefore, a linear model may predict
market shares that are greater than 1 or less than 0. A logistic transformation of the market share
is used to limit the estimates between zero and one (log(sit / 1-sit), where s is the market share for
county i in time period t). Each market share does not represent the same number of mortgages
(1)
where nit is the total number of loans for county i in time period t. To reflect this non-constant
loan originations receive more weight; and the closer a county’s market share is to 50% the
This type of estimation approach is coined grouped logit by STATA because it conducts
a logistic transformation of the dependent variable and the dependent variable represents a group
of binary observations that are aggregated to the county level. To calculate the estimated market
(2)
Two factors complicate presenting results in a readily accessible form: 1) the logistic
transformation makes it difficult to interpret coefficients, and 2) because five different dependent
variables are used, specification tests generate more than two dozen sets of results. Tables 3-7
8
provide the estimated coefficients and their level of significance for each of the five market
shares. Each table provides three different specifications that are designed to display coefficient
In general the results are consistent with expectations and prior theoretical and empirical
research on mortgage selection. For example, in locations where house prices have been
increasing more people use adjustable rate mortgages. This is true whether measured over the
prior year or the prior five years. Locations with more wealth and lower credit scores are also
more likely to use adjustable rate loans. However, locations with smaller down payments use
more adjustable rate loans. These results are consistent with the theory that higher risk
borrowers self select in a separating equilibrium to adjustable rate loans. There is also evidence
that borrowers are more likely to use adjustable rate loans in locations where housing is more
expensive. A steep yield curve is positively associated with the use of adjustable rate loans,
indicating short run thinking and financing typical of subprime borrowing. However, not all
results meet expectations. For example, our proxy for mobility is negatively associated with
adjustable rate use. In addition, measures of risks in the housing market are mixed. In general
they show that individual house price volatility reduces adjustable rate use while metropolitan
Tables 4-7 also provide the point estimates for each explanatory variable for the other
four market share results. For the sake of brevity we will not review all these results but instead
examine sensitivity tests to illustrate the results. Five sensitivity tests are presented in the
Appendix: Sensitivity Tests. While holding all other variables constant and evaluating them at
their means, a single explanatory variable is varied +/- three standard deviations from its mean
9
Four variables can have large impacts on the share of that use an adjustable rate loan –
income growth, credit scores, LTV, and wealth. The results indicate that more adjustable rate
loans will be used in locations displaying one or more of the following characteristics:
substantial wealth, increasing income, low credit scores, and small down payments. The
sensitivity tests for the use of hybrid loans show that the empirical results have not been able to
estimates. Interest only loans are a larger part of the subprime market in locations where
borrowers have high credit scores, low down payments are made, and/or income is declining.
Again, many of the variables have large effects on the use of non-amortizing loans, but two
variables have an especially large effect. Non-amortizing loans are a much bigger part of the
subprime market in locations where borrowers have high credit scores and income is declining.
The primary drivers of balloon use are high debt to income ratios and good credit scores.
In summary, the sensitivity tests show strikingly consistent results. Good credit scores
are used to compensate for other risks associated with an alternative mortgage product. For
balloons the risk is driven by high payments. For non-amortizing loans and interest only loans
the risks are declining income and low down payments. However, for adjustable rate loans
higher wealth and house price appreciation are used to compensate for low credit scores, low
Forecasts are generated to test the predictive power of the results and to test for any large
changes in how the market allocated mortgage types over time. To generate the market shares
we use the coefficients (see Appendix – Sample Specification Tests for coefficient estimates and
10
precision of the results) to estimate the market share for each county in each year and then
calculate the average market share for the nation as a whole by weighting each observation by
the number of loans made in that location. The Appendix: In Sample Forecast shows that the
empirical estimates perform very well at a national level tracking the growing market shares for
non-amortizing loans and the rising use of adjustable rate loans through 2004 and the decline in
Given the ability of the model to perform well in the sample, next we estimate the model
using only the years 2000, 2001, and 2003 and use the estimated coefficients to predict market
shares for 2004, 2005, 2006, and 2007. This break time period of 2003/2004 was chosen
because in 2004 the volume of subprime loans originated more than doubled and continued to
grow at a very rapid pace through 2005. Therefore, we are roughly trying to estimate a model
over a time period before subprime’s explosive growth and the subsequent increases in the use of
interest only and non-amortizing loans. The Appendix: Out of Sample Forecast 2003-2007
shows that under the earlier lending regime economic and financial conditions cannot explain the
increased use of interest only loans and adjustable rate loans and only a little over one-half of the
increase in use of non-amortizing loans. In contrast, the model predicts that more loans with
balloons would have been used. The balloon result may reflect the increased regulation of
balloons by state level anti-predatory lending laws that were being introduced over this time
period.
The extent of the deviation between the predicted product usage and actual usage also
varies by region of the country. For example, while interest only loans become almost 30
percent of the market for the nation as a whole, in the Pacific region they were almost 45 percent
of the market. By 2007 the use of interest only loans in the Pacific region was approximately
11
four times higher than economic and financial conditions predict. In addition, non-amortizing
loans were being used in 2006 and 2007 approximately 20 percentage points more than
predicted. In contrast, the deviations are not nearly as large in the North East (Mid-Atlantic and
The primary finding is that the allocation of product types shifted dramatically in 2004
and 2005 and these shifts were not driven by economic or financial fundamentals but by other
factors in underwriting or borrower demand. This shift was largest in the Pacific region.
VI. Conclusion
Throughout the last seven years the subprime mortgage market became a $600 billion
industry that contributed trillions of dollars in the financial world through Mortgage Backed
Securities. Gaining an understanding of the mortgage products being purchased has become
increasingly important for policymakers in the midst of a dramatic subprime crisis and ever
While there are many factors that have contributed to the subprime debacle, the
increasing use of exotic and new mortgage products helped to set the stage by increasing the
conditions (Ho and Pennington-Cross, 2005). This paper examines the selection of the types of
loans that have been strongly associated with the rapid deterioration of subprime lending –
adjustable rate loans, hybrid rate loans, interest only loans, non-amortizing loans, and loans with
balloon payments.
The estimation results indicate that all of these loan types tend to respond to financial and
economic conditions as predicted by theory and prior empirical evidence in the prime market.
12
Therefore, there is some evidence of a rational allocation of product types to locations and
households where they make the most sense. However, the results also indicate that model
underestimate the rapid increased use of interest only loans and non-amortizing loans. This
deviation is even larger in the Pacific region. It is our interpretation that these deviations are not
caused by changes in the economic or financial conditions that traditionally determine mortgage
product use. Rather, the deviations are likely the result of loosened underwriting standards,
increased steering by originators and perhaps other members of the real estate industry, or a
change in borrower preferences in favor of more exotic loan types.. Whatever the root cause of
the dramatic shift toward exotic loans types, their over-use very likely contributed to the
13
References
Brueckner, J. K. (1995). "Borrower Mobility, Adverse Selection, and Mortgage Points." Journal
of Financial Intermediation, vol 3 issue 4, 416-441.
Buist, H., & Yang, T. (2000). "Housing Finance in a Stochastic Economy: Contract Pricing and
Choice." Real Estate Economics, vol 28 issue 1, 117-39
Campbell, J. Y., & Cocco, J. F. (2003). "Household Risk Management and Optimal Mortgage
Choice." The Quarterly Journal of Economics , vol 118 issue 4, 1449-1494.
Chambers, M., Garriga, C., & Schlagenhauf, D. (2007). "Accounting for Changes in the Home-
ownership Rate." Working Paper, Florida State University .
Dodd, R. (2007). "Subprime: Tentacles of a Crisis." Finance & Development, vol 44 issue 4, 15-
19.
Dunn, K., & Spatt, C. (1985). "An Analysis of Mortgage Contracting: Prepayment Penalties and
the Due-on Sales Clause." Journal of Finance , vol 40, 293-308.
Gramlich, E. M. (2007). "Booms and Busts: The Case of Subprime Mortgages." Economic
Review - Federal Reserve Bank of Kansas, 105-113.
Ho, G., & Pennington-Cross, A. (2007). "The Varying Effects of Predatory Lending Laws on
High-Cost Mortgage Applications." Federal Reserve Bank of St. Louis Review, 39-59.
Ho, G., & Pennington-Cross, A. (2006). "The Termination of Subprime Hybrid and Fixed Rate
Mortgages.” Federal Reserve Bank of St. Louis Working Paper, 2006-42.
LaCour-Little, M., & Yang, J. (2008). "Pay Me Now or Pay Me Later: Alternative Mortgage
Products and Housing Consumption." Working Paper, Presented at the American Real Estate
and Urban Economics Mid-Year Conference in 2007, 1-58.
Nichols, J., Pennington-Cross, A., & Yezer, A. (2005). "Borrower Self-Selection, Underwriting
Costs, and Subprime Mortgage Credit Supply." The Journal of Real Estate Finance and
Economics , vol 30 issue 2, 197-219.
Piskorski, T., & Tchisty, A. (2006). "Optimal Mortgage Design." Working Paper, NYU .
14
Posey, L., & Yavas, A. (2001). “Adjustable and Fixed Rate Mortgages as a Screening
Mechanism for Default Risk.” Journal of Urban Economics, vol 49 issue 1, 54-79.
Stanton, R., & Wallace, N. (1998). "Mortgage Choice: What's the Point?" Real Estate
Economics, vol 26 issue 2, 173-205.
Szerb, L. (1996). "The Borrower's Choice of Fixed and Adjustable Rate Mortgages in the
Presence of Nominal and Real Shocks." Real Estate Economics ,vol 24 issue 1, 43-55.
15
Table 1: Mortgage Selection Variable Definitions and Sources
Variable Definition Source
Market Share
Adjustable Rate Percentage of adjustable rate mortgages LPABS
Hybrid Rate Percentage of hybrid mortgages LPABS
Balloon Payment Percentage of balloon mortgages LPABS
Interest Only LPABS
Payment Percentage of interest only mortgages
Non‐Amortizing LPABS
Payment Percentage of non‐amortizing mortgages
Explanatory
%ΔHPI Metropolitan area housing price index one year percent change OFHEO
5yr‐%ΔHPI Metropolitan area housing price index five year percent change OFHEO
Svol County one year standard deviation of the housing price index growth OFHEO
Vol OFHEO calculated state parameter measures for volatility OFHEO
Unemp County average yearly unemployment rate BLS
5yr‐%ΔI County metropolitan five year income per capita growth rate BLS
HP/I County average subprime house price divided by income per capita LPABS/Census
LTV County average loan to assessed house value LPABS
D/I County average amount of total debt as a ratio of pre‐taxed income LPABS
Wealth County average income multiplied by state median age Census
Pop County population per square mile Census
Mobility State median age multiplied by state average household size Census
FICO County average FICO score LPABS
Yield Curve Difference of average 10 year treasury bill and 1 year treasury bill SLFRB
LPABS: Loan Performance Asset Backed Securities, OFHEO: Office of Federal Housing Enterprise
Oversight, SLFRB: St. Louis Federal Reserve Bank, BLS:.Bureau of Labor Statistics, Census: Census
Bureau.
16
Table 2: Summary Statistics
Variable Mean Standard Deviation
Market Share
Adjustable Rate 0.45 0.14
Hybrid Rate 0.98 0.03
Balloon Payment 0.10 0.08
Interest Only Payment 0.07 0.10
Non‐Amortizing
Payment 0.16 0.15
Explanatory
%ΔHPI 5.88 4.79
5yr‐%ΔHPI 29.73 18.83
Svol 1.37 1.08
Vol 1.30 0.19
Unemp 4.91 1.52
5yr‐%ΔI 21.29 6.05
HP/I 4.34 1.76
LTV 83.61 3.53
D/I 36.43 4.48
Wealth 10,811.78 2,034.16
Pop 271.60 488.19
Mobility 90.18 2.33
FICO 632.81 24.23
Yield Curve 0.68 0.78
17
Table 3: Adjustable Rate Results
Specification 1 Specification 2 Specification 3
Variable Coefficient t‐stat Coefficient t‐stat Coefficient t‐stat
%ΔHPI 0.015** 15.60 0.018** 22.29 0.020** 22.94
5yr‐HPI 0.005** 7.92 0.005** 7.88 0.006** 10.28
Svol 0.020** 2.63 0.019** 2.81 0.027** 3.99
Vol ‐0.480** ‐6.89 ‐0.439** ‐5.86 ‐0.222** ‐3.16
Unemp 0.005 0.65 0.035** 4.04 0.028** 3.84
5yr‐I ‐0.026** ‐12.48 ‐0.024** ‐10.86 ‐0.022** ‐8.65
HP/I 0.031** 2.78 0.079** 4.73
LTV 0.028** 4.43 0.045** 6.20
D/I ‐0.015* ‐1.71 ‐0.011 ‐1.27
Wealth 0.062** 6.87 0.087** 6.35
Pop ‐0.007 ‐0.28
Mobility ‐0.041** ‐8.83
FICO ‐0.005** ‐3.61
Yield 0.088** 7.02
Constant 0.826** 8.12 ‐2.104** ‐3.67 2.395** 2.67
Adjusted R‐squared 0.55 0.49 0.55
**5 percent level of significance & *10 percent level of significance
18
Table 5: Interest Only Results
Specification 1 Specification 2 Specification 3
Variable Coefficient t‐stat Coefficient t‐stat Coefficient t‐stat
%ΔHPI 0.010** 2.480 0.013** 3.720 0.029** 13.350
5yr‐HPI 0.011** 7.080 0.004** 2.420 0.012** 8.950
Svol 0.095** 5.820 0.060** 4.230 0.036** 2.470
Vol ‐0.605** ‐2.630 ‐0.743** ‐3.630 ‐0.196 ‐1.320
Unemp ‐0.115** ‐5.370 ‐0.097** ‐3.310 0.017 1.210
5yr‐I ‐0.049** ‐6.630 ‐0.039** ‐5.430 ‐0.029** ‐5.340
HP/I 0.220** 6.140 0.097** 4.700
LTV 0.067** 2.680 0.073** 4.340
D/I ‐0.011 ‐0.370 ‐0.044** ‐2.160
Wealth 0.122** 4.170 0.083** 5.240
Pop 0.000 1.130
Mobility ‐0.065** ‐5.590
FICO 0.031** 15.450
Yield ‐0.183** ‐5.660
Constant 0.110 0.370 ‐7.499** ‐2.870 ‐22.074** ‐9.510
Adjusted R‐squared 0.36 0.49 0.73
**5 percent level of significance & *10 percent level of significance
19
Table 7: Balloon Results
Specification 1 Specification 2 Specification 3
Variable Coefficient t‐stat Coefficient t‐stat Coefficient t‐stat
%ΔHPI ‐0.027** ‐8.260 ‐0.023** ‐5.460 ‐0.016** ‐7.060
5yr‐HPI 0.011** 16.220 0.009** 7.920 0.008** 9.180
Svol ‐0.041** ‐2.060 ‐0.025 ‐1.560 ‐0.082** ‐7.000
Vol ‐0.357** ‐2.960 ‐0.250** ‐2.110 ‐0.264** ‐2.440
Unemp ‐0.115** ‐5.000 ‐0.102** ‐4.510 ‐0.017 ‐1.570
5yr‐I ‐0.012** ‐3.860 0.003 0.810 0.005** 2.050
HP/I ‐0.023 ‐1.200 ‐0.075** ‐4.540
LTV 0.062** 6.020 0.026** 3.450
D/I 0.223** 17.780 0.132** 9.820
Wealth ‐0.027** ‐2.820 ‐0.058** ‐4.730
Pop 0.000** ‐2.790
Mobility 0.006 0.890
FICO 0.014** 10.880
Yield ‐0.545** ‐22.850
Constant ‐0.790** 3.760 ‐14.667** ‐13.71 ‐17.091** ‐13.050
Adjusted R‐squared 0.33 0.48 0.67
**5 percent level of significance & *10 percent level of significance
20
Appendix – County Level Distribution of Mortgage Types in 2007
ARM Percentage Breakdown
hybr i d_p 0%t o 25% >25%t o 50% >50%t o 75% >75%t o 100%
21
Interest Only Breakdown
22
Balloon Breakdown
bal l oon_p 0%t o 25% >25%t o 50% >50%t o 75% >75%t o 100%
23
Appendix – Sensitivity Tests
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐10 ‐7.5 ‐5 ‐2.5 0 2.5 5 7.5 10 12.5 15 17.5 20 2 3 4 5 6 7 8 9
One Year House Price Appreciation Unemployment Rate
1
1
0.9 0.9
0.8 0.8
ARM Market Share
House Price to Income Ratio Five Year Income Growth
1 1
0.9 0.9
0.8 0.8
ARM Market Share
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐20 ‐10 0 10 20 30 40 50 60 70 80 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
1 1
0.9 0.9
0.8 0.8
ARM Market Share
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
100
200
300
400
500
600
700
800
900
0
1000
1100
1200
1300
1400
1500
1600
1700
505 520 535 550 565 580 595 610 625 640 655 670 685 700 715
Population Per Square Mile Average FICO Score
1 1
0.9 0.9
0.8 0.8
ARM Market Share
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92 94 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Loan‐to‐value Ratio Average Debt‐to‐Income Ratio
1 1
0.9 0.9
0.8 0.8
ARM Market Share
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
85 86 87 88 89 90 91 92 93 94 95 96 97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
1 1
0.9 0.9
0.8 0.8
ARM Market Share
ARM Market Share
0.7 0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3 0.2
0.2 0.1
0.1 0
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
24
Hybrid Sensitivity Tests
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐10 ‐7.5 ‐5 ‐2.5 0 2.5 5 7.5 10 12.5 15 17.5 20 2 3 4 5 6 7 8 9
One Year House Price Appreciation Unemployment Rate
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 ‐20 ‐15 ‐10 ‐5 0 5 10 15 20 25 30 35 40
House Price to Income Ratio Five Year Income Growth
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐20 ‐10 0 10 20 30 40 50 60 70 80 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
1 1
0.9 0.9
0.8 0.8
ARM Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
100
200
300
400
500
600
700
800
900
0
1000
1100
1200
1300
1400
1500
1600
1700
505 520 535 550 565 580 595 610 625 640 655 670 685 700 715
Population Per Square Mile Average FICO Score
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92 94 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Loan‐to‐value Ratio Average Debt‐to‐Income Ratio
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
85 86 87 88 89 90 91 92 93 94 95 96 97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
1 1
0.9 0.9
0.8 0.8
Hybrid Market Share
Hybrid Market Share
0.7 0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3 0.2
0.2 0.1
0.1 0
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
25
Interest Only Sensitivity Tests
1 1
0.9 0.9
0.8 0.8
0.7 0.7
IO Market Share
IO Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐10 ‐7.5 ‐5 ‐2.5 0 2.5 5 7.5 10 12.5 15 17.5 20 2 3 4 5 6 7 8 9
One Year House Price Appreciation Unemployment Rate
1 1
0.9 0.9
0.8 0.8
0.7 0.7
IO Market Share
IO Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 ‐20 ‐15 ‐10 ‐5 0 5 10 15 20 25 30 35 40
House Price to Income Ratio Five Year Income Growth
1 1
0.9 0.9
0.8 0.8
0.7 0.7
IO Market Share
IO Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐20 ‐10 0 10 20 30 40 50 60 70 80 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
1 1
0.9 0.9
0.8 0.8
0.7
IO Market Share
0.7
IO Market Share
0.6 0.6
0.5 0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
100
200
300
400
500
600
700
800
900
0
1000
1100
1200
1300
1400
1500
1600
1700
505 520 535 550 565 580 595 610 625 640 655 670 685 700 715
Population Per Square Mile Average FICO Score
1 1
0.9 0.9
0.8 0.8
0.7 0.7
IO Market Share
IO Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92 94 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Loan‐to‐value Ratio Average Debt‐to‐Income Ratio
1 1
0.9 0.9
0.8 0.8
0.7 0.7
IO Market Share
IO Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
85 86 87 88 89 90 91 92 93 94 95 96 97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
1 1
0.9 0.9
0.8 0.8
IO Market Share
0.7 0.7
IO Market Share
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2 0.1
0.1 0
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
26
Non Amortization Sensitivity Tests
1 1
0.9 0.9
Non‐Amort Market Share
0.8 0.8
ARM Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐10 ‐7.5 ‐5 ‐2.5 0 2.5 5 7.5 10 12.5 15 17.5 20 2 3 4 5 6 7 8 9
One Year House Price Appreciation Unemployment Rate
1 1
0.9 0.9
Non‐Amort Market Share
Non‐Amort Market Share
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 ‐20 ‐15 ‐10 ‐5 0 5 10 15 20 25 30 35 40
House Price to Income Ratio Five Year Income Growth
1 1
0.9 0.9
Non‐Amort Market Share
0.8 0.8
0.7 0.7
Non‐Amort Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐20 ‐10 0 10 20 30 40 50 60 70 80 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
1 1
0.9 0.9
Non‐Amort Market Share
Non‐Amort Market Share
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
1000
1100
1200
1300
1400
1500
1600
1700
100
200
300
400
500
600
700
800
900
0
505 520 535 550 565 580 595 610 625 640 655 670 685 700 715
Population Per Square Mile Average FICO Score
1 1
0.9 0.9
Non‐Amort Market Share
Non‐Amort Market Share
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92 94 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Loan‐to‐value Ratio Average Debt‐to‐Income Ratio
1 1
0.9 0.9
Non‐Amort Market Share
Non‐Amort Market Share
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
85 86 87 88 89 90 91 92 93 94 95 96 97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
1 1
0.9
Non‐Amort Market Share
0.9
Non‐Amort Market Share
0.8 0.8
0.7 0.7
0.6
0.6
0.5
0.5
0.4
0.4 0.3
0.3 0.2
0.2 0.1
0.1 0
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
27
Balloon Sensitivity Tests
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐10 ‐7.5 ‐5 ‐2.5 0 2.5 5 7.5 10 12.5 15 17.5 20 2 3 4 5 6 7 8 9
One Year House Price Appreciation Unemployment Rate
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 ‐20 ‐15 ‐10 ‐5 0 5 10 15 20 25 30 35 40
House Price to Income Ratio Five Year Income Growth
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
0.7 0.7
Balloon Market Share
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
‐20 ‐10 0 10 20 30 40 50 60 70 80 0 0.4 0.8 1.2 1.6 2 2.4 2.8 3.2 3.6 4 4.4
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0
0
100
200
300
400
500
600
700
800
900
0
1000
1100
1200
1300
1400
1500
1600
1700
505 520 535 550 565 580 595 610 625 640 655 670 685 700 715
Population Per Square Mile Average FICO Score
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
60 62 64 66 68 70 72 74 76 78 80 82 84 86 88 90 92 94 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50
Loan‐to‐value Ratio Average Debt‐to‐Income Ratio
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
85 86 87 88 89 90 91 92 93 94 95 96 97 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
1 1
0.9 0.9
0.8 0.8
Balloon Market Share
Balloon Market Share
0.7 0.7
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3 0.2
0.2 0.1
0.1 0
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
28
Appendix – In Sample Forecast
Actual = , Predicted = ------
ARM Market Share Hybrid Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Non‐Amortizing Market Share Interest Only Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Balloon Market Share
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2000 2001 2002 2003 2004 2005 2006 2007
29
Appendix – Out of Sample Forecast 2003-2007
Actual = , Predicted = ------
ARM Market Share Hybrid Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Non‐Amortizing Market Share Interest Only Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Balloon Market Share
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2000 2001 2002 2003 2004 2005 2006 2007
30
Appendix – Out of Sample Forecast 2003-2007 North East
New England and Middle Atlantic Census Divisions
Actual = , Predicted = ------
ARM Market Share Hybrid Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Non Amortizing Market Share Interest Only Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Balloon Market Share
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2000 2001 2002 2003 2004 2005 2006 2007
31
Appendix – Out of Sample Forecast 2003-2007 West Coast
Pacific Census Division
Actual = , Predicted = ------
ARM Market Share Hybrid Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Non Amortizing Market Share Interest Only Market Share
1 1
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
2000 2001 2002 2003 2004 2005 2006 2007 2000 2001 2002 2003 2004 2005 2006 2007
Balloon Market Share
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2000 2001 2002 2003 2004 2005 2006 2007
32
Appendix – Sample Specification Tests
33
Sample Specification Tests: Non Amortizing Results
2000‐2002 Sample 2003‐2007 Sample
Variable Coefficient t‐stat Coefficient t‐stat
%ΔHPI ‐0.004 ‐0.560 0.016** 12.760
Unemp 0.043 1.810 0.008 0.700
HP/I 0.000 0.010 0.047** 2.700
5yr‐I ‐0.016** ‐2.640 ‐0.026** ‐6.690
5yr‐HPI 0.013** 5.290 0.014** 12.470
Svol ‐0.079** ‐2.740 ‐0.029** ‐3.990
Pop 0.003 0.080 0.013 0.430
FICO 0.014** 6.230 0.033** 23.060
LTV 0.024** 2.170 0.074** 5.510
D/I ‐0.012 ‐0.970 0.073** 4.010
Mobility 0.003 0.360 ‐0.050** ‐5.530
Vol ‐1.020** ‐4.010 ‐0.328** ‐2.540
Yield ‐0.311** ‐6.120 ‐0.382** ‐13.760
Wealth 0.016 0.800 0.040** 3.890
Constant ‐11.602** ‐7.910 ‐26.980** ‐18.140
Adjusted R‐squared 0.34 0.88
**5 percent level of significance & *10 percent level of significance
34
Sample Specification Tests: Balloon Results
2000‐2002 Sample 2003‐2007 Sample
Variable Coefficient t‐stat Coefficient t‐stat
%ΔHPI ‐0.004 ‐0.690 ‐0.011** ‐5.040
Unemp 0.050** 2.050 ‐0.021 ‐1.640
HP/I ‐0.023 ‐0.850 ‐0.117** ‐6.720
5yr‐I ‐0.011* ‐1.910 ‐0.012** ‐3.510
5yr‐HPI 0.011** 4.490 0.011** 10.040
Svol ‐0.078** ‐3.130 ‐0.076** ‐7.030
Pop ‐0.015 ‐0.480 ‐0.044** ‐4.560
FICO 0.012** 5.440 0.018** 11.960
LTV 0.030** 2.840 0.033** 3.580
D/I ‐0.004 ‐0.380 0.212** 11.860
Mobility 0.010 1.100 ‐0.001 ‐0.110
Vol ‐1.008** ‐4.060 ‐0.157 ‐1.360
Yield ‐0.383** ‐7.590 ‐0.487** ‐15.830
Wealth 0.000 0.550 0.000** ‐4.230
Constant ‐11.805** ‐7.730 ‐22.659** ‐12.360
Adjusted R‐squared 0.28 0.75
**5 percent level of significance & *10 percent level of significance
35