Fraudulent Income Overstatement by Home Buyers During the Mortgage Credit Expansion
of 2002 to 2005
Atif Mian Princeton University
NBER
Amir Sufi University of Chicago Booth School of Business
NBER
Mian and Sufi (2009) Result in Chicago
(.51,.69](.35,.51](.23,.35][.14,.23]
Fraction subprime borrowers, 1996
Englewood
Garfield Park Downtown/MagMile
Hyde Park
Lakeview Lincoln Park
Mian and Sufi (2009) Result in Chicago Englewood,
Garfield Park Rest of Chicago
Fraction subprime 68% 37% Median household income, 2000 $24K $44K Poverty rate, 2000 35% 17% Mortgage credit growth, 02 to 05 55% 27% IRS income growth, 02 to 05 1.9% 4.0% Growth in income of home buyers, 02 to 05 7.7% 4.3% IRS income growth, 05 to 12 0.8% 2.3% Fraction of zips making top fraud list 75% 2%
Mian and Sufi (2009) Result in Chicago Englewood,
Garfield Park Rest of Chicago
Fraction subprime 68% 37% Median household income, 2000 $24K $44K Poverty rate, 2000 35% 17% Mortgage credit growth, 02 to 05 55% 27% IRS income growth, 02 to 05 1.9% 4.0% Growth in income of home buyers, 02 to 05 7.7% 4.3% IRS income growth, 05 to 12 0.8% 2.3% Fraction of zips making top fraud list 75% 2%
Mian and Sufi (2009) Result in Chicago Englewood,
Garfield Park Rest of Chicago
Fraction subprime 68% 37% Median household income, 2000 $24K $44K Poverty rate, 2000 35% 17% Mortgage credit growth, 02 to 05 55% 27% IRS income growth, 02 to 05 1.9% 4.0% Growth in income of home buyers, 02 to 05 7.7% 4.3% IRS income growth, 05 to 12 0.8% 2.3% Fraction of zips making top fraud list 75% 2%
1. What caused the unprecedented growth in mortgage credit for home purchase in low credit score zip codes? (MS09)
• This is about extensive margin, does not explain rise in total debt, use county FE
2. What caused the rise in total household debt in the United States? (MS12, MS14)
• Home equity-based borrowing, not just by low income, utilize across-county variation
Our Research: Two Separate Questions
Their summary stats show poor zips had: Higher mortgage credit growth from 02 to 05 Lower IRS income growth from 02 to 05
They do not dispute our research on: home equity borrowing and rise in aggregate
household leverage role of household debt in Great Recession
Their average mortgage size result is perfectly consistent with MS09
What Is in Dispute? Not Much
MS09 Result Shows Up in Adelino et al
MS09 Result Shows Up in Adelino et al
Any result in Adelino et al using income information from mortgage applications of home-buyers should be heavily discounted We have no problem with any result not
using mortgage application income (i.e., average mortgage size result) But, we believe MS12, MS14 already
explain aggregate rise in household debt, not cited or discussed in Adelino et al
Our Bottom Line
Visualization of Main Issue in Dispute
fraud?199820012011
2005
1
1.5
2
2.5
3
Buy
er a
pplic
atio
n in
com
e to
IRS
inco
me
ratio
0 1 2 3 4 5 6 7 8 9 10Average IRS income decile of zip codes
Buyer application incometo IRS income ratio
Dispute in a Table
Mortgage originations for home purchase growth 2002 to 2005, annualized
(1) (2) (3) (4) IRS income growth, 2002 to 2005, annualized -0.662** -0.705** (0.089) (0.087) Buyer income growth, 2002 to 2005, annualized 0.420** 0.433** (0.038) (0.038) Buyer income overstatement, 2002 to 2005 0.473**
(0.035) N 3,014 3,014 3,014 3,014 R2 0.380 0.394 0.407 0.406
Buyer income overstatement, 2002 to 2005 = Income growth according to mortgage applications – Income growth according to IRS in zip code
What neighborhoods saw high buyer income overstatement? The same subprime zip codes analyzed
extensively in Mian and Sufi (QJE 2009)! According to Adelino et al, mortgage boom
was high income individuals moving into low credit score, poor neighborhoods
Some Background
Overstatement, Credit Scores, Income
-.02
0
.02
.04
.06
Buy
er in
com
e ov
erst
atem
ent,
2002
to 2
005
Most prime 2 3 Most subprimeZip code credit scores, 1996
By 1996 credit scores
-.02
0
.02
.04
.06
Buy
er in
com
e ov
erst
atem
ent,
2002
to 2
005
Lowest income 2 3 Highest incomeAverage IRS income, 2002
By 2002 income
Overstatement Characteristics
(1) (2) (3) (4) (5) (6) Fraction
subprime, 1996
Ln[Median household income],
2000
Poverty rate, 2000
Fraction with less than high
school education,
2000
Fraction unemployed,
2000
Household debt default rate, 2000
Buyer income overstatement 0.193** -0.253** 0.178** 0.249** 0.168** 0.205** (0.026) (0.023) (0.021) (0.025) (0.019) (0.024) N 3,014 3,014 3,014 3,014 3,014 3,014
1. Aggregate evidence from extant research 2. Direct measures of fraud higher in zip
codes with high overstatement 3. Zip codes in question see terrible
contemporaneous and ex post outcomes – no evidence of gentrification
4. Break down in correlation between IRS and mortgage application income growth
Application Income is Not True Income
1. Aggregate Evidence
FCIC: “about $1 trillion of the loans made during the [2005 to 2007] period were fraudulent” State of Illinois: “60% of the [stated loan]
income amounts were inflated by more than 50%”
Credit Boom and Rampant Fraud
“At the downtown L.A. branch [of mortgage lender Ameriquest], some of Glover's coworkers had a flair for creative documentation. They used scissors, tape, Wite-Out, and a photocopier to fabricate W-2s, the tax forms that indicate how much a wage earner makes each year. It was easy: Paste the name of a low-earning borrower onto a W-2 belonging to a higher-earning borrower and, like magic, a bad loan prospect suddenly looked much better. Workers in the branch equipped the office's break room with all the tools they needed to manufacture and manipulate official documents. They dubbed it the ‘Art Department.’”
Michael Hudson, WSJ Reporter
On fraudulent income overstatement: Avery, Bhutta, Brevoort, and Canner (2013) Blackburn and Vermilyea (2012) Jiang, Nelson, and Vytlacil (2014)
Other types of fraud: Griffin and Maturana; Piskorski, Seru, and Witkin; Ben David; Garmaise One lesson: fraud concentrated in non-
GSE private label securitization market
Academic Research
Easy to See Fraud in Aggregate Data
-.01
0
.01
.02
.03
Ann
ualiz
ed re
al in
com
e gr
owth
HMDA, 2000 to 2005 ACS/Census, 2000 to 2005
HMDA versus ACS/Census
-.03
-.02
-.01
0
.01
.02
Ann
ualiz
ed re
al in
com
e gr
owth
HMDA, 2002 to 2005 AHS, 2002 to 2005
HMDA versus AHS
HMDA shows strong increase in homebuyer income from 02 to 05, ACS/Census and AHS show decline
“users of the HMDA data should be aware that borrower income was likely significantly overstated during the peak of the housing boom, particularly in some areas of the country. One potential implication of this finding is that lending to lower-income borrowers, as measured in the HMDA data, may be attenuated around the peak of the housing market”
Avery, et al (2013)
2. Direct Evidence of Fraud
Fraud in High Overstatement Zip Codes
(1) (2) (3) (4) (5) (6) (7) InterThinx Measures of
Fraud Piskorski, Seru, and Witkins (2014)
Measures of Fraud Change in
non-agency share of
mortgages, 2002-2005
Change in low-doc share of
mortgages, 2002-2005
Zip code makes top mortgage fraud list,
2010
Zip code makes top mortgage fraud list, 2010-2014
Misreported non-owner-
occupant
Misreported second lien
Either misreported
Buyer income overstatement 0.121** 0.100** 0.051** 0.123** 0.030** 0.034** 0.051** (0.020) (0.016) (0.018) (0.045) (0.009) (0.009) (0.009) N 2,981 2,981 3,014 3,014 2,969 2,969 2,969 R2 0.483 0.598 0.071 0.067 0.271 0.321 0.245
3. No Gentrification
Negative Gentrification
Growth in IRS Income from time x to time y (1) (2) (3) (4) (5)
Period: x = 1991 y = 1998
x = 1998 y = 2002
x = 2002 y = 2004
x = 2004 y = 2005
x = 2005 y = 2006
Ln(Buyer income) – Ln(IRS income), at time x 0.004* 0.043** -0.014** -0.040** -0.023** (0.002) (0.002) (0.002) (0.005) (0.003) N 2,590 3,013 3,014 3,014 3,014 R2 0.226 0.405 0.226 0.213 0.185
High Overstatement Zips Fall Apart
(1) (2) (3) (4) (5) (6) (7) IRS income
growth, 2005 to
2012
IRS wage growth, 2005 to
2012
Census income growth 2000 to
2010
Change in poverty rate,
2000 to 2010
Change in unemp rate,
2000 to 2010
Change in mortgage
default rate, 2005 to
2007
Change in mortgage
default rate, 2005 to
2010 Buyer income overstatement -0.094** -0.149** -0.121** 0.041** 0.024** 0.059** 0.126** (0.020) (0.018) (0.022) (0.007) (0.006) (0.007) (0.012) N 3,011 3,011 3,011 3,011 3,011 3,014 3,014 R2 0.325 0.372 0.412 0.333 0.252 0.331 0.529
4. Correlation Breakdown
Breakdown in Low GSE Share Zips
Buyer income growth from mortgage applications, annualized (1) (2) (3) (4)
Time period 1991 to 1998 1998 to 2002 2002 to 2005 2005 to 2007 IRS Income growth*Quartile 1 GSE share 0.481** 0.550** -0.075 0.269** (0.039) (0.136) (0.064) (0.052) IRS Income growth*Quartile 2 GSE share 0.298** 0.422* 0.134 0.172* (0.056) (0.174) (0.094) (0.075) IRS Income growth*Quartile 3 GSE share 0.297** 0.555** 0.272** 0.279** (0.060) (0.165) (0.087) (0.083) IRS Income growth*Quartile 4 GSE share 0.238** 0.330 0.314** 0.307** (0.081) (0.181) (0.107) (0.085) N 2,590 3,013 3,014 3,014 R2 0.535 0.210 0.367 0.188
Arguments in Adelino et al
“In 2005, buyer income 75% more than IRS income – too large to be fraud” Incorrect comparison: fraud buyers
overstate 20-30% relative to regular buyers, not to average IRS income HMDA 2005 home buyer income: $99K ACS 2005 home buyer income: $83K Average IRS income: $53K
Fraud easily explains difference
“Fraud Too Small to Explain Gap”
“Buyer income positively related to credit growth even within high GSE share zip codes where fraud was less likely” We agree, but this is because there was
no fraud and no big credit supply expansion in these zip codes We would expect positive correlation between
true income growth and credit growth where credit supply expansion less important
“GSE versus Non-GSE Results”
“Change in average mortgage size from 2002 to 2005 conditional on origination positively related to IRS income” Yes, we agree, but this is perfectly
consistent with MS09 Marginal individuals in subprime zip codes
getting mortgages obtain smaller mortgages Even if mortgage amounts constant in prime
zip codes, we would expect this result
“Average Mortgage Size Matters”
Mortgage Size Result and MS09
Low Credit Score Zip High Credit Score Zip
Individual: Subprime Prime Prime Prime
Mortgage size, 2002 - 100 100 100
Mortgage size, 2005 50 100 100 100
Mortgage origination growth
(150-100)/100 = 50% (200-200)/200 = 0%
Growth in average mortgage size, conditional on origination
(75 – 100)/100= -25% (100 – 100)/100= 0%
Other Issues
We agree: expansion to low income extensive margin cannot explain rise in total debt (MS09 never claimed this) Home equity-based borrowing much more
important, was present even among middle to upper-middle households: MS12, MS14 (not really addressed here) But, aggregate results using fraudulent
application income are incorrect
Aggregate Increase in Household Debt
Aggregate Share of Mortgages, Defaults
0
.1
.2
.3
.4
Most prime 2 3 Most subprime
Fraction of all originations, 2002 and 2005
2002 2005
0
.1
.2
.3
Most prime 2 3 Most subprime
Fraction of all defaults, 2007, 2008, 2009
2007 2008 2009
In MS09, we used sample of zip codes with house price data available (45% of 2002 debt outstanding) Adelino et al claim negative correlation
between credit growth and income growth not robust when using full sample Major flaw: when moving to full sample,
one must weight by zip code population because of extremely small zip codes
Robust to Full Sample?
MS09 Result Robust to Full Sample
Mortgage originations for home purchase growth, 2002 to 2005, annualized (1) (2) (3) (4) (5) (6) (7) IRS Income growth, 02 to 05, annualized 0.143** 0.005 -0.124* -0.203** -0.123** -0.304** -0.662** (0.049) (0.051) (0.051) (0.052) (0.046) (0.065) (0.089) Sample Full Full Full Full Full >5000 House
Price Winsorized 1% 5% 10% Population weights No No No No Yes No No Fraction of 2000 population 100% 100% 100% 100% 100% 80% 29% Fraction of 2000 mortgage debt outstanding
100% 100% 100% 100% 100% 85% 45%
N 18,336 18,336 18,336 18,336 18,336 7,622 3,014 R2 0.356 0.407 0.428 0.432 0.378 0.452 0.380
Conclusion
Main argument in Adelino et al is: “lending technology in early 2000s did not change fundamentally – no real credit supply shift” There are more than 10 published
academic articles (excluding us) contradicting this argument Top regulators also disagree – Bernanke
and Dudley on record saying housing boom mainly due to lending change
No Change in Lending Technology?
Levitin and Wachter (2013) : “the bubble was, in fact, a supply-side phenomenon, meaning that it was caused by excessive supply of housing finance … it was the result of a fundamental shift in the structure of the mortgage finance market from regulated to unregulated securitization.”
Justiniano, Primiceri, and Tambalotti (2014): “the housing boom that preceded the Great Recession was due to an increase in credit supply driven by looser lending constraints in the mortgage market.”
Landvoigt, Piazzesi, and Schneider (2014): “cheaper credit for poor households was a major driver of prices, especially at the low end of the market.”
Demyanyk and Van Hemert (2011) show that “loan quality – adjusted for observed characteristics and macroeconomic circumstances – deteriorated monotonically between 2001 and 2007.”
Mayer, Pence, and Sherlund (2009): “lending to risky borrowers grew rapidly in the 2000s. We find that underwriting deteriorated along several dimensions: more loans were originated to borrowers with very small down payments and little or no documentation of their income or assets, in particular.”
Academic Research
MS09 main result confirmed: expansion of mortgage credit to low credit score zip codes was unrelated to fundamental improvements in income Credit supply shift, driven by private label
securitization, was the culprit Maybe fraud exploded in subprime zip
codes because brokers responding to same credit supply shift? Future research
Final Thoughts