Credit Growth and the Financial Crisis: A New Narrative Stefania Albanesi, University of Pittsburgh Giacomo DeGiorgi, University of Geneva Jaromir Nosal, Boston College
DIW Berlin Seminar, December 7, 2016
Introduction • Prevailing narrative about the financial crisis credit growth during boom concentrated in subprime segment defaults during financial crisis also concentrated in this segment ) expansion of subprime credit viewed as a leading cause for the crisis
Introduction • Prevailing narrative about the financial crisis credit growth during boom concentrated in subprime segment defaults during financial crisis also concentrated in this segment ) expansion of subprime credit viewed as a leading cause for the crisis • Mechanism: Defaults&foreclosures ! Drop in house prices ! Contraction in credit&consumption ! Recession
(Justiniano & al. 2016, Berger & al. 2015, Lorenzoni & Guerreri 2015, Kehoe, Midrigan & Pastorino 2014, Mian & Sufi 2014, Midrigan & Philippon 2016, Kaplan, Mittman &Violante 2016, Hedlund & Garriga 2016, etc.)
Our Contribution • Study behavior of household credit and delinquency, 1999-2013 • Using large administrative panel: FRBNY CCP/Equifax data debt positions, delinquencies, credit scores Findings:
Our Contribution • Study behavior of household credit and delinquency, 1999-2013 • Using large administrative panel: FRBNY CCP/Equifax data debt positions, delinquencies, credit scores Findings: 1. Credit growth during boom and defaults during the crisis concentrated at mid/top of credit score distribution for all debt categories challenges prevailing narrative (consistent with Adelino, Shoar & Severino 2015 and Foote, Loewenstein & Willen 2016 findings for home debt) raises questions on drivers of default risk & policy responses
Our Contribution • Study behavior of household credit and delinquency, 1999-2013 • Using large administrative panel: FRBNY CCP/Equifax data debt positions, delinquencies, credit scores Findings: 1. Credit growth during boom and defaults during the crisis concentrated at mid/top of credit score distribution for all debt categories challenges prevailing narrative (consistent with Adelino, Shoar & Severino 2015 and Foote, Loewenstein & Willen 2016 findings for home debt) raises questions on drivers of default risk & policy responses 2. Also true within zip codes, including those with large subprime population Empirical approach based on geographical variation distorts evidence on distribution of debt and defaults, overstates role of subprime debt
Data • Federal Reserve Bank of New York Consumer Credit Panel/Equifax Data quarterly, 1999:Q1-2013:Q4 all individuals with an Equifax credit report (anonymous) • Use 1% sample – about 2.5 million individuals each quarter • Over 600 variables all aspects of financial liabilities: by type of account, balances, numbers delinquent behavior: by severity, type of debt public record items: court judgements, collections, etc. credit score, age, ZIP code • For 2009, matched to payroll data
Existing Evidence • Due to lack of demographic data in credit file, credit scores typically used to rank individuals or geographical aggregates (zip codes, MSAs)
Existing Evidence • Due to lack of demographic data in credit file, credit scores typically used to rank individuals or geographical aggregates (zip codes, MSAs) • What is a credit score?
proprietary measure, ranks individuals by probability of 90 day+ delinquency in subsequent 8 quarters uses only past borrowing behavior (FCRA, CCRRA), intended to capture idiosyncratic default risk
Existing Evidence • Due to lack of demographic data in credit file, credit scores typically used to rank individuals or geographical aggregates (zip codes, MSAs) • What is a credit score?
proprietary measure, ranks individuals by probability of 90 day+ delinquency in subsequent 8 quarters uses only past borrowing behavior (FCRA, CCRRA), intended to capture idiosyncratic default risk
• Individuals or geographical areas ranked by initial credit scores 1996 (Mian & Sufi 2009, for zip codes)
1997 (Mian & Sufi 2016, for individuals)
New Evidence • Our approach: Rank individuals by credit score at time of borrowing recent score, though not contemporaneous with delinquencies
New Evidence • Our approach: Rank individuals by credit score at time of borrowing recent score, though not contemporaneous with delinquencies • Our findings: Credit growth during boom concentrated in prime segment Defaults concentrated in middle of credit score distribution share of new defaults to subprime drops during crisis
New Evidence • Our approach: Rank individuals by credit score at time of borrowing recent score, though not contemporaneous with delinquencies • Our findings: Credit growth during boom concentrated in prime segment Defaults concentrated in middle of credit score distribution share of new defaults to subprime drops during crisis • Our analysis: Debt growth for low initial credit scores mostly explained by life cycle Geographical aggregation further distorts role of initial credit score
Prevailing Narrative: Mortgage Balances • Ranking individuals and zip codes by initial credit scores or fraction of subprime, following Mian&Sufi 2016 and 2009 • Individuals with low initial credit score and zip codes with high initial fraction of subprime borrowers exhibit stronger debt growth Individuals by 1999 Credit Score
Zip Codes by 2001 Fraction of Subprime 2.4
3.6 3.2
2
2.8 2.4
1.6
2 1.6
1.2
1.2 0.8
0.8 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Per capita real mortgage balances, ratio to 2001Q3. Deflated by CPI-U. Individual quartile cuto↵s: 615, 720, 791, 840. Source: Authors’ calculation based on FRBNY’s Consumer Credit Panel/Equifax Data.
Outline – Why initial credit score rankings are flawed – A better approach, based on a lender’s perspective – New findings and puzzles Role of investors – Broader implications Should we use geographically aggregated evidence?
The Problem with Initial Credit Score Rankings • Initial credit score rankings mechanically magnify debt growth for low credit score borrowers due to life cycle demand for credit
The Problem with Initial Credit Score Rankings • Initial credit score rankings mechanically magnify debt growth for low credit score borrowers due to life cycle demand for credit • Low credit score individuals disproportionately young young experience future credit growth due to life cycle life cycle credit growth related to life cycle income growth credit score at time of borrowing higher than when young
The Problem with Initial Credit Score Rankings • Initial credit score rankings mechanically magnify debt growth for low credit score borrowers due to life cycle demand for credit • Low credit score individuals disproportionately young young experience future credit growth due to life cycle life cycle credit growth related to life cycle income growth credit score at time of borrowing higher than when young • Credit growth for subprime individuals in 1999 driven by life cycle demand
Credit Score and Age
Younger than 24
25-34
35-44
45-54
55-64
Older than 65
0
.1
.2
.3
.4
• Low credit score individuals disproportionately young
1
2
3
4
Fraction in each age bin in 1999 by Equifax Risk Score quartile in 1999. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Score and Age • Low credit score individuals in 1999 exhibit credit score growth over time Ratio to 2001Q1 (3QMA) 1.12 1.1 1.08
1.12
Quartile 1 (Lowest) Quartile 2 Quartile 3 Quartile 4 (Highest)
1.1 1.08
1.06
1.06
1.04
1.04
1.02
1.02
1 0.98 2001
1 0.98 2003
2005
2007
2009
2011
2013
Current credit score as ratio to 1999, by Equifax Risk Score quartile in 1999. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Score, Borrowing and Age • Estimated age e↵ects Credit Score
Debt Balances
160
$40,000
140
$30,000
120
$20,000
100
$10,000
Total Debt Balances
80
$0
60
-$10,000
40
-$20,000
20
-$30,000
21
26
31
36
41
46
51
56
61
66
71
-$40,000
0 21 -20
Mortgage Balances
26
31
36
41
46
51
56
61
66
71
76
81 -$50,000
Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
76
81
Credit Score, Borrowing and Age
• Age e↵ects in credit score =) young borrowers rise by 1-2 quartiles in credit score ranking over 10 year period • Age e↵ects in debt growth =) young borrowers experience $30,000 increase in balances over 10 year period • E↵ects of life cycle di↵er across quartiles of the initial credit score distribution due to di↵erent age structure ! isolate role of age distribution and role of life cycle
Life Cycle and Credit Demand I. Role of di↵erence in age distribution across quartiles Di↵erence from Quartile 4 in 2001Q3-2007Q4 balance growth accounted for Quartile 1: 25%, Quartile 2: 20%, Quartile 3: 14% Quartile 1
Quartile 2
3.75
3.75
3.25
3.25
2.75
2.75
2.25
2.25
1.75
1.75
1.25
1.25
0.75
0.75 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Actual
CF: Quartile 1 age distribution
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
CF: Quartile 4 age distribution
Actual
Quartile 3
CF: Quartile 4 age distribution
Quartile 4
3.75
3.75
3.25
3.25
2.75
2.75
2.25
2.25
1.75
1.75
1.25
CF: Quartile 1 age distribution
1.25
0.75
0.75 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Actual
CF: Quartile 1 age distribution
CF: Quartile 4 age distribution
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Actual
CF: Quartile 1 age distribution
CF: Quartile 4 age distribution
Real mortgage balances by 1999 Equifax Risk Score quartile. Counterfactuals set the age distribution equal to the one for quartile 1 and quartile 4. Source: Authors’ calculation based on Federal Reserve Bank of New York’s Consumer Credit Panel/Equifax Data.
Life Cycle and Credit Demand II. Removing life cycle e↵ects Absent lifecycle e↵ects, credit growth by initial credit score similar across quartiles during boom Actual
Counterfactual
3.6
3.6
3.2
3.2
2.8
2.8
2.4
2.4
2
2
1.6
1.6
1.2
1.2 0.8
0.8 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Real mortgage balances by 1999 Equifax Risk Score quartile, actual and counterfactual. Ratio to 2001Q3. Counterfactual assigns to each 1999 age bin, in each quarter, debt balances of those who currently are in that age bin. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Scores, Debt Balances and Income • Alternative to initial credit scores? Recent credit scores
strongly positively related to income, conditional on age better indicators of default risk
Credit Scores, Debt Balances and Income • Alternative to initial credit scores? Recent credit scores
strongly positively related to income, conditional on age better indicators of default risk
• Use supplementary payroll data for 2009, merged with Equifax CCP 11k observations, comparable with CPS, ACS
Credit Scores, Debt Balances and Income • Alternative to initial credit scores? Recent credit scores
strongly positively related to income, conditional on age better indicators of default risk
• Use supplementary payroll data for 2009, merged with Equifax CCP 11k observations, comparable with CPS, ACS
• Recent credit scores strongly positively related to income, given age Slope declines with age
• Estimates consistent for 8Q, 4Q lagged, current, and 4Q, 8Q ahead credit scores
Credit Scores, Debt Balances and Income • Recent credit scores strongly positively related to income, given age Slope declines with age 850
800
750
700
650
600
550
25
30
35
40
45
50
55
60
65
Predicted 8Q lagged Equifax Risk Score by age and 2009 Worknumber total annual labor income, for age specific 1-99 percentile of income range. See specification. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Debt and Income Over the Life Cycle • How do credit scores and borrowing by 1999 age vary with 2009 income? highest growth for young in 1999 with high income in 2009
Debt and Income Over the Life Cycle • How do credit scores and borrowing by 1999 age vary with 2009 income? highest growth for young in 1999 with high income in 2009 25-34 yo in 1999 by their income quintile in 2009
2003
2005
2007
2009
2001
2.5 1
1.5
2
Ratio to 2001 2 2.5 1.5 1
60 45 30 15 0
0 2001
Quintile 1 (Lowest) Quintile 5 (Highest)
Difference from 2001 15 30 45 60
Quintile 5 (Highest)
3
Quintile 1 (Lowest)
3
Total debt balances 75
75
Credit score
[35-44 in 1999] [45-54 in 1999]
2003
2005
2007
2009
Equifax Risk Score and total debt balances for 25-34 yo in 1999 by their 2009 Worknumber total annual labor income quantile. Di↵erence with 2001 (credit score) and ratio to 2001 (balances). Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Debt and Income Over the Life Cycle • How do income and borrowing by 1999 credit score vary with 2009 score? Highest growth in both income and borrowing for high 2009 credit scores
Debt and Income Over the Life Cycle • How do income and borrowing by 1999 credit score vary with 2009 score? Highest growth in both income and borrowing for high 2009 credit scores
• Quartile 1 in 1999 credit score ranking: subsequent credit score growth associated with income and balance growth
2009 credit score Debt balances Income
Quartile 1 $38k $39k
Quartile 2 $74k $47k
Quartile 3 $126k $57k
Quartile 4 $213k $62k
Mean income and total debt balances by 2009 Equifax Riskscore quartile for individuals in the first quartile of the 1999 Equifax Risk Score distribution. Worknumber total annual labor income for restricted Worknumber sample. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Life Cycle: Summary • Di↵erences in credit growth by 1999 credit score largely explained by life cycle factors
Role of Life Cycle: Summary • Di↵erences in credit growth by 1999 credit score largely explained by life cycle factors • Credit score and debt growth for young/low score individuals in 1999 during the credit boom related to income growth Consistent with PSID analysis on relation between income and borrowing
Role of Life Cycle: Summary • Di↵erences in credit growth by 1999 credit score largely explained by life cycle factors • Credit score and debt growth for young/low score individuals in 1999 during the credit boom related to income growth Consistent with PSID analysis on relation between income and borrowing • Strong correlation between recent credit scores and income suggests recent credit scores better indication of default risk
Credit Growth and Defaults by Credit Score • How does credit growth vary with credit scores at time of borrowing? • Approach: Lender’s perspective
Credit Growth and Defaults by Credit Score • How does credit growth vary with credit scores at time of borrowing? • Approach: Lender’s perspective • Baseline specification: i Bt,t+h =
X
↵(j
1)
+ ⌘ CSti
1,t 1 k
+ time fe + age fe + "it
j=1,2,3,4
i Bt,t+h = change in balances between t and t + h
i = individual, t = quarter, h 2 {4, 8, 12} horizon
1 ) = e↵ect for 1Q lagged quartile of credit score distribution CSti 1,t 1 k = change in credit score between t 1 and t 1
↵(j
also includes time ⇥ quartile interactions
k
Credit Growth and Defaults by Credit Score • How does credit growth vary with credit scores at time of borrowing? • Approach: Lender’s perspective • Baseline specification: i Bt,t+h =
X
↵(j
1)
+ ⌘ CSti
1,t 1 k
+ time fe + age fe + "it
j=1,2,3,4
i Bt,t+h = change in balances between t and t + h
i = individual, t = quarter, h 2 {4, 8, 12} horizon
1 ) = e↵ect for 1Q lagged quartile of credit score distribution CSti 1,t 1 k = change in credit score between t 1 and t 1
↵(j
also includes time ⇥ quartile interactions • Findings: Highest growth for prime borrowers during boom
k
Credit Growth by Credit Score: Mortgage Debt • 8Q ahead change in mortgage debt increased with 1Q lagged credit score • Negligible contribution of past credit score change to future borrowing • Similar findings at 4Q and 12Q horizon and for mortgage balances Dependent Variable: 8Q Ahead Mortgage Balance Change (USD) 1
1Q lagged CS Quartile E↵ects 2 3 4
3,182
9,559
9,291
4,803
4,129
10,164
9,787
5,173
CS 4Q
1
6Q
50 51
Estimated 1Q lagged Equifax Riskscore quartile e↵ects and coefficients for 4Q, 6Q past change from 1Q lagged score in balance change regressions. Baseline specification. All estimates significant at 1% level. Sample period 2001Q1-2011Q4. Number of obs. 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Growth by Credit Score: Mortgage Debt • Estimated time e↵ects suggest little growth for quartile 1 during boom 9000 7000 5000 3000 1000 -1000 -3000 -5000 -7000 -9000 -11000 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Estimated time e↵ects by 1Q lagged Equifax Riskscore quartile from balance change regressions. Baseline specification. Dependent variable is the 8Q ahead change in per capita mortgage balances in USD. Sample period 2001Q1-2011Q4. Number of obs. (baseline) 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Growth by Credit Score: Mortgage Debt • Sizable and highly significant di↵erence in time e↵ects across quartiles • Similar for 4Q ahead and 12Q ahead and for mortgage balances Quart2-Quart1
Quart3-Quart1
Quart4-Quart1
7000
7000
7000
5000
5000
5000
3000
3000
3000
1000
1000
1000
-1000
-1000
-1000
-3000
-3000
-3000
Estimated time e↵ects by 1Q lagged Equifax Riskscore quartile from balance change regressions. Baseline specification. Dependent variable is the 8Q ahead change in per capita mortgage balances in USD. Dashed lines denote 5% confidence intervals. Sample period 2001Q3-2011Q4. Number of obs. (baseline) 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Growth by Credit Score: Role of Age • Sizable estimated age e↵ects, only for quartiles 2-4 =)
no life cycle growth in debt for young borrowers who remain in quartile 1 25000
20000
15000
10000
5000
0
-5000 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Estimated age e↵ects from balance change regressions. Baseline specification. Dependent variable is the 8Q ahead change in per capita mortgage balances in USD. Sample period 2001Q3-2011Q4. Number of obs. (baseline) 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Additional Evidence: Extensive Margin • Fraction with first mortgages (homeownership rate) and new mortgage originations virtually constant for quartile 1 during boom Fraction with First Mortgages
Fraction with New Originations
0.6
0.3
0.5 0.4
0.2 0.3 0.2
0.1
0.1
0
0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Fraction with first mortgages and fraction with new mortgage originations by 8Q lagged Equifax Risk Score quartile . Quartile cuto↵s: 615, 720, 791, 840. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Credit Scores at Origination
.2 0
.2 0 2001
.6
Quartile 2 Quartile 4 (Highest)
.4
Quartile 1 (Lowest) Quartile 3
.4
.6
• No growth in share of originations for quartile 1 during boom
2003
2005
2007
2009
2011
2013
Individuals with a new mortgage origination. Fraction in each quartile of the 4Q lagged Equifax Risk Score distribution. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Defaults by Credit Score: Delinquent Balances • Change in delinquent balances falls with 1Q lagged credit score on average • Negligible contribution of past credit score change to delinquencies • Similar findings at 4Q and 12Q horizon and for mortgage balances Dependent Variable: 8Q Ahead 90+ Days Delinquent Debt Balance Change (USD) 1
1Q lagged CS Quartile E↵ects 2 3 4
505
635
227
194
993
856
404
318
4Q
CS 6Q
1
33 34
Estimated 1Q lagged Equifax Risk Score quartile e↵ects and coefficients for 4Q, 6Q past change from 1Q lagged Riskscore in balance change regressions. Baseline specification. All estimates significant at 1% level. Sample period 2001Q3-2011Q4. Number of obs. 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Defaults by Credit Score: Delinquent Balances • Estimated time e↵ects suggest large rise for quartiles 2-3 during crisis 5800 3800 1800 -200 -2200 -4200 -6200 -8200 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Estimated time e↵ects by 1Q lagged Equifax Risk Score quartile from balance change regressions. Baseline specification. Dependent variable is the 8Q ahead change in per capita 90+ days delinquent debt balances in USD. Sample period 2001Q3-2011Q4. Number of obs. 64,588,488. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Defaults by Credit Score: Delinquencies • Quartile 1 share of new delinquencies drops during crisis
2001 2003 2005 2007 2009 2011 2013
.6 .4 .2
Quartile 4 (Highest)
0
.6
Quartile 2
Quartile 3
Share (3QMA) .2 .4
Quartile 1 (Lowest)
0
.02 .01
.015
Quartile 4 (Highest)
.005
Quartile 3
Share of
0
Quartile 2
Fraction (3QMA) .005 .01 .015
Quartile 1 (Lowest)
0
.02
Fraction with
2001 2003 2005 2007 2009 2011 2013
New 90 days+ delinquencies by credit score quartile, 8Q lagged Equifax Risk Score. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Defaults by Credit Score: New Foreclosures • Quartile 1 share of new foreclosures drops during crisis
.8
Quartile 1 (Lowest)
.8
Share of .006
.006
Fraction with Quartile 2
2001
2003
2005
2007
2009
.6 Quartile 2 Quartile 3
.4
Quartile 1 (Lowest)
.2
.2 0
Quartile 4 (Highest)
0
.004 0
0
.002
Fraction (3QMA) .002 .004
Quartile 4 (Highest)
Share (3QMA) .4 .6
Quartile 3
2001 2003 2005 2007 2009 2011 2013
New foreclosures by credit score quartile, 8Q lagged Equifax Risk Score. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Summary • Initial credit score ranking overstates credit growth for low credit scores • Using lender’s approach based on recent credit scores: credit growth during boom concentrated in prime quartiles rise in defaults concentrated in middle of credit score distribution share of new delinquencies/foreclosures to subprime drops during crisis ! rise in foreclosures 50% lower with no rise for prime borrowers
Summary • Initial credit score ranking overstates credit growth for low credit scores • Using lender’s approach based on recent credit scores: credit growth during boom concentrated in prime quartiles rise in defaults concentrated in middle of credit score distribution share of new delinquencies/foreclosures to subprime drops during crisis ! rise in foreclosures 50% lower with no rise for prime borrowers • Why did individuals with ’good credit’ experience defaults during crisis? Rise in investors (multiple first mortgages)
Risky loans (non-conforming loans, Alt-A loans, ARM etc) Large aggregate shock (decline in house prices, rise in unemployment)
Role of Investors • Investors (2+ first mortgages) rise during boom for mid/top quartiles
2009
2011
.96 .92 .84
2005
2007
2009
2011
2013
2013
.84
.05 2001
.96
.2
.96
2+ First Mortgages
.15
1 First Mortgage (Right Axis)
.1
.92 2007
0
.88 2005
.8
.2 .15 .1 .05
2003
2003
Quartile 4
2+ First Mortgages
0 2001
.8
.05 2001
Quartile 3 1 First Mortgage (Right Axis)
.88
.15
.84
0
2013
.92
2011
.88
2009
.84
2007
2+ First Mortgages
.8
2005
.88
.92 2003
.8
0 2001
1 First Mortgage (Right Axis)
.1
.2 .15 .1
2+ First Mortgages
.05
1 First Mortgage (Right Axis)
.2
Quartile 2 .96
Quartile 1
2003
2005
2007
2009
2011
2013
Fraction with 1 and 2+ first mortgages by 8Q lag Equifax Riskscore quartile. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Investors: Mortgage Balances • Share held by investors grows during boom, especially for mid/top 1
.8 .7
.25 2007
2009
2011
2001
.6 2003
2005
2007
2009
2011
2013
.35
1
2+ First Mortgages
2001
.7
.25 2005
2007
2009
2011
2013
2001
.6 .5
.6
.15 .1
.2 2003
.5
.1
.15
.2
.7
.25
.8
.3
.9
1 First Mortgage (Right Axis)
.8
.3
.4
Quartile 4
2+ First Mortgages
.35
.4
Quartile 3 1 First Mortgage (Right Axis)
.5
.6
.15 .1
2013
1
2005
.9
2003
.5
.2
.7
.25 .2 .15 .1 2001
1 .9
.35
2+ First Mortgages
.3
.9
1 First Mortgage (Right Axis)
.8
.3
.4
Quartile 2
2+ First Mortgages
.35
.4
Quartile 1 1 First Mortgage (Right Axis)
2003
2005
2007
2009
2011
2013
Share of mortgage balances by number of first mortgages by 8Q lag Equifax Riskscore quartile. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Investors: Mortgage Delinquencies • Rise in fraction with 90+ days delinquent mortgage balances only for investors during crisis Quartile 1
Quartile 2
2009
2011
Quartile 3
2013
.5 .3 .2 .1
.1 0
0
.1 0
2001
.2
.3
.4
.5
2+ First Mortgages
.5
.5 2011
2013
.4
.4 2009
2011
.3
.3 2007
2009
1 First Mortgage
.2 2005
2007
.2
.4 .3 .2 .1
2003
2005
Quartile 4
0 2001
2003
2+ First Mortgages
.5
1 First Mortgage
.1
.1 2001
0
.1
0
2013
.4
.5 .4
.4 2007
.3
.3 2005
0
.2 2003
2+ First Mortgages
.2
.4 .3 .2 .1 0 2001
1 First Mortgage
.5
2+ First Mortgages
.5
1 First Mortgage
2003
2005
2007
2009
2011
2013
Fraction with 90+ days delinquent mortgage balances by 8Q lag Equifax Riskscore quartile. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Investors: Mortgage Delinquencies • Share of 90+ days delinquent mortgage balances rises for investors in prime segments during crisis .35
.35
Quartile 1
Quartile 2 1 First Mortgage (Right Axis)
2+ First Mortgages
1 .25
2003
2005
2007
2009
2011
2013
2001
.9 .8 .7 2003
2005
2007
2009
2011
2013
Quartile 4 1 First Mortgage (Right Axis)
2+ First Mortgages
2001
1 .25
.9
.2
.8
.1
.15 .8
0
2003
2005
2007
2009
2011
2013
2001
.7
.05 .7
0
.05
.1
.15
.9
.2
.25
1
.3
2+ First Mortgages
.3
1 First Mortgage (Right Axis)
.35
Quartile 3
.35
2001
.2 .1
.15 .8
0
.05 .7
0
.05
.1
.15
.9
.2
.25
1
.3
2+ First Mortgages
.3
1 First Mortgage (Right Axis)
2003
2005
2007
2009
2011
2013
Share of mortgage holders with 90+ days delinquent mortgage balances by 8Q lag Equifax Riskscore quartile. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Broader Implications • Aggregate consequences of growth in subprime lending: Defaults&foreclosures ! Drop in house prices
! Contraction in credit&consumption ! Recession
• Causal link based on geographical variation (zip code, MSA, county, state) (Mian & Sufi 2014, Mian, Rao & Sufi 2013, Kehoe, Midrigan & Pastorino 2014, Mian, Sufi & Trebbi 2014, Midrigan & Philippon 2016, Justiniano, Primiceri & Tambalotti 2016, etc )
Broader Implications • Aggregate consequences of growth in subprime lending: Defaults&foreclosures ! Drop in house prices
! Contraction in credit&consumption ! Recession
• Causal link based on geographical variation (zip code, MSA, county, state) (Mian & Sufi 2014, Mian, Rao & Sufi 2013, Kehoe, Midrigan & Pastorino 2014, Mian, Sufi & Trebbi 2014, Midrigan & Philippon 2016, Justiniano, Primiceri & Tambalotti 2016, etc ) • New findings challenge validity of evidence based on geographical variation
Geographical Variation: Zip Codes • Mortgage credit grows more for zip codes with high share of subprime 2.4 2 1.6 1.2 0.8 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Quartile 1
Quartile 2
Quartile 3
Quartile 4
Zip code level per capita real mortgage balances, ratio to 2001Q3, by fraction of subprime borrowers in 2001. Deflated by CPI-U. Source: Authors’ calculation based on Federal Reserve Bank of New York’s Consumer Credit Panel/Equifax Data.
Geographical Variation: Zip Codes • Higher credit growth in all zip codes for prime borrowers Quartile 1
Quartile 2
2.5
2.5
2.25
2.25 2
Ratio to 2001
Ratio to 2001
2 1.75 1.5
1.75 1.5 1.25
1.25
1
1
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
0.75
0.75 Above 660
Above 660
Below 660
Quartile 3 2.5
2.5
2.25
2.25
2
2
Ratio to 2001
Ratio to 2001
Below 660
Quartile 4
1.75 1.5 1.25
1.75 1.5 1.25
1 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 0.75
1 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 0.75
Above 660
Below 660
Above 660
Below 660
Mortgage debt growth for prime&subprime individuals by quartile of share of subprime in 2001. Based on 8Q lagged individual credit scores. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Age Distribution • Zip codes in quartile 4 exhibit faster growth in mortgage balance, for both prime and subprime borrowers • Quartile 4 has larger share of young individuals Fraction in each age bin, 2001Q1-2013Q4 20-24
25-34
35-44
45-54
55-64
65-85
Quartile 1
0.063
0.157
0.200
0.218
0.171
0.192
Quartile 2
0.070
0.184
0.200
0.205
0.161
0.181
Quartile 3
0.074
0.201
0.206
0.200
0.152
0.168
Quartile 4
0.081
0.212
0.210
0.199
0.145
0.153
Average age distribution in 2001Q1-2013Q4 by quartile of fraction of subprime in 2001. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Role of Age Distribution • To assess role of age distribution, construct counterfactual mortgage balance growth with age distribution set equal to quartile 1 for all quartiles • Contribution of di↵erences in age distribution to di↵erences in 2001Q1-2007Q4 debt growth relative to quartile 1: Mortgage Balances Quartile 2 0.44
Quartile 3 0.43
Quartile 4 0.84
• Di↵erence in age distribution accounts for most of the di↵erence if mortgage balance growth between quartile 4 and quartile 1
Geographical Variation: Zip Codes • Prime individuals contribute more to growth in delinquent balances and foreclosures during crisis in all zip codes
2005
2007
2011
2013
.7 .3
.4
.5
.6
1 (Lowest) 2 3 4 (Highest)
.1 0
.1 0 2001
.2
.7 .5
Quartile Quartile Quartile Quartile
.2
Share .3 .4
.6
.7 2009
.1
.2 2003
0
.1
.2
.3
Share .3 .4
.5
.6
1 (Lowest) 2 3 4 (Highest)
0 2001
Foreclosures
.4
Quartile Quartile Quartile Quartile
.5
.6
.7
Delinquent Balances
2003
2005
2007
2009
2011
2013
Share of 90+ days delinquent balances and new foreclosures for prime individuals, based on 8Q lagged individual credit score. Quartiles of subprime share in 1999. Source: Authors’ calculations based on FRBNY CCP/Equifax Data.
Geographical Variation: Zip Codes • Why did zip codes with high subprime share experience more severe recession? Young, low income/education, high inequality, high minority share, urban
Median age 2001 fraction subprime (med) Associate+ degree (2012) Percent white Percent black Mean Income $200K Mean Income
(2006-11)
Pop per sq mile Average UR 2001-2007 Average PDI 2001-2007 PDI Growth 2001-2007 HPI Growth 2001-2007 Average UR 2007-2010 HPI Growth 2007-2010
Quartile 1
Quartile 2
Quartile 3
Quartile 4
50 19% 0.45 93% 1.7% 6.4 1214 4.94% $ 41,045 25% 29% 6.93% -21%
49 32% 0.31 90% 3.6% 7.9 1380 5.19% $30,442 16% 37% 7.30% -30%
48 44% 0.23 83% 7.6% 9.4 1386 5.38% $25,692 10% 42% 7.51% -27%
46 60% 0.17 63% 24.6% 11.8 2322 5.72% $21,019 4% 47% 7.81% -36%
Quartiles of subprime share in 2001. Source: Authors’ calculations based on FRBNY CCP/Equifax Data, IPUMS, IRS, BLS, ACS data.
Discussion and Ongoing Work • Reassessment of role of subprime credit in the crisis need to rethink driving factors crucial for policy responses and prevention • Why rise in defaults for individuals with good credit? large role of investors large income shocks, unrealistic house price expectations? alternative default risk indicators? • Ongoing work: further analysis of geographical variation reassessment of collateral channel/role of ’home equity’ borrowing
