high risk consumer credit markets

39
High Risk Consumer Credit Markets Jonathan Levin Gaston Eyskens Lectures November 2013 1

Upload: traci

Post on 05-Jan-2016

32 views

Category:

Documents


0 download

DESCRIPTION

High Risk Consumer Credit Markets. Jonathan Levin Gaston Eyskens Lectures November 2013. Roadmap. Lectures Technology and Asymmetric Information High Risk Consumer Credit Markets Measuring Inefficiencies from Adverse Selection Can Markets for Health Insurance Work?. Introduction. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: High Risk Consumer Credit Markets

1

High Risk ConsumerCredit Markets

Jonathan Levin

Gaston Eyskens LecturesNovember 2013

Page 2: High Risk Consumer Credit Markets

Roadmap

Lectures

1. Technology and Asymmetric Information2. High Risk Consumer Credit Markets3. Measuring Inefficiencies from Adverse Selection4. Can Markets for Health Insurance Work?

Page 3: High Risk Consumer Credit Markets

3

Introduction Consumer credit markets long viewed as highly imperfect due to

informational frictions, especially for high-risk borrowers.

Classic asymmetric information theories => too little credit.

How should we understand the boom in consumer lending in the early 2000s, especially for high-risk or “subprime” credit?

Start with a review of relevant background and evidence. Then describe series of studies on subprime consumer lending looking at – whether there was a high latent demand for credit; and what pricing /

credit availability changes might have triggered borrowing.– informational problems facing lenders and whether they were relaxed

in early 2000s by credit scoring, access to credit, etc.

Page 4: High Risk Consumer Credit Markets

4

Asymmetric Information in Credit Markets

Theory of asymmetric information in credit markets (Jaffee & Russell, 1976; Stiglitz & Weiss, 1981).

– Adverse selection: high risk individuals more willing to borrow as they don’t expect to repay higher interest rates or larger loans disproportionally attract bad risks.

– Moral hazard: borrowers either can’t repay or have less incentive to repay loans with high interest rates or high principal lenders may refuse to make higher-risk loans.

Page 5: High Risk Consumer Credit Markets

5

Household Credit Constraints

Evidence of household liquidity constraints– Excess income sensitivity of household consumption, or even “pay-

check to pay-check” spending: Kuchler, 2013.

– Tax rebate studies: Parker 1999; Souleles, 1999; Johnson et al. 2006, Agarwal et al., 2007, many more.

– Credit card borrowing limits: Gross & Souleles, 2002;

– Labor markets: Card et al. 2007; Chetty, 2008.

Why does it matter? Presence of household liquidity constraints has wide-ranging implications.– Fiscal policy, unemployment and other forms of social insurance,

housing policy, life-cycle models, etc.

– Issues particularly relevant for low-income households.

Page 6: High Risk Consumer Credit Markets

6

Behavioral Economics of Borrowing Behavioral economics considerations

– Structure of borrowing --- get money today, pay tomorrow --- conducive to myopic behavior (over-optimism, self-control,...)

– Opposite concern from asymmetric info: too much borrowing !

Examples– “Payday” lending and high fee credit arrangements.

– Usury laws and other forms of consumer protection

Particularly if high fees or interest rates, need to be cautious in assuming households act in their own best interest.

Page 7: High Risk Consumer Credit Markets

7

Rise of Subprime Lending

US household debt/income: 90% to 120% in 2000-07.

Growth of “subprime” lending in the US, 1995-2005.– Payday loan outlets: 1,000 to 25,000

– Americredit auto loans: $230m to $8,454m

– Subprime mortgage originations: $65bn … $625bn

Housing market had the most macroeconomic consequence, but expansion of consumer credit was much broader.

Page 8: High Risk Consumer Credit Markets

8

Why the Rise of Subprime Lending?

Credit supply– Securitization, high demand for bonds, collateral seen as safe.– Lenders thought that technology had overcome frictions.

Credit demand– Change in price? Interest rates fell, but not by that much ...

perhaps demand response was to another changing “price”?– Alternatively, outward “shift” in demand, e.g. due to rising

inequality and stagnant incomes – Rajan (2011).

Page 9: High Risk Consumer Credit Markets

9

Background on Subprime Study

Describe research with Will Adams, Liran Einav and Mark Jenkins, using company records from a big subprime lender.

Attempt to provide:– Evidence of un-met credit demand among subprime borrowers.– Evidence of adverse selection and moral hazard problems.– Evidence on effects of credit scoring, risk-based pricing. – Insight into operation of credit markets for low income

population: how they work and what are the key features.

References: Adams, Einav and Levin (2009 AER); Einav, Jenkins and Levin (2012 EMA and 2013 RJE).

Page 10: High Risk Consumer Credit Markets

10

Setting for the Study Large integrated auto sales and finance company

– Purchases cars at auction and re-sells with financing

Customers: employed but low-income, bad credit– Applicant incomes: majority are $20-35k– About 15% are homeowners– One-third have no bank account.

Very detailed and complete data– Applications, originations from June 2001-Dec 2004– Payments and recoveries tracked through April 2006

Page 11: High Risk Consumer Credit Markets

11

A Subprime Auto Lender

Page 12: High Risk Consumer Credit Markets

12

Description of a Typical Transaction Buyer arrives on lot and applies for credit.

– Credit score determines car and financing offer.

Key offer terms are price (not int. rate) and minimum down.– Minimum down payment of $400-1,500– Cars priced at $10-11k, with blue book value $5-6k.– Loans are 3-4 years, most at state APR caps (25-30%)– “Effective” interest rates are higher, up to 80%.

Buyers finance heavily and default often– Typical down payment < $1,000, with loan of $10k.– Majority of loans (60+%) end in (early) default.– Recoveries are low – 1/3 of defaults are “skips”.

Page 13: High Risk Consumer Credit Markets

13

Defaults Happen Quickly

00.010.020.030.040.050.060.070.080.090.10.110.120.130.140.150000010.160.170.180000010.190.20.209999990.220.230.239999990.250.259999990.270000010.280.289999990.300000010.310.319999990.330000010.340.349999990.360000010.370.380.389999990.400000010.410.419999990.430000010.440.449999990.460000010.470.479999990.490000010.50.509999990.519999980.529999970.540000020.550000010.560.569999990.579999980.589999970.600000020.610000010.620.630.639999990.649999980.660000030.670000020.680000010.690.699999990.709999980.720000030.730000020.740000010.750.759999990.769999980.779999970.790000020.800000010.810.819999990.829999980.839999970.850000020.860000010.870.880.889999990.899999980.910000030.920000020.930000010.940.949999990.959999980.970000030.980000020.9900000110.0

0.5

1.0

1.5

2.0

2.5

Low RiskMedium RiskHigh Risk

Fraction of Loan Paid

De

ns

ity

Page 14: High Risk Consumer Credit Markets

14

Returns on Loans are Bimodal

Page 15: High Risk Consumer Credit Markets

15

Predictive Risk Scores

Strong incentive to identify bad loans in advance! Predictive scoring achieves significant risk stratification.

Risk Category Default RateLow Risk 44%Medium Risk 64%High Risk 71%

Page 16: High Risk Consumer Credit Markets

16

Evidence of liquidity constraints: Demand Probit model of purchase decision for applicants

Compare sensitivity to immediate and deferred payments (i.e. minimum down payment and car price). If buyers aren’t liquidity constrained, total payment is what matters

Calculate assuming rational expectation of default,– With 5% annual discount rate, is approximately 0.75.– With 50% annual discount rate is around 0.5.

Page 17: High Risk Consumer Credit Markets

17

Estimating demand: Down payment changes

Page 18: High Risk Consumer Credit Markets

18

Pricing Variation

Minimum down– Multiple changes of $100-500 for a subset of credit

categories; and discrete jumps at credit score thresholds.

Car price– Rely on variation in list prices: two major changes in

markup schedule; plus discrete jumps at cost thresholds.

Lots of robustness checks in the papers.

Page 19: High Risk Consumer Credit Markets

19

Probit Estimates of Purchase Decision Estimated marginal effect on purchase probability

Demand extremely sensitive to minimum down: $100 increase in minimum down => 9% fall in demand

To get same effect, need $900 price increase ($50 a month)! Implied annual discount rate: 427%

dProb/dx Std Err.

Car Price ($000) -0.030 (0.006)

Minimum Down ($000) -0.340 (0.005)

Page 20: High Risk Consumer Credit Markets

20

Evidence of Liquidity Constraints: Seasonality

Demand spikes during tax season: Cash sales but not trade-ins!

Page 21: High Risk Consumer Credit Markets

21

Evidence of Liquidity Constraints: Seasonality

Tax rebates can be large: up to $4,500 due to EITC. Figure classifies customers by EITC eligibility, plots demand spike.

Page 22: High Risk Consumer Credit Markets

22

Asymmetric Information

Could asymmetric info be the cause of liquidity constraints?

Try to test for moral hazard and/or adverse selection– Moral hazard: larger loan leads to higher default risk– Adverse selection: higher default risks take larger loans

Both imply correlation of loan size + default.

Ideal experiment to distinguish MH/AS: mandate participation, 1. Randomize loan size to measure MH2. Allow choice of loan size to measure MH + AS.

Page 23: High Risk Consumer Credit Markets

23

Larger Loans More Likely to Default

Page 24: High Risk Consumer Credit Markets

24

Down Payment as Signal of Repayment

Page 25: High Risk Consumer Credit Markets

25

Identifying Moral Hazard and Adverse Selection Starting point: Cox hazard model of default:

Goal: identify causal effect of loan size L on default (MH).

Problem: direct estimate will be confounded if unobservables affect both default and down payment (i.e. coefficient will measure MH + AS, not MH)

Solution: jointly model and estimate loan size (using tobit):

Key exclusion restriction: default depends on not directly on or ; exogenous variation in identify .

Implement with two-step control variable method.

Page 26: High Risk Consumer Credit Markets

26

Estimates of Moral Hazard

Page 27: High Risk Consumer Credit Markets

27

Expected Return and Loan Size

Page 28: High Risk Consumer Credit Markets

28

Decomposing Moral Hazard and Adverse Selection

Page 29: High Risk Consumer Credit Markets

29

Contract Pricing under MH and AS Einav-Jenkins-Levin: model of consumer demand that includes

– Decision to purchase, how much to finance, repayment

Paper derives estimating equations from “standard” model of consumer behavior, but do not require that interpretation.

Key features of the model ‒ Larger loans to dampen repayment incentives (MH) ‒ Consumers differ along observable and unobservable

dimensions that affect all three decisions (capture AS).

Estimate “complete” model using variation in min down and car prices described above to identify AS/MH parameters.

Page 30: High Risk Consumer Credit Markets

30

Model-Based Results

Effect of car prices (eq. interest rates): repayment incentives…– Higher car prices => larger stream of payments, but doesn’t last

as long. Doesn’t have much impact on originations.– Perhaps surprisingly, little incentive to offer interest rate

discounts to low risk applicants (little demand response!).

Effect of down payment requirements: control selection…– Higher down payment requirement => fewer originations, but

higher quality and smaller loan sizes, so less defaults.– Use risk score aggressively: for high risk applicants, use high min

down to avoid bad loans. But want to attract low risk applicants.

Page 31: High Risk Consumer Credit Markets

31

Setting Down Payment Requirements

Page 32: High Risk Consumer Credit Markets

32

Setting Down Payment Requirements

Page 33: High Risk Consumer Credit Markets

33

Credit Scoring and Adverse Selection

Adoption of credit scoring and risk-based pricing has been a major innovation in credit markets.

Model-based analysis (EJL, 2012)– Risk-based down payment requirements limits AS.– Also competitive effects (barrier for rivals)

Natural experiment analysis (EJL, 2013)– Look at adoption of credit scoring in 2000-2002.– Match equivalent borrowers before and after adoption.

Page 34: High Risk Consumer Credit Markets

34

Effect of Credit Scoring: Model-Based

Without scoring, lender is exposed to adverse selection

Page 35: High Risk Consumer Credit Markets

35

Effect of Credit Scoring: Model-Based

Risk-based pricing forces higher risks to pay more down.

Page 36: High Risk Consumer Credit Markets

36

Effect of Credit Scoring: Model-Based

And lowers default rates via screening + smaller loan sizes.

Page 37: High Risk Consumer Credit Markets

37

Effect of Credit Scoring: Natural Experiment

Adoption of credit scoring in 2000-2002‒ Match equivalent borrowers before and after adoption.‒ Study change in offered/originated loans to different groups.

Credit scoring allowed two forms of tailored lending– Higher down payments for riskier borrower (as above)– Larger loans (wider set of cars) offered to low risks.

Overall, very large effect of credit scoring on profitability.‒ About $1,000 per loan!‒ Importance of down payments and loan sizes roughly equal.

Page 38: High Risk Consumer Credit Markets

38

Welfare and Behavioral Issues

We were not able to say much about consumer welfare.

Expansion of low-income lending presumably– Helped some households that were credit-constrained– Allowed other households to get into debt trouble.

Welfare tricky once we move away from revealed preference assumption that agents act in own (ex ante) best interest.

To make progress, need information beyond choice data.

Page 39: High Risk Consumer Credit Markets

39

Conclusions Striking feature of the high-risk consumer credit markets

– Willingness of borrowers to take large, high-interest loans– Can’t rely on borrowers to avoid bad loans => lending is difficult.

Ability to sort borrowers using predictive models of default extremely important for profitable lending.

Down payment requirements, rather than interest rates / prices appear to be the key screening barrier for buyers.

Similar features seem to apply in subprime mortgages– Lower down payment requirements => rapid increase in originations,

with risky borrowers taking on substantial leverage.– Of course, had the additional feature that collateral value deteriorated

sharply once defaults began and housing market cooled.