Credit Cards and the Great Recession: The Collapse of Teasers · Inspired by evidence documented by Ausbel and Shui (2005) and Agrawal at al. (2015), showing consumers choose ex post

Credit Cards and the Great Recession:The Collapse of Teasers

Lukasz A. Drozd1 Michal Kowalik2

1Federal Reserve Bank of Philadelphia 2Federal Reserve Bank of Boston

Nov, 2018

The views expressed in this paper are those of the authors and do not necessarily reflectthose of the Federal Reserve Bank of Philadelphia, Federal Reserve Bank of Boston, or the

Federal Reserve System.

Drozd and Kowalik Why Tease? The Role and Ramifications of Credit Card “Teaser” Rates


Big picture question: Transmission of 2007 financial crisis to real economy

Current view on the issue puts households and household credit in the spotlight(Mian, Rao and Sufi, 2013; Mian and Sufi, 2014)

Was deleveraging a product of declining net worth or the effect of tightening oflending standards by distressed financial institutions?

(Gilchrist Zakrajsek, 2017; Mondragon, 2015; Greenstone, Mas, Nguyen, 2012)

This paper: A detailed look at deleveraging on credit cards

Document and explore a plausible mechanism for supply-driven deleveraging

Show the mechanism consistent with deleveraging on credit cards

Assess contribution to MS aggregate consumption-demand channel


Big picture question: Transmission of 2007 financial crisis to real economy

Current view on the issue puts households and household credit in the spotlight(Mian, Rao and Sufi, 2013; Mian and Sufi, 2014)

Was deleveraging a product of declining net worth or the effect of tightening oflending standards by distressed financial institutions?

(Gilchrist Zakrajsek, 2017; Mondragon, 2015; Greenstone, Mas, Nguyen, 2012)

This paper: A detailed look at deleveraging on credit cards

Document and explore a plausible mechanism for supply-driven deleveraging

Show the mechanism consistent with deleveraging on credit cards

Assess contribution to MS aggregate consumption-demand channel

Key stylized facts we build on


Fact 1.

Prior to the crisis many borrowers relied on promotional offers to, in effect, borrow forthe long term on promo rates

fact 1: 35%+ debt (43% prime) on 10pp+ promo discount s.t. expiration 12 months away

fact 2: promo balance transfer volume ≈ flow of expiring promo debt

Fact 2.

Availability of promotional offers vanished in 2008, resulting in a collapse of balancetransfer and almost 50% decline in the share of promotional debt by 2011

fact 3: cc solicitations and balance transfer volume fell by 70%

fact 4: share of debt with promo flag declined by 50 percent by 2011

Fact 3.

Collapse of promo activity coincident with deleveraging on credit cards

fact 5: 32% peak-to-trough decline in credit card debt relative to trend from late 2008 onwardfact 5: 20% peak-to-trough decline relative to consumer credit excluding student loans

What We Do


Develop equilibrium theory of promotional pricing featuring hyperbolic discounting

Inspired by evidence documented by Ausbel and Shui (2005) and Agrawal at al.(2015), showing consumers choose ex post suboptimal offers

Consumers are myopic, but Gabaix and Leibson (2017), show a rational model withnoisy information gives rise to “as-if” hyperbolic discounting

Show theory consistent with U.S. credit market prior to the crisis

Ask whether a withdrawal of promo offers account for deleveraging and related facts

What We Do







What We Do







Literature: Empirics


Mian and Sufi (2010-IMF)

Show reliance on credit cards across the U.S. counties prior to the crisis a strongpredictor of the decline in auto sales after 2008 even when controlling for householdleverage

Brown, Haughwout, Lee, Van der Klaauw (2013), Demyanyk and Koepke (2012)

Question importance of supply forces based on decline in credit in inquiries

Our answer: inquiries a function of offers received, a primary tool of solicitation

Agrawal, Chemisengphet, Mahoney, Stroebel (2015)

Show evidence that CARD Act of 2009 was unlikely the culprit

Literature: Theory


Intentionally left blank.

Outline


Data description and summary results

Mechanisms: contract theory of promo offers

Quantitative model

Quantitative analysis and results

Outline




Quantitative model


Data description and sources

1The data is on an account level with a monthly frequency and is provided by bank holding companies subject toDFAST. The sample before 2013 is limited to several largest banks and it comes from OCC merged data with Y14Mreporting. We focus on this sample here. Data after 2013 covers a broader sample of banks.

2The credit bureau data summarizes credit history of 200,000 credit market participants: the first 100,000 recordsare representative as of 2001 and the second one is representative as of 2013. We use observations from both panels.


1. Aggregate data from the Board of Governors of the Federal Reserve System, such as thestock of revolving debt (consumer credit - G.19) and net charge-off rate on credit cards forall banks

2. Supervisory OCC/Y14M account level micro-data focusing on general purpose creditcards from 6 largest credit card lenders tracked between 2008 and 2017, and eight in total,having an approximate market share of over 50 percent in 2007 (accounting for 30 percentof general purpose card credit card accounts)1

2. Experian credit bureau data comprising of a representative panel of 200,000 credit recordstracked between 2001 and 2013.2

Credit card market prior to the crisis


1. A large fraction of debt had promo status with a median duration of 12 months

Statistic 2008Q1

1. Use of promotional debt:Promo debt to total debta [%] 35Promo debt with 670+ FICO to total debt [%] 43Promo debt with at least 50% APR discount to promo debtb [%] 68Median duration of promo spell (originated in 08)c [months] 10Average duration of promo spell (originated in 08)c [months] 12Median duration of promo spell (all accounts)c [months] 12Average duration of promo spell (all accounts) c [months] 16

aDebt are credit card balances carried over for at least one billing cycle, hence 2008Q1 effectively starts in Feb.bPromo debt on low APR is the promo debt for which the promotional APR is lower than the step-up APR by at least 50 percent.cThe spell is a number of months for which an account has a positive promotional balance, among accounts originated in 2008. We find equalmedian and higher mean for all accounts, which suggests accounts originated prior to 2008 had a longer promotional spell.



2. Promo debt provided a major discount relative to non-promotional and step-up rates

Statistic 2008Q1

2. Interest rates (in APR):Median promo APR [%] 3.5Average promo APR [%] 4.3Average promo APR with discount 50%+ debt [%] 2.6Average non-promo APR [%] 15.5Average step-up APR on promo accounts w/ debt [%] 17.3Median step-up APR on promo accounts w/ debt [%] 16.0



3. A large volume of balance transfers sustained the stock promotional debt consistent with“chaining” of promotional offers to, in effect, borrow for the long-term

Statistic 2008Q1

3. Refinancing and balance transfers:Balance transfers (BT) per annum to promo debt [%] 131BT to promo cards to BT total [%] 92BT to flow of promo debt nearing expiration (last quarter) [%] 104Average transferred amount per BT [$] $4,290

Collapse in promo activity


4. Promo activity collapsed during the crisis.

(BT decline consistent with 70% decline in solicitations.)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

2007 2009 2011 2013 2015 2017

Year

Promo debt as a fraction of total debt

45%

NBER recession

Promo debt as a fraction of total debt for FICOs 670+

(2008q1)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

2007 2009 2011 2013 2015 2017

Year

Annual balance transfers to total debt

74%

Annual promo balance transfers to total debt

NBER recession

(2008q1)

Deleveraging (all banks)


5. Collapse in promo activity coincident with deleveraging.

10%

13%

15%

18%

20%

1996 1999 2002 2005 2008 2011 2014 2017

Card debt per adult to median income

Year

NBER Recession

32 percent decline

0.4

0.42

0.44

0.46

0.48

0.5

0.52

0.54

2006 2008 2010 2012 2014 2016

Year

Card debt to consumer credit less student loans 16 percent

decline

Deleveraging (all banks)


5. Collapse in promo activity coincident with deleveraging.

5060708090

100110120130140150

Mortgage

HE Revolving

Auto Loan

Credit Cardcredit card market peak in 2009Q1= 100

Year

Orthogonality to mechanical composition


6. Decline in BT orthogonal to measure risk composition of consumer pool.

5060708090

100110120130140150

Mortgage

HE Revolving

Auto Loan

Credit Cardcredit card market peak in 2009Q1= 100

Year

Outline




Quantitative model


Environment


Life-cycle setup with continuum of consumers and a large number of lenders:

Consumers:

face random income (and access to market)

borrow from lenders to smooth consumption

can default on debt at a fixed utility cost (stigma)

Lenders:

have unlimited access to funds at exogenous cost of funds r

extend unsecured open ended credit lines to consumers

compete in the market in Bertrand fashion (max U s.t. zero pf)

Consumer preferences


Borrowers discount the future hyperbolically and are naive about it:

Preferences as of current period t

u(ct) + ηβ[u(ct+1) + βu(ct+2) + β2u(ct+3) + ...]

Actual preferences in future period

u(ct+1) + ηβ[u(ct+2) + βu(ct+3) + ..]

Consumer preferences


Borrowers discount the future hyperbolically and are naive about it:

Preferences as of current period t

u(ct) + ηβ[u(ct+1) + βu(ct+2) + β2u(ct+3) + ...]

Actual preferences in future period

u(ct+1) + ηβ[u(ct+2) + βu(ct+3) + ..]

⇒ Consumers overestimate how fast their future self will pay down debt

Lending protocol


Credit line is: F ≤ R - promo rate, R - step-up rate, L - pre-authorized credit limit

Borrowers can switch by refinancing with a new lenders (without recall)

Incumbent lenders continually reprice under CARD Act of 2009 restriction:

Rates cannot be raised above R (can be lowered)

Cannot slash credit limits below debt

Refinancing subject to a friction: Even when consumer refinance she continues to payinterest for a ρ fraction of the next period

Lending protocol








Lending protocol








Lending protocol








Timing


Period t

Period t+1

Consumers Lenders

1. Markov random variable s resolves alluncertainty within the period (income; arrival of offers)

3. Consumer decides whether to accept market offer M or reject and stay with the incumbent with no recall

5. Consumers strategically decides whether to repay and chooses consumption c and current borrowing b accordingly

2. Lending market opens: Consumer receives market offer M and repriced offer from incumbent I

4. Incumbent lenders reprice again rest of the period

Initial endogenous state is Debt , Credit line

…

Consumer problem: refinancing decision λ


Consumer first chooses whether to refinance or stay by solving:

V ηt (C, B, s) = maxλ=0,1

Uηt (Mηt (C, B, s), Iηt (C, B, s;λ), B, s;λ)

where

Mηt (C, B, s) is market offer

Iηt (C, B, s;λ) is incumbent’s repriced offer

(Note: λ = 0 if refinancing option not available under s. )

Consumer problem: default decision δ


Consumer strategically plans default by solving:

Uηt (CηM , CηI , B, s;λ) = max

δ=0,1Uηt (CηM , C

ηI , B, s;λ, δ)

where

CηI := Iηt (C, B, s;λ) = (F ηI , RηI , L

ηI ) repriced contract from incumbent

CηM := Mηt (C, B, s) = (F ηM , R

ηM , L

ηM ) market offer (active or inactive)

(Note: λ = 0 if market offer inactive. )

Consumer problem: consumption c and borrowing b


Consumer chooses consumption and current borrowing by solving:

Uηt (CηM , CηI , B, s;λ, δ) = max

(c,b)∈Γ{u(c)− χ(s)δ+

ηβEs[δV 1t+1(C−1, 0, s

′) + (1− δ)V 1t+1(λC1

M + (1− λ)C1I , b, s

′)]}

subject to budget constraint given by

c ≤ Yt(s)−B + b− (1− δ)[λF ηM + (1− λ)(ρF ηI + F ηM )

]b+

b ≤ (1− λ) min{LηM , LηI }+ λLηI

where C−1 = (r−1, 0, 0) exogenous seed contract.







′) + (1− δ)V 1t+1(λC1

M + (1− λ)C1I , b, s

′)]}



]b+









′) + (1− δ)V 1t+1(λC1

M + (1− λ)C1I , b, s

′)]}



]b+



Lemma

LηM never binds and LηI can be assumed tight without loss.







′) + (1− δ)V 1t+1(λC1

M + (1− λ)C1I , b, s

′)]}



]b+

b ≤ (1− λ)LηM + λLηI


Lemma

LηM never binds and LηI can be assumed tight without loss.

Lender Problem: Market offer M


The equilibrium market offer solves:

Mηt (C, B, s) = argmax

CηM

Uηt (CηM , CηI , B, s)

subject to

ΠMt (CηM , CηI , B, s) = 0

where CηI is equilibrium repriced offer (simultaneous game).

Lender Problem: Incumbent’s repriced offer I


The equilibrium repriced offer solves:

Iηt (C, B, s;λ) = argmaxCηI

ΠIt (CηM , CηI , B, s;λ)

subject to

RηI ≤ R, F ηI ≤ R, LηI ≥ B

and

Uηt (CηM , CηI , B, s;λ) ≥ Uηt (CηM , C

ηI , B, s;λ)

where CηI = (R,R,B) and CηM is equilibrium repriced offer.

Lender profit function Π


Omitted.

Equilibrium


Recursive equilibrium comprises consumer’s policy functions

cηt , bηt , δ

ηt

lender pricing policiesMηt , I

ηt

and consumer and lender value functions

V ηt , Uηt ,Π

It ,Π

Mt

such that they are consistent with consumer problem and lender problem.

Outline




Quantitative model


Assumptions


Three-periods T = 3 (can generalize to any finite T )

Two-state income process: Y (s = 1) := Y > Y (s = 0) = Y /

Extremely convex cost of defaulting χ(s = 1) =∞, χ(s = 0) = 0.

Cost of funds normalized to zero

Equivalent problem of geometric consumers η = 1


Transformed problem solves:max

(F,R,L)∈ΘU(F, R, L)

whereU(F, R, L) := max

b1≤L,b2u1(c1) + β(1− p)[u2(c2) + β(1− p)u3(c3)]

c1 :=Y −B + b1 − Fb+1c2 :=Y − b1 + b2 − Rb+2c3 :=Y − b2

and subject toΠ(F, R, L) = (F − p)b+1 + (1− p)(R− p)b+2 = 0.

Equilibrium contract for geometric consumers η = 1


Proposition

If local monotonicity holds: F = p = R, L is nonbinding, and the consumer does not refinance.

Equilibrium contract for geometric consumers η = 1


Proposition

If local monotonicity holds: F = p = R, L is nonbinding, and the consumer does not refinance.

1. Allocation satisfies MRT = MRS in high state MRT = −(1− p) while MRS = −(1− p)βu′(c2)

u′(c1),

which implies:

u′(c1) = βu

′(c2).

2. Consumer’s Euler equation dictates F = 0 :

(1 − F )u′(c1) = β(1 − p)u

′(c2)

3. Binding L also implies F = p.

Equilibrium contract for hyperbolic consumers η < 1


Proposition

If local monotonicity holds in the ex ante problem at F = p = R, L slack, then F < p < R.

Equilibrium contract for hyperbolic consumers η < 1


Proposition

If local monotonicity holds in the ex ante problem at F = p = R, L slack, then F < p < R.

1. Allocation satisfies MRT = MRS in high state MRT = −(1 − p)bη2b2

while

MRS = −(1 − p)βu′(c2)

u′(c1), which implies:

u′(c1) = βu

′(c2)

b2

bη2.

2. Consumer’s Euler equation dictates F = 0 :

(1 − F )u′(c1) = β(1 − p)u

′(c2)

3. Binding L also implies F = p.

Numerical example


0

0.1

0.2

0.3

0.4

1 0.95 0.9 0.85 0.8

0 APR Refinancing1

No refinancing0

0

0.1

0.2

0.3

0.4

0.5

0.6

1 0.95 0.9 0.85 0.8

0 APR Refinancing1

No refinancing0

Figure: A numerical example: Equilibrium contract as a function of η (β = 1).

Notes: The figure illustrates equilibrium contract for a range of values of hyperbolic discount factor η, assuming Yl =1/2, Yh = 1, B = 1, ρ = .5, p = .1, β = 1 and u(c) = log(c). F is restricted to be non-negative. The shadedarea indicates when refinancing occurs on the equilibrium path. The right panel shows the wedge between ex ante and ex postborrowing that creates incentives to set promotional terms.

Outline




Quantitative model


Parameterization


Log-utility, hyperbolic discount from Ausubel and Shui (2005): η = 0.81

Cost of defaulting is parameterized by χ0: χ(y) = χ1 max(y − χ0, 0).

Income of a working age in economy state ω = {R,E} is:

yt(ω) = etktzt(ω)

where

yt - agent’s income at age t

et - deterministic age-dependent income profile

kt - a 3 state discrete i.i.d. process

zt(ω) - 6x6 state Markov process that depends on ω

Individuals start life at the age of 24 years, retire at the age of 65 year, and die at the ageof 80 years and period length is l (parameter we calibrate)

Demographics simulated starting from 2010 population structure and using deathprobability tables and .9 population growth

Calibration


Table: Data targets and calibrated values of jointly selected parameters.

Data Model

A. Targeted moments1. Credit card debt of card holder to median personal income [%] 22 222. Net charge-off rate [%] 4 43. Average duration of promo offers [months] 12 124. Average step up rate on promo accounts [%] 17 205. Average rate on credit card debt [%] 12 12

B. Jointly calibrated parametersDiscount factor β 0.926Cost of defaulting χ0 0.867Period length l [months] 20Refinance delay ρ 0.4Lender cost of funds r 0.07

C. Preset parametersHyperbolic discount factor β 0.81Income process (see Online Appendix and supp. files)

Model Validation


Classification of promo accounts in mapping model onto data:

0% 20% 40% 60% 80%

90%

80%

70%

60%

50%

40%

30%

20%

10%

0%

As a fraction of accounts with F<R [in %]

Prom

o discou

nt size

: 1‐F/R [in %]

Classified as promotional in the model

Model Validation


Model does remarkable job matching moments we did not target:

Table: Data targets and calibrated values of jointly selected parameters.

Statistic (in percent % unless otherwise noted) Data Model

Promo debt as a fraction of total debt 35 33Annual balance transfers as fraction of debt 39 44Average interest rate on promo debt (+3 in data) 7 6Median interest rate on promo debt (+3 in data) 6 6Share of revolvers among card users 59 60

Calibration of collapse of promo


10%

15%

20%

25%

30%

35%

40%

45%

50%

2007 2010 2013 2016

Model: recession + transition + collapse of promo

Year

fitted collapse of promo schock

Promo card debt to total card debt

Data

Model: recession + transitionModel: recession

NBER Recession

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

2007 2010 2013 2016

Model: transition + recession + collapse of promo

Year

Annual balance transfers to debt

Data

Model: transition + recession Model: transition

NBERRecession

Figure: Collapse of promotional activity: model via-a-vis the U.S. data.

Notes: The figure illustrates the decline in the share of promotional credit card debt to total debt (left panel) and the collapseof balance transfers (promotional balance transfers) as a fraction of debt. Solid lines correspond to the model and the dottedline is the data. We consider three models that incrementally add shocks. The total contribution of the collapse of promoshock is the difference between green line with circles and the orange line with squares.

Quantitative results: Deleveraging


60

70

80

90

100

110

120 Model: recession + transition + collapse of promo

Year

2007=100

Card debt per adult to median income

Data

Model: recession + transisionModel: recession

NBERRecession

60

70

80

90

100

110

120 Model: recession + transition + collapse of promo

Year

2007=100

Real card debt per adult

Data (detrended)


NBERRecession

Figure: Deleveraging on credit cards: model via-a-vis the U.S. data.

The figure illustrates deleveraging on credit cards relative to median income and in absolute terms (in data real value detrendedusing the 1996-2006 linear trendline). Solid lines correspond to the model and the dotted line is the data. We consider threemodels that incrementally add shocks. The total contribution of the collapse of promo shock is the difference between greenline with circles and the orange line with squares.

Quantitative results: Charge-off rate


0%

2%

4%

6%

8%

10%Model: recession + transition + collapse of promo

Year

2007=fitted

Net chargeoff rate on card debt

Data(all banks)


0%

2%

4%

6%

8%

10%

12%

14%

16%

18%

20%

2007 2010 2013 2016

Model: recession + transition + collapse of promo

Year

2007=fitted

Average APR on card debt

Data


NBER Recession

Figure: Charge-off rate and interest rate on card debt: model via-a-vis the U.S. data.

Note: The figure illustrates the net charge-off rate on card debt (fraction of debt defaulted on) and the average interest ratepaid on credit card debt estimated using our account level dataset. Solid lines correspond to the model and the dotted line isthe data. The charge-off rate is for all banks and comes from FRB. We consider three models that incrementally add shocks.The total contribution of the collapse of promo shock is the difference between green line with circles and the orange line withsquares.

Quantitative results: Aggregate implications


80

85

90

95

100

105

110

115Model: just transition + collapse of promo

Year

2007=100

Real consumption per adult

Data (detrended)

Model: recession

NBER Recession 95

96979899

100101102103104105

Model: just transition + collapse of promo

Year

2007=100

Consumption to disposable income ratio

Data

NBERRecession

Figure: Charge-off rate and interest rate on card debt: model via-a-vis the U.S. data.

Note: The figure illustrates the net contribution of the collapse of promo shock to decline in consumption in the model (orangeline with squares). Solid lines correspond to the model and the dotted line is the data. The left-panel looks at total consumptionand the right panel looks at consumption income ration (aka average propensity to consume). The orange line here refers tonet added contribution of having promo shock (with transition) and isolates out this shock from recession and demographicchanges.

Conclusions


Credit card market played a quantitatively significant role in HH deleveraging

A supply-driven shock likely at play: withdrawal of promotional offerings from the market

The shock is sufficient to account for deleveraging in a consistent way, and has sizableimplications on aggregate assuming Mian-Sufi demand driven view of Great Recession

Documents

Credit Cards and the Great Recession: The Collapse of Teasers · Inspired by evidence documented by Ausbel and Shui (2005) and Agrawal at al. (2015), showing consumers choose ex post