how does unemployment insurance aect consumer spending? · spender-saver [campbell-mankiw 89] 0.80...

How Does Unemployment Insurance A�ect ConsumerSpending?

Peter Ganong and Pascal Noel

UChicago Harris, Harvard

December 15, 2016

1

How Does Unemployment Insurance A�ect ConsumerSpending?

Data: Chase bank accounts with direct deposit of UI benefits

Empirics: Estimate path of spending for UI recipients. Why?

ModelsInconsistent with canonical bu�er stock modelEstimate alternative behavioral models [Gabaix 16, Campbell-Mankiw 89]

Consumption-smoothing gains from UI

1

Data: Description

Checking accounts – transaction type aggregated by month

Oct 2012 through May 2015

210,000 UI recipients

Concern: 28% of households have checking accounts at multiplebanks [Consumer Financial Life Survey 14]

1 Sum over family’s linked accounts2 Select families that do most of their banking with Chase

Restriction: Ø5 monthly outflows

2

Data: Building Spending From Outflows

0.0

0.1

0.2

0.3

Debit C

ard Swipe

sCas

hBills

(Utilit

ies)

Chase

CC BillOuts

ide

CC Bill

Instal

lmen

t

Debt

Other

Paper

Checks

Saving

Shar

e Pr

ior t

o U

I Ons

et

Checking Account Outflows [Median $3520]

3

Data: Representativeness

Sample Bank Benchmark SourcePretax Fam Inc * Prior to UI Receipt $4,580 $5,080 SIPP Figure

Age § Prior to UI Receipt 44 41 SIPP Figure

Ckg Balance * Employed $1,460 $1,500 SCF Figure

Spending § Selected Categories $1,799 $1,912 CEX Table

Geography All 23 states 50 states Chase Map

* median, § mean

4

●

●

●●

●

●

●

●

●

●

Why Does Spending Drop at Onset?

Spending Changeat Exhaustion

0.80

0.85

0.90

0.95

1.00

−3 0 3 6Months Since First UI Check

Spen

ding

Rel

ative

to t

= −2

Spending While Unemployed

Equation Ex Post Exhaustees Standard Errors

5

Why Does Spending Drop? Low Current Income

● ●● ●

●

●

●●

Higher Spend By3.3% ($87)

0.94

0.96

0.98

1.00

−6 −3 0

Months Since First UI Check

Rat

io to

t = −3

UI_Benefits

● High

Low

Spending

6

Exhaustion: State-Level Comparison: Spending

● ● ● ●

●

●

● ●●

●

● ●

●●

●●

●●

●

●●

●

● ●

●

0.80

0.85

0.90

0.95

1.00

0 10Months Since First UI Check

Spen

ding

Rel

ative

to t

= −3

sample

● Florida (Max Weeks = 14)

Other States (Max Weeks = 26)

Benefit Exhaustion −− Spending Sample: Duration >= 14 Weeks

7

Exhaustion: Is Consumption Really Dropping?

Pre-Exhaust Post-Exhaust �%Large % Drop

Drug Stores $38 $31 -18%Medical Copay $28 $24 -14%Food At Home $289 $253 -13%Entertainment $25 $22 -11%

Small % Change

8

Exhaustion: Is Consumption Really Dropping?

Pre-Exhaust Post-Exhaust �%Large % Drop

Drug Stores $38 $31 -18%Medical Copay $28 $24 -14%Food At Home $289 $253 -13%Entertainment $25 $22 -11%

Small % ChangeAuto Loan $76 $71 -7%Mortgage $148 $142 -4%Insurance $141 $138 -2%Any Credit Bureau Delinquency 18% 19%

9

Model: Data vs Calibrated Model

Job Loss

0.8

0.9

1.0

−2 0 2 4 6Months Since First UI Check

Spen

ding

Rel

ative

to t

= −2

Model

Spending During Unemployment Spell Data and Buffer Stock Model

10


Exhaust Benefits

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

Model

Spending During Unemployment Spell Data and Buffer Stock Model

11


●

●

●●

●

●

●

●

●

●

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Model: Buffer Stock

Spending During Unemployment Spell: Data and Buffer Stock Model

12

Five Close Cousins Do Not Predict Drop At Exhaustion

◊ = arg min◊

ÿ

t(ct ≠ ct(◊))2

Estimate discount factor and risk aversion Slide

Alternative borrowing technologyCredit card borrowing Figure

Natural borrowing limit Figure

Alternative time preferencesHeterogeneous impatience [Krusel Smith 98] Figure

Naive present-bias [Laibson 97] Figure

Other Models

13

Sparse model [Gabaix 16]

●

●

●●

●

●

●

●

●

●

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Model: Sparsity

Spending During Unemployment Spell Misperception parameter m = 0.71 (0 is myopia and 1 is full rationality)

14

Spender-Saver [Campbell-Mankiw 89]

●

●

●●

●

●

●

●

●

●

0.80

0.85

0.90

0.95

1.00


Spen

ding

Rel

ative

to t

= −2

● Data

Mean from Spender−Saver Model

Estimated Shares: 25% Hand−to−Mouth, 40% Permanent Income, 35% Buffer Stock

15

Consumption-Smoothing Gains

Gain Relative to a 1% Increase in Life-time Income

Model UI Level ø 1.6% UI Dur ø 1 Mo Ratio (2)/(1)Bu�er stock 0.044% 0.122% 2.78Spender-saver 0.054% 0.202% 3.72Sparsity 0.071% 0.205% 2.89

16

Lessons for Consumer Financial Regulation

Question: How to teach people to prepare for bad events?Failure to prepare creates possible opportunities for welfareimprovement

This paper: Spend as if they don’t expect exhaustion to happenCFPB: Payday lending rule

Borrowers expect to repay sooner than they do [Bertrand and Morse2011, Mann 2013]

17

Conclusion

Monthly spending tracks UI benefits1 Onset: MPC of 43 cents2 Exhaustion: Spending drops 11%

Consequences of the drop at exhaustionReject rational modelConsistent with sparsity and spender-saverConsumption-smoothing gains of extending UI benefits >�> raising UIbenefits

18

Must believe job-finding rate is 74% [Spinnewijn 15]

● ● ● ● ● ●

●

0.00

0.25

0.50

0.75

1.00

0 2 4 6Months Since First UI Check

Expe

cted

Job−F

indi

ng P

roba

bilit

y ● Calibrated to Job−Finding Rate

Over−optimism Estimated to Fit Spending Path

Monthly Job−Finding Expectations Under Different Beliefs

19

Need to believe job-finding rate is 74% [Spinnewijn 15]

●

●

●●

●

●

●

●

●

●

0.75

0.80

0.85

0.90

0.95

1.00


Valu

e R

elat

ive to

t = −2

● Data

Model: Believe 73% Chance of Job at Exhaustion

Spending Assuming 74.0% Job−Finding Rate At Exhaustion

20

Model: Estimated Parameters

◊ =I

— discount factor“ risk aversion

◊ = arg min◊

ÿ

t(ct ≠ ct(◊))2

21


●

●

●●

●

●

●

●

●

●

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Natural Borrowing Limit

Unsecured Borrowing

Zero Assets at Onset, Cannot Borrow

A. Alternative Borrowing Technologies

Back 21

Other Models

Durables CommitmentsModel: Mortgage default should raise nondurables spendingData: Spending falls at exhaustion

Rational inattention Assets

Model:Agents with most at stake are most rational (low-asset group andlow-income group)smaller drop for these groups

Data: larger drop for these groups

Illiquid asset with transaction costKaplan and Violante 14 estimate a $1,000 costTwo-year loss from unemployment is mean $10,000 ∆ agent shouldliquidate to smooth consumptionComplicating factor: dynamic uncertainty from unemployment

Back22

Model: Permanent Income Consumer

● ●

●

●

●

●

●

●

●

●

0.6

0.7

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Buffer Stock

Hand to Mouth

Permanent Income

Three Agent Types in Spender−Saver

Back22

Model: Credit Card Borrowing

●

●

●●

●

●

●

●

●

●

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Natural Borrowing Limit

Unsecured Borrowing

Zero Assets at Onset, Cannot Borrow

A. Alternative Borrowing Technologies

Back 22

Models with persistent over-optimism

●

●

●●

●

●

●

●●

●

0.6

0.7

0.8

0.9

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Model: Believe 41% Chance of Job in Every Month

F. Persistent Over−optimism About Job−finding Rate

Back 22

Alternatives: Time Preferences

Heterogeneous impatience [Krusell and Smith 98, Carroll, Slacalek, Tokuokaand White 15, Parker 15, Auclert 16]

Three types. Impatient type has ” = 0.9 and no assets.

ct(◊) =w1 c

impatientt + w2c

perm inct + (1 ≠ w1 ≠ w2)cbu�er-stock

t

◊ = arg min◊

ÿ

t(ct ≠ ct(◊))2

23

Heterogeneous impatience slide – 0.9

●

●

●●

●

●

●

●

●

●

0.80

0.85

0.90

0.95

1.00


Spen

ding

Rel

ative

to t

= −2

● Data

Model: Krusel−Smith

D. Heterogeneous Agents Estimated Weights: 10% Impatient, 30% Permanent Income, 60% Buffer Stock

Back 24

Naive present bias [Laibson 97]

max{ct}

E

T≠tÿ

n=0—”n

u(ct+n)

To build intuition, show spending with — = 0.8 and — = 0.6

Estimate — which best fits data

25

Quantitative Evaluation of Models]

●●

● ● ●

●●

●●

●

0.6

0.8

1.0


Spen

ding

Rel

ative

to t

= −2

● Data

Model: Hyperbolic, Beta = 0.3

Model: Hyperbolic, Beta = 0.6

E. Hyperbolic Time Preferences

Back 26

What Models Can Fit Path of Spending? (Worst to Best)

Fit =ÿ

t

(ct ≠ ct(◊))2

Back

Model Comment # Params Fit

Permanent income Spending drops 0 1.00Baseline Should cut spending before benefit exhaust 0 0.22Borrow on credit Spending drop too large at benefit exhaust 0 0.11Sparse agent Act as if income loss at exhaust is 71% as

big as true loss1 0.09

Estimate params Spending drop too large at benefit exhaust 2 0.08Heterogeneousimpatience

Even highly impatient (e.g. ” = 0.9) cutbefore exhaustion

2 0.06

Sparse agent, esti-mate ”

Act as if income loss at exhaust is 71% asbig as true loss

2 0.03

Spender-saver 25% of agents hand-to-mouth 2 0.03Over-optimism 68% job-finding rate in exhaust month 2 0.01 27

Model Technical Details

Concern Spec Change PlotChanges That Don’t Matter

Duration Dependence {e6, e7, . . .}: 0.25 æ 0.15 Figure

Spending Mismeasured Spending = All Outflows Figure

Consumption Commitments “: 2 æ 4 Figure

Changes That Matter

Alternative Asset Values a0 = {0, 12} Figure

Alternative Discount Factor — = {0.98, 1} Figure

28

Model Failure: Too Little Liquid Asset Holdings

Liquid Asset HoldingsData: Survey of Consumer Finances 0.7 months

Model: Match drop through exhaustion 0.8 monthsModel: Steady state a 2.4 months

Gournichas and Parker (02) ~12 months

29

Implications for UI Policy

Fact: UI surprisingly important for consumptionIncrease UI to help families smoothIncrease UI for macro stabilization

Fact: People who cut spending more find a job faster. Some peoplewon’t search until benefits are exhausted.

Decrease UI

30

Implications beyond UI Policy

Fact: Spending drops sharply at benefit exhaustionPolicy: Help families prepare for exhaustion

Fact: People who cut spending more find a job faster. Some peoplewon’t search until benefits are exhausted.

Policy: Encourage “worrying” early on to motivate job search

Fact: Spending very sensitive to incomePolicy: Encourage larger bu�ers. Dedicated accounts?

31

Empirics. Onset. Family Income Recovers Quickly

●●●●●●●●●●●●●●●●●●●●

●●

●

●●●

●●●●●●●●●●●●●●●●●●●●●●

$0

$1,000

$2,000

$3,000

−20 −10 0 10 20


Mea

n ($

)

Income Source● Gov

LaborLabor + Gov

Labor and Gov Transfers −− UI Receipt Beginning in Month 0

32

Empirics. Onset. Concept Di�erences Explain QuickIncome Recovery

Jacobson, LaLonde and Sullivan (93) report a 30% permanent income lossfor displaced workers. Why?

See also Couch and Placzek (10), von Wachter, Sullivan and Manchester (09),Davis and von Wachter (11), Jarosch (15), Flaaen et al. (15)

Sources:

1 Gov’t transfersRecovery of family labor income matches SIPP Figure

2 Family income vs individual income3 All UI recipients vs high-tenure JLS Figure Back

33

Empirics. Why Drop at Onset? Work-Related ExpensesSteps

1 Identify expenditure categories containing work expenses cwork

Method: drop at retirement for people with enough assets to smooth

Aguiar and Hurst (13) �c > median at retirementFood Away From Home Food Away From HomeTransportation Fuel, Auto, Flights/HotelsClothing Department Stores

Small Durables, Online31% of nondurables 29-41% of nondurables

2 Estimate impact of change in employment status Details

E (cwork(y , e = 1)) ≠ E (cwork(y , e = 0))

Back

34

Spending At Onset By Expenditure Category

●

●

●

●

●

●

●

●

●

●

●

0.91

0.94

0.97

1.00

−5 −4 −3 −2 −1 0 1 2 3 4 5


Spen

ding

: Rat

io to

t = −5

● Card Non−Work−Related

Card Work−Related

Cash and Bills

Spending If Stay Unemployed

Back 35

Empirics. Onset. Work-Related Expenses ¥ 1/3

−150

−100

−50

0

Counterfactual 1: dNon−Work−Related

Categories@ Onset

Counterfactual 2: dWork−Related

Categories@ Exhaust

All Spendingin Work−Related

Categories

Total Change

Cha

nge

From

t=−3

to t=−1

($)

Role of Work−Related Expenses in Spending Drop at Onset

Back 36

Spending Change at Reemployment

0.80

0.85

0.90

0.95

1.00

−20 −10 0 10 20


Spen

ding

: Rat

io to

t = −3

Completed UI Duration

1

2

3

4

5

36

Families Smooth Income Loss Over Several Months

37

Reemployment Spending Recovery Details

Focus on families that get UI for exactly three months

1 Prior work overstated spending drop during unemployment Figure

2 Spending recovers slowest for low-asset families Figure

38

Bank UI Families Have Incomes Similar to SIPP

●

●

●

●

●

●

●●

●

●

●●

●● ●

● ● ●

● ●

●

0

5

10

15

0 50,000 100,000 150,000 200,000

Annual Family Income

Perc

ent o

f Sam

ple

Data Source

● Survey (SIPP)

JPMCI Ckg Inflow

Distribution of Income Prior to UI Receipt

Back39

Representativeness by Age

●

●

●

●

●

●

0

10

20

20 30 40 50 60 70

Age of Family Head or Primary Customer

Perc

ent o

f Sam

ple

Data Source

● Survey (SIPP)

JPMCI

Age Distribution

Back 40

Representativeness by Labor Income Before and AfterSeparation

Before Sep Drop at Sep SourceExternal Benchmark

(1) Labor Share of Total 85% 52% Rothstein and Valetta (14)(2) Use Payroll DD 80% 80% SCF(3) = (2) * (1) 68% 42%Bank 69% 38%

Back

41

Assets External Benchmark

●

● ●

●

●

●

●

●●

●

●

●●

●●

● ● ● ● ●

●

● ● ●

●

0

10

20

30

0 5,000 10,000

Checking Account Balance

Perc

ent o

f Sam

ple

key

● Survey (SCF) Employed

JPMCI Employed

JPMCI UI Recip

Checking Account Balance

Medians: SCF Employed $1,500, Bank Employed $1,520, Bank UI $1,060 Back 42

Chase Branch Footprint

Back 43

Monthly Spending Compared to External Benchmarks

Category Bank Ratio to CEX Ratio to BEAFood At Home 478 1.44 0.82Food Away From Home 291 1.33 0.62Utilities 371 1.19 –

Ratio to SCFMortgage 1536 1.12Auto Loan 484 1.04Credit Card 1010 0.63Back

44

Benchmarks: Receipt of UI Direct Deposit

External Benchmarks SourceWeekly UI payments 2.9 million FREDConsumer Units 125 million CEXConsumer units getting UI 2.2% –UI recipients getting DD 45% Natl Consumer Law CenterConsumer units getting UI DD 1.0% –

Bank families getting UI DD 0.8%Back

45

Control Group Nondurables Spending Rises $7/month

●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

1800

1900

2000

2100

2013−01 2013−07 2014−01 2014−07 2015−01

Mea

n

Secular Trend In Spending

Back

0.0

0.1

0.2

0.3

Debit C

ard Swipe

sCas

hBills

(Utilit

ies)

Chase

CC BillOuts

ide

CC Bill

Instal

lmen

t

Debt

Other

Paper

Checks

Saving

Shar

e Pr

ior t

o U

I Ons

et

Checking Account Outflows [Median $3520]

46

Control Group Income Rises $15/month

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

1900

2000

2100

2200

2300

2400

2013−01 2013−07 2014−01 2014−07 2015−01

Mea

n

Secular Trend In Labor Income

Back 47

Calendar Adjustment for Income

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

−400

−200

0

200

2013−01 2013−07 2014−01 2014−07 2015−01

Adju

stm

ent T

erm

● Annual Income 30K−80K

Ever Receive UI

Pay DD in 20−31 of 32 Mos

Monthly Adjustment Constant for Labor Income

Back 48

Calendar Adjustment for Spending

●●

●

● ●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

−200

0

200

2013−01 2013−07 2014−01 2014−07 2015−01

Adju

stm

ent T

erm

● Annual Income 30K−80K

Ever Receive UI

Pay DD in 20−31 of 32 Mos

Monthly Adjustment Constant for Spending

Back 49

Empirics: Introduction – Families Remaining Unemployed

●

●●

●

●

● ●

●

●

●

●

0.91

0.94

0.97

1.00

−5 −4 −3 −2 −1 0 1 2 3 4 5


Rat

io to

t = −5

Spending If Stay Unemployed

Back50

Equation for Spending Drop

0.91

0.94

0.97

1.00

−5 −4 −3 −2 −1 0 1 2 3 4


Spen

ding

: Rat

io to

t = −3

Completed UI Duration

1

2

3

4

5

6

Spending by Completed UI Duration

�ct = 1n

ÿ

iœUI Duration >tci ,t ≠ ci ,t≠1

Back51

Nonparametric spending series

Back52

Exhausted vs Did Not Exhaust

●

● ●

●

●

●

●

●●

●

0.91

0.94

0.97

1.00

−5 −4 −3 −2 −1 0 1 2 3 4 5


Rat

io t

o t=

−5

● Did Not Exhaust UI

Exhausted UI

Spending At Onset

Back53

Labor Income in high vs low benefit states

● ● ●●

●

●

●

●●

●

●

●

●

●●

●●

● ● ● ●● ● ●

● ● ● ● ●● ●

● ●● ●

0.5

0.7

0.9

−6 −3 0 3 6 9 12 15 18 21 24 27


Rat

io to

t = −3

UI_Benefits

● High

Low

Labor

Back54

Spending (long-term) in high vs low benefit states

● ● ● ●

●

●

● ●

●

● ●

●

●● ● ●

● ●● ●

●

● ●● ●

●

●● ●

●

● ●● ●

●

0.900

0.925

0.950

0.975

1.000

−6 −3 0 3 6 9 12 15 18 21 24 27


Rat

io to

t = −3

UI_Benefits

● High

Low

Spending

Back55

Spending At Unemployment OnsetPre-Onset Post-Onset �%

(t ≠ 3) (t ≠ 1)Cut a Lot

Any Student Loan 12.4% 10.9% -16%Food Away From Home $185 $164 -11%Auto $181 $162 -11%Any Doctor Copay 24.6% 22.4% -9%

Cut a LittleRetail $358 $337 -6%Food At Home $300 $291 -3%Any Auto Loan Pay 17.0% 16.6% -2%

StableUtilities $164 $163 -1%Any Entertainment 43.7% 44.3% 1%Any Mortgage Pay 15.0% 15.3% 2%

Back

56

Age < MedianAnnual Income < MedianAny Mortgage Payments

CC utilization > 50%Chase Assets in Bot QuintChase Assets in Top Quint

Debt / Income > MedianHas Chase Credit Card

No Revolving CC BalancePenalty Fees > $5/month

SingleTotal Assets in Bot QuintTotal Assets in Top QuintUI Ben / Inc in Bot QuintUI Ben / Inc in Top Quint

0.0 0.1 0.2 0.3 0.4 0.5Marginal Propensity to Consume

MPC Heterogeneity at Onset

Back

57

Empirics. Onset. Mean Duration and Spending Drop

AZ

CO

FL

GA

ID

IL

KY

LA

MI

NJ

NV

NY

OH

OK

OR

TX

UT

WAWI

WV

100

150

200

250

12.5 15.0 17.5 20.0Mean UI Duration (Weeks)

Spen

ding

Dro

p at

Ons

et ($

)Onset −− Falsification Test with UI Duration

Back58

Borrowing on All Credit Cards

●

●●

●

● ●

●●

●

● ●● ●

●● ●

●●

●

● ●●

●

●● ●

●●

6900

6950

7000

7050

7100

−6 −3 0 3 6 9 12 15 18 21


Rev

olvi

ng B

alan

ce −−

All C

redi

t Car

dsCredit Card Borrowing in Credit Bureau

UI Receipt Beginning in Month 0

Back to Onset Back to Model59

Herkenho�, Phillips and Cohen-Cole 2016

Back to Model60

Baseline Chars: Pre-Onset Medians By UI Duration

Duration Income Spending Ckg Assets1 2788 2236 9492 2894 2239 10113 2811 2181 10514 2737 2164 9835 2685 2147 9976 2612 2110 982Exhaust 2564 2112 1045

Back

61

Onset: Heterogeneity By Duration

●●●●●●●●●●●●●●●●●●●

●

●

●●●

●●

●

●

●●●●●●

●●●●●●●

$1,000

$1,500

$2,000

$2,500

$3,000

−20 −10 0 10 20


Mea

n Am

ount

($)

key

● Exhaust

Find Job in 2−6 Mos

Find Job in < 2 Mos

Labor Income By Benefit Duration

Back62

Onset: Heterogeneity By Duration

●●●●●●●●●

●●●●●●●●●●

●●●

●

●

●

●

●

●●

●

●●

●●●●●

●●●

●

$2,200

$2,400

$2,600

−20 −10 0 10 20


Mea

n Am

ount

($)

key

● Exhaust

Find Job in 2−6 Mos

Find Job in < 2 Mos

Spending By Benefit Duration

Back63

Onset: Heterogeneity By Total Liquid Assets

●●

●

●

●

●

● ●

●

●●

● ●

●● ●

● ● ●● ●

●

0.80

0.85

0.90

0.95

1.00

−6 −3 0 3 6 9 12 15


Inco

me:

Rat

io to

t = −3

● Liquid Asset Bottom Tercile

Liquid Asset Medium Tercile

Liquid Asset Top Tercile

Income Event Study for Nonexhaustees

Back64

Onset: Heterogeneity By Total Liquid Assets

●

● ● ●

●

● ●

● ● ●●

●●

● ● ●

● ●● ●

● ●

0.90

0.95

1.00

−6 −3 0 3 6 9 12 15


Spen

ding

: Rat

io to

t = −3




Spending Event Study for Nonexhaustees

Back65

Onset MPC By State

AZ

CO

FLGA

IL

LAMI NJ

NY

OH

OR

TX

UT

WV

0.2

0.3

0.4

0.5

0.2 0.3 0.4 0.5 0.6 0.7

Fraction of UI Recipients Getting Direct Deposit

MPC

= S

pend

ing

Dro

p / I

ncom

e D

rop

Usage of Direct Deposit and MPC at Onset

Back

66

Spending: Long-Run Trends

●

●

●

●●●●●●●

●●●●●●●●●

●●●●●●

●

●

●

●

●

●●●

●●●●●●●

●●●●

●●

●●●●

●●

●●

●

−400

−200

0

−20 −10 0 10 20Months Since Start of UI Spell

Amou

nt R

elat

ive to

t−3

($)

key● Income

Spend.LoSpend.Hi

Income and Spending

Back 67

Linked and Unlinked Accounts

Most families with multiple checking accounts have linked theiraccounts together under a single primary customer

About 10% of UI recipients have multiple accountsNot linkedMatched by same last name and addressCould arise if two Chase customers got married, decided to keepseparate accounts

PlotsOnset: Income and SpendingExhaustion: Income and Spending Back to Onset Back to Exhaustion

68

Linked and Unlinked Accounts – Income at Onset

● ● ● ●● ● ● ● ● ● ●

−1500

−1000

−500

0

−5 −4 −3 −2 −1 0 1 2 3 4 5


Inco

me

Rel

ative

to t

= −3

($)

Account

● Did not receive UI deposit

Received UI deposit

Labor Income at UI Onset −− Families with 2 accts

69

Linked and Unlinked Accounts – Spending at Onset

●●

●

●

●

● ●

●

● ●

●

−300

−200

−100

0

−5 −4 −3 −2 −1 0 1 2 3 4 5


Spen

ding

Rel

ative

to t

= −3

($)

Account


Received UI deposit

Spending at UI Onset −− Families with 2 accts

70

Linked and Unlinked Accounts – Income at Exhaustion

●

● ● ●● ● ●

●●

● ●

0

200

400

600

−5 −4 −3 −2 −1 0 1 2 3 4 5

Months Since Last UI Check

Inco

me

Rel

ative

to t

= −1

($) Account


Received UI deposit

Labor Income at UI Exhaustion −− Families with 2 accts

71

Linked and Unlinked Accounts – Spending at Exhaustion

●

● ● ●

● ●

● ●

●

● ●

−300

−200

−100

0

100

−5 −4 −3 −2 −1 0 1 2 3 4 5


Spen

ding

Rel

ative

to t

= −1

($)

Account


Received UI deposit

Spending at UI Exhaustion −− Families with 2 accts

72

Income Recovery: Comparison to SIPP

●●●●●●●●●●●●●●●●●●

●

●

●

●

●

●

●

●

●

●●●●●●●

●●●

●●●●

●●●●

0.4

0.5

0.6

0.7

0.8

0.9

1.0

−20 0 20 40Months Since UI Start

Rat

io to

Prio

r Ear

ning

s

key

● JPMCI

SIPP

Earnings Drop in JPMCI and Survey Data

Back 73

Income Recovery: JLS Mass Layo�

Back 74

Income Recovery: JLS All UI Recips

Back 75

Empirics. Work-Related Expenses Details

Methodology for estimating impact of change in employment statuson spending in work-related categories

E (cwork(y , e = 1)) ≠ E (cwork(y , e = 0)) =E (cwork(yemp, e = 1)) ≠ E (cwork(yunemp, e = 0))¸ ˚˙ ˝

Total drop in work categories

≠

E (cwork(yUI Benefit, e = 0)) ≠ E (cwork(yUI Exhaust, e = 0))¸ ˚˙ ˝

Drop in work categories due to lost incomeEstimate final term using two methods:

Drop in spending on work-related categories at benefit exhaustionMPC

workexhaust(yEmp ≠ yUI Benefit)

Drop in spending on non-work-related categories at onsetMPC

nonworkonset (yEmp ≠ yUI Benefit)

Back76

Annual Spending Data Miss Monthly Smoothing

● ● ● ●

●

●

●

●

●

●

●●

●● ●

●●

● ●

0.85

0.90

0.95

1.00

−6 −3 0 3 6 9 12


Spen

ding

: Rat

io to

t = −3

Monthly Spending Data vs Prior Work Using Annual Data Completed UI Duration = 3 Months

Back77

Path of Income Similar By Asset Holdings

●

●

●

●

●

● ●

● ●

●

● ●●

●

●●

●

● ●

0.80

0.85

0.90

0.95

1.00

1.05

−6 −3 0 3 6 9 12


Inco

me:

Rat

io to

t = −3

Assets at Onset● Low Asset

Medium Asset

High Asset

Income. Sample has Completed UI Duration of 3 Months.

Back78

Spending Recovers Slowly For Low Asset Types

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

0.92

0.94

0.96

0.98

1.00

−6 −3 0 3 6 9 12


Spen

ding

: Rat

io to

t = −3

Assets at Onset● Low Asset

Medium Asset

High Asset

Spending. Sample has Completed UI Duration of 3 Months.

Back

79

Exhaustion: Robustness Checks for Internal Validity

●● ●

● ●

●

●

●

●●

●

●●

●●

●●

●

●

●

$−500

$0

0 5 10 15Months Since Last UI Check

Amou

nt R

elat

ive to

t−1

($)

key● Income

Spending..Lower.Bound.Spending..Upper.Bound.

Spending at Exhaustion: Sharp Inc Change

Back 80

Age < MedianAnnual Income < MedianAny Mortgage Payments

CC utilization > 50%Chase Assets in Bot QuintChase Assets in Top Quint

Debt / Income > MedianHas Chase Credit Card

No Revolving CC BalancePenalty Fees > $5/month

SingleTotal Assets in Bot QuintTotal Assets in Top QuintUI Ben / Inc in Bot QuintUI Ben / Inc in Top Quint

0.0 0.1 0.2 0.3 0.4 0.5Marginal Propensity to Consume

MPC Heterogeneity at Exhaustion

Back

81

Exhaustion MPC By State

AZ

COFL

GA

IL

LA

MI

NJNY

OH

OR

TX

UT

WV

0.3

0.4

0.5

0.6

0.2 0.3 0.4 0.5 0.6 0.7

Fraction of UI Recipients Getting Direct Deposit

MPC

= S

pend

ing

Dro

p / I

ncom

e D

rop

Usage of Direct Deposit and MPC at Exhaustion

Back

82

Exhaustion: Time Aggregation for Spending

●

●●

●

●

●

●

● ●

●

●

●

●●

●

●●

●

●

−200

−100

0

0 5 10


Spen

ding

Rel

ative

to t

= −1

($)

key

● All Exhaustees

Last UI Ck >= 25

Spending at Benefit Exhaustion

Back 83

Income Drops by $1200 At Exhaustion

●

●

● ● ●●

● ● ● ● ● ● ● ● ● ● ● ● ● ●$0

$500

$1,000

$1,500

$2,000

$2,500

−5 0 5 10


Mea

n Am

ount

($)

● UI

Labor

Labor + UI

Income at Benefit Exhaustion

Back84

Exhaustion: Heterogeneity by Income Drop

●●

●

●

●

● ● ●●

●

●

●

●

0.4

0.6

0.8

1.0

−3 0 3 6 9


Inco

me:

Rat

io to

t = −3

● UI / Inc Bottom Tercile

UI / Inc Medium Tercile

UI / Inc Top Tercile

Income Event Study for Exhaustees

Back to Empirics Back to Model85

Exhaustion: Heterogeneity by Income Drop

●●

●

● ●

●

● ●●

●

●●

●

0.80

0.85

0.90

0.95

1.00

−3 0 3 6 9


Spen

ding

: Rat

io to

t = −3

● UI / Inc Bottom Tercile

UI / Inc Medium Tercile

UI / Inc Top Tercile

Spending Event Study for Exhaustees


Exhaustion: Heterogeneity by Liquid Assets

●●

●

●

●

● ●●

●

●

●

●

●

0.6

0.8

1.0

−3 0 3 6 9


Inco

me:

Rat

io to

t = −3




Income Event Study for Exhaustees


Exhaustion: Heterogeneity by Liquid Assets

● ●

● ●

●

●●

●

●

●

●●

●0.80

0.85

0.90

0.95

1.00

−3 0 3 6 9


Spen

ding

: Rat

io to

t = −3




Spending Event Study for Exhaustees

Back88

Exhaustion: Heterogeneity by State

● ● ● ● ● ●

●

●●

●

●

● ● ● ● ● ● ● ●0.0

0.1

0.2

0.3

0.4

−6 −3 0 3 6 9 12


Rat

io to

Tot

al In

com

e at

t−3

state

● FL

IL

MI

NY

TX

Govt Payments −− Sample: UI Exhaustees

Back89

Exhaustion: Heterogeneity by State

● ●●

●

●

●

● ● ●●

●●

● ● ●

●

● ●

●

0.80

0.85

0.90

0.95

1.00

−6 −3 0 3 6 9 12


Rat

io to

t−3

state

● FL

IL

MI

NY

TX

Spending −− UI Exhaustees

Back90

Details on Equivalent Variation CalculationsFuchs-Schuendeln and Hassan (15) calculate z which solves

u(y + x + z)¸ ˚˙ ˝

MPC=1

+11u(y + z) = 12 u(y + x

12)¸ ˚˙ ˝

perm incomefor CRRA utility with “ = 2. We calculate

15ÿ

t=1u(cPIH

t ) =15ÿ

t=1u(cdata

t + z)

15ÿ

t=1u(chand≠to≠mouth

t ) =15ÿ

t=1u(cdata

t ≠ z)

15ÿ

t=1u(cbu�er≠stock

t ) =15ÿ

t=1u(cdata

t + z)

using a 15-month horizon. For cdata, we assume that agent behavesoptimally after date 7 and aggregate over all possible job-finding histories

back91

Welfare Loss Under Di�erent Models

−0.10

−0.05

0.00

0.05

Gain over c=y Policy

Loss from PIH Policy

Loss from Buffer Stock Policy

Equi

vale

nt V

aria

tion

(% o

f ann

ual c

)Welfare Loss for Agent Choosing u( cdata ) Under Different Models

Back92

Distribution of Spending Changes

0.00

0.05

0.10

0.15

−1.8 −1.6 −1.4 −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

Ratio of Two−Month Change in Spending to Monthly UI Benefit

Shar

e

Two−Month Window

Control (Before Onset)

Treatment (Exhaustion)

Distribution of Spending Change Before Onset and At Exhaustion

93

Distribution of Change in Spending at Benefit Exhaustion

−0.02

0.00

0.02

−1.8 −1.6 −1.4 −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

Ratio of Two−Month Change in Spending to Monthly UI Benefit

Diff

eren

ce in

Sha

re B

etwe

en

Mon

th o

f Exh

aust

ion

and

3 M

onth

s Be

fore

Ons

etDistribution of Spending Change at UI Exhaustion

Back94

Schmieder, von Wacther and Bender (15) Figure 6Reemployment wages in German UI data

Back95

People Who Cut Spending Sooner Find a Job Faster –Heterogeneity

Back

96

how does unemployment insurance aect consumer spending? · spender-saver [campbell-mankiw 89] 0.80...

Documents