DefaultOptionsand Retirement Saving Dynamics TahaChoukhmane - - PowerPoint PPT Presentation

DefaultOptionsand Retirement Saving Dynamics

TahaChoukhmane

NBER(2019-2020),MITSloan(2020- )

October2019

1

High stakes setting: retirement savings plans

Default = non-participation Default = participation Call provider to enroll Call provider to opt-out ~50% participate after 1yr >90% participate after 1yr fOpt-in regimef fAutoenrollmentf

Autoenrollment (AE) is a˙ecting ~100 million people worldwide:

I NZ (’07), UK (’12), Turkey (’17): all private sector workers I US: the majority of 401(k) plans already implements AE . . .... ............. ....... .....

.... 5 states are extending AE to workers without a 401(k)

Motivation

Keyinsightfrom behavioral economics: defaultoptionsmatter

1

Autoenrollment (AE) is a˙ecting ~100 million people worldwide:

I NZ (’07), UK (’12), Turkey (’17): all private sector workers I US: the majority of 401(k) plans already implements AE . . .... ............. ....... .....

.... 5 states are extending AE to workers without a 401(k)

Motivation

Keyinsightfrom behavioral economics: defaultoptionsmatter Highstakessetting: retirementsavingsplans

Default=non-participation Default=participation Call providertoenroll Call providertoopt-out ~50%participateafter1yr >90%participateafter1yr fOpt-inregimef fAutoenrollmentf

1

. . .... ............. ....... ..... ....

Motivation

Keyinsightfrom behavioral economics: defaultoptionsmatter Highstakessetting: retirementsavingsplans

Default=non-participation Default=participation Call providertoenroll Call providertoopt-out ~50%participateafter1yr >90%participateafter1yr fOpt-inregimef fAutoenrollmentf

Autoenrollment (AE) isa˙ecting~100million people worldwide:

I NZ(’07),UK(’12), Turkey(’17): all privatesector workers I US:themajority of401(k)plansalreadyimplementsAE

5 statesareextendingAEtoworkerswithouta401(k)

1

Challenge: no long-run data because AE is a recent policy This paper:

1 Identify the mechanism through which AE a˙ects behavior 2 Build and estimate a lifecycle model to study AE long-run e˙ect

ThisProject

ManystudiesonAEshort-runimpactbutlong-rune˙ectunknown:

Q:What isthee˙ect ofautoenrollmenton lifetimesavingsandwelfare?

2

ThisProject

ManystudiesonAEshort-runimpactbutlong-rune˙ectunknown:

Q:What isthee˙ect ofautoenrollmenton lifetimesavingsandwelfare?

Challenge: nolong-rundata because AE is arecent policy Thispaper:

1 Identifythemechanism through which AEa˙ects behavior 2 Buildandestimatealifecycle model tostudyAElong-rune˙ect 2

Outline

1 Three Factsabout Autoenrollment 2 A LifecycleModelwithDefaultE˙ects

Model Estimation

3 Results

Long-terme˙ect Optimal policies

4 Conclusion

3

TwoDatasets

U.S.401(k)Data:

New proprietarydataset I obtainedfrom a large US pension provider Monthlycontributions,balances,andassetallocationfor4mworkersbtw. 2006-17

U.K.NationallyRepresentativeData:

ASHE2006-16 : nationallyrepresentative 1%panel Followsworkersacrosssuccessivejobs

3

⇒ need a model to extrapolate e˙ect over many job switches

Fact II: Increasing AE default ↓ participation

=> model speci†cation w/ opt-out costs

One known fact w/ a new interpretation: Fact III: Median non-AE catch-up to AE over 3yrs

=> small opt-out cost → large default e˙ects

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savinginnextjob

4

FactI:AEReducedSavinginNextJob

Mandatory Autoenrollement forallU.K. privatesectoremployees Policyroll-out by employersize between2012-2017

Policy rollout

Identifcation:

Treated Employer (subject to AE) Untreated Employer New Employer (AE or nonAE) New hire 1 New hire 2 Previous employer j-1 New employer j Year x Firm Fe 𝛾 = 𝑡1,𝑘 − 𝑡2,𝑘

5

tenure, age contr

FactI:AEReducedSavinginNextJob

AEreduced participation by 11% innextopt-injob! Existingwithin-jobestimatesmay overstateAEe˙ectonlifetimesavings

Policy Actual startdate 2012

Panel A- Participationrate

AEtonon-AE

• 0.109**

(0.052) AEtoAE 0.013 (0.017)

Panel B- Contribution in (% ofpensionablepay)

AEtonon-AE

• 0.472**

(0.185) AEtoAE

• 0.048

(0.066)

Observations 35,651 Sizej−1 X Sizej X Employerej X Year X Robuststandarderrorsclustered by currentemployer ;*** p<0.01,** p<0.05,* p<0.1

Sample: 22-60y &≤1ytenure in ASHE 2006-17. Additional controls: total pay, previoustotal pay,tenure, previous

• ls,gender

6

tenure, age contr

FactI:AEReducedSavinginNextJob

AEreduced participation by 11% innextopt-injob! Existingwithin-jobestimatesmay overstateAEe˙ectonlifetimesavings

Policy Actual startdate 2012

2005 2006 2007 2008 2009 2010 2011

PanelA- Participationrate

AEtonon-AE

• 0.109**

0.073 0.022

• 0.003

0.022 0.046 0.008

• 0.056

(0.052)

(0.062) (0.041) (0.055) (0.054) (0.066) (0.055) (0.073)

AEtoAE 0.013 (0.017)

Panel B- Contribution in (% ofpensionablepay)

AEtonon-AE

• 0.472**

0.023

• 0.092

0.161

• 0.123

0.021

• 0.234
• 0.137

(0.185)

(0.219) (0.173) (0.489) (0.214) (0.224) (0.213) (0.300)

AEtoAE

• 0.048

(0.066)

Observations 35,651 35,651 35,651 35,651 35,651 35,651 35,651 35,651 Sizej−1 X Sizej X X X X X X X X Employerej X Year X X X X X X X X Robuststandarderrorsclustered by currentemployer ;*** p<0.01,** p<0.05,* p<0.1

Sample: 22-60y &≤1ytenure in ASHE 2006-17. Additional controls: total pay, previoustotal pay,tenure, previous

• ls,gender

6

Fact II: Increasing the AE default ↓ participation

=> ... w/ an opt-out cost

One known facts w/ a new interpretation: Fact III: Median non-AE catch-up to AE over 3yrs ...

=> opt-out cost is small

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodeltoextrapolatee˙ectaftermanyjobswitches

6

=> ... w/ an opt-out cost

One known facts w/ a new interpretation: Fact III: Median non-AE catch-up to AE over 3yrs ...

=> opt-out cost is small

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodel...

Fact II: IncreasingtheAEdefault↓ participation

6

Fact II:IncreasingDefault ↓ Participation

Compare workershired before/after 86U.S.†rmsincreasedtheirdefault

Example: 3%→ 6%

• 5%
• 4%
• 3%
• 2%
• 1%

0% 1% 1 2 3 4

Δ in percentage pts AE default increased by x% of salary

Participation rate

(i.e. contributions > 0%)

Controls: plan, year,andageFEs,logtenure Sample: 86US 401kplans.159,216workersw/≤1yoftenure postgrace-period

7

Fact II:IncreasingDefault ↓ Participation

Compare workershired before/after 86U.S.†rmsincreasedtheirdefault

Example: 3%→ 6%

• 5%
• 4%
• 3%
• 2%
• 1%

0% 1% 1 2 3 4

Δ in percentage pts AE default increased by x% of salary

Participation rate

(i.e. contributions > 0%)

Controls: plan, year,andageFEs,logtenure Sample: 86US 401kplans.159,216workersw/≤1yoftenure postgrace-period

7

Fact II:IncreasingDefault ↓ Participation

Compare workershired before/after 86U.S.†rmsincreasedtheirdefault

Example: 3%→ 6%

• 5%
• 4%
• 3%
• 2%
• 1%

0% 1% 1 2 3 4

Δ in percentage pts AE default increased by x% of salary

Participation rate

(i.e. contributions > 0%)

0% 1% 2% 3% 4% 5% 6% 1 2 3 4

AE default increased by x% of salary

Positive contrib < initial default

(e.g. contributions at 1% or 2%)

Δ in percentage pts

Controls: plan, year,andageFEs,logtenure Sample: 86US 401kplans.159,216workersw/≤1yoftenure postgrace-period

7

Fact II:IncreasingDefault ↓ Participation

Nudgingworkerstocontributemorew/higherdefault.... ... ledmoretodrop-outandcontributeatthelowestrates! Opt-outcost: †tsthisevidence

• Ex. worker preferedcontirbutionrate 1%
• 3% default: stayat 3% (not worth bearingopt-outcost)
• 6% default: dropto 1% (farenoughfrom preferedrate)

Othertheories(lossaversion,anchoring): opposite prediction

8

One known facts w/ a new interpretation: Fact III: Median non-AE catch-up to AE over 3yrs ...

=> opt-out cost is small

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodel...

Fact II: IncreasingtheAEdefault↓ participation

=>... w/anopt-outcost

8

=> opt-out cost is small

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodel...

Fact II: IncreasingtheAEdefault↓ participation

=>... w/anopt-outcost

Oneknownfactsw/anewinterpretation: Fact III: Mediannon-AEcatch-uptoAEover3yrs...

8

Static setting

|{z}

y

Gains from switching:

• Tax bene†t
• Generous employer match

.

V

a

Large opt-out cost:

DellaVigna (’06,’18): min. $1,200 Bernheim et al (’15): avg. $2,200

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

Median worker

Opt-in 3% AE

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

9

Static setting

|{z}

y

Gains from switching:

• Tax bene†t
• Generous employer match

.

V

a

Large opt-out cost:

DellaVigna (’06,’18): min. $1,200 Bernheim et al (’15): avg. $2,200

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

Opt-in 3% AE

Median worker

9

|{z}

y

a

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

Opt-in 3% AE

Median worker

Staticsetting

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

.

Gainsfromswitching: Largeopt-outcost:

• Tax bene†t

V

DellaVigna(’06,’18): min. $1,200

• Generousemployermatch

Bernheimetal (’15): avg. $2,200

9

|{z}

y

a

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

Opt-in

3% AE

Median worker

Staticsetting

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

.

Gainsfromswitching: Largeopt-outcost:

• Tax bene†t

V

DellaVigna(’06,’18): min. $1,200

• Generousemployermatch

Bernheimetal (’15): avg. $2,200

9

|{z}

y

a

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

Opt-in

3% AE

Median worker

Staticsetting

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

.

Gainsfromswitching: Largeopt-outcost:

• Tax bene†t

V

DellaVigna(’06,’18): min. $1,200

• Generousemployermatch

Bernheimetal (’15): avg. $2,200

9

|{z}

y

a

.

Fact III:Median non-AECatch-up to AE

Workershired in the 12months before/after AE at 3% in 34†rms

0% 3% 6% 9% 12% 12 24

Tenure

• Contrib. Stock

(Cumul. employee contrib. % of salary) (months)

Opt-in

3% AE

Median worker

Dynamicsetting

.

Gainsfromswitching:

• Tax bene†t
• Generousemployermatch

V Smalleropt-outcost: Inalifecyclemodel I estimatean

• pt-outcostof∼ $250

9

... but heterogeneity matters

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodel...

Fact II: IncreasingtheAEdefault↓ participation

=>... w/anopt-outcost

Oneknownfactsw/anewinterpretation: Fact III: Mediannon-AEcatch-uptoAEover3yrs...

=>opt-outcostissmall

9

Three Factsabout Autoenrollment

Two newfacts: FactI: AEincurrentjob↓ savingsinnextjob

⇒ needamodel...

Fact II: IncreasingtheAEdefault↓ participation

=>... w/anopt-outcost

Oneknownfactsw/anewinterpretation: Fact III: Mediannon-AEcatch-uptoAEover3yrs...

=>opt-outcostissmall

...butheterogeneity matters

9

• Heterogeneity Matters

FirmA Choietal’04

Intheshortrun: largetreatmente˙ectsonlyatthe bottom...

0% 3% 6% 9% 12% 12 24 0% 3% 6% 9% 12% 12 24

Median participant

3% AE Opt-in

0% 3% 6% 9% 12% 15% 12 24

• Contrib. Stock

(Cumul. employee contrib. % of salary)

Tenure

(months)

Tenure

(months)

Tenure

(months)

25th Percentile

3% AE Opt-in

75th Percentile

3% AE Opt-in

... willthesesavingsincrease persist inthelongrun?

10

Outline

1 Three Factsabout Autoenrollment 2 A LifecycleModelwithDefaultE˙ects

Model Estimation

3 Results

Long-terme˙ect Optimal policies

4 Conclusion

11

... parsimonious speci†cation of preferences (3 parameters):

Time preferences: standard (eis + exponential discount factor) Opt-out cost: utility cost every time agent deviates from the default

TheModel

I buildandestimateadetailledlifecyclemodelwithdefaulte˙ects Featuresricheconomicenvironment(8statevariables)...

1 Assets: realisticretirementaccount,liquidsaving,andunsecureddebt 2 Labormarket:incomeandemploymentriskvarieswithageandtenure(SIPPdata) 3 Government: progressive tax and beneftsystem (SocialSecurity & UI) 4 Demography: mortality risk,andchanginghouseholdcompositionoverlifecycle 1 2 11

TheModel

I buildandestimateadetailledlifecyclemodelwithdefaulte˙ects Featuresricheconomicenvironment(8statevariables)...

1 Assets: realisticretirementaccount,liquidsaving,andunsecureddebt 2 Labormarket:incomeandemploymentriskvarieswithageandtenure(SIPPdata) 3 Government: progressive tax and beneftsystem (SocialSecurity & UI) 4 Demography: mortality risk,andchanginghouseholdcompositionoverlifecycle

... parsimonious speci†cation of preferences (3 parameters):

1 Time preferences: standard(E.I.S. & exponentialdiscountfactor) 2 Opt-outcost: utility costeverytimeagentdeviatesfromthedefault 11

DataandEstimation

EstimationSample: 34plansw/a50%matchupto6%andnoautoescalation Workershired in the 12months before/after AE at 3% SimulatedMethod ofMomentsresults: Estimates(quarterlyfreq.) EIS

• disct. fact. opt-outcost

σ δ k 0.455 0.987 $254 (0.013) (0.001) (11)

χ2 stat. (41df): 586

12

Robustness:

WeightingMatrix Opt-inonly AEonly

Extensions:

PresentBias ProportionalCost

Sensitivity:

Andrews,Gentzkow,Shapiro’17

EstimationMoments

12

DistributionofContributionRates

Employeesintheir1st yearoftenure

0% 20% 40% 60% 80% 100% 0% 1-2% 3% 4-5% 6% 7-9% >10%

Freq.

Opt-in (0% default)

0% 20% 40% 60% 80% 100% 0% 1-2% 3% 4-5% 6% 7-9% >10%

Auto-enrollment (3% default)

Data Model

Freq. Contribution rate (% of salary) 13

DistributionofContributionRates

Employeesintheir1st yearoftenure

0% 20% 40% 60% 80% 100% 0% 1-2% 3% 4-5% 6% 7-9% >10%

Freq.

Opt-in (0% default)

0% 20% 40% 60% 80% 100% 0% 1-2% 3% 4-5% 6% 7-9% >10%

Auto-enrollment (3% default)

Data Model

Freq. Contribution rate (% of salary) 13

Evolutionover Tenure

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Participation - Opt-in

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Participation - 3% Data

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Share at the 3% default

14

Evolutionover Tenure

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Participation - Opt-in

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Participation - 3% Data Model

0% 20% 40% 60% 80% 100% 4 8 12 16

Quarter of tenure

Share at the 3% default

14

Outline

1 Three Factsabout Autoenrollment 2 A LifecycleModelwithDefaultE˙ects

Model Estimation

3 Results

Long-terme˙ect Optimal policies

4 Conclusion

15

Out-of-Sample validation I: Model estimated using the introduction of AE at 3% ... ... predicts response to increasing the default Out-of-Sample validation II: Preference estimates from U.S. 401(k) plans ... ... predict the response to a national policy in the U.K.

Externalvalidity

Whyshould we believethemodellong-run predictions?

Advantageofstructuralestimation: extrapolate toanother policy, population, institutionalsetting,time-frame

results results 15

Out-of-Sample validation II: Preference estimates from U.S. 401(k) plans ... ... predict the response to a national policy in the U.K.

Externalvalidity

Whyshould we believethemodellong-run predictions?

Advantageofstructuralestimation: extrapolate toanother policy, population, institutionalsetting,time-frame Out-of-SamplevalidationI:

results

ModelestimatedusingtheintroductionofAEat 3%... ... predictsresponse to increasing thedefault

results 15

Externalvalidity

Whyshould we believethemodellong-run predictions?

Advantageofstructuralestimation: extrapolate toanother policy, population, institutionalsetting,time-frame Out-of-SamplevalidationI:

results

ModelestimatedusingtheintroductionofAEat 3%... ... predictsresponse to increasing thedefault Out-of-Samplevalidation II:

results

PreferenceestimatesfromU.S.401(k)plans... ... predict the response to anational policy in the U.K.

15

For most people: ↑ saving early-on ↓ saving later in life BUT large e˙ects at the bottom of the lifetime earnings distrib.

AE↑ LifetimeSavingsattheBottom

Typical AE policy at 3%adopted by allemployers

• 2%

0% 2% 4% 6% 8% 10% 12% 14% 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th

% change relative to opt-in Deciles of lifetime earnings

Incidenceonworkers AE6pct AE10pct HighPresentBias LowPresentBias ProportionalCost 16

AE↑ LifetimeSavingsattheBottom

Typical AE policy at 3%adopted by allemployers

Formost people: ↑ savingearly-on↓ savinglaterinlife BUTlargee˙ects at the bottom ofthelifetimeearningsdistrib.

• 2%

0% 2% 4% 6% 8% 10% 12% 14% 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th

% change relative to opt-in Deciles of lifetime earnings

Incidenceonworkers AE6pct AE10pct HighPresentBias LowPresentBias ProportionalCost 16

Subject to employers’ budget constraint: Total profts + Wages + Matching costs = Constant

Optimal Policy

Plannerselectsdefaulttomaximizesocialwelfare:

(selecteddefaultadopted by all employersover alifetime)

can bemorepatientthanindividuals (paternalistic) can putmore weighton low-income (inequality-averse) Saez’02 treatonlyafractionofopt-outcostaswelfarerelevantGoldin,Reck’18

16

Optimal Policy

Plannerselectsdefaulttomaximizesocialwelfare:

(selecteddefaultadopted by all employersover alifetime)

can bemorepatientthanindividuals (paternalistic) can putmore weighton low-income (inequality-averse) Saez’02 treatonlyafractionofopt-outcostaswelfarerelevantGoldin,Reck’18

Subjecttoemployers’budgetconstraint: Total profts + Wages + Matchingcosts =Constant

16

Inequality averse AE 6% AE 5% AE 4% Paternalistic AE 6% AE 6% AE 6%

Utilitarian Policymaker

Utilitarianpolicymaker prefersthe opt-inregime...

Matchandtaxincentives⇒ savemorethanimplied by preference

• AEshiftcons. evenmoretowardretirement⇒ ↓ welfare

Employers Matching Wages profts rate adjustment

Levels

Utilitarian Opt-in Opt-in Opt-in

ProportionalCost HighPresentBias LowPresentBias 17

Utilitarian Policymaker

Utilitarianlifetimeutilitydecreasesformost... ...butincreasesatthe bottom(ex. 6%AE)

• 0.2%
• 0.1%

0.0% 0.1% 0.2% 0.3% 0.4% 1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th

Consumption-equivalent change relative to opt-in Deciles of lifetime earnings

18

Inequality-Averse/Paternalistic Policymaker

Inequality-averseorpaternalisticpolicymaker setsdefaultnearmatchthreshold

Employers Matching Wages profts rate adjustment

Levels

Utilitarian Opt-in Opt-in Opt-in Inequalityaverse AE6% AE5% AE5% Paternalistic AE6% AE6% AE6%

ProportionalCost HighPresentBias LowPresentBias 19

Outline

1 Three Factsabout Autoenrollment 2 A LifecycleModelwithDefaultE˙ects

Model Estimation

3 Results

Long-terme˙ect Optimal policies

4 Conclusion

20

....... .... ................

SummaryofmyFindings

Peoplecatchupovertime... workersundomuch of AE positivee˙ect by savingless lateron AEincurrentjobcausesworkerstosavelessattheirnextopt-injob ... therefore,a$250opt-outcostcanexplaindefaulte˙ect Notsocostlytoremainat default becausecancompensatelate AEincreaseslifetime welfare/savingsonly at the bottom

• ptimaldefaultiseither0%oremployermatchthreshold

(dependsonsocialplanner™s preferences)

20

Whathavewelearned

LifeCycleHypothesis(LCH):

I AEe˙ectseenasamajorchallengetotheLCH I I showthat w/small frictionLCH performsremarkably well

Nudges:

I inadynamicsettingsavingsnudgesarelesse˙ective... I ... butcanstillhaveimportantdistributionale˙ects 21