Social Networks, Reputation and Commitment: Evidence from a Savings - - PowerPoint PPT Presentation

Social Networks, Reputation and Commitment: Evidence from a Savings Monitors Experiment

Emily Breza† Arun G. Chandrasekhar‡

†Columbia Business School ‡Stanford

Under-savings ubiquitous

◮ Evidence of large benefits of savings ... yet low

◮ e.g., Dupas and Robinson ‘13, Schaner ‘13, Beaman et al ‘14

◮ Access not necessarily the problem in India

◮ RBI-led expansion of rural branches, no-frills accounts ◮ Low rates of account opening and use

◮ Psychological “frictions” make saving hard:

◮ Can’t commit to save/procrastination? (e.g., Ashraf et al ‘06) ◮ Can’t remember/inattention? (Karlan et al ‘12, Kast et al ‘13)

This paper: can we use social reputation to overcome such frictions and encourage savings?

Informal finance uses social reputation

Peer-driven financial institutions are thought to rely on this:

◮ RoSCAs, SHGs, VSLAs, Microfinance groups

In theories of MF/ROSCAs,“social reputation” often assumed “the contributing member may admonish his partner for causing him or her discomfort and material loss. He might also report this behavior to others in the village, thus augmenting the admonishment felt. Such behavior is typical

• f the close-knit communities in some LDCs.”

– Besley and Coate (1995)

What we do

Encourage savings by assigning a unique monitor to each saver.

◮ Basic idea:

◮ Make a bet with self about ability to save over 6 months. ◮ Stakes: reputation gain/loss from progress in front of some other

member of village.

◮ Monitor assigned to a saver for the duration of experiment.

◮ Informed about savings in target account. ◮ Simply told about progress (bi-weekly). ◮ Monitor need not do anything!

Why should a saver care about the monitor?

“A person may save more if it is an important person knowing they might get more benefits from this person later on.” – Subject 1 “The monitor will feel that if in the future he or his friends gives her some job or tasks or responsibilities, the saver may not fulfill them” – Subject 2 “They would speak less to the saver and feel ‘cheated to trust’ [sic]. They may tell others...” – Subject 3 “People will only reach their goals if their monitors are family, friends, neighbors, or important people.” – Subject 4

What we do

Encourage savings by assigning a unique monitor to each saver.

◮ Basic idea:

◮ Make a bet with self about ability to save over 6 months. ◮ Stakes: reputation gain/loss from progress in front of some other

member of village.

◮ Monitor assigned to a saver for the duration of experiment.

◮ Informed about savings in target account. ◮ Simply told about progress (bi-weekly). ◮ Monitor need not do anything!

◮ Not all monitors created equal...

◮ Central monitors?

Can spread more info; more important in future interactions

◮ Proximate monitors?

Info typically goes to people saver will run into.

Setting

◮ 60 villages in rural Karnataka,

India

◮ 1.5 to 3 hour’s drive from

Bangalore

◮ Experimental participants

aged ∼18-45

◮ 1,300 savers who expressed

desire to save more

◮ 1,000 monitors

◮ Primary occupations:

agriculture and sericulture

Village network data

◮ ∼16,500 households

surveyed across 75 villages

◮ Relationships:

relatives, friends, creditors, debtors, advisors and religious company

◮ Undirected,

unweighted OR network

A simple model social reputation flow

Record savings

Report to Monitor (Low Centrality)

Only a few people hear gossip

Report to Monitor (High Centrality)

Many more people hear gossip

Report to Monitor (Low Proximity)

Only a few (distant) people hear gossip

Report to Monitor (High Proximity)

Only a few (close) people hear gossip

Who would make a good monitor?

low centrality high proximity > high centrality high proximity > low centrality low proximity

◮ greater motivation to save if more people are likely to hear about

your good/bad deeds (centrality)

◮ more relevant if people informed of your good/bad deeds are

those you are likely going to meet in the future (proximity)

Questions

• 1. Can we encourage savings using central/proximate monitors?
• 2. Does information flow? Where is savings coming from?
• 3. What happens in the medium term (15+ mo. later)?
• 4. When given choice of monitor, do individuals pick well or unwind?

Design

Treatments: 1300+ savers, 1000+ monitors, 60 villages

• 1. No Monitor (BC): in all 60 villages
• 2. Researchers Choose Monitor at Random (R): 30 villages
• 3. Savers Choose Monitor Endogenously (E): 30 villages

All received bundle of services (resembles business correspondent)

◮ Account opening ◮ Goal elicitation (conducted at pre-screen home visit) ◮ Bi-weekly visits (reminders and weak monitoring)

Treatments and Roll-Out

Village

◮ Sample villages selected (based on networks data)

Treatments and Roll-Out

Pure Control Potential Savers Potential Monitors

◮ Potential savers & monitors visited, savings goals elicited

Treatments and Roll-Out

Pure Control Savers Attend Meeting Monitor Pool Monitor Dropouts Savers Not Interested

◮ Interested monitors and savers attend village meeting

Treatments and Roll-Out

Pure Control Monitored +BC Saver Chosen Monitors Monitor Dropouts Savers Not Interested Excess Monitors BC Saver

◮ Some savers randomly chosen to receive monitors

Treatments and Roll-Out

Pure Control Monitored +BC Saver Endogenous Chosen Monitors Monitor Dropouts Savers Not Interested Excess Monitors BC Saver Pure Control Monitored +BC Saver Random Chosen Monitors Monitor Dropouts Savers Not Interested Excess Monitors BC Saver

Village A Village B

◮ Random vs. Endogenous Monitor assignment randomized at

village level

◮ Random Matching (30 villages)

◮ Savers randomly assigned to a monitor from pool

◮ Endogenous Matching (30 villages)

◮ Savers choose monitor from pool in random order

Timeline

Saving Period Begins:

• Baseline Survey
• Bi‐weekly visits start
• Monitors start to get info

Village Meeting Account Opening:

• Bank or PO

Follow‐Up Survey Saving Period Ends:

• Endline Survey

6 Months ~15 Months

Compensation

◮ Pure Control (no contact until end of 6 mos.)

◮ No compensation

◮ Savers (takers only)

◮ In Kind: Account opening services ◮ Direct: Rs. 50 ($1) deposited into account

◮ Monitors

◮ Payment: ◮ Rs. 50 if saver reaches half of goal

[helps in a robustness exercise]

◮ Rs. 150 if saver meets goal ◮ Rs. 0 otherwise

Results

• 1. Can we encourage savings using central/proximate monitors?

Do randomly assigned monitors help?

Results: Log Total (Form. + Inform.) Savings

7.4 7.5 7.6 7.7 7.8 7.9 8 8.1 BC, Random Monitor, Random

Mean log savings balances across all accounts

Endline

.2 .4 .6 .8 1 Density

• 4
• 2

2 4 6 log(Total End Savings/Savings Goal) Random Monitor No Monitor

Does the network position of random monitors matter?

7.4 7.5 7.6 7.7 7.8 7.9 8 8.1 8.2 BC Low Centrality Monitor High Centrality Monitor

Mean log savings balances across all accounts

7.2 7.4 7.6 7.8 8 8.2 8.4 8.6 BC Far Saver‐Monitor Close Saver‐Monitor

Mean log savings balances across all accounts

Monitor effectiveness & graph position

log (Form.+Inform. Sav.)iv = α + βCentmon(i) + γProxi,mon(i) + δ′Xiv + ǫiv

(1) (2) (3) (4) (5) (6) Dependent Variable Log Total Savings Log Total Savings Log Total Savings Log Total Savings Log Total Savings Log Total Savings Monitor Centrality 0.178** 0.134* 0.153** (0.0736) (0.0729) (0.0675) Saver-Monitor Proximity 1.032*** 0.865** 1.108*** (0.352) (0.334) (0.294) Model-Based Regressor 1.450** 1.819*** (0.693) (0.632) R-squared 0.150 0.155 0.161 0.148 0.101 0.080 Fixed Effects Village Village Village Village Controls Saver, Monitor Saver, Monitor Saver, Monitor Saver, Monitor Double- Post LASSO Double- Post LASSO ◮ Increasing monitor centrality by one standard deviation increases total

savings by 14%

◮ Increasing social proximity by one standard deviation increases total

savings by 16%

• Regs. conditional on demographics (e.g., caste, wealth, age, geo.)

Endline

.2 .4 .6 .8 1 Density

• 4
• 2

2 4 6 log(Total End Savings/Savings Goal) R Monitor: High Centrality R Monitor: Low Centrality

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

Did beliefs change?

Respondents’ beliefs about savers

◮ 560+ random respondents chosen 15 mo. after end of intervention ◮ asked about 8 savers who had monitors ◮ asked if each saver was responsible (e.g., “good at meeting goals”) ◮ is respondent more likely to say “Yes” when the saver truly did

meet her savings goal (or “No” when the saver didn’t) when the random monitor is more central?

Dependent Variable: Beliefs about Saver Monitor Centrality Respondent-Monitor Proximity Observations R-squared Fixed Effects Controls (4) (5) (6) Reached Good at Meeting Goals Good at Meeting Goals Good at Meeting Goals 0.0389 0.0374 0.0353 (0.0144) (0.0140) (0.0148) 0.0476 0.0181 0.0360 (0.0422) (0.0366) (0.0342) 4,743 4,743 4,743 0.030 0.023 0.314 Respondent No Village Respondent r r r Saver Saver Saver

Central monitor causes beliefs to be updated in direction of actual goal attainment (13.3%)

Where did the savings come from?

Retrospective

(1) (2) (3) (4) (5) (6) Dependent Variable Increased Labor Supply Business Profits Cut Unnecessary Expenditures Money from Family and Friends Reduced Transfers to Others Took a Loan Random Monitor 0.0712 0.0202 0.0787

• 0.0227

0.0148

• 0.0222

(0.0332) (0.0156) (0.0422) (0.0346) (0.0120) (0.0190)

• Dep. Var. Mean

0.15 0.03 0.15 0.19 0.01 0.04 Observations 1,026 1,026 1,026 1,026 1,026 1,026 R-squared 0.055 0.026 0.020 0.056 0.016 0.014 Fixed Effects Village Village Village Village Village Village Controls Saver Saver Saver Saver Saver Saver

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

◮ Evidence of reputation updating 15 months later ◮ Save from labor/business, cut unnecessary expenditure

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

◮ Evidence of reputation updating 15 months later ◮ Save from labor/business, cut unnecessary expenditure

• 3. What happens in the medium term (15+ mo. later)?

What happens 15 months out?

↓ in (inability to respond to) shocks

(1) (2) (3) (4) Dependent Variable: Shocks Total Number Total Number Greater than Median Greater than Median Monitor Treatment: Random Assignment

• 0.199
• 0.249
• 0.0757
• 0.0944

(0.128) (0.131) (0.0416) (0.0441) Observations 1,153 1,153 1,153 1,153 R-squared 0.021 0.021 0.019 0.016 Mean of Dep. Var (Control) 1.769 1.769 0.577 0.577 Fixed Effects Village No Village No

Asked about not having enough money to cover necessary expenses in response to:

◮ Health shock, livestock health shock, other urgent consumption

need etc.

15 Months Later

.2 .4 .6 .8 1 Density

• 10
• 5

5 log(Total EL2 Savings/Savings Goal) Random Monitor No Monitor

15 Months Later

.2 .4 .6 .8 1 Density

• 10
• 5

5 log(Total EL2 Savings/Savings Goal) R Monitor: High Centrality R Monitor: Low Centrality

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

◮ Evidence of reputation updating 15 months later ◮ Where does savings come from?

• 3. What happens in the medium term (15+ mo. later)?

◮ 10%-20% decline in being unable to respond to shocks ◮ Monitor benefits persist: 34% increase in total savings

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

◮ Evidence of reputation updating 15 months later ◮ Where does savings come from?

• 3. What happens in the medium term (15+ mo. later)?

◮ 10%-20% decline in being unable to respond to shocks ◮ Monitor benefits persist: 34% increase in total savings

• 4. When given choice of monitor, do individuals pick well or unwind?

Can savers get there on their own in endogenous villages?

Endogenous treatment

Goal: Benchmarking exercise

◮ Policy-relevant alternative, naturalistic implementation

◮ recall MF and ROSCAs often have endogenous group formation ◮ Stickk.com has individuals pick a “referee” to verify their progress

toward their goals and to provide motivation

◮ an MFI has actually approached us to try to implement

something similar in an urban customer population

◮ Experimental design allows for this measurement

What should we expect? Lots of possible outcomes:

◮ savers could pick enablers, unwind any benefits of a monitor ◮ savers could pick savings-maximizing allocation of monitors ◮ anything in between

Note: Experiment not designed to unpack choice

7.4 7.5 7.6 7.7 7.8 7.9 8 8.1 8.2 8.3 BC, Random Monitor, Random

Mean log savings balances acr

7.4 7.5 7.6 7.7 7.8 7.9 8 8.1 8.2 8.3 BC, Random Monitor, Random

Mean log savings balances acr

BC, Endogenous Monitor, Endogenous

es across all accounts

Results

• 1. Can we encourage savings using central/proximate monitors?

◮ ↑ 1 · σ in centrality =

⇒ > ↑ 14% total savings

◮ ↑ 1 · σ in proximity =

⇒ > ↑ 16% total savings

◮ receiving a monitor =

⇒ >↑ 35% total savings

• 2. Does information flow? Where is savings coming from?

◮ Evidence of reputation updating 15 months later ◮ Where does savings come from?

• 3. What happens in the medium term (15+ mo. later)?

◮ 10% Lower incidence of being unable to respond to shocks ◮ Monitor benefits persist: 34% increase in total savings

• 4. When given choice of monitor, pick well enough.

Conclusions

◮ Leveraging reputational force can encourage savings

◮ Particularly powerful if “right” partner is chosen

◮ Community doesn’t unwind it; chosen monitors can be used ◮ Reputational channel may be an important driver of behavior in

RoSCAs, SHGs, MFI groups, etc.