7/28/2019 Borrow Less Tomorrow
1/18
BORROW LESS
TOMORROWBEHAVIORAL APPROACHES TODEBT REDUCTION
PRELIMINARY
Dean Karlan, Yale University
Jonathan Zinman, Dartmouth College
June 2011
7/28/2019 Borrow Less Tomorrow
2/18
Problem: Household DebtManagement
Low household savings rate (sometimes)
High debt loads/reliance in emergencies
Highest, safest return for many households:pay down expensive debt
But getting/staying out of debt is hard,
psychologically, cognitively, etc.
o Consumers may need helpo Yet many face limited, unattractive options in
market for debt management services
7/28/2019 Borrow Less Tomorrow
3/18
Potential Solution:Borrow Less Tomorrow (BoLT)
Design Strategy:o Use behavioral insights to develop a new product
that helps people reduce/manage debt
o Save More Tomorrow as a guideo SMaRT: 401k enrollments
o Important institutional/market differences between retirementsaving and debt reduction
Kitchen sink/behavioral engineering approacho
Design product to counter multiple potential biases:o In price perceptions, attention, preferences
o Product/choice architecture approach
o (Ideally/eventually complemented with robust advicemarket)
7/28/2019 Borrow Less Tomorrow
4/18
BoLT Features
Motivate: Help consumers identify whetherthey should borrow less
Plan: Help consumers make a concrete plan
to borrow less Commit: Offer client option(s) to incentivize
sticking to the plano Current options include peer supporters/referees
o Harder commitment options to be added soon
Communicate: Send feedback/reminders tostay on track
7/28/2019 Borrow Less Tomorrow
5/18
Pilot Testing So Far
Small-sample concept test: Jan 2010- April2011o Convenience sample: tax preparation clients, Community Action
Project (CAP), in Tulsa Pilot addresses two threshold questions
Will borrowers use BoLT if offered simple, freeversion?
(Also: who wants BoLT? Demand analysis)
Potential for cost-effectiveoperation/administration?
Also trying to build large enough sample totest BoLTs effectiveness
7/28/2019 Borrow Less Tomorrow
6/18
Operations/Field Experiment Flowfor Pilot Testing
Person enters tax prep site Offered survey, consent form for soft pulls of credit
report Randomly assigned to Treatment (some version of
BoLT) or Control (no BoLT, or BoLT later) BoLT group:
Counselor works with client to identify whether suitabledebt
Uses repayment calculator to come up with a plan Offers opportunity to enlist peer support Peer supporters contacted
Later: Regular reminders to make payments Client notification and peer notification if fall off-track
7/28/2019 Borrow Less Tomorrow
7/18
Pilot Sample Characteristics:Year 1 N =505 with auto loan or credit card debt
From baseline credit report:Credit score: mean 603, standard deviation (95)
Credit card balances: mean $2,408, SD ($5,103)Auto loan balance: mean $5649, SD (7626)
From baseline survey: 70% at least some college
>20% report using fringe loan product in last 2years
High prevalence of various financial distressproxies
But 57% say financial situation >= OK
7/28/2019 Borrow Less Tomorrow
8/18
Pilot Threshold Question 1:Demand?
Are borrowers interested in using BoLT? Yes!
38% take-up rate among CAPs tax prepclients with a credit card or auto loan at
baselineo This is without doing any pre-screening or target
marketing for high-interest debt
Among BoLT takers, 48% enlist peer support Given that have some BoLT takers, and lots of
baseline data on everyone, can do somedemand analysis
7/28/2019 Borrow Less Tomorrow
9/18
Question 1a, Demand Analysis:Behavioral Demand Factors?
Baseline data on N = 173 assigned to BoLT offer
Univariate corrs with behavioral measures: Often regret spending: 0.19 (0.08)
Standard time-inconsistency: 0.04 (0.09) Exponential growth bias: 0.03 (0.08)
Think about financials < a lot: -0.11 (0.08)
Denial proxy: 0.08 (0.08)
Similar multivariate results when include all ofabove behavioral factors
Will add other RHS variables as sample sizegrows.
7/28/2019 Borrow Less Tomorrow
10/18
Question 1b, Demand Analysis:Description of BoLT plans
58% of takers use for a credit cardo Estimated mean interest rate: 18%
o Estimated principal amount: $4,000
o
Mean plan to accelerate repayment: $23/montho Median payoff time with vs. without BoLT: 17 vs. 35
months
42% of takers use for auto loano Estimated mean interest rate: 12%
o Estimated principal amount: $10,000
o Mean plan to accelerate repayment: $41/month
o Median payoff time with vs. without BoLT: 32 vs. 45months
7/28/2019 Borrow Less Tomorrow
11/18
Pilot Threshold Question 2:Delivery?
Can BoLT be marketed cost-effectively? Probably yes. More speculation on this later.
Feasible to monitor whether someone is on track
with paydown schedule?o Yes, using credit report soft pulls (with client consent
obtained up front)o Finding cost efficiencies key to scale-up
Peer supporters actually reachable when a BoLTclient falls off track?o Yes, though there is some room for improvement;
e.g., give a heads-up to peers up front
Text message reminders deliverable? Yes.
7/28/2019 Borrow Less Tomorrow
12/18
Beyond Threshold Questions:Is BoLT Effective?
Why is random assignment important for answeringthis?
After 6 months, 29% of BoLT users were onschedule
59% of BoLT users stayed on schedule for >= 1month
Are these proportions low or high? Unknown!
o Depends on the counterfactual what wouldbalances look like in absence of BoLT?
o If difficult to pay down expensive debt thencounterfactual = continued high balanceso Analogy to smoking: most attempts to quit fail
o So we implement BoLT as part of field test that
7/28/2019 Borrow Less Tomorrow
13/18
Measuring Effectiveness:Some Outcomes of Interest
Outcome measurement from follow-up softpulls of credit report. So can measure:
Credit score
Balances *1/0 plausible-sized reduction in balances*
In other settings might be able to do some
short follow-up surveys by web or phoneMeasure self-assessed financial condition, etc.
7/28/2019 Borrow Less Tomorrow
14/18
o ec veness on ear sample:
12-month treatment effects OLS-estimated effects of BoLT offer on: 5%, 10% reduction in card debt: -0.05(0.04), -
0.04(0.04)
5%, 10% reduction in car debt: 0.03(0.05),0.03(0.04)
Propensity score matching (person-level)estimates: 5%, 10% reduction in card debt: 0.00(0.05),
0.00(0.05)
5%, 10% reduction in car debt: 0.14(0.06),0.15(0.06)
Small sam le is a challen e but:
7/28/2019 Borrow Less Tomorrow
15/18
sample:Preliminary thoughts on
interpretationWhy might this version of BoLT work for car, notcard?
Most people have many cards, one car loan
Explore this hypothesis going forward using Loan-level matching estimators: that way can look at
whether BoLT takers pay down targeted card, butcharge up other(s)
Heterogeneous treatment effects: conditional on otherbaseline characteristics, is BoLT less effective forthose with more cards
If this whats going on, suggests important goingforward to offer BoLT across multiple loans
7/28/2019 Borrow Less Tomorrow
16/18
Going Forward: DesignPermutationsMuch more development and testing worth doing!
Consumermotivation through direct marketing
Expand and improve set of potential plans. Examples:o Plans over aggregated balances (e.g., across all cards)
o Plans to restrict new borrowing (e.g., no new trades)
o Extend to products not covered in traditional bureaus?o Improve repayment planning using algorithm-based advice to set
default options
Expand set of potential commitments. Examples:o Performance bonds
o
Cut me offso Link peer support to social networking
Refine communication strategy Content, timing, frequency, duration, channel, client customization
7/28/2019 Borrow Less Tomorrow
17/18
Going Forward: Business Model
Reach scale with partners who have comparativeadvantage in deliveryo Bureaus/credit report monitoring services
o Credit counseling agencies
o Online financial management aggregators/advisors
o Debt buyers/collectors (back-end for clients whorepay)
o Payments platformso Mobile walletso Bill payers
Identify revenue modelso Subscription (as part of a larger bundle?)
o Cross-sell/bundling
7/28/2019 Borrow Less Tomorrow
18/18
BoLT and Beyond:Debt Management Market Opportunities
BoLT but one example of a focus on market solutions for
Helping consumers take control of their finances
In particular on the liability side of balance sheet Huge asset management industry
Wheres the debt management industry? Trillions of dollars in U.S. household debt
Return (cost) dispersion much bigger on liability than on assetside Hall and Woodward (2010); Stango and Zinman (2011a, b)
Helping consumers reduce borrowing levels, costs is where themoney is: hundreds of basis points
Existing solutions limited in scope (CCS), value (debt settlement,brokers)
(Wheres household balance sheet management industry?) Private banking for Main Street