Borrow Less Tomorrow: Behavioral Approaches to Debt Reduction
by Dean Karlan
and Jonathan Zinman
– Visit Original Article
Mounting evidence suggests that behavioral biases and cognitive limitations (which we referred to loosely as “behavioral factors”) depress wealth accumulation.
Most applications of this research have focused on asset accumulation: using behavioral insights to develop product and policy innovations, thus far largely delivered through U.S. workplaces, that facilitate increased saving or investment for retirement.
There is also mounting evidence that behavioral factors may lead households to “over borrow” as well as “under save,”3 but little of this work has developed or tested potential solutions. This is not for wont of need. For many households debt decumulation is a more efficient route to increase net worth; e.g., many more households hold debt than financial assets, and historical credit card and auto loan interest rates exceed historical equity returns. But the mass market for debt reduction advice and products is thin.4
This opportunity motivates us to develop and test Borrow Less Tomorrow (BoLT), a behavioral approach to debt reduction that combines a simple planning/goal-setting process, commitment, and reminders. The planning/goal-setting process in the present study involved interested clients working with a surveyor/marketer to identify a suitable auto loan or credit card and come up with a realistic repayment schedule that would accelerate repayment. Marketers worked with the client to identify a relatively high-APR debt and used a simple spreadsheet to help illustrate the impact of different monthly payment amounts on total repayment time and interest paid. The voluntary commitment option here was social in nature: a BoLT client could name one or more peer supporters who would be notified if the
client fell off-track. Once signed up, the research team monitored repayments using “soft pulls” of the client’s credit report (i.e., credit checks that do not impact the credit score because they are not used for underwriting purposes) and delivered clients monthly reminder messages (email or phone). We emphasize that the version of BoLT tested here is very much of the “proof of concept” variety: there is almost certainly room for improvement and adaptation in the product design, delivery, and administration, as we discuss in the Conclusion.
BoLT’s “kitchen-sink” approach of using several levers to address multiple behavioral factors is modeled on Save More Tomorrow™ (SMaRT). But the institutional particulars are different. SMaRT piggybacks on the extensive workplace intermediation of retirement saving created by employers tax subsidies. Debt markets are far less intermediated in the U.S. (and elsewhere), and so we pilot-tested BoLT using a “direct-to-consumer” distribution channel. SMarT also has a key component in which it shifts the default action to saving more unless someone comes in and actively decides to reverse their decision, whereas with BoLT we help borrowers set a new plan but do not have a mechanism in this institutional setting to change the default action.
Our pilot sample of 465 individuals is drawn from the clientele of free tax-preparation services offered by the Community Action Project of Tulsa, Oklahoma. From January-April 2010, the research team approached persons waiting to get their taxes done at three of CAP’s sites with an invitation to take a fifteen-minute survey in return for a $5 gift card to QuikTrip, a local gas station and convenience store.5Those who completed the survey and a consent form permitting credit report soft pulls were then randomized to either receive a BoLT offer or not. Baseline credit reports identify those with a potentially suitable debt.
The pilot results are merely suggestive, but they point to strong demand for new debt reduction products/services with behaviorally-motivated features. Among those randomly assigned to a BoLT offer, 41% signed up for some version of it. Conditional on BoLT take up, the take-up rates for escalating repayment schedules, peer support, and reminders were 41%, 27%, and 81%.
Estimates of 12-month treatment effects offer some hope that BoLT-like approaches can produce their intended effects, although we caution that most of our results here are imprecisely estimated zeros. We measure impacts on financial condition using the random assignment to identify the causal effects of BoLT and credit report data to glean some unbiased measures of financial condition. Unobtrusively captured administrative data has several potential advantages over survey data relying on subject self-reports. It can be cheaper to obtain. It can be less prone to biased attrition. It is less prone to reporting biases that may be correlated with treatment and unobserved determinants of outcomes (Karlan and Zinman 2008; Zinman 2009). In all we find some statistically weak evidence that BoLT reduces credit card balances over a 12-month horizon, and little evidence that BoLT affects auto balances or broader outcomes (credit scores, delinquency, line of credit utilization, or number of active debts).
This pilot informs several lines of inquiry in (household) finance, behavioral economics, and intertemporal choice. It is the first study we know of that takes the behavioral financial engineering approach used to interesting effect in asset markets (Benartzi and Thaler 2004; Ashraf, Karlan, and Yin 2006; Gine, Karlan, and Zinman 2010) and applies it to debt reduction. It provides novel data points on the demand for a new debt reduction product and for a social commitment device. And it is the first paper we know of that uses a randomized-control design to measure the impact of a debt reduction initiative.6
The paper proceeds as follows. Section 2 describes our pilot setting and sample, BoLT product design and implementation, and research design and implementation. Section 3 describes BoLT demand and usage in the pilot. Section 4 estimates whether BoLT actually succeeded in helping people reduce debt over a 12-month horizon. Section 5 concludes by discussing prospects for improving BoLT’s design and implementation, including some speculation about various business models for offering BoLT on a large scale.
