A Personal Touch: Text Messaging for Loan Repaymentjzinman/Papers/RepaymentReminders.pdf · We...

1

A Personal Touch: Text Messaging for Loan Repayment*

July 2015

Dean Karlan Melanie Morten Jonathan Zinman

Yale University Stanford University Dartmouth College Department of Economics Department of Economics Department of Economics

New Haven, CT 06520 Stanford, CA 94305 Hanover, NH 03755 [email protected] [email protected] [email protected]

ABSTRACT We worked with two microlenders to test impacts of randomly assigned reminders for loan repayments in the “text messaging capital of the world”. Messages only improve repayment when they include the loan officer’s name. This effect holds for clients serviced by the loan officer previously but not for first-time borrowers. Taken together, the results highlight the potential and limits of communications technology for improving loan repayment effort, and suggest that personal connections between borrowers and bank employees can be harnessed to help overcome market failures. Word count: 4432 Keywords: microfinance, microcredit, loan repayment, debt management, text messages, ICT4D JEL codes: D12, G21

*Karlan: [email protected], Yale University, Innovations for Poverty Action, M.I.T. J-PAL, and NBER; Morten: [email protected], Stanford University and NBER; Zinman: [email protected], Dartmouth College, Innovations for Poverty Action, M.I.T. J-PAL, and NBER. We appreciate the cooperation of Greenbank and Rural Bank of Mabitac in designing and implementing this study. We thank the editor and anonymous reviewers, John Owens and the staff at MABS in the Philippines for help with data and implementation mechanics, and Tomoko Harigaya, Rebecca Hughes, Mark Miller, Megan McGuire and Junica Soriano for excellent field work. We thank conference participants at the 2011 Advances with Field Experiments and 2013 IZA/WZB Field Days conference for helpful comments. Financial support from the Bill and Melinda Gates Foundation is gratefully acknowledged. All opinions and errors are our own.

2

I. Introduction

We worked with two large Philippine banks, Greenbank and Mabitac, to test whether and

how text message reminders can induce timelier loan repayment in a country that has been

dubbed the “text messaging capital of the world”.1 Each bank sent borrowers weekly text

message reminders about their weekly repayment obligation. Additional levels of randomization

varied the timing and content of the messages across borrowers. The timing treatment varied

whether each message was sent on the due date, or on one or two days prior to the due date. The

content treatments varied whether the message mentioned the loan officer’s name, and whether

the message was framed in loss or gain terms (Table 1).

In testing whether and which messages affect loan repayment, this design addresses two

deeper questions. One is what drives default. Ineffective repayment messaging would be

consistent with default being out of the borrower’s proximate control—e.g., bad luck instead of

bad behavior— and hence inconsistent with several varieties of market failure where incentive

problems (what economists refer to as moral hazard, a change in the amount of effort when the

effort cannot be fully monitored, such as a borrower forgetting to make loan repayments after

taking out a loan, or failing to put enough money aside when a bit of planning could have done

so easily) create market failures. Effective repayment messaging would suggest that borrowers

1 See, for example, http://www.businesswire.com/news/home/20100823005660/en/Research-Markets-Philippines---Telecoms-Mobile-Broadband. Accessed 19 February 2012.

3

do exercise some discretion about whether to repay and hence support the hypothesis that moral

hazard is a problem. The second question is how to mitigate any moral hazard and thereby

improve repayment (and market efficiency). As we detail below, our results on messaging

content lead to the surprising inference that high-touch connections between loan officers and

borrowers are key for crafting strategies for low-touch communications like SMS.

We do not find an overall treatment effect: there is no evidence that the average message

improved repayment performance relative to the control group. The timing treatments do not

have significant effects relative to the control group, nor significant differences from each other.

Nor does loss or gain framing produce significant improvements relative to the control group, or

robustly significant differences from each other.

We do find that including the loan officer’s name significantly improves repayment. For

example, we estimate that it reduces the likelihood that a loan was unpaid 30 days after maturity

by 5.1 percentage points (a 38% reduction on a base of 0.135). The effect of mentioning the loan

officer’s name are only significant for borrowers who entered the study having borrowed from

the bank before-- and hence having been serviced by the same loan officer. The effect is

significantly different for prior borrowers compared to first-time borrowers. These results have

implications for several different aspects of research and practice.

First, showing that repayment can be swayed by mere wording of a text message adds

evidence that default in credit markets is at least partially due to the borrower’s decision to

repay, not her proximate inability to repay. This implies that improved enforcement strategies

could reduce default (e.g., Adams, Einav, and Levin 2009; Karlan and Zinman 2009).

4

Second, the results shed some light on why or why not some types of messages might

influence voluntary repayment decisions. We start with why not. Most message varieties here do

not work, and there are only minor textual differences across the different varieties (see Table 1

for the scripts). Hence it seems doubtful that the messages here mitigated limited attention as a

pure reminder, as postulated by the most closely related study.2,3 Nor does it seems likely that the

messages provided informative signals about bank enforcement intentions; if that were the case,

we might expect that all messages (all of which mention the bank name) would improve

repayment relative to the no-message condition. And we might expect that including the loan

officer’s name in the message would have (relatively) strong effects on first-time borrowers, who

probably have less precise expectations about bank and loan officer enforcement practices,

whereas instead we found the reverse.4 Nevertheless we cannot rule out the possibility that prior

borrowers inferred closer monitoring from the loan officer message, perhaps because such a

message signals a departure from business-as-usual.

2 Cadena and Schoar (2011) randomize whether individual microcredit clients in Uganda were sent an SMS, which in most cases was a picture of the bank, three days before each monthly loan installment was due. They do not randomize timing or content. Their messages improved timely repayment by 7-9% relative to the control group, an effect they benchmark, using pricing randomizations, as commensurate with effects of reducing the cost of capital by 25%. Karlan et al (2012) and Kast et al (2012) find that text message reminders increase savings deposits among microfinance clients in four different banks across four different countries. 3 One could square our results with a limited attention interpretation as follows: messages mentioning the borrower’s name reduce repayment relative to getting a message with no one’s name mentioned (we did not test a nameless message), and mentioning the loan officer’s name is no more effective than sending a nameless message. I.e., mentioning the borrower’s name is somehow off-putting to the borrower—this does not seem especially plausible given that bank correspondence typically addresses a borrower by name-- but aside from that, messages serve as effective reminders. 4 One could square this result with a limited attention interpretation, if text messages are a stronger signal of increased monitoring when they come from a recognizable individual employee of the bank, and repeat borrowers are more likely to recognize, or attend to, the name of their loan officer.

5

We think the most plausible interpretation of hinges on personal relationships between

borrowers and loan officers. Perhaps borrowers feel indebted, both financially and socially, to

their loan officer, and the message triggers feelings of obligation and/or reciprocity (e.g.,

Cialdini and Goldstein 2004; Charness and Dufwenberg 2006) that increase repayment effort.

This mechanism is distinct from the ones studied in the long literature on how information

acquired by bank employees can help improve loan performance by helping banks price and

enforce better. That literature (e.g., Boot 2000; Agarwal et al. 2011) focuses on how additional

information improves screening and/or monitoring. In contrast, our results suggest that

messaging can improve repayment even without obtaining additional information on the

borrower. This holds whether the underlying mechanism is providing information to the

borrower (via signaling) and/or priming a personal relationship between loan officer and

borrower.

Third, our results shed light on optimal design of information and communications

technology-driven development efforts (Information and Communication Technologies for

Development, or “ICT4D”). Research on ICT4D is only in its youth (e.g., Aker and Mbiti 2010;

Donner 2008; Jack and Suri 2011), so unsurprisingly there has been relatively little focus on the

mechanics (content, timing, etc.) of communications. Our findings suggest that content is

critical, even within the constraints of a 160-character text message limit. Another interesting

question is how technological innovations interact with local institutions. E.g., it seems intuitive

to expect that, at least to some extent, economies of scale in Information and Communication

Technology (ICT) would favor larger, (trans-)national, transaction-based institutions over

6

smaller, more local, relationship-based institutions like the ones conducting this experiment (our

partner banks here are large by Philippine microlending standards, but much smaller than very

large financial institutions one tends to think of benefiting from economies of scale). Our results

suggest that properly crafted ICT-based innovations can actually buttress relationship lending

and the smaller institutions that rely on it. Low-touch messaging strategies and high-touch

interactions can interact in important ways.

Our results come with some important caveats. Most of our null results are imprecise; our

confidence intervals often do not rule out economically large effects in either direction, although

we are able to reject equality between the loan officer message and the other ones. We can only

examine effects on the first loan with messages, and hence cannot say whether borrowers

become desensitized or more sensitized to messages over multiple loan cycles.5 Extrapolating to

other settings requires caution; e.g., this is one of only two loan repayment message studies we

know of, and taken together the two studies already present an important puzzle. Cadena et al

(2011) find that an SMS image of their bank does increase repayment on average. We do not find

that text messages mentioning the bank increase repayment on average. Is this difference due to

variation across the two studies in borrower characteristics? In credit market characteristics? In

ICT market characteristics? In lender practices? These questions highlight the need for

formulating and testing different theories about mechanisms underlying messaging effects, and

for testing these theories within settings when theory predicts such heterogeneity, and across 5 We find no evidence that the effect of a given type of text message changes over time (as clients cumulatively receive more messages) when we have multiple observations of payment or nonpayment per loan recipient (figures available on request). See Calzolari and Nardotto (2011), Altman and Traxler (2014), Stango and Zinman (2014), and Alan et al (2015) for evidence on reminder/messaging dynamics.

7

different settings when theory predicts differential effects across underlying contextual factors

and market conditions.

II. Experimental Setting and Design

We worked with two for-profit banks, Green Bank and Mabitac, to design and implement an

SMS loan repayment experiment on individual liability microloan borrowers. Green Bank

operates in both urban and rural areas of the Visayas and Mindanao regions. It is the 5th largest

bank by Gross Loan Portfolio in the Philippines.6 Mabitac operates in both urban and rural areas

of the Luzon region. It is the 34th largest bank in the Philippines. Both banks are among the

leading microlenders in the Philippines.

The Philippines is a suitable environment for a study of responses to SMS messaging.

Anecdotally known as the “texting capital of the world”, cellphone use is widespread: 81% of the

population had a cellphone subscription in 2009,7 and texting is an especially popular method of

communication because of its low cost, generally about 2 cents a message. Approximately 1.4

billion SMS are sent by Filipinos each day.

We lack much demographic data on the specific borrowers in our sample, but prior work

with a different bank with similar microfinance operations suggests that borrowers are likely

predominantly middle-aged female microentrepreneurs, and likely about average with respect to

education and household income (Karlan and Zinman 2011). This other bank however is located

6 Source: www.mixmarket.org, accessed 2/7/2012. 7 Source: World Bank Development Indicators Database; accessed 8/27/2011.

8

in urban and peri-urban Manila, versus the urban and rural setting of the two partner banks from

this study.

Summary statistics for the loans in our sample are presented in Table 2. The average loan is

approximately $400 USD, repaid weekly over a 16 to 20 week term at around 30% APR.

Microloan charge-off rates are typically around 3% for the banks in this study.

Late payments are common (Table 3): 29% of weekly loan payments are made at least one

day late in the control group, and 16% are a week late. 14% of loans are not paid in full within

30 days of the maturity date. Late payments and delinquencies are costly for the banks because

they trigger additional monitoring, accounting, and legal actions. Mabitac begins actively

following up on late payments 3 days after the due date, while Greenbank begins after 7 days8.

Late payments and delinquencies can also be costly for the borrower: borrowers are charged late

fees, may be subject to legal action, and their creditworthiness and ability to secure future loans

is decreased. So the focus of the experiment was to improve timely repayment of the weekly loan

installments.

We designed the experiment to test the effects of three different dimensions of a messaging

strategy. One dimension is whether to send messages at all: borrowers were assigned to either

treatment (receive a message weekly, for the entire loan term) or control (no message). 8 Bank staff at Mabitac communicated to us that their standard procedure was that a call was made to the client the day that the payment was missed, and the loan officer notifies their supervisor. After 3 days, the account officer visits the client again, accompanied by their supervisor. If the client cannot pay then the first demand letter to the client is issued. For Greenbank, there is a grace period of maximum 7 days from when the payment is due. The account officers make a follow up visit after the payment is first missed. If the customer cannot pay, then the first demand letter is issued. The second visit the head account officer accompanies the loan officer to the meeting, and then a second demand letter is issued. The third visit the branch manager accompanies the loan officer and a third demand letter is issued.

9

A second dimension is messaging content. Each message was randomly assigned a 2x2

combination of “loss vs. gain framing” and “personalization”, producing the four scripts shown

in Table 1. All message variants included some boilerplate content: the bank name, “pls pay your

loan on time”, and “If paid, pls ignore msg. Tnx”. Loss-framed messages started with “To avoid

penalty…”. Gain-framed messages started with “To have a good standing...”. The

personalization treatment varied whether the client's name (“From [bankname]: [client name],…)

or the account officer’s name (“From [account officer’s name] of [bankname]:….”) was included

in the first 2 or 3 words of the message.

The third dimension tested here is timing: we randomly and independently varied whether

each of the borrower’s messages were sent 2 days before the scheduled payment date, the day

before the scheduled payment date, or the day of the scheduled payment date.9 Borrowers

received the same content and timing each week, for the term of the loan; i.e., we randomized at

the loan level.10

9 The randomization was set to 33% for control and 66% to treatment, equally divided between the 4 treatment groups. The timing treatment was independently randomized, with each three treatments equally likely. However, due to a coding error the final breakdown of randomization was 34% control, and then 12%, 25%, 14%, 15% to each of the four treatments instead of 17% each treatment group. There was no error in coding the independently randomized timing treatment. Table 2 tests for balance across treatment arms in baseline loan characteristics, and we find no evidence of imbalance. 10 Other mechanics of the randomization: the research team received weekly reports of clients with payments due in the following week from each participating branch. We randomized clients the first time they appeared in these weekly reports. Once randomized into treatment, clients received weekly text messages until their loan maturity date. The text messages were automatically sent using SMS server software.

10

Our study sample includes 943 loans originated by Greenbank and Mabitac between May

2008 and March 2010,11 and captures about half of the individual liability microloans made by

the two banks during this period where the client provided a cellphone number to the lender. We

include only the first loan per client during this time period because a treatment that affects

repayment on the first loan in turn affects the likelihood of subsequent loans. We assign clients

as either new or repeat borrowers based on their loan history prior to the commencement of our

study. We exclude first loans where bank reporting errors made it impossible for us to randomize

messaging or to match randomly assigned loans to bank data on loan repayment.12

Table 2 verifies that assignment to treatment was orthogonal to key loan characteristics: loan

size, loan term, and number of weeks the loan was part of the study (to verify there was not

differential attrition). None of the pairwise comparisons between control and treatment produces

a statistically significant difference (thus the lack of stars in rows (2)-(9)). We also regress each

of the eight different measures of treatment assignments on the three baseline loan variables and

fail to reject equality at the 90% statistical significance threshold in only one of the eight

regressions, which is about what would expect to find by chance (i.e., one expects one false

rejection for every ten regressions).

11 We stopped randomizing in March 2010, after Green Bank changed its database and its reporting of loans to the research team for random assignment (see next footnote) was disrupted. 12 We randomized a total of 1703 loans. We could only randomize loans that were reported to us in timely fashion by the participating bank branches. 138 loans were not reported to us until less than two weeks before the end of their repayment cycle and are dropped from the analysis. Another 305 loans that did get random assignments could not be matched to repayment data because one of the banks changed its database midstream and did not create a unique identifier for tracking/matching loans across the old and new database. Treatment assignment is uncorrelated with whether we could match to administrative data (p-value 0.53). One loan was randomized twice, and 316 loans were subsequent loans of clients already in our study and hence dropped from the final analysis. This leaves a final sample of 943 loans.

11

III. Treatment Effects of Messaging on Loan Repayment

Table 3 presents simple means comparisons for five different measures of late loan

repayment. Stars indicate a pairwise significant difference between a treatment arm and the

control group. These comparisons preview one of our main regression results: only the positive

messages containing the account officer’s name reliably reduce delinquency relative to the

control group.13

We also estimate treatment effects using an ordinary least squares regression with the

following specification:

Yit = α + βTi + δXi + εit

where Y is a measure of late payment for loan (or, equivalently, client) i at time t. T is either an

indicator for whether messages were sent for this loan, or the complete vector of treatment

categories capturing assignment to one of the four content arms (loss or gain X client name or

account officer name) and to one of the three timing arms. In either setup the control group is the

omitted category for T. X is a vector of account officer and month-year (of the Y observation)

fixed effects. In specifications where we have multiple weekly observations per loan we cluster

the standard errors by loan to allow for correlation in payment activity for the same person.

Table 4 and 5 present results for three different measures of late repayment (Appendix Table

1 presents results for three other outcome measures, with similar results). Weekly Payment 13 We also use an alternative measure of late payment: whether the client missed a payment in the calendar week, defined as Sunday-Saturday. If loan payments are made late they are applied to the most outstanding installment first, so this alternative measure could capture whether a client is making regular payments even if they remain in arrears. Under this measure, no payment is made at all in 19% of weeks when a payment is due. The empirical results are robust to using this alternative measure of late payment (Appendix Table 1).

12

Received Late and More Than 7 Days Late are based on the timeliness of the required weekly

payments; for these outcomes the unit of observation is the loan-week, and we only include

weeks starting with the week a bank first reported a loan to us and we randomly assigned that

loan to treatment or control. The other outcome, Late 30 Days After Loan Maturity, is measured

at a single point in time, and hence the unit of observation is the loan14.

Table 4 Column 1, 4 and 7 present the estimated treatment effect of receiving any messaging

on each of the three outcomes. None of the point estimates is statistically significant, but all three

are negative (indicating reduced delinquency) and the confidence intervals do not rule out

economically meaningful effects (the control group means are reproduced near the bottom of the

table for reference/scaling). Appendix Table 2 groups the gain- and loss-frame messages to test

these relative to each other and to the control group. It shows a bit of evidence that loss-frame

messages are more effective at reducing late payments on average: the coefficient on loss-frame

is more negative than the coefficient on gain-frame for each of the three outcomes, and the p-

values on the difference between loss and gain frames are 0.09, 0.18, and 0.25.

Table 4, Panel A, Column 2, 5 and 8 present the results for the content treatments for each of

the three outcome measures, respectively. Neither of the client name messages (whether gain- or

loss-framed) produces statistically significant effects, and four of the six coefficients on these

variables are actually positive (indicating increased delinquency). Conversely, the account

officer name messages have negative point estimates in each of the six cases, with the positive- 14 Late is a dummy variable equal to one if the weekly payment was not received on the day it was due; more than 7 days late is a dummy equal to one if the weekly payment was not received by 7 days after the weekly payment due; Late 30 days after loan maturity is a dummy variable equal to one if the loan balance was not repaid was not fully repaid 30 days after the loan maturity date.

13

framed message in particular producing statistically significant reductions in each of the three

delinquency measures.

In Columns 2, 5, and 8 (but not the other columns) some adjustment for multiple-hypothesis

testing is in order: with 12 cells, one would expect to find about one significant result purely by

chance. A conservative way to do this would be to simply inflate each p-value by 12,

thereby reducing the likelihood that any individual treatment variable is statistically

significant by a factor of 12. This correction would move the p-values on the Positive frame-

account officer name from 0.025 to 0.3 in Column 2, from 0.091 to 1 in Column 5, and from

0.002 to 0.022 in Column 8. This correction is too conservative in our view because each of the

three outcomes is meant to measure the same thing: loan default as a proxy for bank profits. In

that sense we are closer to running a single regression with 4 treatment variables-- forced to

choose, the best proxy outcome is probably the books-closing measure in Column 8-- than we

are to running 3 separate regressions with 12 treatment variables.

Hence Appendix Table 3 aggregates our 3 outcomes into a single outcome-- a "summary

index" in statistical parlance-- and reruns the specification used in Table 4 Columns 2, 5, and 8.

Appendix Table 3 has three columns of its own in order to show results for three different

aggregation rules. We find results similar to Table 4: a statistically significant result on Positive

frame-account officer name in each case, and no other statistically significant results. The one

difference from Table 4 is that in Appendix Table 3 we find statistically significant differences

between positive- and negative-framed account officer messages.

14

Table 5 explores whether the content messages perform differently for new versus veteran

clients. There are competing hypotheses here. One on hand, new borrowers know less about

account officer monitoring than experienced borrowers, and hence new borrowers may infer

more from the account officer message than experienced borrowers about how closely account

officer’s will monitor the client. On the other hand, two theories suggest that the messages will

be more effective on prior borrowers. First, if heightening sentiments of reciprocity, the more

experienced borrowers may be more responsive to messages, as they have built up more of a

relationship with the account officer than the new borrowers. Second, past borrowers may see

these messages as a signal of an increased diligence to enforce repayment given that no such text

messages have been sent before, whereas new borrowers had no prior experience against which

to compare.

Table 5 shows that the account officer-signed text messages are effective only for the prior

borrowers. The difference between prior and first-time borrowers is significant for all three

dependent variables (p-values at the bottom of Table 5).

The evidence on whether framing matters for the account officer message is mixed. In the

full sample, (Table 4) only the positive-frame account officer message is statistically significant,

but the p-values on the difference between the loss vs. gain framed account officer messages are

0.22, 0.41, and 0.11 for our three main outcomes. In Table 5 the p-values between the loss vs

gain framed account officer messages are 0.23, 0.12, 0.04, 0.65, 0.87, and 0.91 for our three

main outcomes x the prior client and first-time client sub-samples. But, when we measure

delinquency using the summary index variables (Appendix Table 3), the positive-frame account

15

officer message does have a stronger effect than the negative frame (p-values of 0.05, 0.04, and

0.07).

Returning to Table 4, Panel B presents treatment effect estimates for the timing

treatments relative to control. Clients assigned to messages were sent them every week, on

either: the day the payment was due, the day before, or 2 days before. This timing treatment was

independently randomized after the initial randomization to either the control group or one of the

four message scripts. We do not find any statistically significant evidence that a specific timing

treatment is effective relative to control, or relatively effective compared to the other timing

treatments.

We then look whether treatment effects vary over the course of the loan by interacting the

number of weeks the loan is in the experiment (and receiving text messages, if assigned to

treatment) with our two key treatment variables. The account officer message effect becomes

slightly stronger over the course of the loan (coefficient -0.06); we find no evidence that the

overall effect of receiving messages changes over the loan. We show regression results in Table

6.

IV. Discussion and Conclusion

We present the results of a simple randomized trial sending text messages to individual

liability microfinance clients from two banks in the Philippines. We do not find an overall

treatment effect: the messages do not improve repayment on average. Nor do we find strong

evidence of (differential) impacts from loss versus gain frames or message timing. However,

16

messages that mention the loan officer’s name significantly, substantially, and robustly improve

repayment rates relative to messages that mention the client’s name and/or to the no-message

control group. This effect is only significant, and significantly stronger, for clients who enter the

sample having borrowed from the bank (and hence been serviced by the loan officer) before.

Conversations with bank management indicate that reductions of these magnitudes would

produce cost savings that greatly exceed the cost of messaging.

What explains the effectiveness of the account officer-signed messages? The fact that

repayment responds to messaging at all suggests the presence of some sort of elasticity of

repayment effort (broadly defined to include strategic default and perhaps project choice; see,

e.g., Karlan and Zinman (2009)) that is not captured by observable risk and that hence is not

priced on the margin. A more precise question is how the account officer message triggers

increased borrower repayment effort and thereby mitigates moral hazard. Prior work suggests

two possible channels. Work on relationship lending suggests that the account officer message

may signal increased intent to monitor the borrower (and note that screening or selection of

borrowers more likely to repay loans is not in play here, since loans are not assigned to treatment

until after origination). Work on social obligation and reciprocity suggests that the account

officer message may trigger better behavior from borrowers who have a relationship with the

account officer in the lay sense (not the asymmetric information sense): a personal or

professional relationship.

We highlight three key lessons for policy, with related directions for future research. First,

human interactions between credit officer and client can help to mitigate credit market

17

inefficiencies, and thus should be considered as informal and quasi-formal financial institutions

become more formalized, technology-driven and automated. Second, behavioral insights that can

be harnessed to change behavior in some contexts—e.g., loss versus gain framing—may not

work in others (a similar lesson comes from Bertrand et al. 2010). Some of this could be due to

difficult in defining precisely the intervention at hand: we interpret the treatment as providing

certain specific information, but naturally there could be alternative constructs. Theory and

replication, testing similar messaging with similar purposes but different wording, for example,

can help bolster the external validity (this is akin to external validity discussed in Lynch (1982;

1999), referred to as conceptual replicability).

Third, and closely related, successfully applying and scaling behavioral insights will require

systematic, randomized, and theory-driven testing. This will help generate the evidence base-- of

what works, what doesn’t, and why-- that is required to develop and apply behavioral insights

out-of-sample. This is important especially as we seek to build knowledge of the external

validity of these results in the sense put forward by Cook and Campbell (1979): under what

contexts, and in which environments, do these effects hold. In terms of messaging for behavior

change, a key next step is developing and testing theories of how different messages will (not)

work in different settings, ideally with large samples that permit sharper inferences on null

effects.

18

References

Adams, William, Liran Einav, and Jonathan Levin. 2009. “Liquidity Constraints and Imperfect Information in Subprime Lending.” American Economic Review 99 (1): 49–84.

Agarwal, Sumit, Brent Ambrose, Souphala Chomsisengphet, and Chunlin Liu. 2011. “The Role of Soft Information in a Dynamic Contract Setting: Evidence from the Home Equity Credit Market.” Journal of Money, Credit and Banking 43 (4): 633–54.

Aker, Jenny, and Isaac Mbiti. 2010. “Mobile Phones and Economic Development in Africa.” Journal of Economic Perspectives 24 (3): 207–32.

Alan, Sule, Mehmet Cemalcılar, Karlan, Dean, and Jonathan Zinman. 2015. “Unshrouding Effects on Demand for a Costly Add-on: Evidence from Bank Overdrafts in Turkey.”

Altmann, Steffen, and Christian Traxler. 2014. “Nudges at the Dentist.” European Economic Review 72 (November): 19–38.

Bertrand, Marianne, Dean Karlan, Sendhil Mullainathan, Eldar Shafir, and Jonathan Zinman. 2010. “What’s Advertising Content Worth? Evidence from a Consumer Credit Marketing Field Experiment.” Quarterly Journal of Economics 125 (1): 263–305.

Boot, Arnoud. 2000. “Relationship Banking: What Do We Know?” Journal of Financial Intermediation 9 (1): 7–25.

Cadena, Ximena, and Antoinette Schoar. 2011. “Remembering to Pay? Reminders vs. Financial Incentives for Loan Payments.” NBER Working Paper 17020, May.

Calzolari, Giacomo, and Mattia Nardotto. 2011. “Nudging with Information: A Randomized Field Experiment.” SSRN Scholarly Paper. Rochester, NY. http://papers.ssrn.com/abstract=1924901.

Charness, Gary, and Martin Dufwenberg. 2006. “Promises and Partnerships.” Econometrica 74 (6): 1579–1601.

Cialdini, Robert, and Noah Goldstein. 2004. “Social Influence: Compliance and Conformity.” Annual Review of Psychology 55: 591–621.

Cook, Thomas D., and Donald T. Campbell. 1979. Quasi-Experimentation: Design & Analysis Issues for Field Settings. Boston: Houghton Mifflin.

Donner, Jonathan. 2008. “Research Approaches to Mobile Use in the Developing World: A Review of the Literature.” The Information Society 24 (3): 140–59.

Jack, William, and Tavneet Suri. 2011. “Mobile Money: The Economics of M-PESA.” NBER Working Paper No. 16721, January.

Karlan, Dean, Margaret McConnell, Sendhil Mullainathan, and Jonathan Zinman. 2012. “Getting to the Top of Mind: How Reminders Increase Saving.” Working Paper.

Karlan, Dean, and Jonathan Zinman. 2009. “Observing Unobservables: Identifying Information Asymmetries with a Consumer Credit Field Experiment.” Econometrica 77 (6): 1993–2008.

———. 2011. “Microcredit in Theory and Practice: Using Randomized Credit Scoring for Impact Evaluation.” Science 332 (6035): 1278–84.

19

Kast, Felipe, Stephan Meier, and Dina Pomeranz. 2012. “Under-Savers Anonymous: Evidence on Self-Help Groups and Peer Pressure as a Savings Commitment Device.”

Lynch, John G. 1999. “Theory and External Validity.” Journal of the Academy of Marketing Science 27 (3): 367–76.

Lynch, John G., Jr. 1982. “On the External Validity of Experiments in Consumer Research.” Journal of Consumer Research 9 (3): 225–39.

Stango, Victor, and Jonathan Zinman. 2014. “Limited and Varying Consumer Attention: Evidence from Shocks to the Salience of Bank Overdraft Fees.” Review of Financial Studies 27 (4): 990–1030.

Positive From [aoname] of [bankname]: To have a good standing, pls pay your loan on time. If paid, pls ignore msg. Tnx

Negative From [aoname] of [bankname]: To avoid penalty pls pay your loan on time. If paid, pls ignore msg. Tnx.

Positive From [bankname]: [name], To have a good standing, pls pay your loan on time. If paid, pls ignore msg. Tnx.

Negative From [bankname]: [name], To avoid penalty pls pay your loan on time. If paid, pls ignore msg. Tnx.

Table 1: Wording of Text Messages

AO name

Client name

Log of loan size (peso)

Loan term (weeks)

Number of weeks in

experimentNumber of

loans

p-‐value of test for random

assignment(1) (2) (3) (4) (5)

Control (1) 9.7 19 10.6 310(0.0) (0.7) (0.3)

Treatment (2) 9.8 19.7 10.6 633 0.713(0.0) (0.5) (0.2)

Content TreatmentsPositive Framing & Client Name (3) 9.8 19 10.2 116 0.432

(0.1) (1.0) (0.4)Negative Framing & Client Name (4) 9.9 20.8 10.6 232 0.072

(0.1) (0.9) (0.4)Positive Framing & Account Officer

Name (5) 9.8 19 10.9 142 0.908(0.1) (1.0) (0.4)

Negative Framing & Account Officer Name (6) 9.8 19 10.6 143 0.584

(0.1) (0.8) (0.5)Timing Treatments

Sent Day of Payment (7) 9.8 20.1 10.6 214 0.719(0.1) (0.8) (0.4)

Sent 1 Day before Payment (8) 9.8 19.6 10.8 196 0.888(0.1) (0.8) (0.4)

Sent 2 Days before Payment (9) 9.8 19.3 10.4 223 0.755(0.1) (0.8) (0.3)

Table 2: Mean Baseline Loan Characteristics

Notes: Standard errors in parentheses. In column 5 each cell shows the p-‐value from the joint test of statistical significance (F-‐test) for an OLS regression of the treatment variable in that row on the three baseline loan characteristics in Columns 1-‐3. Each regression also includes controls for account officer and month-‐year fixed effects, which are not shown the table.

Proportion of weekly

payments made late


payments made more than 1 day late


payments made more than 7 days late

Any unpaid balance at maturity

Any unpaid balance 30 days past maturity

Number of loans

(1) (2) (3) (4) (5) (6)Control 0.288 0.253 0.156 0.235 0.135 310

(0.018) (0.018) (0.016) (0.024) (0.019)Any Treatment 0.295 0.251 0.152 0.216 0.098* 633

(0.013) (0.013) (0.011) (0.016) (0.012)Content TreatmentsPositive Framing & Client Name 0.302 0.27 0.175 0.241 0.121 116

(0.032) (0.031) (0.029) (0.040) (0.030)Negative Framing & Client Name 0.334 0.279 0.17 0.259 0.134 232

(0.022) (0.021) (0.019) (0.029) (0.022)Positive Framing & Account Officer Name 0.231* 0.190** 0.110* 0.134** 0.035*** 142

(0.026) (0.025) (0.022) (0.029) (0.016)Negative Framing & Account Officer Name 0.291 0.249 0.148 0.21 0.084 143

(0.027) (0.026) (0.024) (0.034) (0.023)Timing Treatments Sent Day of Payment 0.292 0.248 0.154 0.21 0.103 214

(0.023) (0.022) (0.019) (0.028) (0.021) Sent 1 Day before Payment 0.3 0.256 0.156 0.209 0.082* 196

(0.023) (0.023) (0.021) (0.029) (0.020) Sent 2 Days before Payment 0.293 0.249 0.148 0.229 0.108 223

(0.023) (0.021) (0.020) (0.028) (0.021)

Table 3: Outcome variables by treatment arm: Simple means comparisons

Notes: Standard errors in parentheses. Stars indicate a statistically significant difference between a treatment arm and the control group: * 90%, ** 95%, *** 99%.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Varying content of text messagesReceived any SMS (1) -‐0.007 0.024 -‐0.001 0.023 -‐0.026 -‐0.002

(0.021) (0.025) (0.018) (0.023) (0.023) (0.027)Received Message with Positive Framing and Client Name (2) 0.007 0.01 0.001

(0.036) (0.031) (0.035)Received Message with Negative Framing and Client Name (3) 0.032 0.029 -‐0.004

(0.029) (0.026) (0.031)Received Message with Positive Framing and Account Officer Name (4) -‐0.062** -‐0.040* -‐0.077***

(0.028) (0.024) (0.025)Received Message with Negative Framing and Account Officer Name (5) -‐0.024 -‐0.019 -‐0.03

(0.028) (0.023) (0.031)SMS signed by account officer -‐0.067*** -‐0.052** -‐0.051**

(0.025) (0.021) (0.024)p-‐value from t-‐test comparing rows (4) and (5) 0.222 0.41 0.105R-‐squared 0.13 0.134 0.133 0.157 0.161 0.161 0.17 0.176 0.174Panel B: Varying timing of text messagesReceived any SMS -‐0.007 0.024 -‐0.001 0.023 -‐0.027 -‐0.002

(0.021) (0.025) (0.018) (0.023) (0.023) (0.027)Received Message Day Payment Due -‐0.012 0.001 -‐0.013

(0.027) (0.024) (0.030)Received Message 1 Day Before Payment Due -‐0.007 -‐0.008 -‐0.04

(0.027) (0.023) (0.027)Received Message 2 Days Before Payment Due -‐0.002 0.002 -‐0.027

(0.027) (0.024) (0.028)SMS signed by account officer -‐0.067*** -‐0.052** -‐0.051**

(0.025) (0.021) (0.024)R-‐squared 0.13 0.13 0.133 0.157 0.157 0.161 0.17 0.171 0.174Mean of dependent variable for control group (both panels) 0.292 0.292 0.292 0.158 0.158 0.158 0.135 0.135 0.135Number of weekly repayments (both panels) 9990 9990 9990 9990 9990 9990 . . .Number of loans (both panels) 943 943 943 943 943 943 943 943 943N (both panels) 9990 9990 9990 9990 9990 9990 943 943 943

Weekly payment received lateWeekly payment received more

than 7 days lateAny unpaid balance 30 days

past maturity

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90%, ** 95%, *** 99%). Dependent variables are dummy variables for measures of late payment or outstanding balance at maturity. Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included in all regressions but coefficients not reported. All treatment effects estimated relative to no-‐message control group.

Table 4: Treatment Effect Estimates for Message, Content, and Timing: OLS Regressions

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Clients who are first-‐time borrowers from the bankReceived any SMS (1) 0.019 0.031 0.024 0.036 -‐0.013 -‐0.003

(0.031) (0.036)(0.028)(0.028) (0.033) (0.035) (0.040)Received Message with Positive Framing and Client Name (2) -‐0.049 -‐0.03 -‐0.018

(0.051) (0.041) (0.047)Received Message with Negative Framing and Client Name (3) 0.065 0.064* 0.003

(0.040) (0.037) (0.046)Received Message with Positive Framing and Account Officer Name (4) -‐0.029 -‐0.025 -‐0.080*

(0.045) (0.036) (0.046)Received Message with Negative Framing and Account Officer Name (5) 0.032 0.036 0.023

(0.042) (0.035) (0.048)SMS signed by account officer -‐0.028 -‐0.029 -‐0.024

(0.038) (0.031) (0.040)p-‐value from t-‐test comparing rows (4) and (5) 0.234 0.118 0.044R-‐squared 0.173 0.178 0.174 0.238 0.244 0.239 0.27 0.277 0.271Mean of dependent variable for control group 0.336 0.336 0.336 0.178 0.178 0.178 0.122 0.122 0.122Number of weekly repayments 4654 4654 4654 4654 4654 4654 . . .Number loans 476 476 476 476 476 476 476 476 476N 4654 4654 4654 4654 4654 4654 476 476 476Panel B: Clients who have borrowed from the bank beforeReceived any SMS (1) -‐0.027 0.019 -‐0.028 0.009 -‐0.056* -‐0.015


(0.046) (0.044) (0.053)Received Message with Negative Framing and Client Name (3) 0 0.001 -‐0.031

(0.037) (0.035) (0.048)Received Message with Positive Framing and Account Officer Name (4) -‐0.083** -‐0.064* -‐0.101***

(0.036) (0.033) (0.034)Received Message with Negative Framing and Account Officer Name (5) -‐0.066* -‐0.070** -‐0.097**

(0.034) (0.029) (0.042)SMS signed by account officer -‐0.095*** -‐0.076*** -‐0.085***

(0.031) (0.029) (0.032)p-‐value from t-‐test comparing rows (4) and (5) 0.651 0.87 0.913R-‐squared 0.14 0.148 0.147 0.13 0.139 0.139 0.175 0.188 0.187p-‐value "SMS signed by account officer" coefficient same across panels 0.007 0.077 0.099Mean of dependent variable for control group 0.248 0.248 0.248 0.137 0.137 0.137 0.151 0.151 0.151Number of weekly repayments 5336 5336 5336 5336 5336 5336 . . .Number of loans 467 467 467 467 467 467 467 467 467N 5336 5336 5336 5336 5336 5336 467 467 467

Table 5. First-‐Time Clients vs Pre-‐Prior Clients: OLS Regressions (same as Table 4 but splitting the sample)

Weekly payment received late Weekly payment received more than 7 days late

Any unpaid balance 30 days past maturity

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90%, ** 95%, *** 99%). Dependent variables are dummy variables for measures of late payment or outstanding balance at maturity. Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included. P-‐value is from F test that the coefficients are the same for first-‐time and pre-‐existing clients, computed by estimating a pooled model with an interaction term. Experience with the bank is measured as of the start of the experiment, because the experiment only covers one loan. All treatment effects estimated relative to no-‐message control group.

(1) (2) (4) (5)

Weeks in experiment 0.004 0.004 0.006** 0.006**(0.003) (0.003) (0.003) (0.003)

Received any SMS X Weeks in experiment -‐0.002 0.001 -‐0.002 0.001(0.003) (0.003) (0.003) (0.003)

Received message with account officer name X Weeks in experiment -‐0.006** -‐0.006**(0.003) (0.003

R-‐squared 0.131 0.133 0.161 0.164Mean of dependent variable for control group 0.292 0.292 0.158 0.158Number of weekly repayments 9990 9990 9990 9990Number loans 943 943 943 943N 9990 9990 9990 9990

Table 6. Treatment Effects Over the Course of the Loan: OLS

Weekly payment received late Weekly payment received more than 7 days late

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90%, ** 95%, *** 99%). Installments since loan start measures the length of time since the loan was disbursed. Weeks in experiment measures the length of time the loan was enrolled in our study and began receiving text messages if assigned to treatment. Messages start one or more weeks after loan disbursement due to lags in bank reporting to the research team, since the research team administered the random assignments. Dependent variables are dummy variables for measures of late payment or outstanding balance at maturity. Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included.All treatment effects estimated relative to no-‐message control group.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Varying content of text messagesReceived any SMS -‐0.006 0.023 0 0.023 -‐0.008 0.022

(0.020) (0.025) (0.016) (0.020) (0.029) (0.035)Received Message with Positive Framing and Client Name 0.017 0.016 0.043

(0.034) (0.027) (0.047)Received Message with Negative Framing and Client Name 0.026 0.026 0.011

(0.028) (0.022) (0.040)Received Message with Positive Framing and Account Officer Name -‐0.057** -‐0.040* -‐0.076**

(0.027) (0.022) (0.038)Received Message with Negative Framing and Account Officer Name -‐0.024 -‐0.011 -‐0.011

(0.026) (0.021) (0.041)SMS signed by account officer -‐0.063*** -‐0.048** -‐0.065*

(0.024) (0.019) (0.034)Panel B: Varying timing of text messagesReceived any SMS -‐0.006 0.023 0 0.023 -‐0.008 0.022

(0.020) (0.025) (0.016) (0.020) (0.029) (0.035)Received Message Day Payment Due -‐0.013 -‐0.005 -‐0.004


(0.026) (0.020) (0.038)Received Message 2 Days Before Payment Due 0.002 0.009 -‐0.009

(0.026) (0.021) (0.037)SMS signed by account officer -‐0.063*** -‐0.048** -‐0.065*

(0.024) (0.019) (0.034)Mean of dependent variable for control group 0.253 0.253 0.253 0.185 0.185 0.185 0.235 0.235 0.235Number of weekly repayments 9990 9990 9990 9990 9990 9990 . . .Number of loans 943 943 943 943 943 943 943 943 943N 9990 9990 9990 9990 9990 9990 943 943 943

Appendix Table 1: Replication of Table 4 with Three Additional Repayment Measures

-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐Weekly payment received late-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐-‐ -‐-‐Weekly payment received more than 7 days late-‐-‐ -‐-‐-‐Any unpaid balance 30 days past maturity-‐-‐-‐

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90%, ** 95%, *** 99%). Dependent variables are dummy variables for measures of late payment or outstanding balance at maturity. Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included. All treatment effects estimated relative to no-‐message control group.

(1) (2) (3)Weekly payment received late Weekly payment received more than 7 days late Any unpaid balance 30 days past maturity

Message had positive framing 0.01 0.01 -‐0.015(0.023) (0.021) (0.026)

Message had negative framing -‐0.032 -‐0.018 -‐0.042*(0.025) (0.022) (0.025)

p-‐value from F-‐test that coefficients are same for negative and positive framing 0.09 0.177 0.247Mean of dependent variable for control group 0.292 0.158 0.135Number of weekly repayments 9990 9990 .Number of loans 943 943 943N 9990 9990 943

Appendix Table 2: Loss vs. Gain Frames

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90\%, ** 95\%, *** 99\%). Dependent variables are dummy variables for measures of late payment or outstanding balance at maturity. Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included. All treatment effects estimated relative to no-‐message control group.

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Varying content of text messagesReceived any SMS (1) -‐0.011 0.047 -‐0.018 0.026 -‐0.018 0.018


(0.092) (0.063) (0.062)Received Message with Negative Framing and Client Name (3) 0.046 0.027 0.014

(0.077) (0.054) (0.052)Received Message with Positive Framing and Account Officer Name (4) -‐0.164** -‐0.130*** -‐0.111**

(0.069) (0.047) (0.045)Received Message with Negative Framing and Account Officer Name (5) 0.004 -‐0.014 -‐0.012

(0.080) (0.054) (0.054)SMS signed by account officer -‐0.125* -‐0.097** -‐0.079*

(0.065) (0.044) (0.043)p-‐value from t-‐test comparing rows (4) and (5) 0.045 0.037 0.069R-‐squared 0.215 0.221 0.218 0.197 0.204 0.202 0.194 0.199 0.197Panel B: Varying timing of text messagesReceived any SMS -‐0.011 0.047 -‐0.018 0.026 -‐0.018 0.018

(0.056) (0.067) (0.039) (0.047) (0.038) (0.046)Received Message Day Payment Due 0.004 -‐0.009 0.001


(0.071) (0.048) (0.047)Received Message 2 Days Before Payment Due -‐0.029 -‐0.027 -‐0.028

(0.073) (0.050) (0.049)SMS signed by account officer -‐0.125* -‐0.097** -‐0.079*

(0.065) (0.044) (0.043)R-‐squared 0.215 0.215 0.218 0.197 0.197 0.202 0.194 0.194 0.197Mean of dependent variable for control group (both panels) 0.58 0.58 0.58 0.424 0.424 0.424 0.292 0.292 0.292Correlation between index components 0.67 0.67 0.67 0.53 0.53 0.53 0.66 0.66 0.66Number of weekly repayments (both panels) . . . . . . . . .Number of loans (both panels) 943 943 943 943 943 943 943 943 943N (both panels) 943 943 943 943 943 943 943 943 943

Appendix Table 3: Replication of Table 4 with summary index measures of late payment

Index: Late, More 7 Days Late, Bal 30 Days Past Maturity

Index: More 7 Days Late, Bal 30 Days Past Maturity

Index: Late, Bal 30 Days Past Maturity

Notes: OLS regressions with standard errors clustered at the loan level. Stars indicate statistical significance (* 90%, ** 95%, *** 99%). Dependent variables are borrower-‐level indices of late payment. Col (1)-‐(3) is an index composed of the average of the three late payments: mean number of payments late, the borrower-‐level mean number of payments more than 7 days late, and any unpaid balance 30 days past maturity. Each component has a maximum value of 1, so the overall index takes a value between 0 (no late payment of any kind at any point in the loan) to 3 (every weekly payment more than 30 days late (and hence also every payment received late), with balance unpaid 30 days past maturity). Col (4)-‐(6) constructs an index with two measures of late payment (more than 7 days late and balance more than 30 days past maturity), and Cols (7)-‐(9) constructs an index with two measures of late payment (late payment and more than 30 days past maturity). Controls for account officer, month-‐year fixed effects, and baseline loan characteristics included in all regressions but coefficients not reported. All treatment effects estimated relative to no-‐message control group.

A Personal Touch: Text Messaging for Loan Repaymentjzinman/Papers/RepaymentReminders.pdf · We...

Documents

Transcript of A Personal Touch: Text Messaging for Loan Repaymentjzinman/Papers/RepaymentReminders.pdf · We...