Estimating the effects of dynamic repayment incentives on loan performance

How to cite

Gupta, Apoorv; Ghosh, Pulak; Mezzanotti, Filippo; Zinman, Jonathan. 2025. "Estimating the effects of dynamic repayment incentives on loan performance." AEA RCT Registry. AEARCTR-0016541. https://doi.org/10.1257/rct.16541-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Apoorv Gupta* , Pulak Ghosh , Filippo Mezzanotti , Jonathan Zinman

* Primary investigator

Registry ID
AEARCTR-0016541
DOI
https://doi.org/10.1257/rct.16541-1.0
First registered
Last updated
Status
On going
Trial dates
2024-09-20 – 2025-08-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16541

Abstract

We plan to identify effects of randomly assigned dynamic repayment incentives to borrowers on the performance of the borrowers on their loans. The messages will sent out by a large digital fintech lender in India to well-performing loans. The lender will randomly assign new loans in its portfolio to either receive a treatment (telling them they will automatically qualify to a receive a loan amount that is 3 times of the original loan upon timely repayment) or a control ( telling them they could qualify to a receive a loan amount that is 1.1 times of the original loan upon timely repayment). Estimates will help shed light on the importance of dynamic incentives (i.e. receiving more credit in the future) on borrowers’ repayment behavior.

Intervention

The experiment will include 50% randomly chosen consumers (“borrowers”) who have borrowed from the lender within the 30 days prior to the date of random assignment.. The lender provides loans for maturity of 3 months, 6 months, 9 months, 12 months and 18 months.

The company plans to experiment with randomized settlement offers on about 1,500 borrowers with newly originated loans on a monthly basis. The sample will include borrowers who took out a new loan with the lender in the last 30 days. Only borrowers with maturity of 3-12 months are considered. The lender will randomly provide half of borrowers in this sample a message informing them that they will automatically qualify for a loan that is 3 times the size of their current loan amount upon timely repayment of their loan.
Borrowers who remain current through their loan term in any month will receive the same message every month during the intervention period per their initially assigned treatment group. That is, the borrower’s treatment assignment is unchanged once assigned. Borrowers that ever become delinquent on their loan stop receiving any message after becoming delinquent.

The lender will communicate offers, per its standard practices, with direct email and text messages that include instructions to contact the company for more information. Offer communications will be harmonized—in frequency, timing, and content—with the routine communications sent to the control group, to the fullest extent possible.

Experimental design

For each borrower in our sample, the lender will communicate treatment messages to half of the sample and control message to the other remaining half of the sample. We worked with the lender to stratify the randomization across four groups. These 4 groups are created by dividing the sample across two variables: (a) 2 bins reflecting the maturity of the underlying loan product; and (b) whether the borrower is a new customer with the company as defined in the month they first become part of the experiment sample (i.e. if the customer is a first-time borrower with the lender versus the customer has taken a loan from the lender in the past).

We will of course use the conditionally assigned discount for any intent-to-treat (ITT) analysis using the following specifications:

yit = a + b1 * Tit + sit + uit

where Tit is takes the value of 1 if borrower i in randomization month t. Control group is the omitted category. sit represent strata x experiment month fixed effect, where strata is one of the 4 groups defined by combination of product maturity and “new to the lender”. We will also analyze heterogeneous treatment effects based on the strata described above.

Design details

Randomization method
Randomization done in office by a computer.
Randomization unit
The unit of randomization is the loan account (borrower) every month.
Clusters
We plan to run the experiment on the new borrowers in the 6 months between September 2024 to Feb 2025. Borrowers who remain current through their loan term in any month will receive the same message during the intervention period message every month per their initially assigned treatment group.
Observations
We expect to have about 1500 new to the lender borrowers per month.
Arms
We split the sample equally across the one treatment and one control arm (i.e. 50% in each of the treatment and control arms).
Primary outcomes
delinquency status and whether the borrower is current
Secondary outcomes
(i) total repayment to the lender made by the borrower on monthly basis. We will look at both the levels of the variable, and the version where this variable is scaled by the total amount owed. We will also examine longer horizons (one month and potentially more), to the best extent possible. (ii) 1(any payment): whether the borrower made any payment to the lender.

Cited by

No citation data available for this registration yet.

Keywords