Text Messaging for Financial Behavior Change in Zambia

This experiment employed six different interventions among three different subsamples of the population: active savers, inactive savers, and borrowers. In addition to looking at the individual treatment arms, we will appropriately pool experimental arms to examine the impact of one-way versus two way treatment effects. The treatments are outlined below:

1) Inactive 2-Way Engagement (inactive savers): received interactive (two-way) Juntos customer engagement service designed specifically to engage/activate inactive account holders.
2) Active 2-Way Engagement (active savers): received interactive (two-way) Juntos customer engagement service designed specifically to increase savings behavior.
3) Salience 1-Way Reminders (active savers): received simple reminders that encourage participants to save more and contain no further information.
4) Planning 1-Way (active savers): received messages reminding participants to make plans to save and to keep track of their savings progress and give them tips on how to do so.
5) Planning 2-Way (active savers): messaging with identical content to the “Planning 1-Way” messaging, but adding the potential for customers to reply to messages, either when prompted or unprompted, to ask questions to a representative from Juntos, who helped advise them on goal setting and tracking strategies.
6) Loan 2-Way (borrowers): received interactive (two-way) messaging that stressed the importance of repaying loans on time for people’s future financial wellbeing.
7) Control groups: three control groups (one for Borrowers, one for Active Savers, one for Inactive Savers) did not receive messaging.

2018-11-01

2019-09-30

Administrative Data (Full Sample)
1) Account balance at 0 and 6 months post experiment (with controls for baseline balance (Twisk, et al., 2018) and with and without winsorization at 99%).
2) Did balance increase by 20% or more? [with/without an adjustment for zero to low balance changes following Ashraf et al. (2006)].
3) Number of deposits - during experiment and 6 months post experiment.

Loans
1) Proportion non-performing (60 days late +) - as at 0 and 6 months post intervention.
2) Days late - as at 0 and 6 months post experiment

Survey Data (Subsample)
1) Financial literacy index
2) Savings attitude index
3) Savings planning index
4) Loan literacy index

Post intervention administrative data was received in 2022.

Administrative Data (Full Sample)
Savings
1) Total amount deposited – during experiment and 6 months post experiment (with and without winsorization at 99%).
2) Proportion of accounts active – as at 0 and 6 months post intervention.

Loans
1) Proportion of loans considered a loss ( 180 days late +)
2) Was the loan repaid early (for loans due within data range)?
3) Number of loans

Survey Data (Subsample)
1) Formal and informal savings/ borrowing increased in last six months
2) Do borrowers report making a late payment?
3) Do borrowers report paying early/more on a formal or informal loan?
4) Savings-related behavioral changes: risk aversion, impulsivity, time preferences (exploratory)

For all experiments (both savings and loans), average treatment effects by demographic subgroup (age, gender, and region) are also included as secondary analyses. We also plan to use machine learning tools to conduct exploratory analysis on heterogeneity of treatment effects, following Athey and Imbens (2015).

We will correct for multiple hypothesis testing using False Discovery Rate (FDR) correction. Following Benjamini et al. (2006) and Anderson (2008), we will report sharpened q values over the set of p values associated, grouping hypotheses, primary outcomes and secondary outcomes in sets and perform adjustments along those sets.

Details of Experimental Design
The sample was first divided into three subsamples: Inactive Savers, Active Savers, and Active Borrowers. Inactive savers (participants having zero or negative balance for four months prior to the start of the experiment) were randomly allocated to either a control group, which did not receive any text messages, or the “Juntos” group, which received the interactive (two-way) Juntos text messages designed to engage inactive account holders. Active Savers were allocated to either a control group, which did not receive any text messages; a treatment receiving the interactive Juntos text messages over eight months; or one of three researcher-designed treatments receiving behaviorally “themed” text message campaigns over six months. For the Active Borrowers (participants with active loans with the bank) a control group received no messages while a treatment group received interactive messages that stressed the importance of repaying loans on time for people’s future financial wellbeing.

Randomization Method: Stratified randomization by user type subgroup, supervised by researchers but completed and implemented by Juntos Global.
First, a valid sample of 78,895 bank clients were ascertained (clients with valid phone numbers, savings/loan accounts). Inactive clients were deemed as zero or negative balance four months before the setup of experiments. Randomisaiton was conducted by Juntos Global and overseen by the researchers. The random allocation was performed as follows:
A) Active savers: Blocking by gender, group age, and balance quartile.
B) Inactive: Blocking by gender and age group.
C) Loan accounts: Blocked by gender, loan status (pass or not), quartile amount of the loan.

Randomization Unit - Individual-level randomization (customer)

No

N/A

Sample Size Sample size: 78,589 Planned Number of Clusters: NA Planned Number of Observations: Loans observations: 6.9 million Transactions observations: 2.9 million Sample Size by treatment arms: Control Juntos Inactive: 14730 Treatment Juntos Inactive: 14728 Control Juntos Active: 10231 Treatment Juntos Active: 10214 Control Themed Weighted: 3849 Saving salience two-way: 7060 Saving planning one-way: 7063 Saving salience one-way:7067 Control Loan: 1135 Treatment Loan: 1140

N/A