Switching Markets: A Field Experiment in Retail Digital Finance in Tanzania

How to cite

Blackmon, William; Annan, Francis; Gine, Xavier; Mwesigwa, Brian; Garz, Seth. 2022. "Switching Markets: A Field Experiment in Retail Digital Finance in Tanzania." AEA RCT Registry. AEARCTR-0010290. https://doi.org/10.1257/rct.10290-1.0

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

William Blackmon* , Francis Annan , Xavier Gine , Brian Mwesigwa , Seth Garz

* Primary investigator

Registry ID
AEARCTR-0010290
DOI
https://doi.org/10.1257/rct.10290-1.0
First registered
Last updated
Status
In development
Trial dates
2022-11-01 – 2023-12-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/10290

Abstract

In many contexts, consumers often fail to switch to more favorable options.[1] A leading an important example is the market for digital financial services (DFS) -- particularly mobile money -- where users rely on agent to complete transactions such as withdrawing cash from their digital wallets. We seek to address two potential inefficiencies related to mobile money agents. First, these agents tend to be concentrated geographically and offer the same basic services, so consumers face a choice of which agent to use. Evidence from mystery shopping shows there is significant variation in the fees these agents charge and the reliability of the services they offer. Consumers do not always use the lowest cost or most reliable agent, suggesting that either other factors are more important to consumers, or that consumers are unaware of agents' cost and reliability. Second, presumably because of a lack of digital literacy, consumers often use agents to complete transactions that they should be able to conduct on their own, such as making transfers or checking their balances.

In this experiment, we seek to address these learning inefficiencies by encouraging consumers to make additional transactions with mobile money agents. To address the first challenge, we encouraging consumers to "shop around," visiting agents they don't regularly use, and explore whether this leads consumers to learn which agents are best in their area, change the agents they use, and ultimately incur lower costs and higher reliability. To address the second challenge, we explore whether making additional transactions with any agent (either their regular agents or new ones) leads consumers to rely less on the assistance of agents and to conduct transactions on their own.

FOOTNOTE:
[1] Lack of consumer switch can be found in markets for energy (Hortacsu et al. 2017; Ito et al. 2017; Office of Gas and Electricity Markets 2019), health insurance (Handel 2013; Polyakova, 2016), credit cards (Stango and Zinman 2015; Galenianos and Gavazza 2020), paid TV (Shcherbakov 2016), mobile phone services (Shy 2002), auto insurance (Kiss 2019), and mortgages (Keys et al. 2016; Andersen et al., 2020), and even attempts to lower switching costs do not result in high switching rates (Office of Gas and Electricity Markets 2020).

Intervention

We will encourage some study participants to complete financial transactions with mobile money agents.

This study leverages a separate descriptive study that involves paying study participants to make "mystery shopping" visits with mobile money agents as a means of assessing the fees these agents charge and other non-monetary costs consumers incur (such as failed transactions).

These mystery shopping visits involve attempting to conduct one of four real transactions: (1) depositing money into a mobile money wallet, (2) withdrawing funds from a wallet, (3) sending money from their wallet to the wallet of another user using a different mobile money provider, or (4) sending cash to the wallet of another user.

The value of the transaction will be either TZS 10,000 or TZS 20,000. Agents will be randomly assigned to one of these transaction values (stratified by market), and all mystery shopping visits with a given agent will be of the value assigned to that agent.

Our study randomizes whether or not respondents are assigned to these mystery shopping visits, and if so, which agents they are assigned to make mystery shopping visits with.

Experimental design

We will include ten public markets where mobile money agents tend to be clustered, five of which will be in Dar es Salaam, Tanzania (the capital and largest city) and five of which will be in Mwanza, Tanzania (the second-largest city). Individual markets will be selected by field team managers.

Within each market, we will conduct a census of agents and include the 20 agents closest to the central point assigned to each market in our study. Of these 20 agents, we will select 10 to be included as recruitment points for consumers who will serve as study participants, through a simple random selection. All mystery shopping visits included in this study will be with this subset of 10 agents per market (100 agents total). We will recruit four consumers from each agent through intercept surveys. Consumers will need to meet the following requirements to be eligible to be included in the study:
1. We will recruit two female and two male consumers from each agent location.
2. Consumers must own a smartphone and pass a basic digital literacy test (to ensure they are able to complete our self-administered, WhatsApp-based surveys independently)
3. Consumers must regularly use the agent they were recruited at.
4. Consumers must have a mobile money account with the provider that the agent they were recruited at was assigned to for mystery shopping visits. (Mobile money agents may serve multiple mobile money providers. As part of the descriptive study being conducted alongside this RCT, agents that serve multiple providers are randomly assigned to one of the providers they serve for all mystery shopping visits, including mystery shopping visits conducted by local consumers).

Consumers will be assigned to one of four treatment arms. Assignment will be at the individual consumer level, stratified by the agent at which consumers were recruited (so each of the four consumers recruited at a given agent will be assigned to a different treatment arm). Treatment assignments will be carried out such that consumers assigned to Treatment Arm 1 and Treatment Arm 2 are different genders and similarly consumers assigned to Treatment Arm 3 and Treatment Arm 4 are different genders.

Treatment Arms 3 and 4 refer to "high quality" and "low quality" agents, respectively. These designations are defined as follows. Based on data from mystery shopping visits conducted by enumerators prior to the baseline of this study, agents will be assigned a score based on their likelihood of successfully completing transactions and fees they charge for completing transactions (conditional on successfully completing the transaction). Higher success rate and lower fees will yield higher scores, and success rate and fees will be given equal weight in generating these scores. Of the ten agents included in this study per market, the five with the lowest scores will be considered "low quality" agents and the remaining five will be considered "high quality" agents.

The treatment arms are as follows:

- Treatment Arm 1 ("Control"): These consumers will not be assigned to any mystery shopping visits

- Treatment Arm 2 ("Regular agents only"): These consumers will be assigned to mystery shopping visits with agents that they use regularly (at least twice in the past 90 days). All consumers assigned to this treatment arm will conduct four visits with the agent they were recruited at. If consumers regularly use more than one agent in the market, they will be randomly assigned to another agent that they use regularly, and will conduct an additional 4 visits with this agent. Consumers will conduct all four of the transaction types discussed in the "Intervention" section with each agent.

- Treatment Arm 3 ("Regular and 'high quality' agents"): These consumers will be assigned to eight mystery shopping visits. Four of these visits will be with the agent they were recruited from. The remaining four visits will be with four different agents, all of which the respondents has not used regularly. Additionally these remaining four visits will be with agents designed as "high quality" based on the above discussion. Finally, these visits will only be with agents that have been assigned to a provider that that consumer has an account with. The four visits with regular agents will include all four types of transactions. The four visits with non-regular agents will also include all four types of transactions, with transaction types being randomly assigned to each agent.

- Treatment Arm 4 ("Regular and 'low quality' agents"): These consumers will be assigned to eight mystery shopping visits. Four of these visits will be with the agent they were recruited from. The remaining four visits will be with four different agents, all of which the respondents has not used regularly. Additionally these remaining four visits will be with agents designed as "low quality" based on the above discussion. Finally, these visits will only be with agents that have been assigned to a provider that that consumer has an account with. The four visits with regular agents will include all four types of transactions. The four visits with non-regular agents will also include all four types of transactions, with transaction types being randomly assigned to each agent.

We have four primary research questions:

Research question 1: Why (or why not) do customers actively shop around for the agent who offers the lowest price or the most reliable service? Are they aware of the distribution of pricing and reliability of agents in the market the frequent? Do they prioritize price, reliability, or other factors (such as proximity or social relationships) when deciding which agents to use?
Research question 1 is descriptive and will be addressed through analysis of data collected from consumers at baseline.

Research question 2: Does exposure to additional agent interactions lead to learning by consumers about how to conduct transfers on their own, without agent assistance?
Research question 2 will be addressed by comparing outcomes measured in the follow-up phone surveys regarding types of transactions conducted and the mode used to conduct these transactions between Treatment Arm 1 and the pooled group Treatment Arms 2-4.

Research question 3: Does exposure to new agents that consumers do not regularly use help consumers identify agents who charge lower fees and/or are more reliable? Does this lead consumers to permanently switch which agents they use to conduct transactions? Do they ultimately experience lower costs and higher reliability?
Research question 3 will be addressed by comparing outcomes measured in the follow-up phone surveys regarding the agents consumers use, the fees they incur, and the success rate of those transactions between Treatment Arm 2 and the pooled group Treatment Arms 3-4.

Research question 4: Does exposure to better agents lead consumers to permanently switch agents at higher rates than exposure to worse agents?
Research question 4 will be addressed by comparing outcomes measured in the follow-up phone surveys regarding the agents consumers use between Treatment Arm 3 and Treatment Arm 4.

Design details

Randomization method
Randomization will be done by a computer using the statistical software Stata.
Randomization unit
Randomization will be at the individual consumer level. Randomization will be stratified by the agent at which participants are recruited from.
Clusters
400 consumers (treatment is assigned at the unit of observation, no clustering.)
Observations
400 consumers
Arms
Treatment Arm 1 ("Control"): 100 consumers Treatment Arm 2 ("Regular agents only"): 100 consumers Treatment Arm 3 ("Regular and 'high quality' agents"): 100 consumers Treatment Arm 4 ("Regular and 'low quality' agents"): 100 consumers
Primary outcomes
We have five primary outcomes of interest: 1. The types of transactions participants choose to conduct and the frequency with which these transactions are made. 2. The mode through which participants choose to conduct mobile money transfers. Typically these transactions will either be carried out independently by participants on their own device, or conducted with assistance from a mobile money agent. 3. The agents that participants choose to use when making mobile money transactions. We will measure if participants use agents different from the agents they were using regularly at baseline, and if so, if these new agents offer better services than the agents they used at baseline in terms of the prices they charge and their reliability (likelihood of successfully completing a transaction). This information will be based on a separate mystery shopping activity conducted prior to baseline by trained enumerators (separate from the mystery shopping visits carried out by study participants). 4. The prices participants pay to complete transactions with mobile money agents and the reliability of service they receive (the likelihood that attempted transactions with agents are successful). 5. Consumers' beliefs (incentivized) about which agents offer comparatively better or worse quality of service in their local market.

Cited by

No citation data available for this registration yet.

Keywords