Financial Inclusion for the Rural Poor

Researchers are conducting a randomized evaluation to explore the impact of allowing JUNTOS beneficiaries to collect their payments though branchless banking agents. In the branchless banking system, local bank agents, typically shopkeepers, serve as deposit and withdrawal points for account holders to access their funds with debit cards. The agent based network, which the project is installing in shops and municipalities will allow the national bank to increase the number of withdrawal points for JUNTOS users, reducing transportation costs and potentially giving users a greater degree of access to their accounts. If this is the case, users may begin to use their account to save more of their JUNTOS payments, making smaller and more frequent withdrawals.

In order to evaluate the effect of branchless banking, a sample of 59 sub-regional districts, each with approximately 400 JUNTOS beneficiaries, will be randomly assigned to one of two groups.

In the first group, branchless banking agents will be established in each district, allowing beneficiaries to access and withdraw funds from their JUNTOS accounts.The second group will serve as a comparison group, where branchless banking agents will not be introduced for the duration of the evaluation period.

Within the 30 treatment districts, 140 villages or "centros poblados" will be randomly assigned into one of two subgroups to evaluate a workshop aimed at increasing trust and empowerment of users in the financial system. 70 villages will receive the workshop and 70 will act as controls.

One year after banking agents are introduced, the researchers will collect information on savings and consumption behavior from household surveys. The study will also incorporate administrative account usage data from Banco de la Nación and the JUNTOS program to examine how beneficiaries use their accounts when they can access them through branchless banking agents.

2014-11-01

2015-01-01

Banco de la Nación administrative data will be used to test hypotheses concerning savings balances and use of savings accounts. The data are available for the universe of JUNTOS users and will be measured in monthly observations for the 12-month intervention period and for the 12 months prior to the intervention. The specific data items on JUNTOS users include:
• Savings account balance
• Number and volume of withdrawals
• Number and volume of transfers made and received
• Number and volume of withdrawals made through agents
• Number and volume of other transactions (balance checks, transfers) made through agents
• Number and volume of all the latter transactions made through agents in comparison to those made at bank branches and ATMs

Household survey collected on a random sub-sample of households 18 months after the start of the intervention. Data from this survey will be used to test hypotheses about the impact on impulse expenditures, savings and investment patterns. The types of items that will be collected include.
• Consumption module: household expenditures on education, health, durables, productive assets, food and soft drinks/tobacco/alcohol (using list randomization to validate response to sensitive questions)
• Savings module: informal savings, formal savings

1. A year long intervention from Sept 2014-Sept 2015 with
a) Installation of Branchless Banking Agents in 30 randomly selected districts in the Southern Sierra of Peru
b) A trust intervention workshop provided to approximately 70 randomly selected villages within these 30 districts


2. Endline survey conducted in October 2015.

3. A final report for the Ministry of Development and Social Inclusion, JUNTOS Programme, State Bank of Peru, Institute of Peruvian Studies, World Bank (CGAP) and the Gates Foundation

We anticipate the intervention itself to bring about the following outcomes

1. Lowering the transaction costs for users via the installation of the Branchless Banking agents closer to the users´ homes will increase the use of the account
2. This effect on the increased use of the accounts will be increased in the case of villages receiving financial education treatment
3. The increased use of the account will also be reflected in increased levels of savings which will in turn affect consumption patterns in favour of spending on more durable goods

Finally, a key outcome of the evaluation is to be able to make policy recommendations on the efficacy of the use of both the Branchless Banking Agents and the kind of messaging in financial education that is most effective to reach users of the government cash transfer programme on a national level. This therefore has implications for some 700,000 JUNTOS beneficiaries directly, and 1,000,000 individuals to include their families.

1. RCT Design: Justification & Explication

The primary motivation for using an RCT to select those receiving the intervention and those not, is to eliminate selection bias, thereby increasing the precision with which we can measure the precision of the intervention. When offering a new programme to all beneficiaries at the same time, it is difficult to know whether any "impacts" - positive or negative - are attributable to the intervention itself (in our case the installation of the Branchless Banking Agent, or the introduction of Training workshops) or due to other reasons, that may or may not be related to the programme. The use of a rigorously designed experiment with randomized selection of districts and villages that receive the two levels of treatment enable us to attribute whatever change our outcome indicators to the intervention directly.

Despite the random assignment of treatment and control groups, it is possible that differences may arise between the two groups. In order to create statistically equal groups, we use control variables to identify the most precise impacts of the treatment.

5.2 The design of the experiment

We shall carry out an experimental evaluation in 59 districts across 3 departments in the sierra of Peru (Puno, Cusco, Apurimac and Ayacucho).

First of all, to select the districts to be randomized, we first needed to establish (1) The type of JUNTOS district in which we want to focus the intervention and (2) Which districts have adequate conditions for the installation of the agents.

Thanks to preliminary research realised in Phase 1 of the research, we have concluded that the most important factors for the adequate functioning of the agents are (a) sufficient data connection of one of the two main network providers (b) sufficient liquidity to manage the JUNTOS payout. In order to establish which districts had these characteristics, we conducted a pre-selection of districts to verify signal and survey potential agents to determine the number of users that each could manage. Only those districts which fulfilled this criteria in the Southern Sierra region were then pre-selected to be randomized. Following this, we also took out districts which had less than 50 users who would be receiving their payout in the bank branch of another district. None of the districts in the sample had (a) a bank branch of their own or (b) an agent or (c) only beneficiaries who received from a transporter. Finally, we took out districts that were located in provinces which had more than 60% of districts with agents installed in order to diminish the possibility of contamination via agent installation in control districts.

Only when we had the list of eligible districts did we then randomize treatment and control. Randomization consisted of two stages.

1.1 Phase 1 Randomization - District Level

The first level of randomization is to offer the installation of the agents, at district level. The sample consists of 59 districts - 30 treatment districts and 29 controls, with approximately 300 beneficiary households per district.

We ensure balance between the two groups of districts on the basis of administrative data on the average distance to district capitals from villages of each district. Furthermore, we balance the number of districts that receive via branch and transporter across treatment and control. Apart from this, we ensure that socio-economic characteristics, such as literacy rates among JUNTOS users, are balanced between the two groups.

In each treatment district, we coordinate with the JUNTOS programme to re-enforce the new, official message of the programme (Resolución N° 044—2013-MIDIS, abril 2013) that beneficiaries may use their accounts without limitations and can take out or leave money in their accounts without risk of suspension from the programme. This message shall be conveyed across treatment and control groups, at the same time as installation is taking place in treatment.

Following this the treatment districts will be assigned to have promoters of the Branchless Banking Agent product, known as the Agente MultiRed, who will go to the district capitals to promote the product in municipalities and shops, that have been identified as having sufficient cash in the pre-screening.

2.2. Stage 2 of randomization: village level

In stage two of the randomization, inside the districts that receive agents, we shall pilot a trust intervention in 70 villages. Each financial education treatment will consist of a workshop of three hours with reminder SMS messages being sent afterwards to re-inforce the message provided in the workshop.

(1) The first treatment will emphasize the benefits of savings via a financial education that focuses on the reasons and justifications for saving.

(2) The second treatment is a second type of financial education, focused on the improvement of the level of trust that users have in the financial system. This second workshop will put more emphasis on creating a culture of use of the technology, by enabling beneficiaries to see the examples of their community leaders using the technology, and having the regulatory system explained to them.

(3) 1 and 2 combined to see the effects of synergies

(4) A control, where beneficiaries will only receive information about the existence of the agent and the mechanics of their use.

We will stratify the second phase of randomization in order to ensure balance between villages, using administrative data provided by BN, SISFOH and JUNTOS. Among the variables here are the distances from villages to district capitals (where agents will be installed) so as to test whether effects will be greater on villages that have a greater time saved in getting to their paypoints. Stratification will assure sufficient power and balance between these sub-populations. BN will grant us data on the universe of JUNTOS beneficiaries, which shall be merged in the analysis with socio-economic information on the same users.

1.3 Control Groups

One of the requirements for the rigourosity of the evaluation is to guarantee that the control groups are not contaminated. For control districts at the first level of randomization, these districts will not have Branchless Banking Agents installed during the period of the intervention. At the second level of randomization, those villages in control will not be provided with any financial education beyond technical training on how to use the machine in the shops themselves.


We work to assure that there is no contamination in the first phase via close communication with BN in Lima and the intervention zone, so as tu ensure that the evaluation does not clash with their expansion plan in the districts which are included in the impact evaluation.

In order to avoid contamination in the second treatment, we make sure that only users in the villages assigned to receive the workshops receive them, via close coordination with local JUNTOS facilitators at a local level, who know where each user comes from. In addition, we ensure good communication with the implementors of financial education - the Institute of Peruvian Studies. IPA will continue to monitor the implementation of the program during the intervention period. At the start of the workshops, IPA will ask users for some data on their friends in other villages. At the endline, IPA will encorporate questions on the knowledge of beneficiaries regarding the content of other treatments to see if there was spill over of knowledge from one to another.

Randomization done in office using STATA with stratification on the travel time to current payment point of JUNTOS users.

There are two levels of randomization. The first randomization will be made on district unit and within the districts of the treatment group a second randomization will be made on villages or "centros poblados" unit.

Yes

The cluster contains 59 districts.
For village level randomization we will randomize to include 4000 users across the three treatments and one control. This will be around 220 villages in total.

About 22,500 households

30 districts treatment, 29 districts control.

[This may need changing] The survey sample will be 6,000 surveys across the three treatment groups, with 2000 surveys per arm. We will have two outcome variables for which we rely on results from survey data. 1. The first is increase in total savings (i.e. bank savings and reported savings in other areas). At district level (i.e. for a comparison between those who received agents and those who did not) we need to see an increase of S./37.5, assuming a control group mean of S./59.5 and a standard deviation of S/.320.03, power of 80% and intracluster correlation of 0.01. At village level (i.e. for a comparison of those who received the training intervention on top of the agent intervention and those who received solely the agent intervention), we need to see an increase of S/.36.5 assuming the same means, SD and power as above but with an ICC of 0.02. 2. The second is we need to see an increase in the spending on durable goods. At district level we need to see an increase of S/.61.08 in spending on durables within a year, assuming a control group mean of S/.272.92, a standard deviation of S/.523.73 and an ICC of 0.01. At village level we expect to see an effect of S./59.08 when there is an ICC of 0.02, but with SD, power and control group means the same as above. For budgeting purposes, we assume that only 90% of attempted surveys will be successfully administered, and we will therefore attempt a total of 6,600 surveys. Some 220 per district.