Experimental Design Details
Our final sample includes $2606$ participants form all the $32$ states in Mexico. Participants have been randomly selected among BBVA Bancomer adult customers (between 17 and 64 year old) with an income equal or lower than USD 400 per month and with a cell phone. Then, they are randomly assign to be part of our treatment (1268 participants) or control group (1338 participants). The length of the experiment was eight weeks (May and June, 2015). Our intervention consists of three blocks, salutation and disclaimer, saving preferences and reminders.
The first and second blocks contain a total of 5 SMSs that receive all participants only once, at the first day of the experiment (May 18th). The first block, SMS 1 to SMS 3, are salutation and disclaimer messages from them to be aware that BBVA Bancomer might communicate with them via SMSs in order to give them some tips for saving and inform them that their data will be protected. The second block, which is sent the day after the first one, includes a short baseline survey with two question (SMS 4A and SMS 4B) asking about their saving habits and their preferences for savings. Questions in the second block are important to capture unobserved heterogeneity between participants and non-participants. In addition, it is useful to control for unobserved effects of our treatment coming from informal ways of saving and to infer whether our treatment have different impact depending pre-treatment saving preferences of individuals.
The third block, which contains three SMSs (SMS 5 to SMS 7), corresponds to our treatment and it is sent just to the treatment group weekly (every Friday between 8am and 9am). Out target group are low income individuals in Mexico who mostly get their salaries weekly, and in cash, every Friday so, we convey to follow this timing. We send the three SMS reminders coinciding with the day that individuals are paid by employers. Since memory and present bias affect the use of reminder systems (Ericson, 2017), ideally, if a treatment is intended to increase savings, individuals should get an automatic deduction for the amount of money that they want to save that will be sent to a different bank account as soon as they get the salary in order to avoid monetary illusion. However, this way of automatize tasks requires at least two necessary conditions, getting wages through a bank account and having smooth income flows over the year (similar and known monthly wages). These conditions do not often hold when it comes to low income people. SMS 5 informs about the importance of saving and also calls for action. SMS 6 reinforce the part of SMS 5 that calls for action. It includes three variations in the wording and tone that allow us to study the existence of a framing bias. Individuals in the treatment group are at the same time randomly assigned to three different groups for receiving positive (gain), neutral or negative (loss) forms of SMS 5. Finally, SMS 6 also call for action but in this case to inform about the benefits of using an alternative access channel to interact with the financial system (i.e. banking agents).
Anytime, participants can write an SMS asking to abandom the treatment. During the intervention, only 6 people requested not to receive the SMSs.
For all participants, we have information about age, gender, estimated income and a range of variables informing their interaction with the bank in terms of savings, bank accounts, use of different channels, etc.