The Difficulties of Financial Inclusion via Large Banks: Evidence from Mexico

The experiment was designed and implemented by the Bank without the participation of the researchers.

We use detailed data from a large commercial Mexican bank and a product (the study card) that accounted for 15\% of first-time loans nation-wide in 2010. Card holders were chosen subject to the additional constraint that they had paid at least the minimum amount due in each of the six months prior to (and including) January 2007. This left the bank with a sampling frame of about one million clients from which the study sample was drawn. The sampling frame was partitioned into nine strata based on tenure with the bank and payment behavior (each taking on three values), which the bank uses internally as predictors of default. The bank then randomly selected a sample of 18,000 clients per stratum. We use stratum weights in all of our analysis to ensure our results are representative of the sampling frame. Within each stratum, clients were randomly assigned to one of nine study arms so there are 2,000 clients per treatment arm within a stratum.

Within each stratum, the bank randomly allocated 2,000 members each to each of 8 intervention arms and one control arm. Each treatment arm is a combination of two contract characteristics: (a) a required minimum monthly payment which is expressed as a fraction of amount outstanding (debt) on the card, and (b) the interest rate on the amount outstanding. The minimum payment was set at either 5% or
10%. The interest rate could take on one of four values: 15%, 25%, 35% or 45%. The interest rate for the study card prior to the study was approximately about 55% so all the experimental interest rates are reductions relative to the status quo. The two different minimum payments and four different interest rates yield 8 unique contract terms. The experimental design thus identifies for each outcome and for each month 8 treatment effects within each of 9 different strata. In addition 2,000 customers within each stratum also served as a control group whose contract terms did not change during the period of the experiment.

2007-03-01

2009-05-01

(a) Default: defined as three consecutive monthly payments that are less than the minimum payment.
(b) Voluntary cancellations by the client
(c) Revenue (net of default and cancellations)

(a) Default: defined as three consecutive monthly payments that are less than the minimum payment.
(b) Voluntary cancellations by the client
(c) Revenue (net of default and cancellations), client level present value of inflows of money to the bank minus the present value of outflows of money. Adjusting for initial and terminal conditions.

(a) Having another loan with study bank
(b) Having loans with other banks
(c) Default in other banks

The experiment was designed and implemented by the Bank without the participation of the researchers.

Within each stratum, the bank randomly allocated 2,000 members each to each of 8 intervention arms and one control arm. Each treatment arm is a combination of two contract characteristics: (a)a required minimum monthly payment which is expressed as a fraction of amount outstanding(debt) on the card, and (b) the interest rate on the amount outstanding. The minimum payment was set at either 5% or 10%. The interest rate could take on one of four values:15%,25%,35% or 45%. The interest rate for the study card prior to the study was approximately about 55%so all the experimental interest rates are reductions relative to the status quo. The two different minimum payments and four different interest rates yield 8 unique contract terms. The experimental design thus identifies for each outcome and for each month 8 treatment effects within each of 9 different strata. In addition 2,000 customers within each stratum also served as a control group whose contract terms did not change during the period of the experiment. In most cases we use the 5% minimum payment and the 45% interest rate group (abbreviated to(45,5)) as the comparison group and we often refer to it as the base arm or base group.
Timeline: Each study client was sent a letter in March 2007 stating the new set of contract terms that would be in force starting in April 2007. Clients were not informed about the study or of any time-line for when then new contract terms would change. The measurement of experimental outcomes began in March 2007 and lasted until May 2009. During this period the interest rate and the minimum payment were kept fixed at their experimentally assigned levels. The experimental terms were not revealed to the risk department (in charge of deciding credit limits).The experiment ended in May 2009 at which point all treatment arm participants received a letter setting out their new contract terms.These terms were the standard conditions with an interest rate of approximately 55% and a mini-mum payment of 4%.

Randomization was done by the Bank in a computer.

The unit of randomization was the credit card.

No

we use 144,000 credit cards, the design was not a clustered design.
It did have strata as we described above

we use 144,000 credit cards

we use 144,000 credit cards

The bank did not provide us with the power calculations