WPS6439


Policy Research Working Paper                         6439




        Why Is Voluntary Financial Education
                  So Unpopular?
                 Experimental Evidence from Mexico

                                  Miriam Bruhn
                                Gabriel Lara Ibarra
                                 David McKenzie




The World Bank
Development Research Group
Finance and Private Sector Development Team
May 2013
Policy Research Working Paper 6439


  Abstract
  Take-up of voluntary financial education programs                                  percent of interested individuals invited to training to
  is typically extremely low. This paper reports on                                  attend. Using a randomized encouragement design, the
  randomized experiments around a large financial literacy                           authors measure the impact of the course on financial
  course offered in Mexico City to understand the reasons                            knowledge and behavior. Attending training results in a
  for low take-up, and to measure the impact of financial                            9 percentage point increase in financial knowledge and
  education. It documents that the general public displays                           a 9 percentage point increase in saving outcomes, but
  little interest in such courses and that participation is low                      no impact on borrowing behavior. Administrative data
  even among individuals who express interest in financial                           indicate that the savings impact is relatively short-lived.
  education. The paper experimentally investigates barriers                          The results suggest people are making optimal choices
  to take-up, and finds no impact of relaxing reputational                           not to attend financial education courses, and point to
  or logistical constraints and no evidence that time                                the limits of using general purpose courses to improve
  inconsistency is the reason for limited participation. Even                        financial behavior for the general population.
  relatively sizeable monetary incentives get less than 40



  This paper is a product of the Finance and Private Sector Development Team, Development Research Group. It is part of
  a larger effort by the World Bank to provide open access to its research and make a contribution to development policy
  discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org.
  The authors may be contacted at mbruhn@worldbank.org, glaraibarra@worldbank.org, and dmckenzie@worldbank.org.




          The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
          issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
          names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
          of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
          its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                        Produced by the Research Support Team
    Why is voluntary financial education so unpopular? Experimental evidence from Mexico #

                                        Miriam Bruhn, World Bank
                                      Gabriel Lara Ibarra, World Bank
                                       David McKenzie, World Bank




Keywords: Financial literacy; financial capability; encouragement design; low take-up.

JEL codes: D14, O12, G28.

#
  We thank the Russian Trust Fund for Financial Literacy for financing this work; Pablo Antón Díaz, Eder González
Ramos and the staff at Innovations for Poverty Action for their assistance in implementing the surveys and impact
evaluation; and the staff at our partnering financial institution for their support for this project. All opinions
expressed in this work are those of the authors alone and do not necessarily represent those of the World Bank,
IPA, or the partnering financial institution.

                                                        1
1. Introduction

Access to finance is strongly associated with poverty reduction worldwide (World Bank, 2008).
For Mexico, evidence suggests that many poor and lower middle-class individuals are credit-
constrained and that relaxing these constraints has high returns. For example, McKenzie and
Woodruff (2008) show that small injections of working capital funding have monthly returns of
15 percent or more for microenterprise owners. Bruhn and Love (forthcoming) find that a rapid
expansion of bank credit serving low-income individuals led to increases in income and
employment. Lack of access to finance has also meant Mexican households have been unable
to smooth consumption in the face of income shocks (Attanasio and Székely, 2004, McKenzie,
2006). These factors illustrate the great potential for increases in access to savings and credit to
benefit low-income individuals.

However, as access to financial services expands around the world, there is also a growing
concern that many consumers may not have sufficient information and financial acumen to use
these new financial products responsibly. Such concerns are particularly pressing in middle-
income countries such as Mexico, which have seen rapid expansions of access to finance in
recent years, bringing many low-income individuals into the formal financial system for the first
time. The number of credit cards has doubled during the past five years, and in urban areas an
increasing number of credit card holders are now poorer and less-educated individuals.
According to the 2008 Mexican Income and Expenditure Survey (ENIGH), 42 percent of
households making a credit card payment had a head with less than high school education, and
one in six had only primary education or less. Lower educated households are more likely to
make unsophisticated decisions (Lusardi and Mitchell, 2011). These unsophisticated decisions
can lead to lower efficiency and competition in the credit card market and, as this market
grows, they could also have implications for the stability of the financial system as a whole.

The recent global financial crisis has emphasized the importance of over-indebtedness on
systemic risk. Low levels of financial literacy were arguably one important factor leading many
homeowners in the U.S. to take out mortgages exceeding their means. 1 During the subprime
mortgage crisis, individuals with lower financial literacy were more likely to be delinquent or to
default on their mortgage (Gerardi, Goette, and Meier, 2010). 2 In Mexico, nearly 10 percent of
all credit card holders fell delinquent on their payments in February 2009, double the U.S. rate.3
Many of these are low-income individuals who are first-time card holders. In addition, the
financial crisis highlighted the need for building up sufficient savings to smooth consumption in

1
  Although see Willis (2011) for a dissenting opinion.
2
  In contrast, individuals with higher financial literacy are more likely to strategically default on underwater
mortgages after the crisis (Burke and Mihaly, 2012).
3
  Source: “Reigning in the Credit Card Industry�, PRI’s The World, May 11, 2009
[http://www.pri.org/theworld/?q=node/26252]
the face of adverse shocks. However, savings rates tend to be low in many countries, including
Mexico (4.2 percent). 4

In response to these concerns, many governments, employers, non-profit organizations and
even commercial banks have started to provide financial literacy courses with the aim to
improve financial education. However, participation rates for non-compulsory financial
education programs are typically extremely low. Willis (2011, p. 430) sums this up as “Voluntary
financial education is widely available today, yet seldom used�. For example, Brown and
Gartner (2007) examine pilot experimental efforts by three credit card providers in the United
States to provide online financial literacy training to delinquent and at-risk credit card holders,
as well as to college students who had newly received credit cards. Target Financial Services
made calls to 80,982 at-risk cardholders, reaching only 6,417 of them, of which half were
invited to use a credit education website: only 684 of these requested a code to log-on, of
which only 28 used the code to log on and only 2 people completed the course. U.S. Bank had
only 384 cardholders out of the 42,000 it attempted to reach complete its online program (0.9
percent). Wells Fargo offered college students a 60-minute phone card as an incentive to do the
training, and had the highest response rate, with 6.7 percent of college student cardholders
offered the treatment logging into the website, and 6.5 percent completing the training.

Thus despite financial education programs becoming increasingly popular among policy-makers
and financial providers, they appear to be deeply unpopular among customers. This raises two
interrelated questions which are important for research and policy. The first is whether there
are economic or behavioral constraints which prevent more individuals participating in such
programs? The second question is whether there are any benefits to these marginal individuals
from doing so, or whether they are rationally choosing not to participate in such training?

We investigate these questions through the context of randomized experiments conducted in
Mexico City. We collaborated with a financial institution to evaluate a free financial literacy
course offered on a large-scale. We document that there is relatively little interest in such a
program among the general population, and then screen recruited subjects on interest in
participating. We then randomly selected half of the interested individuals to be invited to the
course. The initial participation rate in the course was low (18 percent) even among the sample
of interested individuals. Motivated by different theoretical and logistics reasons why
individuals may not attend training, we randomized the treatment group into different
subgroups, which received treatments designed to provide evidence on some key barriers to
take-up. These treatments included monetary payments for attendance, deferred payments,
4
 This number is based on data from the 2008 Mexican Income and Expenditure Survey (ENIGH). It corresponds to
the average household savings rate (total income-total expenditures/total income) using sampling weights and
dropping those with savings rates below -100 percent and above 100 percent. The average household saving rate
using monetary income and expenditures is 11.4 percent.

                                                     3
free cost transportation to the training location, and a video CD with positive testimonials
about the training. We find that the latter two incentives did not increase take-up, but both
current and deferred monetary payments boosted attendance rates by about 10 percentage
points (from 18 to between 27 and 33 percent, depending on the payment size). There was no
significant difference in attendance between those who got the payment immediately and
those who received a deferred payment, suggesting that time-inconsistency or high discount
rates are not the main constraint to participation.

A follow-up survey conducted six months after the course is then used to measure the impacts
of the training on financial knowledge, behaviors and outcomes. We find attending training
results in a 9 percentage point increase in financial knowledge, and a 9 percentage point
increase in saving outcomes, but no impact on credit card behavior, retirement savings, or
borrowing. Moreover, administrative data suggests that the savings impact is relatively short-
lived. Data on credit card balances and repayment rates show no systematic differences across
the treatment and control groups related to the course. In addition to limited overall impacts of
training, we find relatively little heterogeneity in impacts. We do not find significantly different
effects by gender or whether the individual is a customer of our partner financial institution,
but we do find stronger savings impacts for individuals with a bachelor’s degree than those
without.

Our results therefore suggest that the answer to the question of why voluntary financial literacy
programs are so unpopular is that they do not offer that much in the way of benefits to those
who are not currently attending them. That is, it appears individuals may be making rational
decisions not to attend such financial literacy training programs. We cannot rule out that such
programs offer benefits to the individuals who voluntarily choose to go without being given any
additional information or incentives, but our results suggest there are limited gains to trying to
encourage more people to attend.

There is a growing literature attempting to assess the causal impacts of financial literacy
trainings, which we review in the next section. Much of the existing literature has looked at
training provided in a compulsory manner, such as to students (Bernheim et al, 2001; Cole and
Shastry, 2008; Lührmann et al, 2012; Bruhn et al, 2013), or otherwise has focused on
specialized content given to specific populations such as migrants and their families (Gibson et
al, 2012; Seshan and Yang 2012; Doi et al, 2012), and farmers (Cai, 2012). We contribute to this
literature by focusing on a large-scale voluntary financial literacy program offered to the
general population in a large urban city – precisely the type of effort that is being increasingly
adopted in a number of countries around the world. As well as evaluating the impact of such a



                                                 4
program, we are able to experimentally test the importance of different constraints to
participation in such programs, which existing literature has not done.

The remainder of this paper is structured as follows. Section 2 provides background information
on financial literacy in Mexico and reviews the existing literature. Section 3 describes the
financial literacy course. Section 4 lays out the experimental design and initial take-up. Section
5 then describes our experiments to overcome barriers to take-up. Section 6 presents estimates
of the impact of the course on financial knowledge, behavior, and outcomes. Section 7
concludes with policy recommendations.



2. Background and Literature Review

2.1 Financial Literacy in Mexico

During the last two decades Mexicans have experienced significant changes in access to
finance, banking products, and exposure to risk. In 1997, the Mexican Pension System switched
from a pay-as-you-go system to one based on individual retirement accounts. Under the new
system, all private sector workers have to choose a pension fund administrator to manage their
account and invest in portfolios with different levels of risk. A similar pension system was
launched for public sector workers in 2007. Banking services and financial product availability
has also increased in recent years. While in 2004 around 25 percent of the population had a
bank account (World Bank 2008), this number had more than doubled by 2009 (Saldivar, 2009),
reaching 58 percent. The number of credit cards has increased from just below 15 million to
almost 28 million during the past five years – so that there are now as many credit cards as
households in Mexico.5

As access to financial services expands and many consumers become users of formal financial
products for the first time, there is a growing concern that many individuals may not have
sufficient information and financial acumen to use these financial products responsibly. The
National Commission for the Protection of Financial Services Users (CONDUSEF) found that 62
percent of Mexicans lack financial education and are not aware of their rights and
responsibilities with respect to financial institutions. 6 In the 2012 Visa International Financial
Literacy Barometer, Mexico ranks in the lowest third of a sample of 28 countries on questions



5
  The INEGI Population Count (Conteo de Población y Vivienda) reports that there were 28 million households in
Mexico in 2010.
6
  http://www.cnnexpansion.com/mi-dinero/2010/10/05/condusef-tras-la-educacion-financiera

                                                        5
related to having a household budget, or savings set aside for an emergency. 7 Coupled with the
evidence of low financial literacy, there is also evidence suggesting that individuals may not be
managing their financial products correctly. For example, a recent study by Ponce, Seira, and
Zamarripa (2009) analyzing credit card use and payment behavior in Mexico suggests that
consumers make financially unsophisticated decisions in the credit card market. Also, Mexican
workers tend to choose their pension fund administrator based on features not related to fees
or returns (Hastings and Tejeda-Ashton, 2008, and Lara-Ibarra, 2011).

The Mexican government has addressed the concern for low levels of financial education in
several ways. CONDUSEF has held a Financial Education Week annually since 2008. The Ministry
of Finance has also launched campaigns to improve financial literacy and financial education
awareness. Furthermore, one of the objectives of the 2008-2012 National Development
Financing Program was “to develop financial culture and consumer protection, promoting that
people realize that they can save, get financing for their productive products, be responsible
with their loans and protect against risks� (Saavedra 2012). The government’s efforts have also
been matched by the development bank BANSEFI and by private sector, e.g. commercial bank,
initiatives. In 2009, a total of 59 programs throughout the country were promoting financial
education among different audiences (Treviño Garza, 2011).

2.2 Literature Review

Financial education programs act under the assumption that lack of knowledge may be
preventing individuals from making sound financial decisions. This is in part due to growing
evidence showing strong associations between financial literacy measures and financial
decisions. For example, Lusardi and Tufano (2008) find that individuals who have low measured
levels of financial literacy tend to pay minimum balances on credit cards, incur late fees on
cards, and use informal sources of credit. Stango and Zinman (2009) show that people who
make mistakes in interest and future value calculations tend to borrow more and save less.
Lusardi and Mitchell (2009) illustrate that people with low levels of financial literacy think less
about retirement and most of them have not planned for retirement at all. And in Mexico,
Hastings and Tejeda-Ashton (2008) conducted a survey that reveals that less literate individuals
tend to choose mutual pension funds with higher fees.




7
 Overall, the Barometer ranked Mexico in the top 3. However, this result may be biased by “taste�-related
questions such as: At what age should the government require financial literacy? How well do you think teens
understand finance? How often do you speak to your kids about money?
http://www.practicalmoneyskills.com/summit2012/decks/bodnar.pdf

                                                        6
There are three main problems with moving from these studies to developing policy
recommendations. The first is that the results are not necessarily causal. They show a
correlation between proxies for financial literacy and outcomes of interest, but these
correlations may simply reflect unobserved characteristics of individuals such as their
numeracy, ability, self-discipline, parental background, or other such features. Second, it is
often conceptually difficult to understand whether individuals are truly under-savings and over-
borrowing, so even if we see increases in savings it is unclear whether this represents a welfare
improvement. Thirdly, even if we knew that individuals were over-borrowing and under-saving,
and that financial literacy had a causal impact on these outcomes, it is still an open question as
to whether and how financial literacy skills can be taught and improved.

A growing literature tries to address the first and third of these issues by relying on quasi
experimental or experimental variation in the provision of financial education programs. The
studies vary widely in terms of the context of the study. In addition, they face the challenge
that the concept of financial literacy is often measured and defined differently (see Xu and Zia,
2012, and Lewis and Messy, 2012).

Compulsory financial education classes taught in high schools have been subject to a number of
studies. Bernheim et al. (2001) use exogenous variation in high school financial education
mandates across U.S. states to show that students exposed to financial education classes save
more as adults. However, Cole and Shastry (2008) cast doubt on these findings, showing that
they are not robust to controlling for state fixed effects and examining effects over time.
Shorter-term evidence come from Bruhn et al. (2013), who conducted a randomized
experiment providing financial education in Brazilian high schools. They find positive effects on
financial knowledge, attitudes and behaviors, and an increase in savings rates, although the
impacts are small in absolute magnitude: a 3 percentage points increase in knowledge, and 1
percentage point increase in savings. In Germany, Lührmann et al. (2012) find teenagers given
financial literacy training show increased interest and knowledge of financial matters, and save
more in a hypothetical task, but they do not measure actual savings.

Other studies have focused on providing financial education to working adults, recognizing the
differences in households’ financial needs and exposure across developed and developing
countries. The literature in developed countries tends to study the impact of financial
education on planning for retirement or investment portfolio choices. Participation in voluntary
seminars for retirement savings also tends to be low. Duflo and Saez (2011) show that offering
monetary rewards can increase attendance substantially, but that this large increase leads to
less than a 1.5 percentage point difference in retirement plan participation.



                                                7
In the developing country context, impact evaluations of financial literacy training have studied
the unbanked, insurance take-up, and migrants. One of the first papers to examine the impact
of financial education in a developing country is Cole, Sampson, and Zia (2009). The authors
implemented a field experiment in Indonesia where they offered randomly selected unbanked
households a financial literacy course geared towards opening a bank savings account. They
find that the financial literacy course had no effect on the likelihood of opening a bank savings
account in the full sample, but it had modest effects for uneducated and financially illiterate
households. Cai (2012) used a randomized experiment to show that farmers in rural China are
more likely to take-up crop insurance and become less price-sensitive after attending financial
education sessions.

Gibson, McKenzie, and Zia (2012), Doi, McKenzie, and Zia (2012), and Seshan and Yang (2012)
analyze how providing information and financial education affects the behavior of migrants and
their households. Gibson et al. (2012) work with migrants in New Zealand and Australia, and
find that financial education increases knowledge about remittance transaction costs, but did
not lead to changes in the amount of remittances sent or use of the cheapest remittance
method. Using a sample of Indonesian migrants, Doi et al. (2012) find that impacts on financial
knowledge, behavior, and savings are largest when both the migrants and their families receive
financial education. The results show that financial education can have large effects when
provided at a teachable moment, but that this impact varies with who receives training. GAO
(2004) has also recognized the importance of providing education during “teachable moments.�
Seshan and Yang (2012) find that Indian migrants in Qatar increase savings after financial
literacy training, but only if they had low financial literacy to begin with.

A common assumption in most financial literacy interventions is that people are not saving
enough. Indeed George Akelof (2001) goes so far as to state in his Nobel address that “it is
common wisdom that people save too little�. He notes that in standard economic models,
savings is the outcome of utility maximization decisions, so saving too little or too much is not
possible. However, behavioral economics offers several reasons for undersaving: present
consumption is more salient than future consumption, so individuals may procrastinate about
saving and have time-inconsistent preferences. He notes that the best evidence of undersaving
is the observation that upon retirement, consumption falls substantially. However, the view
that people do not save enough for retirement has been challenged by Aguiar and Hurst (2005),
who show that although expenditure falls dramatically upon retirement, consumption does not
as people increase home production. Nevertheless, evidence from lab experiments (e.g. Brown
et al., 2009) supports two views of undersaving: that people don’t know how to save optimally
in complex environments; and that even when they do know how to save, they can’t resist
short-term temptations to spend. Financial education in principle can reduce under-saving by

                                               8
helping with these two problems: by giving individuals more information and knowledge about
savings strategies, and by providing tools and ideas to help resist temptations to spend.

In a recent review, Gale and Levine (2010) find that none of the traditional approaches 8 have
generated unambiguous evidence of positive impacts of financial literacy efforts. The literature
is now moving towards exploring whether innovative channels for providing financial education
may affect behavior. Ongoing studies in India, Peru, South Africa and in the U.S. (among others)
are testing whether the provision of information via videos, radio, mass media or video games
are effective in improving individuals’ financial decisions.

This study builds on and complements this existing literature in several ways. First, we work
with a population that is already banked. Thus, we do not study whether financial education
makes individuals more likely to use formal financial products, such as savings accounts or
credit cards, but rather on whether they use these products to their advantage and responsibly.
Second, the financial education course we analyze covers topics that are relevant for a range of
users of financial services and is delivered at scale by a financial institution, rather than being a
pilot program or government program. Third, we examine interest for such training in a general
population, and then focus on individuals who state an interest in attending a financial literacy
course. We show that even among this population training attendance rates are low and we
examine the barriers to participation experimentally through the use of different incentive
treatments. Finally, we obtain administrative data on savings and credit card usage from a
financial institution to validate the conclusions drawn from survey data.



3. The Financial Literacy Training Product

The financial literacy training course studied in this paper is a large-scale program offered to the
general public. The goal of the program is to convey basic knowledge and tools that individuals
may need to manage their personal finances responsibly. It is targeted at adults and is offered
free of charge. The program was first launched a few years ago and has trained over 300,000
individuals. In 2011, the course won a regional award for innovativeness in fostering financial
education.

During the time of our study, the course was being taught at several locations in Mexico City, in
a number of other cities in Mexico, as well as through an online platform. Our study focuses on

8
 The authors mention four approaches: employer-provided financial education, state-mandated financial and
consumer education for students, credit and mortgage counseling, and community-based programs providing
general financial education.

                                                      9
the training locations in Mexico City. Courses are offered on a continuous basis, with one or two
sessions per day Monday through Saturday. Each session has capacity for twenty participants,
although typical attendance at the most central location is only four or five people per day on
weekdays and more on weekends; and attendance is even lower at the other locations, with
participation rates of about one person per day during our pre-treatment monitoring. Training
is administered via individual computer terminals, with an instructor present to show videos
and facilitate interactive exercises that are used to strengthen the concepts taught in the
material. Participants also receive workbooks that contain the information being presented, as
well as exercises to be completed during the course. At the end of the session, participants take
a short test and receive a certificate conditional on completing the test successfully. They also
receive a CD to take home. This CD includes the tools used in the exercises performed during
the course.

The course lasts half a day and consists of modules on saving, retirement, credit cards, and
responsible use of credit. The course explains why savings is important and discusses different
savings instruments and steps individuals can take to increase the amount they save, such as
setting savings goals and keeping a household budget. Saving for retirement and pension funds
are also covered. The course then discusses the use of credit cards, associated fees, and how to
decipher a credit card statement. Finally, information is provided on good credit card debt
management practices, an individual’s credit score and credit history, and steps individuals can
take to preserve and improve their credit management.



4. Experimental Design and the Low Demand for Financial Literacy Training

To estimate the impact of the training on financial knowledge, behavior, and outcomes, we
conducted a randomized experiment. This section describes the design and implementation of
the initial experiment, including sample selection, randomization, and take-up. The next section
then discusses our approaches to experimentally examining the barriers to take-up, while
follow-up data collection is discussed in Section 6. Appendix A provides a timeline to summarize
when each project phase took place.

Many programs in finance and private sector development face low take-up (McKenzie, 2010),
and prior voluntary financial literacy programs have also experienced low take-up issues (Brown
and Gartner, 2007). While the 300,000 individuals taking part in the financial literacy training
program studied here is a substantial number, this is country-wide and over four years.
Assuming at most 100,000 were trained in Mexico City in any one year, this represents less than



                                               10
0.6 percent of Mexico City’s adult population obtaining the training annually. 9 Low take-up has
severe consequences for statistical power. We therefore tried several approaches to screen
individuals for their interest in participating in financial literacy training and thereby measure
the impacts of such training on these individuals. This process gives insights into the demand
for financial literacy training, and provides training impacts for a policy relevant group: since it
is difficult to make adult financial education mandatory, a key policy question is whether policy
efforts should try and encourage more people who are interested in attending such programs
to actually attend them.

4.1 Approach One: Low Demand for Financial Literacy Demonstrated through a Mailing
Campaign

Our first approach to obtaining a sample of people interested in financial literacy training was
to send a screener survey to clients of our partner financial institution. The institution agreed to
partner with us and to provide us with a de-identified list of all their clients in Mexico City who
have a savings account and a credit card. By conditioning on these characteristics we intended
to identify individuals for whom debt management and saving advice is likely to be relevant and
who have credit card and savings behavior that we could study. We further narrowed down the
list of clients to those who lived in a municipality with a financial literacy training location or
less than 5 km away. From this sample, we randomly selected 40,000 clients to receive a
mailing with the screener survey. The randomization was stratified by gender and age. The
mailings were sent by our partner institution through their usual provider between January 7
and 12, 2011. The delivery company confirmed delivery of 98.8 percent of the letters. Non-
deliveries were due to clients having moved or the address not being found.

The mailing contained a letter informing clients we are partnering with their financial institution
to help research ways people are managing their savings and credit card debt, and that we
would like to see if they are interested in participating in a financial education training session.
The letter only mentioned the training in general terms and did not refer to the specific
program we are studying or the locations where this training was being offered. The mailing
also contained a two-page screener survey that clients could mail back to us in a pre-paid
envelope to indicate their interest in the training. This short screener survey collected
information on name, address, phone number, sex, age, education level, occupation, household
income and expenditure (in bins), as well as basic usage of savings accounts and credit cards.
Clients also had the option of responding to the survey by going to a website or by calling a toll-
free number. In order to increase response to the screener survey, half of the letters (20,000)

9
 CONAPO projects that in 2012 there were 17,372,952 individuals 17 years old or older living in the Metropolitan
Zone of Mexico City. http://www.conapo.gob.mx/es/CONAPO/De_las_Entidades_Federativas_2010-2050

                                                       11
were randomly selected to include an offer for a monetary payment of 75 Pesos (about US$5)
to the first 200 clients who submitted their answer.

The total number of letters sent (40,000) was chosen based on information from our partner
financial institution that typically only 2-3 percent of their clients reply to any sort of mail offer
sent by them. This expected response rate would yield 800-1,200 clients who would form the
sample for our randomized experiment. However, we received much fewer responses than
anticipated – only 42 responses. That is, only 0.1 percent of clients expressed interest in a
financial literacy training program.

We suspect that this low response rate is in part due to the climate of insecurity that has
prevailed in Mexico City during the past few years and that has made people distrustful of
unsolicited requests for data. Around the time of our study, fake phone and mail extortions had
become a threat to the general population. In fact, both the federal government and
commercial banks launched a campaign asking people to avoid giving personal information to
strangers, especially through phone or mail communications offering money or prizes in return
for personal data. Our letters were sent out by our partner financial institution, bearing their
logo, but people may still have been suspicious of the request for data. 10

4.2 Approach Two: Low Demand for Financial Literacy Demonstrated through an Online
Campaign

Our second strategy for obtaining a study sample was to conduct a screener survey through
Facebook. We created a Facebook page for financial literacy and launched a Facebook ad that
pointed to this page. The Facebook page included the same information as the letters sent in
the mail, mentioning the importance of financial literacy. The page invited people to indicate
their interest in participating in a (generic) financial literacy course by clicking on a link that
redirected them to a page where they could answer our screener survey online. This survey
contained the same questions as one mailed to our partner financial institution’s clients. We
did not offer a financial incentive for completing the online survey.

The Facebook ad was targeted to individuals residing in Mexico City and ran for two months,
from mid-February to mid-April, 2011. 11 It was displayed about 16 million times. We obtained a
total of 1,240 fans of our Facebook page and 119 responses to the online survey. Since this

10
   All letters included a toll free number where people could make enquiries. We did not receive any phone calls
enquiring about training in response to this mailer, suggesting a low demand for such training.
11
   At the time, Facebook had 7,743,220 registered users who reside in Mexico City (approximately 87 percent of
the population).


                                                        12
sample was still not large enough for our study, we implemented a third approach to screening
people for an interest in a financial literacy course, as described below.

4.3 Approach Three: Street and Branch Surveys

We conducted screener surveys on the streets of Mexico City and outside branches of our
partner institution. Surveyors were placed in busy locations within the city during a period of
eight weeks (from April 25 to June 25, 2011) where they tried to interview people passing by.
We also placed surveyors outside branches of our partner institution between July 6 and 19,
2011, where they approached exiting customers and people waiting in line.

For this approach, interviewers asked people if they would be interested in participating in a
financial literacy course, providing the same information as stated in letters sent during the
mailing campaign. As in the letters, the name of the course was not disclosed. If the
respondents expressed interest in the course 12, interviewers asked them to fill out the screener
survey, so that we could contact them later with further information about it. The
questionnaire included the same set of question as the mail and online surveys. We did not
offer a monetary incentive for completing the surveys, but people who answered the survey
were offered cookies and a pen as a small thank you token. We obtained a total of 6,945
completed questionnaires from the street survey and 2,294 from the branch survey.

4.4 Treatment Randomization and Balance

For all individuals who had expressed interest in a financial literacy course either through the
mail, online, street or branch screener survey, we conducted phone audits to verify the contact
information they had provided. This eliminated about half of the respondents. We further
dropped respondents who lived outside the Mexico City metropolitan area or who had
participated in a financial literacy program in the past. We also dropped observations that had
missing answers to the questions we stratified on in the randomization, as described below.
Our final sample includes eight respondents from the mail survey, five from the online survey,
2,490 from the street survey and 1,000 from the branch survey, giving a total sample of 3,503
people. We divided this sample into a control group of 1,752 individuals, and a treatment group
of 1,751 individuals, using stratified randomization. The randomization was conducted by the
authors by computer.




12
  Subject recruiters reported that, roughly, only 2 out of every 5 people approached expressed an interest in the
course and agreed to fill out the survey.

                                                        13
We stratified the randomization by whether we obtained the respondent through the branch
vs. the mail, online or street survey, by gender, by having at least a bachelor’s degree or not,
and by whether the person was (i) a client of our partner financial institution, (ii) a client of
another financial institution, (iii) neither. Within the sample of financial institution clients, we
further stratified by whether they made a deposit into their savings account during the past
month and by whether they have a credit card. For clients with a credit card, we stratified by
whether they made more than the minimum payment each month during the past six months.
For individuals who were not financial institution clients, we stratified by whether they lived
closer than 8 km away from a training location or not.

Our original intention with the mail screener survey was to study only individuals who are
financial institution clients to learn whether the financial literacy course can improve their
financial behavior and outcomes. Through the online, street, and branch surveys, we obtained
1,325 respondents who were not financial institution clients. We decided to keep these in the
sample since the course material could in principle also be relevant for them and they may start
financial institution relationships as a result of taking the course. However, the percentage of
non-financial institution clients in the treatment group who ended up attending the financial
literacy training was very low (18.1 percent), making it difficult to detect any effects on this
sample. Since the take-up rates were higher among financial institution clients (28.1 percent),
as discussed further below, we decided to conduct our follow-up survey only among financial
institution clients and we drop non-financial institution clients from the impact analysis,
although they are included in our experiments on inducing attendance.

Table 1 shows baseline variables collected through the screener survey for the sample of
financial institution clients. About half of the individuals were clients of our partner financial
institution as opposed to clients of another institution. Close to 65 percent made a savings
deposit during the past month. About 40 percent had a credit card at baseline and only half of
them made more than the minimum payment in all previous months. In fact, about 20 percent
made a late payment on their credit card during the past six months. The demographic
variables show that about half of the sample is female and 40 percent has at least a bachelor’s
degree. The average age is 33 years. As expected given the random assignment, all baseline
variables are balanced across the treatment and control groups.

4.5 Initial Take-Up: Low Attendance for Training Even among Those Who Say They Are
Interested in Attending a Financial Literacy Course

Starting on August 1, 2011, each person in the treatment group was contacted by telephone
and invited to participate in the financial literacy training program. If the participant confirmed

                                                14
interest, they were offered the opportunity to choose a training location, date and schedule
that best suited them. The phone operator then enrolled the participant based on this
information. The operator proceeded to confirm the details of the appointment with the
participant before ending the call.

A second call was made the day prior to the participant’s appointment as a reminder to
increase the probability of attendance, and a third and final call was done the day after their
appointment to confirm attendance and inquire about their level of satisfaction with the
courses. If during this third call, if participants responded that they had missed the
appointment, they were offered the opportunity to reschedule for a future date, if they claimed
they were still interested in attending.

During this phase we signed up 1,049 out of 1,751 individuals (59.9 percent) for the course.
About a third of the people who signed up for the course actually attended (312 people).
Participants gave a range of reasons for not attending the session they had signed up for,
including difficulties attending due to work and family commitments, sickness, and in some
cases, issues with instructors turning up late or them arriving late and being turned away.

The overall attendance rate for the 1,751 treatment group individuals who had been screened
for interest in attending a financial literacy course was thus only 17.8 percent. This number is
low compared to attendance rates for business training courses. A study in Mexico found a 65
percent attendance rate for business training. Attendance rates for business training
interventions in other countries range from 39 to 92 percent (McKenzie and Woodruff, 2012).



5. Overcoming Barriers to Attendance through Experimental Interventions

We first examine theoretical reasons why individuals may choose not to participate in financial
literacy training, and then describe the experiments designed to test these barriers.

5.1 Why People Might Not Want to Participate in Financial Literacy Training

Let c be the cost of attending the financial literacy program. While the program itself is free,
individuals would incur transportation costs in getting to and from the program, as well as the
opportunity cost of lost income (or lost leisure time). Let bt be the benefits the individuals will
realize in period t from participating in the course, such as better financial outcomes, and Ebt
the certainty-equivalent expectation of these benefits. Then theory will predict that an



                                                15
individual will choose to attend a financial literacy course if the expected discounted benefits of
the course exceed the costs of attending. i.e. if

                      ∑𝑇       𝑡
                       𝑡=0 𝛽𝛿 𝐸�𝑡 > �                                             (1)

Revealed preference would suggest that anyone whom it would benefit to take the financial
literacy course would have done so, while anyone who chooses not to take the course is doing
so because they do not view the benefits as exceeding the costs.

This theory then suggests several barriers that may prevent individuals from participating in the
financial literacy course, even if it has positive benefits to them (∑𝑇
                                                                       𝑡=0 �𝑡 > 0). A first reason is
just that they face costs of attending, so that c is large in magnitude. A second set of reasons
concern the timing of when costs are incurred relative to when benefits are received. In
particular, individuals may not participate because the costs are experienced immediately,
while the benefits may take time to accrue. Individuals with high discount rates (low 𝛿 ) may
find the discounted value of benefits is less than the current costs. Individuals who are present-
biased (𝛽 < 1) can have time-inconsistent preferences, and so while they would like to have
attended financially literacy training in the future, because the benefits occur in the future and
attendance occurs today, they keep putting the course off. Thirdly, individuals may not know
the benefits of participating, potentially undervaluing them. Even if their expected benefits are
accurate, with risk aversion, uncertainty as to these benefits will still cause Ebt < bt. Finally, one
could also imagine that liquidity constraints prevent individuals from paying the costs today,
even if they see positive net expected benefits. This final explanation seems less relevant in our
case where many individuals have credit cards and most have savings.

5.2 Experimental Interventions to Overcome Barriers to Attendance

We designed a second stage of interventions to explore if we could tease out whether high
uncertainty about benefits, transport costs, high opportunity costs, impatience or low appeal
for the course could explain individuals’ low attendance and thereby learn what policy efforts
could spur attendance. The treatment group was divided further into six different groups – one
group who received no further assistance, and the following five booster treatment groups:

       1. Offered 1,000 pesos (US$72) for completing the training: participants were given a
          Walmart gift card of 1,000 pesos if they attended 13,
       2. Offered 500 pesos (US$36) gift card for completing the training,


13
     For comparison, at follow-up, median monthly income in our sample was about 9,000 pesos (US$650).

                                                        16
   3. Offered 500 pesos (US$36) gift card that they would receive one month after
      completing the training,
   4. Offered a free taxi ride to and from the course location,
   5. Provided a video CD containing positive testimonials from people who had attended the
      course.

Treatments 1 and 2 enable us to examine whether individuals are more likely to attend the
course as the benefits of attending the course increase. This helps get at the issue of whether
individuals are making rational decisions by responding to changes in the net benefits, as well
as to measure how the demand for training varies with these benefits. The comparison of
treatments 2 and 3 enables us to see whether high discount rates or present bias is a reason for
a lack of attendance – people might think the course has benefits, but because these benefits
occurs in the future and attendance occurs today, keep putting the course off. If this was the
issue, we would expect much greater response to Treatment 2 than to Treatment 3. Treatment
4 lowers the costs of attending the training, which enables examination of whether one
important component of c is the constraint. Treatment 5 aims to reduce informational
constraints that may prevent people from attending the course if they are not sure if it will be
helpful, thereby attempting to reduce the difference between Ebt and bt.

The incentive treatments were assigned through stratified randomization using the entire
treatment group. We stratified by whether the respondent was screened through the branch
vs. mail, online, or street survey, by income group, by whether the person was (i) a client of our
partner financial institution, (ii) a client of another financial institution, (iii) neither, and by
attendance status thus far. The attendance status categories where (i) attended, (ii) was
scheduled, but didn't attend, (iii) was reached, but didn't want to be scheduled and, (iv) can't
be reached. Individuals who had already attended the course did not receive the incentives, but
they were included in the randomization to allow us to report post-incentive treatment take-up
rates by incentive type for the complete treatment group.

5.3 Impacts of Experiments to Increase Attendance

Figure 1 shows the overall percentage attending in each incentive treatment group (after the
initial treatment offer and the booster treatments). Some participants were unreachable by
phone, so we show results as a percentage of all participants assigned to a treatment, and of all
who could be reached to actually offer them the treatment. Table 2 provides regression
estimates of the impacts of the various treatment arms on attendance.




                                                17
As a result of the booster interventions, we succeeded in getting an additional 114 individuals
to attend the course, giving a total of 426 attendees out of 1,751 treatment group individuals
(24.3 percent). Offering a monetary incentive of $36 increased the take-up rate from about 18
percent to 27 percent of those assigned to treatment, while the $72 incentive increased take-
up further to 33 percent. While the difference between the two monetary amounts is not
statistically significant, they both suggest that individuals are rationally responding to higher
benefits of training by being more likely to attend. The treatment impact is exactly the same
when US$36 is offered immediately at the completion of training, or one month after training.
This suggests that high discounts or present-bias are not the main barriers to participation in
training.

In contrast to the monetary incentives, we find that transportation assistance and the
testimonials did not significantly increase attendance. One reason for the lack of impact of
transportation assistance were security concerns in Mexico City, with people reluctant to take a
taxi cab that came to their home. The lack of impact of the testimonials could reflect people
updating their beliefs about the efficacy of the training very little after the receipt of this
treatment, or that lack of information about the benefits and quality of the training was not the
main reason for non-participation.

Finally note that even when participants were offered US$72 to attend, it is still the case that
the majority of individuals who had initially expressed interest in attending financial literacy
training do not attend. 14

5.4 Which Individuals Attend Training?

If we break down the take-up rate by individuals who are clients of a financial institution and
those who are not, we find that the attendance rate was 28.1 percent among clients vs. 18.1
percent among non-clients. Since we do not expect to be able to detect treatment effects with
a take-up rate of 18.1 percent and since our randomization was stratified by being a financial
institution client, we decided to collect follow-up data only on the 2,178 individuals who are
clients of a financial institution. From this point on, all tables and analysis in this paper cover
only these individuals.

Using data from the screener survey, we analyze which individual background characteristics
are correlated with training attendance in the treatment group. Table 3 displays this analysis,
using two different measures of take-up (i) a dummy variable for whether the individual

14
  This dissociation of intentions and follow through has been documented in studies of retirement seminars
(Madrian and Shea, 2001).

                                                       18
attended the course prior to our booster incentive interventions and (ii) a dummy variable for
whether the individual attended either before or after we offered the incentives.15 The bottom
row of table 3 shows that the take-up rate increased from 20.8 percent to 28.5 percent after
the incentive intervention (in our sample of financial institution clients). Two characteristics are
strongly and positively correlated with attendance: education and age. Individuals who have a
bachelor’s degree or higher are up to 14 percentage points more likely to have attended the
training. The take-up rate among this group was 38 percent after the incentive intervention.
Older individuals are also more likely to have participated in the training. In addition, we find
weak evidence that females and individuals who are clients of our partner financial institutions
as opposed to another institution are slightly more likely to have attended the training.



6. Impacts of Financial Literacy Training

We first discuss our data and estimation methodology, and then turn to impacts of the financial
literacy training on financial knowledge, behaviors and outcomes.

6.1 Follow-Up Survey

We conducted a follow-up survey between February and July 2012 to measure post-training
financial knowledge, behavior and outcomes. We kept the questionnaire relatively short (about
15 minutes) to encourage participation. The questions focused on concepts and behaviors
taught in the course. We discuss specific questions and outcome variables below.

For logistical reasons, we first attempted to conduct the follow-up survey over the phone. If the
person did not respond to the survey during the first attempt, we offered them a 500 pesos
(US$36) Walmart gift card for completing the survey during the second attempt. If we were still
not able to interview the person over the phone, a surveyor visited their house to conduct a
face-to-face interview. If the participant was not at home, the surveyor delivered a letter with
information about our study and instructions for how to contact us to participate in the survey
and to receive the Walmart gift card. Surveyors made two more attempts (three attempts in
total) to conduct a face-to-face interview if a respondent was not at home.

We were able to interview 72.8 percent of our sample during the follow-up survey. The attrition
rate was slightly higher in the treatment group (29 percent) than in the control group (25.3

15
  We do not include income among the variables displayed in Table 3 since this variable has many missing values
and including it would thus reduce the sample size. When we add income to the analysis, it is not correlated with
take-up and the coefficients on the other variables remain similar to the ones shown in Table 3.

                                                        19
percent). The last two columns of table 1 show baseline characteristics for the sample of
individuals interviewed at follow-up. The characteristics are very similar to the full sample and
we do not find any statistically significant differences between control and treatment group
means in the follow-up sample. This shows the difference in attrition rates between the two
groups is not leading to imbalance on observable characteristics.

6.2 Administrative Data

We obtained administrative data on saving account balances and credit card outcomes from
our partner institution. Due to confidentiality reasons, our partner cannot disclose individual
level data, but they offered to generate summary statistics at the treatment and control group
level. This was going to be straightforward with the sample we screened through the mail
survey since the list of individuals in this sample came from our partner financial institution,
meaning that they could easily find these individuals in their records. The low response rate to
this survey implied, however, that our sample now almost entirely consists of individuals who
were screened through the street or branch surveys. About half of the current sample (1,034
individuals) reported being clients of our partner financial institution and 470 of these
individuals could be found in the institution’s records based on name, address and phone
number. 16 The last two columns of table 1 show our baseline statistics for this sample of 470
clients. Overall, this group is similar to the full sample. Also, the characteristics of the treatment
and control group clients who were matched with administrative data were not statistically
different at baseline.

6.3 Estimating Treatment Impacts

We estimate the impact of the financial literacy training with the following intention-to-treat
(ITT) equation

yi,s,m = α + βTrainingInvitei,s,m + ∑ γs ds + ∑ δm dm + εi,s,m                                              (2)

where yi,s,m is a follow-up survey measure of the financial knowledge, behavior, or outcome of
individual i, in randomization strata s, who was surveyed in month m. The variable
TrainingInvitei,s,m indicates whether an individual was invited to the course or not and is thus
equal to one for the treatment group and equal to zero for the control group. We control for
randomization strata dummies ds , as well as month of follow-up interview dummies dm . The
main coefficient of interest is β, which represents the treatment effect of being invited to a

16
  For ethical reasons and given the security concerns in Mexico City, we did not ask for date of birth or national ID
numbers in the survey.

                                                         20
financial literacy course. The coefficient β in equation 2 is also equal to the difference in means
of the outcome variable yi,s,m across the treatment and control group, conditional on strata and
interview month dummies.

In addition, we estimate the following local average treatment effect (LATE) regression

yi,s,m = α + βAttendedTraining i,s,m + ∑ γs ds + ∑ δm dm + εi,s,m                           (3)

where AttendedTraining i,s,m is equal to one for treatment group individuals who attended a
course and zero otherwise. We instrument this variable with our indicator variable for whether
an individual was invited to the training or not (TrainingInvitei,s,m ). The coefficient of β here
represents the local average treatment effect: that is, the impact of the financial literacy course
on the individuals who attended a course as a result of being invited to the course but who
would not have otherwise attended.

6.4 Impact on Financial Knowledge

We measure financial knowledge through eight follow-up survey questions based on the
material that was taught in the course. For each question, we create a dummy variable
indicating whether the respondent gave the correct answer to the question. These eight
questions are then aggregated into a financial knowledge index that is the average of the eight
dummy variables and thus represents the fraction of questions that the respondent answered
correctly. Finally, we asked individuals to rate their own level of financial education on a scale
from one to five (1 = excellent, 2 = good, 3 = satisfactory, 4 = unsatisfactory, 1 = no knowledge
of the topic). Based on the responses to this question, we coded a dummy variable indicating
whether the self-assessed level of financial literacy is satisfactory or higher.

Table 4 lists the eight dummy variables used to measure financial literacy, as well as the
knowledge index and self-assessed knowledge level. The table shows the control group mean
for each variable and the ITT and LATE estimates that correspond to the coefficients β in
equations (1) and (2).

The results in table 4 show that the course had a positive and statistically significant impact on
financial knowledge. The knowledge index increased from 0.31 in the control group to 0.34 in
the treatment group due to the training, meaning that control group individuals answer 31
percent of the eight questions correctly, while treatment group individuals answer 34 percent
correctly. The size of this impact is relatively small, but the effect is larger for some individual
questions. For example, only 13 percent of control group individuals know that bank deposits

                                                21
are insured up to 400,000 UDIs (currency units), but the training increased this number to 20
percent in the treatment group. Similarly, 24 percent of the control group knows what CAT
(total annual cost of credit) is, and this number increased to 28 percent in the treatment group
due to the training. The training also increased the self-assessed level of financial literacy, with
63 percent of the treatment group saying their knowledge is satisfactory or higher, compared
to only 58 percent in the control group.

When interpreting the magnitude of the effects, it is important to keep in mind that only about
30 percent of the treatment group attended the financial literacy course. The LATE estimates in
the last column of table 4 take into account this relatively low take-up rate. They show much
larger impacts on individuals who actually went to the training as a result of being invited
through the intervention. The knowledge index increased by 8.7 percentage points due to the
course for these individuals and the probability of answering some of the specific knowledge
questions correctly increased by up to 20 percentage points.

6.5 Impact on Savings Behavior and Outcomes

The financial literacy course emphasized specific behaviors that may help individual save more
money, including: checking financial institution transactions regularly and keeping track of
expenses, making a budget and setting a savings goal, identifying necessary and unnecessary
expenses to reduce overall expenditures. The follow-up survey asked whether individuals
engage in these behaviors and we code five dummy variables indicating that they do, as listed
in panel A of table 5. We aggregate these behaviors into a savings behavior index by taking an
average of the five dummies, giving the fraction of five behaviors that individuals engage in.

The control group means in panel A of table 5 show that many individuals check their financial
institution transactions regularly (69 percent), keep track of expenses (79 percent), and make a
budget (77 percent), even without having taken the financial literacy course. We do not detect
a significant treatment effect of the training on these behaviors, perhaps because these
behaviors were already quite common in the control group. The remaining two savings
behaviors we study are not as frequently used in the control group: 57 percent of individuals
have a savings goal and 59 percent have cut expenses in the past three months. We find that
the course increased the percentage of individuals who cut expenses during the past three
months to 63 percent, but this effect is only statistically significant at the ten percent level.

In addition to studying savings behavior, we also examine whether the training had an impact
on savings outcomes. Panel B of table 5 lists three measure of personal savings (i) a dummy
variable indicating whether the individual has any type of savings/money set aside, (ii) a dummy

                                                22
variable for whether the individual reports saving at least some fraction of their income during
the past six months, and (iii) a dummy variable indicating whether the respondent said they
save more each month than they did a year ago. We construct a savings outcomes index by
taking the average of these three dummy variables. All three savings outcomes are higher in the
treatment group than in the control group at follow-up, although the differences are not
statistically significant for the individual variables. We do, however, find a positive and
significant impact of the course on the savings outcome index. Control group individuals reach
on average 65 percent of all three savings outcomes, increasing to 68 percent in the treatment
group.

Recall that we provided monetary incentives of 500 or 1,000 pesos ($36 or $72) to some
randomly chosen treatment group individuals in order to encourage them to attend the
financial literacy course. These amounts are equivalent to 5.5 and 11 percent of median
monthly income in our sample. To check whether the positive effect on savings outcomes is
driven by the monetary payments instead of the course itself, we add a dummy variable to
estimation questions (1) and (2), indicating whether the individual received an incentive
payment for course participation. Panel C of table 5 replicates the results in panel B, controlling
for the monetary incentive dummy variable. The estimated impact of the course on savings
outcomes is slightly smaller, but remains statistically significant.

The course included material on retirement savings and pension funds. Table 6 shows the
impact of the course on retirement savings behavior and outcomes. In the control group, 57
percent of individuals have a pension fund and the course did not change this percentage in the
treatment group. Pension funds are typically provided through an individual’s employer,
implying that we did not necessarily expect to find an impact of the course on this outcome
since acquiring a pension fund may involve switching jobs. The other variables reported in this
table are based on follow-up survey questions that were only answered by individuals who
have a pension fund. In order to account for possible selection bias, we fill the variables in with
“0� for individuals who do not have a pension fund, so that they are defined for the complete
sample, except for individuals who did not know whether they have a pension fund or not.

In terms of retirement savings behaviors, we measure whether (i) individuals choose a pension
fund administrator based on fees or returns, as opposed to using the default option, relying on
friends’ advice etc., (ii) they check their pension fund statement regularly and, (iii) they have
calculated how much money they will need upon retirement. We construct an index of
retirement saving behavior that is the average of the three behavior dummy variables. As the
retirement savings outcome, we examine whether individuals report saving money for



                                                23
retirement (on top of their pension fund). 17 We do not find an impact of the training on any of
the retirement savings behaviors or outcomes listed in table 6.

6.6 Impact on Borrowing Behavior and Outcomes

The course also discussed responsible use of credit cards and healthy borrowing behavior. Table
7 shows that 48 percent of the control group had a credit card at follow-up. This percentage
was not statistically different in the treatment group. For individuals with a credit card, we
define several measures of credit card behavior and outcomes, as listed in panels A and B of
table 7. The measures are based on questions that were only posed to individuals with a credit
card and refer to their most frequently used card. The variables are filled in with “0� for
individuals who do not have a credit card to account for possible selection bias.

We examine six credit cards behaviors. Three are dummy variables indicating whether the
person (i) knows their credit limit, (ii) knows the credit card interest rate, and (iii) checks the
credit card statement every month. The other three variables are coded as the fraction of the
past six months, where the individual (i) paid the balance in full, (ii) made only the minimum
payment, and (iii) got cash through the credit card. We construct a credit card behavior index
by first converting all variables to z-scores using the control group mean and standard deviation
and then taking the average of these z-scores. The two behaviors that are last on the list are
coded with a negative sign in the average since they represent undesirable credit card behavior.

We study three credit card outcomes (i) whether the issuer blocked the credit card during the
past six month, (ii) the fraction of the past six months where the individual was charged late
payment fees, and (iii) the fraction of past six months where they were charged overdraft fees.
The incidence of all three evens is quite low in the sample, ranging from 1 percent to 4 percent.
We also construct on index of these three outcomes, using the average of z-scores. The results
in table 7 show no impact of the course on credit card behavior or outcomes.

Table 8 examines borrowing behavior more broadly. We look at three loan behaviors, coded as
dummy variables that indicate whether the individual (i) applied for a loan from any source
during the past six months, (ii) went to a pawn shop to get credit during the past six months,
and (iii) stopped servicing outstanding debt during the past six months. We aggregate these
behaviors into a loan behavior index by taking the average of the three dummy variables. In the


17
  We intended to also ask questions on having calculated how much money is needed upon retirement and
whether the individual is saving money for retirement in the complete sample, independent of whether the person
has a pension fund or not. However, by mistake, these questions were only posed to individuals who reported
having a pension fund.

                                                      24
control group, 23 percent of individuals applied for a loan during the past six months, 10
percent went to a pawn show, and 13 percent stopped serving outstanding debt. These
percentages are not statistically different in the treatment group. We also find no effect of the
course on the index of loan behavior.

Panel B of table 8 shows the impact of the course on loan outcomes. One third of control group
individuals have a loan from any source and their total outstanding debt represents about 15
percent of annual income. These numbers are statistically identical in the treatment group,
implying that the training had no effect on loan outcomes.

6.7 Heterogeneous Treatment Effects

We now ask whether the impact of the course was different for different groups of people. The
take-up regressions in table 3 show that individuals with a bachelor’s degree, as well as females
and clients of our partner financial institution were more likely to attend the training compared
to other people in the treatment group. The treatment effects may be larger for these groups
since more of them were exposed to the course material. It could also be the case that
individuals who know that they will benefit more from the course are more likely to attend in
the first place.

The treatment randomization was stratified by gender, being a client of our partner institution,
and having a bachelor’s degree. We can thus examine the effect of the training on these groups
of people separately by adding an interaction term between the treatment group dummy
(TrainingInvitei,s,m) and a dummy variable indicating whether the individual was female, a
client of our partner institution, or had a bachelors, to equation (2).

Table 9 reports the heterogeneous treatment effect regression, using all indices from tables 4
through 8 as the outcome variables. We do not study the index components here to minimize
multiple hypothesis testing. The results in panel A show that the course had a similar impact on
clients of our partner financial institution and on clients of other institutions. The positive effect
on financial knowledge and on savings outcomes is not statistically different across these two
groups. Panel B shows similarly no statistically significant differences in treatment effects by
gender.

The only indication of heterogeneity in impacts is found in panel C, which shows that individuals
with a bachelor’s degree are more likely to improve savings behavior and retirement savings
behavior than individuals without this education level. Since the take-up rates for training were
also higher for the more educated, this could just reflect the fact that this group was more likely

                                                 25
to attend training. We therefore estimate LATE impacts by education level to see whether the
effect of actually receiving training varies with education status. We find that receiving training
had a larger effect on individuals with a bachelor’s degree for both the savings behavior and
retirement savings behavior index (although the heterogeneous treatment effect on savings
behavior is only statistically significant at the 10.4 percent level).

6.8 Comparison of Treatment and Control Group Outcomes with Administrative Data

Our partner financial institution provided monthly data on savings and credit card outcomes at
the treatment and control group level for the 470 clients found in their data base, from
December 2010 through May 2011. This data covers several pre-intervention months since we
started inviting clients to the financial literacy course in August 2011. It also overlaps with the
follow-up survey, which was conducted between February and July 2012.

Figures 2 to 4 plot the administrative data for the treatment and control group over time.
Figure 2 shows that the median savings account balance in the treatment and control group
followed a more or less parallel trend before our intervention started in August 2011. In
October 2011, the savings balance starts rising in the treatment group, going from about 900
pesos in September 2011 to 1,600 pesos in December 2011, while the savings balance in the
control group stays relatively constant. After January 2011, the savings balance in the
treatment group starts falling again and goes back to its original level in April 2012. Although
the aggregate data does not allow us to conduct a test of statistical significance in the
difference of medians, the observed pattern is consistent with the course having a positive
effect on savings account balances. However, this effect appears to be temporary.

Figures 3 and 4 displays median credit card balances and average percentage of credit card
debt paid off each month, respectively. The treatment and control group values follow a similar
pattern throughout the period, indicating no impact of the course on credit card balances or
debt paid off each month.

Overall, the patterns in the administrative data are consistent with the follow-up survey results.
The course seems to have had no effect on credit card behavior or outcomes, but it led to a
small increase in savings. The administrative data suggests that this increase was temporary
and dissipated within a few months.




                                                26
7. Conclusions and Policy Implications

Despite the popularity of financial literacy workshops among policymakers and financial
institutions, voluntary participation rates are typically very low for these programs. Working
with a half-day long training course in Mexico City that aims to teach participants the
importance of savings and responsible debt and credit card use, we use randomized
experiments to investigate the lack of demand.

We find very little interest in participating in financial education among samples of financial
institution clients and Facebook users. Screening our sample to focus on those individuals who
express interest in attending training, we find the majority of them also do not attend.
Experiments to increase take-up suggest that this low participation rate is not mainly due to
high discount rates, time-inconsistency, or lack of information, but rather appear to be due to
individuals thinking that the benefits of such training are not high enough. Monetary incentives
which increase the benefits lead to more attendance. A follow-up survey conducted about six
months after the course enable us to measure just what these benefits are. We find
participating in the course leads to increases in financial knowledge and savings, but this
increase in savings appears to dissipate quickly, and there is no evidence that the training
changed credit card usage or borrowing behavior. The lack of interest in training therefore
appears to be a rational choice, since users see relatively little benefit from it.

One natural response to the modest impacts of training measured here is to note that the
training is only a few hours long, and to thus argue that much longer and intensive training
sessions are needed to really affect financial behavior. However, our study shows that most of
the general population has very little interest in attending even a short financial literacy course,
and that it is very difficult to even get individuals who state they are interested in participating
to show up and attend. Monetary incentives did help boost attendance, but even an immediate
payment of $72 got fewer than 40 percent of people who had expressed interest in a financial
literacy program to attend. Longer and more intensive programs would appear likely to have
even more trouble attracting participants. Moreover, Kim et al. (2003) provides an example to
suggest that even 18 months of adult credit counseling has not been effective.

We caution that our study measures local treatment effects, showing the effect of training for
people who are induced to attend it as a result of our interventions, but would not attend it
without the information, logistical support, and incentives we provided. It seems plausible that
the impacts for these individuals who have self-selected into not attending a program that is
available to anyone who wants to participate will have less benefit from such training than the
people who choose to attend of their own accord. However, as our study notes, fewer than two

                                                27
percent of the eligible population voluntarily choose to take part in training. Even if the training
has large benefits for them, the results of our study suggest that the benefits of encouraging
more people to participate in such training are likely to be slight. This concurs with recent
skepticism about the value of such training (Willis, 2011; The Economist, 2013).

Finally, we note two issues with the increasing focus of financial institutions and policymakers
in extending financial education to the masses through general multi-purpose financial literacy
courses. First, the classroom setting for the provision of financial education may not be
appealing to the general public. We find higher take-up rates among the population of bachelor
graduates who may be more comfortable in such a setting, or may be better at digesting
information provided through this channel. Alternative methods to educate the general public
should be tested. For example, Berg and Zia (2013) provide evidence of an increase in financial
knowledge and reductions in certain types of borrowing following financial education taught
through a Soap Opera in South Africa. Second, the variety of topics covered in the course
conflicts somewhat with the idea that some concepts are best taught at teachable moments.
Workshops about pension savings provided to workers at the time of making their pension
savings allocations, or to college students or young graduates when they receive their first
credit card may be more effective in influencing these outcomes. In contrast to credit and
retirement savings, there is less of an obvious teachable moment for education about savings.
Indeed, since individuals make spending and savings decisions on a very regular basis, it could
be argued that general financial literacy courses are likely to work better for this topic since
individuals have the ability to quickly translate the concepts learned into practice. This might
help explain why we find impacts (albeit transitory ones) only on savings outcomes in our study,
and be consistent with the savings impacts of financial education programs taught in schools.




                                                28
References

Aguiar, Mark and Erik Hurst (2005) “Consumption versus Expenditure�, Journal of Political Economy,
113(5): 919-48.

Akerlof, George (2001) “Behavioral Macroeconomics and Macroeconomic Behavior�, Nobel Prize
Lecture, http://www.nobelprize.org/nobel_prizes/economics/laureates/2001/akerlof-lecture.pdf

Attanasio, Orazio and Miguel Székely (2004) “Wage shocks and consumption variability in Mexico during
the 1990s�, Journal of Development Economics 73(1): 1-25.

Berg, Gunhild and Bilal Zia (2013) “Financial Literacy through Mainstream Media: Evaluating the Impact
of Financial Messages in a South African Soap Opera�, Mimeo. World Bank.

Bernheim, B. Douglas, Daniel M. Garrett, and Dean M. Maki (2001) “Education and Saving: The
Long Term Effects of High School Financial Curriculum Mandates� Journal of Public Economics
80(3): 435-65.

Bertrand, Marianne, Dean Karlan, Sendhil Mullainathan, Eldar Shafir and Jonathan Zinman
(2010) “What’s Advertising Content Worth? Evidence from a Consumer Credit Marketing Field
Experiment�, Quarterly Journal of Economics, 125(1): 263-306.

Brown, Alexander, Zhikang Chua and Colin Camerer (2009) “Learning and Visceral Temptation
in Dynamic Savings Experiments�, Quarterly Journal of Economics, 124(1): 197-231.

Brown, Amy and Kimberly Gartner (2007) “Early Intervention and Credit Cardholders.�

http://cfsinnovation.com/system/files/imported/managed_documents/earlyintervention.pdf
[accessed 15 March, 2013].

Bruhn, Miriam, Luciana de Souza Leão, Arianna Legovini, Rogelio Marchetti, and Bilal Zia (2013)
“Financial Education and Behavior Formation: Large-Scale Experimental Evidence from Brazil�.
Mimeo. World Bank.

Bruhn, Miriam and Inessa Love (forthcoming) “The Economic Impact of Banking the Unbanked:
Evidence from Mexico.� The Journal of Finance.

Burke, Jeremy and Kata Mihaly (2012) “Financial Literacy, Social Perception and Strategic
Default�. RAND Working Paper WR-937. March 2012.

Cai, Jing (2012) “Social networks and the Decision to Insure: Evidence from Randomized
Experiments in China� Mimeograph.




                                                 29
Cole, Shawn A., Thomas Sampson, and Bilal Zia (2009) “Financial Literacy, Financial Decisions,
and the Demand for Financial Services: Evidence from India and Indonesia.� Harvard Business
School Working Paper, No. 09-117, February.

Cole, Shawn A., and Gauri Kartini Shastry (2008) “Smart Money: The Effect of Education,
Cognitive Ability, and Financial Literacy on Financial Market Participation.� Harvard Business
School Working Paper, No. 09-071, December.

Doi, Yoko, David McKenzie, and Bilal Zia (2012) “Who You Train Matters. Identifying
complementary Effects of Financial Education on Migrant Households� World Bank Policy
Research Working Paper 6157. August.

Duflo, Esther and Emmanual Saez (2011) “The Role of Information and Social Interactions in
Retirement Plan Decisions: Evidence from a Randomized Experiment� Quarterly Journal of
Economics 118 (3) (April 13): 815-842

The Economist (2013) ““Teacher, leave them kids alone�, February 16,
http://www.economist.com/news/finance-and-economics/21571883-financial-education-has-
had-disappointing-results-past-teacher-leave-them

Gale, William G. and Ruth Levine (2010) “Financial Literacy: What Works? How Could it be More
Effective?�, Mimeograph, October.

Gerardi, Kristopher, Lorenz Goette, and Stephan Meier (2010) “Financial Literacy and Subprime
Mortgage Delinquency: Evidence from a Survey Matched to Administrative Data� Atlanta Fed
Working Paper 2010-10. April 2010.

Gibson, John, David McKenzie, and Bilal Zia (2012) “The Impact of Financial Literacy Training for
Migrants� World Bank Policy Research Working Paper 6073. May.

Government Accountability Office (2004) “The Federal Government’s role in improving financial
literacy: highlights of a GAO forum� http://www.gao.gov/assets/210/202486.pdf [accessed
02/08/2013]

Hastings, Justine and Lydia Tejeda-Ashton (2008) “Financial Literacy, Information, and Demand
Elasticity: Survey and Experimental Evidence from Mexico.� National Bureau of Economics
Working Paper No. 14538, December.

Kim, Jinhee, Thomas Garman, and Benoit Sorhaindo (2003) “Relationships among Credit
Counseling Clients’ Financial Well- Being, Financial Behaviors, Financial Stressor Events, and
Health.� Financial Counseling and Planning, 14(2): 75–87.



                                               30
Lara-Ibarra, Gabriel (2011) “The Effects of Framing on Retirement Management Decisions:
Evidence form a Field Study in Mexico� Mimeograph

Lewis, S. and Messy, F. (2012) “Financial Education, Savings and investments: An Overview�
OECD Working Papers on Finance, Insurance and Private Pensions, No. 22.

Lührmann, M., Serra-Garcia, M., & Winter, J. (2012) “The effects of financial literacy training:
Evidence from a field experiment with German high-school children�, University of Munich
Working Paper 2012-24.

Lusardi, Annamaria (2004) “Saving and the Effectiveness of Financial Education� mimeograph,
Dartmouth College, January.

Lusardi, Annamaria and Olivia S. Mitchell (2009) “How Ordinary Consumers make Complex
Economic Decisions: Financial Literacy and Retirement Readiness.� Mimeo, March.

Lusardi, Annamaria and Olivia S. Mitchell (2011) “Financial Literacy and Retirement Planning in the
United States�. CeRP Working Papers. Center for Research on Pensions and Welfare Policies, Turin
(Italy), February.

Lusardi, Annamaria and Peter Tufano (2009) “Debt Literacy, Financial Experience and
Overindebtedness.� Dartmouth College Working Paper, March.

Madrian, Brigitte and Dennis Shea (2001) “Preaching to the Converted and Converting Those
Taught: Financial Education in the Workplace.� Working Paper, University of Chicago

McKenzie, David (2006) “The consumer response to the Mexican Peso crisis�, Economic
Development and Cultural Change, 55(1): 139-172

McKenzie, David (2010) “Impact Assessments in Finance and Private Sector Development: What
have we learned and what should we learn?�, World Bank Research Observer, 25(2): 209-33.

McKenzie, David and Christopher Woodruff (2008) “Experimental Evidence on Returns to
Capital and Access to Finance in Mexico�, World Bank Economic Review 22(3): 457-82

McKenzie, David and Christopher Woodruff (2012) “What are we learning from business
training and entrepreneurship evaluations around the developing world?� World Bank Policy
Research Working Paper No. 6202.

Ponce, Alejandro, Enrique Seira, and Guillermo Zamarripa (2009) “Do consumers borrow on
their cheapest credit card? Evidence from Mexico.� Working paper available at SSRN
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1364722



                                                31
Saavedra, Ana Luisa (2012) “Developing National Strategies for financial education: the Mexican
case� Colombia-OECD-World Bank Conference on Financial Education “Progress of global
policies and practices and Latin American Experiences� http://www.banrep.gov.co/educacion-
economica/OECD/009543/documentos/saavedra_oecd_2012.pdf [accessed 01/16/2013]

Saldívar, German (2009) “Mexican National Strategy on Financial Education� SHCP
Presentation.  http://www.oecd.org/finance/financialeducation/44265373.pdf [accessed
01/16/2013]

Seshan, Ganesh and Dean Yang (2012) “Transnational household finance: A field experiment on
the cross-border impacts of financial education for migrant workers�, Mimeo. University of
Michigan.

Stango, Victor and Jonathan Zinman (2009. “Exponential Growth Bias and Household Finance.�
Journal of Finance 64(6): 2807-2849)

Treviño Garza, Luis (2011) “Financial Capabilities Measurement� Presentation at AFI�BNM
Financial                  Inclusion                 Policymakers                Forum.
http://www.bnm.gov.my/microsites/fipf2011/day2_session_5/06_FIPF%20Session%205_Luis%
20Trevino.pdf [accessed 02/08/2013]

Willis, Lauren (2011) “The Financial Education Fallacy�, American Economic Review Papers and
Proceedings 101(3): 429-34.

World Bank (2008) Finance for All? Policies and Pitfalls in Expanding Access. World Bank Policy
Research Report.

Xu, Lisa and Bilal Zia (2012) “Financial Literacy Around the World. An Overview of the Evidence
with Practical Suggestions for the Way Forward� World Bank Policy Research Working Paper
6107. June.




                                              32
                         Figure 1: Financial Literacy Training Take-Up Rates by Incentive Group

       45
                                                             39
       40
       35                                           33                                          32                            32
       30                                                                              27                            27
                                                                                                                                                                25
       25                                                                                                                                              21                                     23
                                21
   %                                                                                                                                                                                 19
       20                 18

       15
       10
        5
        0
                         No extra                  $72 now                         $36 now                          $36 later                      Free                           Testimonials
                         incentive                                                                                                            transportation

                                      Full treatment group                                  Only individuals who could be reached


Note: Full treatment group includes 1,751 individuals. The sub-sample who could be reached to
offer them the treatment includes 1,485 individuals.



                                                   Figure 2: Median Savings Account Balance

                        3,500

                        3,000

                        2,500
        Mexican Pesos




                        2,000

                        1,500

                        1,000

                         500

                           -
                                 Dec-10




                                                                              May-11
                                                                     Apr-11




                                                                                                                                              Dec-11




                                                                                                                                                                                           May-12
                                          Jan-11


                                                            Mar-11




                                                                                       Jun-11


                                                                                                         Aug-11




                                                                                                                                                                                  Apr-12
                                                                                                                            Oct-11
                                                                                                                                     Nov-11


                                                                                                                                                       Jan-12


                                                                                                                                                                         Mar-12
                                                   Feb-11




                                                                                                Jul-11


                                                                                                                   Sep-11




                                                                                                                                                                Feb-12




                                                                     Control group                                Treatment group


Note: Administrative data. Sample includes 470 individuals who are clients of our partner
financial institution.

                                                                                                 33
                                                                     Figure 3: Median Credit Card Balance

                       9,000
                       8,000
                       7,000
       Mexican Pesos




                       6,000
                       5,000
                       4,000
                       3,000
                       2,000
                       1,000
                             -
                                          Dec-10




                                                                                           May-11




                                                                                                                                                                                                                 May-12
                                                                                  Apr-11




                                                                                                                                                               Dec-11
                                                     Jan-11


                                                                         Mar-11




                                                                                                    Jun-11


                                                                                                                          Aug-11




                                                                                                                                                                                                      Apr-12
                                                                                                                                             Oct-11
                                                                                                                                                      Nov-11


                                                                                                                                                                        Jan-12


                                                                                                                                                                                           Mar-12
                                                                Feb-11




                                                                                                               Jul-11


                                                                                                                                    Sep-11




                                                                                                                                                                                 Feb-12
                                                                                  Control group                                    Treatment group


Note: Administrative data. Sample includes 470 individuals who are clients of our partner
financial institution.

                         Figure 4: Average Percentage of Credit Card Debt Paid Off Each Month

               35.0
               30.0
               25.0
               20.0
      %
               15.0
               10.0
                       5.0
                       0.0
                                 Dec-10




                                                                                      May-11




                                                                                                                                                               Dec-11




                                                                                                                                                                                                                    May-12
                                                                             Apr-11
                                            Jan-11



                                                                   Mar-11




                                                                                                Jun-11



                                                                                                                        Aug-11




                                                                                                                                                                                                        Apr-12
                                                                                                                                             Oct-11

                                                                                                                                                      Nov-11



                                                                                                                                                                        Jan-12



                                                                                                                                                                                             Mar-12
                                                       Feb-11




                                                                                                             Jul-11



                                                                                                                                   Sep-11




                                                                                                                                                                                  Feb-12




                                                                                  Control group                                  Treatment group


Note: Administrative data. Sample includes 470 individuals who are clients of our partner
financial institution.




                                                                                                                 34
                                                 Table 1: Confirming Randomization Using Baseline Data

                                                                  Full Sample in Baseline   Sample Interviewed at Follow-up       Administrative Data Sample
                                                                  Control Treatment           Control        Treatment            Control       Treatment
                                                                   Mean      Difference        Mean          Difference            Mean         Difference
Stratification Variables
Baseline survey conducted in branch                                0.35       -0.0058          0.37            -0.0045              0.62         -0.0005
Client of partner financial institution (vs. other institution)    0.48       -0.0010          0.48             0.0015
Made savings deposit during past month                             0.64       0.0012           0.64             0.0011              0.70          0.0534
Has credit card                                                    0.41       -0.0039          0.42             0.0161              0.56         -0.0005
Paid more than credit card minimum in all past 6 months1           0.51        0.0172          0.52            0.0169               0.56          0.0085
Has bachelor’s degree or higher                                    0.40        0.0016          0.41            0.0195               0.45         -0.0386
Female                                                             0.47        0.0064          0.50            0.0115               0.48          0.0336

Other Baseline Variables
Age                                                               32.69       0.6308           32.97           0.7372              36.14          0.9789
Occupation is employee                                            0.51        -0.0171          0.49            0.0031              0.50          -0.0282
Paid credit card late in past 6 months1                            0.23       0.0124           0.22             0.0214              0.21          0.0087
Monthly household income is above MXP 6,500                        0.64       -0.0072          0.63            -0.0111              0.67          0.0367
Monthly household expenditure is above MXP 6,500                   0.54       -0.0081          0.54            -0.0141              0.54         -0.0394

Sample Size                                                      1090        1088               814              772                 243            227
Notes: *, **, and *** indicate statistically different from control mean at the 10, 5 and 1% levels respectively. Differences for baseline variables not
used in the stratification control for randomization strata. Administrative data sample includes clients of our partner financial institution that were
found in their database.
1
    Conditional on having a credit card




                                                                                35
    Table 2: Impact of Incentive Treatments on Take-Up in Treatment Group


                                                      Dependent variable:
                                                         Attended course
                                                     (1)                (2)
$72 now dummy                                    0.1480***          0.1758***
                                                  (0.0352)           (0.0401)
$36 now dummy                                    0.0931***          0.1123***
                                                  (0.0343)           (0.0393)
$36 later dummy                                  0.0918***          0.1110***
                                                  (0.0343)           (0.0392)
Free transportation dummy                          0.0330             0.0407
                                                  (0.0326)           (0.0376)
Testimonials dummy                                 0.0142             0.0153
                                                  (0.0320)           (0.0367)

F-test p-value: $72 now = $36 now                  0.1435             0.1357
F-test p-value: $36 now = $36 later                0.9725             0.9761
Control group mean of outcome variable             0.1815             0.2129

Observations                                        1751               1457
                                                                 Only individuals
                                               Full treatment
Sample                                                            who could be
                                                    group
                                                                    reached
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistical
significance at the 10, 5 and 1 percent levels respectively. OLS regressions of a
dummy for having attended the course on a set of dummies indicating to which
incentive group the individual was randomly assigned (the omitted category is the
control group, i.e. received no extra incentives). Regressions control for
randomization strata dummies.




                                       36
                Table 3: Determinants of Program Take-Up in Treatment Group

                                                                    Dependent variable:
                                                                  Attended    Attended course
                                                                course w/o        w/ or w/o
                                                                 incentives       incentives
                                                                      (1)              (2)
Baseline survey conducted in branch                                 0.0437           0.0250
                                                                  (0.0298)         (0.0330)
Client of partner financial institution (vs. other institution)     0.0232         0.0540*
                                                                  (0.0271)         (0.0300)
Made savings deposit during past month                              0.0284           0.0359
                                                                  (0.0258)         (0.0285)
Has credit card                                                    -0.0450          -0.0296
                                                                  (0.0355)         (0.0393)
Paid more than credit card minimum in all past 6 months             0.0570           0.0309
                                                                  (0.0392)         (0.0435)
Has bachelor’s degree or higher                                 0.1189***         0.1420***
                                                                  (0.0254)         (0.0282)
Female                                                            0.0445*            0.0405
                                                                  (0.0245)         (0.0272)
Age                                                             0.0035***         0.0039***
                                                                  (0.0011)         (0.0012)
Occupation is employee                                             -0.0127          -0.0137
                                                                  (0.0243)         (0.0270)
Paid credit card late in past 6 months                              0.0524           0.0503
                                                                  (0.0454)         (0.0467)
Observations                                                         1081             1081
Mean of outcome variable                                            0.2077           0.2849
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistical significance
at the 10, 5 and 1 percent levels respectively.




                                                  37
                                  Table 4: Impact on Financial Knowledge

                                                                                       ITT              LATE
                                                              Sample Control       Treatment         Treatment
                                                                Size  Mean         Difference        Difference
Panel A: Knowledge index and components
Knowledge index (average of 8 components below)                1586       0.31     0.0307***         0.0871***
                                                                                    (0.0094)          (0.0261)
(1) Knows what UDI is ("Unidad de Inversion")1                 1580       0.10       0.0044            0.0125
                                                                                    (0.0150)          (0.0426)
(2) Knows deposit insurance exists up to 400,000 UDIs          1578       0.13     0.0732***         0.2073***
                                                                                    (0.0188)          (0.0533)
(3) Knows what a credit report is                              1575       0.39     0.0518**          0.1464**
                                                                                    (0.0245)          (0.0684)
(4) Knows credit card cycle is 30 days                         1568       0.46       0.0190            0.0535
                                                                                    (0.0251)          (0.0701)
(5) Knows they have 20 days to pay credit card w/o
interest                                                       1568       0.12       0.0143            0.0402
                                                                                    (0.0168)          (0.0472)
(6) Knows that what CAT is ("Costo Anual Total")2              1557       0.24      0.0369*          0.1036*
                                                                                    (0.0216)         (0.0602)
(7) Knows what an AFORE (pension fund) is                      1559       0.72      0.0499**         0.1409**
                                                                                    (0.0219)         (0.0613)
(8) Knows retirement age is 65                                 1551       0.29       0.0253           0.0714
                                                                                    (0.0234)         (0.0659)
Panel B: Self-assessed financial literacy
Says their financial knowledge is satisfactory or higher       1550       0.58      0.0537**         0.1500**
                                                                                    (0.0248)         (0.0683)
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistically different from control
mean at the 10, 5 and 1 percent levels respectively, after controlling for randomization strata and month
of follow-up interview dummies.
1
    Unidad de Inversion (UDI) is an inflation adjusting currency unit.
2
    Costo Annual Total (CAT) is total annual cost of credit, including all interest rates and fees




                                                        38
                           Table 5: Impact on Savings Behavior and Outcomes

                                                                                           ITT             LATE
                                                                Sample     Control     Treatment        Treatment
                                                                  Size      Mean       Difference       Difference
Panel A: Savings behavior index and components
Savings behavior index (avg. of 5 components below)               1586       0.68        0.0133            0.0376
                                                                                        (0.0124)         (0.0348)
(1) Checks financial institution transactions regularly           1586       0.69       -0.0235           -0.0666
                                                                                        (0.0226)         (0.0644)
(2) Keeps track of expenses                                       1586       0.79        0.0072            0.0204
                                                                                        (0.0206)         (0.0582)
(3) Makes a budget                                                1585       0.77        0.0264            0.0748
                                                                                        (0.0211)         (0.0596)
(4) Has a savings goal                                            1582       0.57        0.0130            0.0367
                                                                                        (0.0250)         (0.0705)
(5) Cut expenses in past 3 months                                 1584       0.59       0.0428*          0.1212*
                                                                                        (0.0247)         (0.0698)
Panel B: Savings outcomes index and components
Savings outcomes index (avg. of 3 components below)               1586       0.65       0.0335**         0.0948**
                                                                                        (0.0147)         (0.0414)
Has any type of savings1                                          1586       0.80        0.0288           0.0814
                                                                                        (0.0200)         (0.0566)
Saved more than zero during past 6 months                         1413       0.83        0.0293           0.0800
                                                                                        (0.0192)         (0.0524)
Saves more each month than a year ago                             1547       0.36        0.0408          0.1151*
                                                                                        (0.0250)         (0.0697)
Panel C: Savings outcomes controlling for monetary
incentives
Savings outcomes index (avg. of 3 components below)               1586       0.65       0.0268*          0.0902*
                                                                                        (0.0151)         (0.0506)
Has any type of savings1                                          1586       0.80        0.0187           0.0629
                                                                                        (0.0205)         (0.0693)
Saved more than zero during past 6 months                         1413       0.83        0.0253           0.0823
                                                                                        (0.0198)         (0.0642)
Saves more each month than a year ago                             1547       0.36        0.0347           0.1161
                                                                                        (0.0255)         (0.0845)
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistically different from control mean at
the 10, 5 and 1 percent levels respectively, after controlling for randomization strata and month of follow-up
interview dummies. Regressions in Panel C additionally include a dummy for whether the individual received a
                                                                                1
monetary incentive payment for participation in the financial literacy course. Includes savings account, caja de
ahorro, tanda and other non-retirement savings




                                                         39
                           Table 6: Impact on Retirement Savings Behavior

                                                                                   ITT           LATE
                                                           Sample    Control    Treatment     Treatment
                                                            Size      Mean      Difference    Difference
Has a pension fund                                          1471      0.57       -0.0031        -0.0088
                                                                                 (0.0263)      (0.0739)
Panel A: Retirement savings behavior index and
components
Retirement savings behavior index (avg. of 3
components below)                                           1471      0.20        0.0114        0.0320
                                                                                 (0.0144)      (0.0404)
Pension fund administrator choice based on fees or
returns                                                     1471      0.16        0.0058        0.0162
                                                                                 (0.0195)      (0.0547)
Checks pension fund statement                               1467      0.34        0.0340        0.0958
                                                                                 (0.0250)      (0.0704)
Has calculated how much money will need upon
retirement                                                  1465      0.11       -0.0026        -0.0074
                                                                                 (0.0158)      (0.0447)
Panel B: Retirement savings outcomes
Is saving money for retirement                              1470      0.17         0.0122        0.0343
                                                                                  (0.0199)      (0.0559)
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistically different from
control mean at the 10, 5 and 1 percent levels respectively, after controlling for randomization strata
and month of follow-up interview dummies. Variables in panels A and B are based on questions that
were only answered by individuals who have a pension fund. To account for potential selection bias,
we fill these variables in with "0" for individuals who do not have a pension fund.




                                                   40
                          Table 7: Impact on Credit Card Behavior and Outcomes

                                                                                     ITT                LATE
                                                                   Sample Control Treatment          Treatment
                                                                     Size  Mean Difference           Difference

Currently has at least one credit card                              1560      0.48      -0.0287        -0.0811
                                                                                        (0.0210)      (0.0596)
Panel A: Credit card behavior index and components
Credit card behavior index (avg. of z-scores of 6 components
below)                                                              1556      0.00      -0.0233        -0.0660
                                                                                        (0.0229)      (0.0652)
(1) Knows credit limit                                              1550      0.45      -0.0266        -0.0756
                                                                                        (0.0209)      (0.0594)
(2) Knows interest rate                                             1519      0.20      -0.0017        -0.0050
                                                                                        (0.0193)      (0.0552)
(3) Checks statement every month                                    1543      0.41      -0.0278        -0.0786
                                                                                        (0.0209)      (0.0594)
(4) Fraction of past 6 months where paid balance in full            1536      0.20      -0.0180        -0.0505
                                                                                        (0.0170)      (0.0479)
(5) Fraction of past 6 months where made only the minimum
payment1                                                            1539      0.12       0.0022        0.0062
                                                                                        (0.0138)      (0.0389)
(6) Fraction of past 6 months where got cash through the
credit card1                                                        1540      0.07      -0.0094        -0.0267
                                                                                        (0.0098)      (0.0281)
Panel B: Credit card outcomes index and components
Credit card outcomes index (avg. of z-scores of 3 components
below)                                                              1554      0.00       0.0434        0.1228
                                                                                        (0.0416)      (0.1184)
(1) Issuer blocked credit card during past 6 months                 1547      0.04       0.0009        0.0026
                                                                                        (0.0096)      (0.0270)
(2) Fraction of past 6 months where was charged late payment
fees                                                                1546      0.03       0.0102        0.0289
                                                                                        (0.0064)      (0.0183)
(3) Fraction of past 6 months where was charged overdraft
fees                                                                1545      0.01       0.0026        0.0075
                                                                                        (0.0034)      (0.0098)
Notes: All variables in Panel A and B refer to the most frequently used credit card. Robust standard errors in
parentheses. *, **, and *** indicate statistically different from control mean at the 10, 5 and 1 percent levels
respectively, after controlling for randomization strata and month of follow-up interview dummies.
1
Included in credit card behavior index with negative sign.




                                                       41
                        Table 8: Impact on Loan Behavior and Outcomes

                                                                                      ITT           LATE
                                                            Sample     Control    Treatment      Treatment
                                                              Size      Mean      Difference     Difference
Panel A: Loan behavior index and components
Loan behavior index (avg. of 3 components below)              1570       0.15       0.0075          0.0212
                                                                                   (0.0118)       (0.0334)
Applied for a loan from any source during past 6 months      1564        0.23      -0.0074         -0.0208
                                                                                   (0.0210)       (0.0594)
Went to a pawn shop to get credit during past 6 months       1568        0.10       0.0054          0.0152
                                                                                   (0.0152)       (0.0429)
Stopped servicing outstanding debt during past 6
months                                                       1434        0.13       0.0205         0.0591
                                                                                   (0.0180)       (0.0523)
Panel B: Loan outcomes index and components
Loan outcomes index (avg. of z-scores of 2 components
below)                                                       1560       -0.01      -0.0132         -0.0372
                                                                                   (0.0427)       (0.1206)
Currently has a loan (from any source)                       1555        0.33      -0.0058         -0.0165
                                                                                   (0.0234)       (0.0660)
Total outstanding debt as percentage of annual income        1209       15.38      -0.6563         -1.7753
                                                                                   (1.1899)       (3.2220)
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistically different from control
mean at the 10, 5 and 1 percent levels respectively, after controlling for randomization strata and month of
follow-up interview dummies.




                                                   42
                                                   Table 9: Heterogeneous Treatment Effects (ITT)
                                                                                          Dependent variable: Index of
                                                                                             Retirement
                                                                   Savings      Savings                     Credit card    Credit card     Loan          Loan
                                                   Knowledge                                   savings
                                                                  behavior     outcomes                      behavior      outcomes      behavior      outcomes
                                                                                              behavior
Panel A: Type of financial institution
Treatment group dummy                               0.0315**      0.0095       0.0343*         0.0099        -0.0550*        0.0157       0.0065       -0.0066
                                                    (0.0130)     (0.0174)      (0.0199)       (0.0194)       (0.0294)       (0.0559)     (0.0161)      (0.0548)
Treatment*Client of partner institution              -0.0016      0.0078        -0.0018        0.0030         0.0658         0.0572       0.0021       -0.0135
                                                    (0.0188)     (0.0246)      (0.0295)       (0.0287)       (0.0460)       (0.0836)     (0.0237)      (0.0859)

F-test p-value: Treatment + Interaction = 0           0.028        0.321         0.135          0.543          0.761         0.240         0.622        0.761

Panel B: Gender
Treatment group dummy                                0.0182       0.0057      0.0546***        -0.0035       -0.0345         0.0587        0.0144       0.0414
                                                    (0.0134)     (0.0169)      (0.0205)       (0.0215)       (0.0331)       (0.0564)     (0.0171)      (0.0626)
Treatment*Female                                     0.0250       0.0150        -0.0421        0.0297         0.0223        -0.0304       -0.0137      -0.1086
                                                    (0.0188)     (0.0246)      (0.0295)       (0.0287)       (0.0459)       (0.0842)     (0.0237)      (0.0858)

F-test p-value: Treatment + Interaction = 0           0.001        0.249         0.554          0.172          0.702         0.648         0.967        0.249

Panel C: Education
Treatment group dummy                               0.0288**     -0.0067        0.0148         -0.0139       -0.0174         0.0585       0.0017       -0.0421
                                                    (0.0116)     (0.0164)      (0.0192)       (0.0179)       (0.0284)       (0.0575)     (0.0161)      (0.0546)
Treatment*Has bachelor or higher                     0.0046      0.0486*        0.0455        0.0613**       -0.0144        -0.0369       0.0141        0.0702
                                                    (0.0196)     (0.0248)      (0.0300)       (0.0297)       (0.0477)       (0.0819)     (0.0237)      (0.0877)

F-test p-value: Treatment + Interaction = 0           0.034        0.025         0.009          0.046          0.406         0.713         0.363        0.681

Observations                                          1586         1586          1586           1471           1556          1554          1570          1560
Notes: Robust standard errors in parentheses. *, **, and *** indicate statistically different from control mean at the 10, 5 and 1 percent levels respectively,
after controlling for randomization strata and month of follow-up interview dummies.


                                                                                43
                                                                  Appendix A: Project Timeline

                                           2010                               2011                                             2012
Project Stage                      A   S    O     N   D   J   F   M   A   M   J    J   A   S   O   N   D   J   F   M   A   M   J   J   A   S   O   N   D
Project setup and preparation      ï?µ   ï?µ    ï?µ     ï?µ   ï?µ   ï?µ
Recruitment: Mailing Campaign                             ï?µ   ï?µ
Recruitment: Online Campaign                                  ï?µ   ï?µ   ï?µ
Recruitment: Street/Branches                                          ï?µ   ï?µ   ï?µ    ï?µ
Invitation to course (Treatment)                                                       ï?µ   ï?µ   ï?µ   ï?µ
Follow-up Survey                                                                                           ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ
Analysis                                                                                                   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ
Administrative data                                   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ    ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ   ï?µ




                                                                              44