Learning Center
Plans & pricing Sign in
Sign Out

111023 Savings as Insurance


									         Insurance Through Savings Accounts
                    Evidence from a Randomized Field Experiment
                   among Low-Income Micro-Entrepreneurs in Chile∗


                     Ronald Abraham†                 Felipe Kast‡             Dina Pomeranz§

                                                   October 2011


           Poverty is often characterized not only by low average income, but also by highly
       variable income and expenditures, and a lack of access to insurance services that
       can help smooth consumption. We investigate whether access to a formal savings
       account can provide a vehicle for self-insurance, by allowing participants to build
       a buffer stock of precautionary savings. In a randomized field experiment in Chile,
       about 3000 low-income micro-entrepreneurs are provided access to a formal savings
       account with no minimum balance or maintenance fees. Evaluating the impact after
       one year, we find that access to such accounts helps participants alleviate the burden
       of economic shocks, both objectively and subjectively. Participants with access to a
       savings account have less informal debt, fewer outstanding payments, and less often
       need to reduce consumption due to economic difficulties. Subjectively, they report
       being less worried about their financial future, and evaluate their recent economic
       situation as less severe. We therefore conclude that formal savings accounts can
       be an effective vehicle to provide a means for consumption smoothing in contexts
       where many other forms of insurance are lacking.

      We are grateful to Fondo Esperanza, Banco Credichile and Microdatos for outstanding collaboration in the implemen-
tation process. We thank Shawn Cole, David Cutler, Michael Kremer, Nicola Fuchs-Schuendeln, Edward Glaeser, Jessica
Goldberg, Daniel Hojman, Lakshmi Iyer, Sandy Jencks, David Laibson, Dean Karlan, Lawrence Katz, Dan Levy, Jeffrey
Liebman, Stephan Litschig, Brigitte Madrian, Sendhil Mullainathan, Rohini Pande, Richard Zeckhauser and participants at
the Harvard Development Lunch for helpful comments and discussions. This project would not have been possible without
the generous support by the following institutions: the Ford Foundation, the Lab for Economic Applications and Policy
(LEAP) at Harvard University and the Woman and Public Policy Program at the Harvard Kennedy School, the David
Rockefeller Center for Latin American Studies, and the Russell Sage Foundation Small Grants Program. The experiment
was carried out in accordance with Harvard GSAS IRB approval.
      Intellectual Capital Advisory Services, Road No. 12, Banjara Hills, Hyderabad, AP 500034, India
      Pontificia Universidad Catlica de Chile
      Harvard Business School, Rock Center 213, Soldiers Field, Boston, MA 02163;

1    Introduction

At low levels of income, economic shocks can have devastating effects. Resources may
fall below subsistence levels, which leads to potentially dire impacts, or costly measures
to avoid them. The severity of this issue is compounded by the fact that poor people are
often faced with highly variable income streams and expenditure shocks. Correspondingly,
worry and anxiety about their economic future often mark the lives of the poor (e.g.
Collins et al., 2009). Building a buffer stock of precautionary savings that facilitates
smoothing of economics shocks therefore has potentially large benefits on their wellbeing.

    While the literature on precautionary savings and buffer stocks has long acknowledged
that incomplete markets may lead to borrowing constraints and an absence of contingent
assets or insurance services, access to a riskless bond or a formal savings account to build
such precautionary savings is usually taken as given (e.g. Zame, 1993). An emerging
literature shows, however, that many individuals in developing countries may not only be
credit constrained, but also savings constrained (e.g. Burgess and Pande, 2005, Ashraf et
al., 2006; Brune et al., 2011; Atkinson et al., 2010; Dupas and Robinson, 2011).

    This paper provides what is to our knowledge the first micro-empirical evidence show-
ing that access to a savings vehicle can facilitate building of precautionary savings, and
reduce participants’ vulnerability to economic shocks, as well as their use of other, po-
tentially more costly smoothing mechanisms. We conduct a randomized field experiment
among around 3000 low-income micro-entrepreneurs in Chile. We find that participants
with access to a formal savings account are less worried about their financial future,
evaluate their recent economic situation as less severe and less often have to cut back
consumption due to economic difficulties. In addition, they report lower levels of informal
debt and fewer arrears on outstanding payments.

    Our exceptional access to detailed savings data of participants also allows us to

track the timing of all deposits and withdrawals. During the follow-up survey, we elicit
information about whether participants experienced any of several categories of economic
shocks or important life events, and if so in which month. Matching this information to
the savings data reveals that participants’ deposit patterns closely match the timing of
these shocks, while no significant correlation is detected with withdrawals.

    This paper contributes both to the growing literature on savings and on (mi-
cro)insurance in developing countries. The literature on savings in developing countries
has followed two strands: a first series of papers studies mechanisms aimed at helping
individuals increase their savings (e.g. Ashraf et al., 2006a Ashraf et al., 2006b; Atkinson,
et al., 2010; Brune et al., 2011; Kast, Meier and Pomeranz, 2011). A second strand of
the literature aims at assessing the impacts of increasing savings or of having access to
a formal savings account on outcomes such as reduction in poverty (Pande and Burgess,
2005), increase in investment and income (Brune et al., 2011; Dupas and Robinson, 2011)
and female empowerment in intra-household bargaining (Ashraf et al., 2010). Several
of the savings vehicles analyzed in these studies have a withdrawal commitment compo-
nent, which impedes discretionary withdrawals in times of need, and may therefore not
be appropriate for building a buffer stock for general precautionary savings. This paper,
in contrast, analyzes a savings account that is fully liquid on the withdrawal margin,
in order to study the potential of formal savings accounts to provide insurance through
precautionary savings.

    Much of the literature on insurance in developing countries has focused on analyz-
ing to what degree individuals are able to smooth shocks even without access to formal
insurance products, for example through loans from their social network (e.g. Townsend,
1994; Murdoch, 1995; Kinnan and Townsend, 2010), and on understanding the reasons
why there are very few functioning formal insurance markets in developing countries,
even for risks that seem to present relatively little problems of moral hazard or adverse

selection, such as weather risks (Cole et al., 2009.) Given the limitations of informal insur-
ance mechanisms, and the challenges faced by formal insurance products, self-insurance
through savings has the potential to play an important role. While the precautionary sav-
ings motive has been long established as a key determinant of savings behavior (Browning
and Lusardi, 1996), this paper provides what is to our knowledge the first empirical evi-
dence establishing the potential of savings to reduce the impact of economic shocks and
improve the wellbeing of the poor.

      The remainder of the paper is organized as follows: Section 2 provides information
about background, data and the study design; Section 3 presents our main empirical
findings; and Section 4 concludes.

2      Background, Data and Study Design

2.1     Background and Data

This paper is based on a randomized field experiment with around 3000 low-income micro-
entrepreneurs in Chile. The study was conducted in collaboration with Fondo Esperanza
(FE), a Chilean microfinance institution, and Banco Credichile (BC), a large commercial
bank. FE’s members are self-employed micro-entrepreneurs (e.g. street vendors, cosmetic
saleswomen), many of whom work in the informal sector. About 90% of FE members are
women, and most live and work in urban areas. FE provides micro-loans to its clients in
3-month cycles, repayment of which is monitored in weekly or biweekly group meetings.

      The credit disbursement and repayment is on a very rigid schedule, and consequently
cannot be used as insurance for emergencies or for unexpected income or expenditure
shocks. In focus groups, many participants therefore expressed the desire for a savings
vehicle that would allow them to build a buffer stock of precautionary savings. The savings

accounts offered to FE’s clients as part of the intervention are with the commercial Bank
Banco Credichile, because FE is not legally licensed to hold savings deposits.

      We draw on three different sources of data. Data on deposits and withdrawals of
those who took up savings accounts is directly obtained from BC, the bank holding the
accounts. It includes information on which participants open an account, whether they
use the account after opening it, and the dates and amounts of all withdrawals and
deposits made after opening the account. The second source of data comes from FE’s
administrative files, which includes information on each participant’s estimated household
size and income and years of education.

      Finally, we complement these two sources of administrative data with an extensive
baseline and follow-up survey. The baseline survey was conducted prior to the introduction
of the savings accounts in April-May 2008 during one of the group meetings. The follow-
up survey, in June-July 2009, was administrated at individuals’ home or work place, so
that those participants who had left FE in the meantime could still be reached. The
surveys include detailed questions about participants’ savings and debt, their economic
situation, recent economic difficulties and consumption patterns, as well as subjective
measures such as participants’ worry about their financial future, regret about not having
saved more, being intimidated by banks, risk aversion and time preferences. Centro de
Microdatos from the University of Chile administered all surveys.

2.2     Experimental Design and Experimental Specification

Prior to any intervention, the baseline survey was conducted in the 196 groups of the
micro-credit organization, Fondo Esperanza (FE). The universe of study participants
consists of the 4175 members who were present in the meeting when the baseline survey
was administered. Two third of the groups were subsequently randomly elected to be

offered a savings account. The accounts were introduced during group meetings in the
weeks following the baseline survey.

    The randomization was conducted at the group level, such that all members of each
group received the same treatment. While take-up of the account was completely vol-
untary (and ended up being about 50%), interested participants were also offered the
opportunity to go open the savings account together with their peers, to overcome the
frequently expressed sentiment of feeling intimidated by entering into a bank.

    The standard savings accounts offered to participants have a real interest rate of
0.3% (similar to the highest available alternative in the Chilean market). The accounts
are particularly suited for the low-income population of this study in that they have no
maintenance fees, no minimum balance, and only a two-dollar minimum opening deposit.
Savings in the accounts are fully liquid for withdrawal at any time. These conditions were
guaranteed for a minimum of two years.

    A subgroup of one quarter of participating groups was randomly assigned to receive
a preferential interest rate of 5%, and in half of the groups, the savings accounts were
accompanied by a peer group commitment mechanism, (see Kast, Meier, and Pomeranz
2011 for an evaluation of the differential savings behavior resulting from these different
sub-treatments). The 5% higher interest rate only very modestly affected savings, if at
all, while the peer group commitment device had a strong positive effect on the number of
deposits and the amounts saved in the accounts. Due to sample size limitations, analyzing
the impacts of the individual treatments is relatively noisy. The main analysis of this paper
therefore focuses on the average impact of having access to a savings account with any of
these characteristics. In Section 3.4 below, we aim to disentangle the differential effects
for those with and without the peer group support.

    About one year after opening of the accounts, the follow-up survey was conducted

to collect data on the impacts of the accounts. To analyze the effect of having a savings
account on various outcomes of interest, we estimate a simple difference in difference
specification comparing those in the treatment group to those in the control group at the
time of the baseline and the follow-up survey:

                           Yit = α + βAccountit + γi + µt +       it                          (1)

where Yit is the outcome variable of interest. Accountit is a dummy variable that takes on
the value 1 if individual i is in the treatment group and period t is the treatment period.
Individual and time fixed effects are represented by γi and µt respectively, while       it   is the
error term.

     We use this specification to analyze the impact of savings accounts on a series of out-
comes, such as participants’ economic difficulties and worry about their financial future,
their need to cut back their consumption as well as their borrowing and lending behavior.
We then analyze whether the savings behavior of individuals who opened an account with
Banco Credichile is correlated to events in their lives that trigger economic shocks. In the
follow-up survey, we elicit information about whether participants experienced any of a
series of possible life events (e.g. loss of a job, birth of a child, etc.), and if so in which
month, and we match this information with data about the account use. We use panel
regressions to estimate whether deposits and withdrawals were affected in the month of
the life event:

                          Yit = α + LifeEventit β + γi + ηt +      it                         (2)

where Yit is a dummy variable indicating whether participant i made a deposit or a
withdrawal in month t. Lif eEventit is a vector of life events for individual i in month t.
Individual and month fixed effects are represented by γi and ηt respectively, while            it   is

the error term.

2.3     Baseline Summary Statistics, Balance of Randomization

        and Use of the Accounts

Table 1 presents baseline summary statistics for the 4175 study participants. As expected,
given the random assignment, characteristics in the treatment group are not statistically
significantly different from the control group.

      In the control group, participants are on average 43.3 years old and have 9.8 years
of schooling. Approximately 31% already had a savings account prior to the study. The
mean monthly income per capita is 116,140 Chilean pesos (about 230 USD), total formal
and informal savings is 63,302 pesos (about 130 USD), and total debt, including the
micro-loan from FE, is 454,872 pesos (about 900 USD). While income is expressed in
per capita terms (household income divided by number of household members, which is
4.3 on average), savings and debt may be the combined for several household members,
including participants’ children. The larger amounts of debt compared to savings is not
surprising given that participants are entrepreneurs and most of their debt is backed
up by inventories and future sales. On a question on whether participants are worried
about their financial future, the mean score was 2.9 out of 4, and evaluating their recent
economic difficulty, participants indicate on average around 5 on a scale of 10.

      Among those offered access to a savings account, the take-up rate was 53%. One
year after opening the account, at the time of the impact evaluation, the average savings
balance by those who had opened an account was about 14,000 pesos. The average
number of deposits in the course of the year was a little over 2, and the average number
of withdrawals was only 0.7.

3      Results

3.1     Impact on worry and self-assessed economic difficulties

The first set of impacts that we analyze is whether participants experienced an overall
insurance effect, both forward and backward looking. We measure the backward looking
effect by asking study participants how much they agreed with the statement that they
were worried about their economic future. Participants chose from four options ranging
from 1 for ’very much disagree’ to 4 for ’very much agree’.

      One year after receiving a savings account, participants in the treatment group were
scored 0.11 points less worried than those in the treatment group (see Table 2). This
difference is statistically significant at the 10% level. Given that this is in response to a
broad question on economic wellbeing, something affected by a host of factors other than
having access to precautionary savings, this result is striking.

      The lower levels of worry about their future might be a result of participants’ expe-
rience of the recent past. In order to capture the retrospective assessment of economic
difficulties, we asked participants, ”In sum, thinking about all economic difficulties in the
last 3 months, how difficult was the situation for you?” The answers ranged from 1 stand-
ing for ’not difficult at all’, to 10 for ’very difficult’. One year after receiving an account,
the treatment group scored 0.24 points lower than the control group. This difference is
statistically significant at the 10% level as well. Therefore the individuals in the treatment
group have a qualitatively better assessment about their economic past.

      We therefore conclude that the treatment increased participants’ economic well-being,
both in terms of improving their self-assessment about their recent economic difficulties
and their outlook to the future. Reducing such economic difficulties and increasing peace
of mind about the future are some of the central benefits that one would expect of an

effective insurance mechanism, so it seems that the savings accounts are fulfilling this
important function.

      We now test for direct evidence of a reduction in fluctuations of consumption due to
economic difficulties for those who were offered a savings account. In order to be able to
test for this, we asked participants whether in the last three months they had to cut back
certain forms of consumption due to economic difficulties. We asked in turn about the
following categories of consumption: meals, meat, medicines, school supplies, clothing,
school snacks, public transport and eating out. This provides us with a series of binary
outcome variables indicating whether the participant had to cut back on consumption for
the category in question.

      Table 4 reports these results. Each represents a different item of consumption, as
indicated by the column heading. The direction of almost all the coefficients on Accountit
is negative, indicating lesser need to cut back consumption for those who have access to
an account. However, only 2 coefficients out of 8 are statistically significant at the 10%
level. A F-test for joint significance of these coefficients provides a p-value of 0.22. Hence,
while there is some evidence of a reduction in consumption variation, the results of all
possible consumption categories taken together is not statistically significant at the 10%

3.2      Impact on lending and borrowing practices

One way in which precautionary savings may affect the way individuals deal with economic
shocks is to replace other, potentially more costly sources to smoothen consumption. In
this sector, we therefore look whether access to a formal savings account affects partici-
pants lending and borrowing practices.

      First, we look at whether participants report less outstanding debt. We capture two

categories of such debt: informal borrowing from relatives and friends, and arrears in
payments to institutions such as utility providers and educational institutions.1 The re-
gression results in Table 3 show that individuals in the treatment group are 6.3 percentage
points less likely to owe money to their relatives or friends (significant at the 1% level)
and 4.2 percentage points less likely to owe money to institutions than the control group
(significant at the 10% level).

       The direction of both these coefficients makes intuitive sense. People often resort to
small informal loans from relatives and friends when they are cash constrained or when
they need money urgently. Those who have access to a savings account will be able to
meet some of these contingencies by dipping into their savings and not having to undertake
an informal loan. Similarly having a pool of savings reduces the likelihood of arrears in
payments to institutions.

       Having a savings account may not only affect participants’ debt, but also their lending
behavior to others. On the one hand, it could lead participants to become less generous
in providing loans to their social network, since they now depend less on loans from their
network for insurance purposes. In addition, an important barrier for savings in low-
income communities is that individuals are not able to refuse loans or gifts to relatives
and friends who ask for small help to manage their economic needs (e.g. Dupas and
Robinson, 2011; Brune, Gine, Goldberg and Yang, 2011). The use of a savings account
may allow individuals to hide their savings and therefore reduce exposure to such requests.
On the other hand, if having a savings account increases total savings or makes the savings
more visible to the social network, participants with access to a savings account may be
better able or more pressured to extend small loans to others.
     Since more specific questions tend to increase the precision of survey responses, these aggregate
outcome variables were each constructed based on a series of more specific underlying questions of the
following format: ”Do you owe money to X? If yes, how much?” with ’X’ successively being replaced by
your sister, brother, father, aunt, etc. We aggregated these responses into a dummy variable indicating
whether the participants had outstanding debt to anyone in one of the two categories.

      To test for this question, we asked each participant whether they were owed money
by their relatives, friends or business contacts.2 Table 3 shows a slightly negative but
not statistically significant effect. So, consistent with the findings of Chandrasekhar et
al. (2010), we do not find evidence that access to a savings account crowds out lending
to others. It is, however, conceivable that the results would have been different in the
long run in a context where participants could be confident that they would be able to
continue to rely on savings in the future. The savings accounts in this study were only
guaranteed for two years.

3.3     Life Events and Savings Behavior

In this section we explore whether life events - such as the birth of a child or a sudden
loss of a job - are reflected in individuals’ savings behavior, in terms of changes in deposit
or withdrawal patterns. This can give us a better understanding of how participants use
their accounts and also allows us to see whether economic difficulties are reflected in real
behavior. If this is the case, we are more confident that the self-reported indicators reflect
real changes in participants’ economic situation.

      In order to measure how life events are reflected in savings behavior, in our follow-up
survey we asked a series of questions on whether X life event happened in the individual’s
life in the past 6 months. ’X’ stood for a series of events, such as loss of a job, business
downturn, accident, significant theft, birth of a child, marriage of self or a child, increase
in household size, partnership breakup or death within nuclear family. Whenever the
participant responded in the affirmative for a particular life event, we asked what month
the event happened in. Thereafter this data was merged with the banking data on monthly
     Similar to the questions on participants debt, we asked a series of questions, such as, ”Have you lent
money to X? If yes, how much?” with ’X’ replaced by ’sister’, ’brother’, etc which were subsequently
aggregated into the categories of relatives and friends, or business partners.

deposits and withdrawals. We only have this data for 1219 individuals since the sample
is restricted to those within the treatment group who opened an account.

    Table 5 shows that life events are highly correlated with deposits by participants but
not with withdrawal patterns. For examples in the months where individuals lose their
job, they are 10.4 percentage points less likely to make a deposit (statistically significant
at the 1% level). This effect is very large in magnitude given that in a month with no ”life
event”, the likelihood of an individual making a deposit in that month is 22.1%. Similarly,
a business downturn, an increase in household size and a death within the nuclear family
brings the likelihood of deposit down by 16.4, 12.7 and 13.6 percentage points respectively
(all coefficients are statistically significant at the 1% level). The only life events that do
not have a statistically significant effect on the deposit patterns are the birth of a child
and breakup with a partner. We also run a specification where Lif eEventit is not a
vector but a dummy variable that takes on the value 1 if any life event has taken place
for individual i in month t. We find that the average effect of any life event is a decrease
in likelihood to make a deposit by 11.3 percentage points.

    However, for withdrawals we do not see significant correlations with these life events.
None of the coefficients on specific life events returned a statistically significant result when
withdrawal was the outcome variable. Perhaps this is related to the fact that withdrawals
overall are relatively rare events: the average withdrawal rate in a month, when there is no
life event, is 5.1 percent, considerably lower than the average rate of deposits. One possible
explanation of these findings may be that participants use savings accounts for long-term
savings rather than for short-term management of economic shocks. This is consistent
with the findings of Dupas and Robinson (2011), who find that individuals manage to
save at home for short periods of time but not over the long run (a few weeks or months).
Therefore, when individuals face a shock, they may dip into their within-household savings
and only rarely draw from their formal savings accounts. Perhaps having a formal savings

account, which ensures long-term security of funds, is an incentive to save small amounts
at home as individuals are aware that they can accumulate the small amounts and deposit
them within the bank.

3.4     Differential effects by types of accounts

As discussed above, half of the sample received a savings with a peer group savings
commitment device, which has been found by Kast et al. (2011) to significantly increase
savings in the study account. It is therefore of interest to understand whether the results
we are finding in this paper are mainly driven by the subgroup who received the peer
group support, or whether they also apply to those who have received an account without
group support. Table 6 addresses this question.

      Looking at the subgroups with and without peer group support reduces the sample
size by half and correspondingly introduces more noise in the analysis. Nevertheless,
Table 6, which reports treatment effects on the treated, shows some interesting suggestive
results. Column 1 shows that the reduction in worry is concentrated among those in the
peer group treatment, and is statistically significantly lower for those with access to a
savings account without the peer group component. On the other hand, the necessity
to cut back on consumption due to economics difficulty is reduced more for those with a
savings account without the peer group treatment. The reduction is significant for those
without the peer group treatment, and smaller and insignificant for those with the peer
group treatment. However, this difference between the two is not statistically significant.
Finally, overall economic difficulties and credit to or from others do not respond differently
for the different sub-treatments.

      These results have several implications. First, they show that the impacts found in
this paper are not generally concentrated on those with the peer group support. The

findings therefore seem to hold more generally everyone who received access to a savings
account in the scope of this study. Due to the noisiness of the results at the subgroup level,
we cannot make very strong interpretations beyond that, but one possible interpretation
of the result that the effect on worry seems to be stronger for those with peer group
support and the effect on consumption cutback weaker is that those with the peer group
treatment save more, and are therefore less worried about the future, while those without
the peer group support use more money for consumption smoothing during the crisis. This
would be consistent with the finding of how transactions track participants’ life events,
where we see that deposits are reduced significantly at times of economic shocks, but the
number of withdrawals does not seem to respond. If participants save small amounts at
home before depositing them into the bank, it is conceivable that during a time of crisis,
those without the peer group treatment use these small savings to smooth consumption
and those with the peer group support continue to save.

3.5     Robustness checks

In the following section, we will analyze possible threats to validity of the analysis. First,
we investigate whether the results might be driven by demand effects, and second, we
discuss attrition in the follow-up survey.

      Demand effects refer to changes in behavior by experimental subjects due to cues
about what constitutes appropriate behavior (Zizzo, 2010). In the context of this study,
one concern is that participants who received access to a savings account through Fondo
Esperanza might report more positive answers in the follow up survey than the control
group, out of gratitude or a sense of indebtedness vis-`-vis the organization. A reason for
why this is not very likely to happen is that participants do not know that the survey is
related to the savings account. Nevertheless, in order to test for possible demand effects,

we included two questions in the survey specifically designed to capture possible demand
effects. These are questions whose answer we expect would be affected by demand effects,
but are not directly affected by the treatment.

       The first question asks participants how complicated they found the process of
scheduling this interview, and the second asks them how satisfied they are with Fondo
Esperanza. Table 7 shows that neither of these questions respond to the treatment. This
indicates that it is very unlikely that the other self-reported effects we are finding in this
paper are driven by demand effects.

       The second series of robustness checks that we are looking at are dealing with the
issue of attrition in the follow-up survey. The baseline survey was conducted in meetings
of FE and by definition had 100% compliance, since we constructed the universe to consist
of all those FE members who were present on the day of the survey. The follow-up survey,
one year later, happened at participants’ home or work place, to ensure we also included
those individuals who were no longer members of FE. Even though we had taken great
care to keep attrition low,3 12.4% of participants could not be found for the follow-up
survey (see Table 8). Since we do not have information for these participants, we are only
able to include individuals in this study who could be reached for the follow-up survey.

       When comparing attrition rates across the treatment and control groups, we find that
attrition is 2.81 percentage points higher in the treatment group. This type of differential
attrition is a potential concern, as the characteristics of the attritors tend to differ from
those who stay in the panel. We control for any permanent differences by using individual
fixed effects. However, we cannot rule out that the trend over time is different for attritors
compared to those who stay in the sample.
    During the baseline survey, we asked participants not only about their own contact information, but
also about the contact details of a close relative or friend. In addition, we chose to work with the survey
agency Microdatos, who has special expertise in following participants for panel studies.

    Using the baseline survey, we can check whether characteristics of attritors and stayers
are indeed different. We find that younger persons have a higher propensity not to be
present in the follow-up survey. Similarly, attritors tend to have a lower amount of
previous savings (see Table 8). This makes intuitive sense: young people are more likely
to move for college, marriage or professional reasons, and are therefore more difficult to
follow in a panel study. They also tend to have lower savings.

    One way to adjust for the potential bias caused by non-balanced attrition is to apply
inverse probability weights (Wooldridge, 2002 and Wooldridge, 2007), which works as
follows. First, we predict the probability that an individual surveyed during the baseline
will be in the follow-up survey. The calculated probability for each individual forms the
propensity score - i.e. the propensity of an individual being in the follow-up survey.
Thereafter, we weigh each individual with the inverse of this propensity score. Therefore,
those who are less likely to be part of the follow-up survey receive a higher weight as
compared to the other participants. This weighing scheme allows us to weigh more heavily
the individuals who are underrepresented in the follow-up survey.

    All results remain qualitatively unchanged when applying inverse probability weights.
In Appendix A1, we present results from section 3.1 once again, both with and without
attrition weights. The data reveals no significant differences between both types of mea-
surements. For example on the question of the participant’s assessment on their recent
economic difficulty, treatment improved the average score by 0.240, while the correspond-
ing coefficient when adjusting with attrition weights is 0.249. Appendix A2 similarly
shows the results on lending and borrowing by participants. Once again, the differences
between results with and without attrition weights were minor. For example, without
attrition weights, being in the treatment group reduces the likelihood of owing money to
relatives and friends by 6.3 percentage points. With attrition weights, the corresponding
coefficient is 6.2 percentage points.

4    Conclusion

This paper investigates the effectiveness of precautionary savings through formal savings
accounts as a means to insure against economic shocks, among a low-income population
in Chile that works in the informal sector and has little access to other formal insur-
ance mechanisms to smooth consumption through the frequent economic shocks. We find
that access to a savings account, coupled with deposit commitment support, leads to a
significant reduction in one of the overbearing elements that often characterize poverty:
worry and anxiety about one’s economic future. This reduction in worry is based on con-
crete economic improvements in participants’ ability to smooth shocks: they experience
recent economic difficulties as less severe, less frequently need to cut back certain types of
consumption due to economic shocks, and have lower outstanding debt to their informal
network or fewer arrears on bills.

    These findings add to the growing of evidence showing the on the benefits of savings
accounts. While the microfinance movement includes micro-credit, micro-insurance and
micro-savings, the latter has traditionally received less attention, possibly because its
potential to generate profits to the microfinance institution is lower. Recent evidence
seems to suggest, however, that microcredit may have relatively limited impacts (Banerjee
et al., 2010), and micro-insurance faces some strong challenges in generating demand,
partly because it requires the client to trust the microfinance institution to fulfill on their
future commitment to pay out in case of a bad state of the world (Cole et al., 2009).
Combined with the other recent findings of positive impacts of access to various forms of
savings vehicles, discussed in the introduction, the findings of this paper therefore suggests
that increasing access to micro-savings vehicles may be a powerful, if less profitable,
intervention for increasing welfare of the poor.

    These findings have important policy implications. First, they suggest that subsidiz-

ing savings accounts, coupled with deposit commitment support, may have large social
returns. Private banks often do not find it in their interest to host savings accounts for
very low account, and correspondingly charge such accounts with large administrative
hurdles, minimum balance requirements, or maintenance fees. Such costs often result in
very large negative interest rates in formal savings accounts for the poor, which can result
in savings constraints. This paper suggests that such constraints can have large welfare

    While there has been a large focus in recent years on expanding access to (expensive)
credit options for the poor, this paper points to the potentially high returns to improving
access to savings vehicles. Free basic current accounts, such as recently introduced by
the Chilean government, for example, can be expected to generate significant benefits.
Similarly, policies that facilitate deposits into savings accounts, such as the recent policy
adopted by the Chilean government to deposit welfare payments into government provided
savings accounts, rather than dispersing them in cash, have potentially large benefits.

    The findings of this paper also raise a number of interesting questions for future
research: If returns to savings are so high, why aren’t individuals building a buffer stock
for precautionary savings at home? Do they have self-control issues? ”Other-control
issues” of having to keep demands by their social network at bay (Brune et al., 2011)?
Concerns about the safety from theft at home? Knowing where the source of the savings
constraint lies will also help determining what characteristics of the savings accounts
have the highest potential to increase welfare. In future analysis for this paper, analyzing
heterogeneous treatment effects between individuals with different degrees of self-reported
self-control and other-control problems and security concerns, we hope to shed more light
on this question.

5    References
Ashraf, Nava, Dean Karlan, and Wesley Yin, ”Tying Odysseus to the Mast: Ev-
   idence from a Commitment Savings Product in the Philippines,” Quarterly Journal
   of Economics, 2006, 121 (1), 635-672.
Ashraf, Nava, Dean Karlan, and Wesley Yin, ”Deposit Collectors.” Art. 5. Special
   Issue on Field Experiments. Advances in Economic Analysis and Policy 6, 2006, no.
Ashraf, Nava, Dean Karlan, and Wesley Yin, ”Female Empowerment: Impact of a
   Commitment Savings Product in the Philippines,” World Development, 2010, 38 (3),
Atkinson, Jesse, Alain de Janvry, Craig McIntosh, and Elisabeth Sadoulet,
   ”Creating incentives to save among microfinance Borrowers: A Behavioral experiment
   from Guatemala,” 2010, Working Paper.
Banerjee, Abhijit, Esther Duflo, Rachel Glennerster and Cynthia Kinnan, ”The
   miracle of microfinance? Evidence from a randomized evaluation,” 2010, Working
Browning, Martin and Annamaria Lusardi, ”Household Saving: Micro Theories and
   Micro Facts,” Journal of Economic Literature, 1996, XXXIV (December), 1797-1855.
Brune, Lasse, Xavier Gin, Jessica Goldberg, and Dean Yang, ”Commitments to
   Save: a Field Experiment in Rural Malawi,” 2011, Working Paper.
Burgess, Robin, and Rohini Pande, ”Do Rural Banks Matter? Evidence from the
   Indian Social Banking Experiment,” American Economic Review, 2005, 95 (3), 780-
Chandrasekhar, Arun, Cynthia Kinnan, Cynthia and Horacio Larreguy, ”Do
   Savings Crowd Out Informal Insurance? Evidence from a Lab Experiment in the
   Field”, 2010, Working Paper.
Cole, Shawn A., Xavier Gine, Jeremy Tobacman, Petia Topalova, Robert
    Townsend, and James Vickery, ”Barriers to Household Risk Management: Evi-
    dence from India.”, Harvard Business School Working Paper, 2009, No. 09-116.
Collins, Daryl, Jonathan Morduch Stuart Rutherford, and Orlanda Ruthven,
    Portfolios of the Poor: How the World’s Poor Live on $2 a Day, 2009, Princeton
    University Press.
Dupas, Pascaline, and Jonathan Robinson, ”Savings Constraint and Microenterprise
   Development: Evidence from a Field Experiment in Kenya,” 2011, Working Paper.
Kast, Felipe, Stephan Meier, and Dina Pomeranz, ”Under-Savers Anonymous:
   Evidence on Self-Help Groups and Peer Pressure as a Savings Commitment Device”,
   2011, Working Paper.

Kinnan, Cynthia and Robert Townsend, ”Kinship and Financial Networks, Formal
   Financial Access and Risk Reduction,” 2011, Working Paper.
Rosenzweig, Mark and Hans Binswanger, ”Wealth, Weather Risk, and the Compo-
   sition and Profitability of Agricultural Investments,” Economic Journal, 1993, 103
   (January), 56-78.
Wooldridge, Jeffrey, ”Inverse probability weighted M-estimators for sample selection,
   attrition, and stratification,” Portuguese Economic Journal, 2002, 1, 117-139.
Wooldridge, Jeffrey, ”Inverse probability weighted estimation for general missing data
   problems,” Journal of Econometrics, 2007, 141 (2007), 1281-1301.
Zame, William, ”Efficiency and the Role of Default When Security Markets are Incom-
   plete,” American Economic Review, 1993, 83 (1), 1142-1164.
Zizzo, Daniel, ”Experimenter demand effects in economic experiments”, Experimental
    Economics, 2010, 12(1), 75-98.

6   Tables

     Table 1: Baseline Summary Statistics and Balance of Randomization

                                                  Control     Account
                                                     (1)          (2)
               Education in years                   9.815       -0.158
                                                   (3.111)     (0.156)
               Age                                  43.29        0.101
                                                  (11.596)     (0.444)
               Income per capita (monthly)        116,117       4,008
                                                  (80,582)     (4,247)
               Household size                       4.273      0.0565
                                                   (1.732)    (0.0661)
               Has prior savings account            0.312     0.00309
                                                   (0.464)    (0.0176)
               Amount of other savings              63260        5,720
                                                 (241,301)     (8,482)
               Amount of former debt              454,659      73,964
                                                 (1659202)    (76,757)
               Worry                                2.899      0.0413
                                                   (0.968)    (0.0428)
               Economic difficulty                    5.026       0.142
                                                  (14.115)      (0.12)
               Group size                           14.79      0.0139
                                                   (4.322)     (0.508)
               Observations                         1,488        2,687
                 Notes: Standard deviation reported in parentheses be-
                 low control means in Column (1). Robust standard er-
                 rors reported in parentheses in Column (2), clustered at
                 the group level. Monetary amounts in Chilean Pesos.
                 500 Pesos = approximately 1 USD. The variables worry
                 and economic difficulty range from 1 to 4 and 1 to 10

Table 2: Effect of Being in Treatment Group on Self-reported Economic Wellbeing

                                         Worry                 Recent Economic Difficulty
                             Intention     Treatment on        Intention     Treatment on
                              to Treat      the Treated         to Treat      the Treated
                                 (1)             (2)                (3)            (4)
  Account                      -0.109*        -0.205**           -0.240*        -0.456**
                               (0.058)         (0.080)           (0.142)         (0.214)
  Constant                    2.930***       2.930***          5.113***        5.113***
                               (0.013)         (0.015)           (0.034)         (0.039)
  Individual fixed effects         Yes             Yes               Yes             Yes
  Year Dummy                     Yes             Yes               Yes             Yes
  Observations                  7,101           7,101             7,097           7,097
  R2                             0.02                              0.01
    Notes: Robust standard errors in parenthesis for all regressions; errors clustered at group
    level for ITT regressions. Outcome variable ”worry” in Columns (1)-(2) ranges from 1 to 4;
    it indicates how worried participants are about their economic future with higher numbers
    signifying more worry. Outcome variable ”recent economic difficulty” in Columns (3)-(4)
    ranges from 1 to 10; it indicates participants’ assessment of their economic difficulties in
    the recent past with higher numbers signifying a more difficult past. Level of significance:
    *** p<0.01, ** p<0.05, * p<0.1.

                             Table 3: Informal Lending and Borrowing Practices

                                    Owes to                 Owes to             Owed by                Owed by
                             relative and friends         institutions    relatives and friends    business contacts
     Panel A (Intention to Treat)
     Account                       -0.063***                 -0.042*              -0.019                  -0.014
                                     (0.021)                 (0.023)             (0.024)                 (0.023)
     Constant                       0.237***                0.357***            0.377***                0.480***
                                     (0.005)                 (0.006)             (0.006)                 (0.006)
     Individual Fixed Effects           Yes                     Yes                  Yes                     Yes
     Year Dummy                        Yes                     Yes                  Yes                     Yes

     R 2                              0.047                    0.05                0.052                   0.017
     Panel B (Treatment on the Treated)
     Account                       -0.118***                -0.078**              -0.037                  -0.027
                                     (0.032)                 (0.037)             (0.037)                 (0.038)
     Constant                       0.236***                0.357***            0.377***                0.480***
                                     (0.006)                 (0.007)             (0.007)                 (0.007)
     Individual Fixed Effects           Yes                     Yes                  Yes                    Yes
     Year Dummy                        Yes                     Yes                  Yes                    Yes
     Observations                     7,086                   7,087                7,094                  7,110
       Notes: Robust standard errors in parenthesis for all regressions; errors clustered at group level for ITT regres-
       sions. Level of significance: *** p<0.01, ** p<0.05, * p<0.1
                                       Table 4: Consumption Variation in Response to Shocks

                                  Meals       Meat      Medicine      School     Clothing     School      Public        Eating     Overall
                                                                     Supplies                 Snack      Transport       Out       Index
                                 (1)     (2)                (3)         (4)         (5)         (6)         (7)          (8)        (9)
     Panel A (Intention to Treat)
     Account                   -0.014  -0.048*            -0.025       -0.008      0.005       -0.004      -0.051*       -0.021      -0.166
                              (0.016)  (0.027)           (0.022)      (0.018)     (0.032)     (0.013)      (0.028)       (0.03)     (0.131)
     Constant                0.107*** 0.408***          0.262***     0.202***    0.513***    0.085***     0.317***     0.416***    2.311***
                              (0.004)  (0.006)           (0.005)      (0.004)     (0.007)     (0.003)      (0.007)      (0.007)     (0.030)
     Individual Fixed Effects    Yes      Yes                Yes         Yes         Yes         Yes          Yes           Yes         Yes
     Year Dummy                 Yes      Yes                Yes         Yes         Yes         Yes          Yes           Yes         Yes
     R2                        0.000    0.002              0.011       0.043       0.002       0.001        0.004         0.003       0.008

     Panel B (Treatment on the Treated)
     Account                   -0.026 -0.090**            -0.047       -0.014      0.010       -0.008     -0.096**       -0.039     -0.311*
                              (0.024)  (0.040)           (0.034)      (0.030)     (0.041)     (0.022)      (0.039)      (0.041)     (0.161)
     Constant                0.107*** 0.408***          0.262***     0.202***    0.513***    0.085***     0.317***     0.416***    2.310***
                              (0.004)  (0.007)           (0.006)      (0.005)     (0.007)     (0.004)      (0.007)      (0.007)     (0.029)
     Individual Fixed Effects    Yes      Yes                Yes          Yes        Yes         Yes          Yes           Yes        Yes
     Year Dummy                 Yes      Yes                Yes          Yes        Yes         Yes          Yes           Yes        Yes
     Observations              7,164    7,164              7,164        7,164      7,164       7,164        7,164         7,164      7,164
      Notes: Robust standard errors in parenthesis for all regressions; errors clustered at group level for ITT regressions. Outcome variables
      in Columns (1)-(8) are dummy variables indicating whether there was a consumption cut back in that particular item. Index in Column
      (9) is an additive score of the number of items that an individual cut back on and hence ranges from 0 to 8. Level of significance: ***
      p<0.01, ** p<0.05, * p<0.1
                 Table 5: Live Events and Savings Behavior

Life Event                      Deposits    Withdrawals      Deposits     Withdrawals
Loss of Job                    -0.104***        0.014
                                 (0.017)      (0.014)
Business Downturn              -0.164***        0.005
                                 (0.025)      (0.012)
Accident                         -0.064*       -0.013
                                 (0.037)      (0.012)
Significant Theft                 -0.087*        0.005
                                 (0.048)      (0.034)
Birth of Child                     0.104       0.079
                                 (0.122)      (0.075)
Marriage of Self or Child       -0.214**        0.095
                                 (0.105)      (0.078)
Increase in Household Size     -0.127***       -0.012
                                 (0.038)      (0.020)
Partnership Break-up              -0.008        0.021
                                 (0.087)      (0.088)
Deaths in Nuclear Family       -0.136***       -0.018
                                 (0.049)      (0.026)
Any Life Event                                               -0.113***        0.004
                                                               (0.015)       (0.006)
Constant                        0.221***        0.051***      0.221***      0.051***
                                 (0.001)         (0.000)       (0.001)       (0.000)
                                   Yes             Yes           Yes           Yes
Individual Fixed Effects
Year Dummy                        Yes             Yes           Yes            Yes
Observations                     1,219           1,219         1,219          1,219
R2                               0.000           0.000         0.000          0.000
  Notes: Robust standard errors in parenthesis, clustered at group level. Level of signifi-
  cance: *** p<0.01, ** p<0.05, * p<0.1

                                  Table 6: Differential Effects of Treatment on the Treated

                                  Worry     Recent Econ.       Owes to        Owes to       Owed by         Owed by    Consumption
                                             Difficulty          relative     institutions     family         business     cutback
                                                             and friends                   and friends      contacts       index
     Account                      -0.083        -0.519*       -0.142***        -0.048         -0.061          -0.028     -0.477**
                                 (0.099)        (0.266)        (0.040)        (0.046)        (0.046)         (0.048)      (0.199)
     Peer Group Treatment       -0.218**         0.112          0.041          -0.055          0.044          0.002        0.298

                                 (0.098)        (0.262)        (0.040)        (0.045)        (0.046)         (0.047)      (0.196)
     Constant                   2.930***       5.113***       0.236***       0.356***       0.377***        0.480***     2.310***
                                 (0.015)        (0.039)        (0.006)        (0.007)        (0.007)         (0.007)      (0.029)
     Individual Fixed Effects       Yes            Yes            Yes            Yes             Yes            Yes          Yes
     Year Dummy                    Yes            Yes            Yes            Yes             Yes            Yes          Yes
     Observations                 7,101          7,097          7,086          7,087           7,094          7,110        7,164
       Notes: Robust standard errors in parenthesis. Level of significance: *** p<0.01, ** p<0.05, * p<0.1
Table 7: Checking for Possible Demand Effects

                Survey Process     Satisfaction with FE
Account              0.040                 -0.013
                    (0.043)               (0.052)
Constant           2.45***                6.38***
                    (0.033)               (0.042)
Observations         3,365                 3,572
R2                   0.001                  0.000
  Notes: Robust standard errors in parenthesis, clustered at
  group level. Level of significance: *** p<0.01, ** p<0.05,
  * p<0.1

                                 Table 8: Checking for the Effect of Attrition

                       Attrition                                      Income                        Has Prior
                     in Treatment      Education          Age        per capita    Household         Savings
                      vs. Control                                    (monthly)       size           Account
     Account            0.028**
     Attrition                             0.210       -1.179**        3,775        -0.160**          -0.006
                                          (0.138)       (0.494)       (5,720)        (0.079)         (0.021)
     Constant          0.124***         9.683***      43.520***     84,456***       4.332***        0.315***
                        (0.009)           (0.076)       (0.229)       (2,796)        (0.035)         (0.009)
     Observations        4,175             4,175         4,175         4,129           4,175           4,175

     R2                  0.001             0.001         0.001         0.000           0.001           0.000
                      Amount of        Amount of      Group Size     Take-up         Number         Number
                     Other Savings    Formal Debt                   of Account     of Deposits   of Withdrawals
     Attrition          32,687*           -74,682        -0.396       -0.053*         -0.008          0.004
                       (18,417)         (107,532)       (0.278)       (0.030)        (0.013)         (0.006)
     Constant         62,306***        512,852***      14.86***      0.533***       0.059***        0.012***
                        (3,914)          (48,546)       (0.235)       (0.018)        (0.008)         (0.002)
     Observations        4,175             4,175         4,175         2,687           4,175           4,175
     R2                  0.002             0.000          0.001        0.001           0.000           0.000
       Notes: Robust standard errors in parenthesis, clustered at group level. Level of significance: *** p<0.01, **
       p<0.05, * p<0.1
A   Appendix

Table A1: Effect of being in Treatment Group on Self-reported Economic Wellbeing

                                                    Worry      Recent econ. difficulty
       Independent Variable
                                                     (1)                (2)
       Panel A: No Weights
       Account                                     -0.109*             0.240*
                                                   (0.058)             (0.142)
       Constant                                   2.930***            5.113***
                                                   (0.013)             (0.034)
       Individual fixed effects                        Yes                 Yes
       Year dummy                                    Yes                 Yes
       R2                                           0.020               0.010
       Panel B: With Attrition Weights
       Account                                     -0.107*             -0.249*
                                                   (0.057)             (0.141)
       Constant                                   2.929***            5.110***
                                                   (0.013)             (0.034)
       Individual fixed effects                        Yes                 Yes
       Year dummy                                    Yes                 Yes
       R2                                           0.020               0.010
       Observations                                 7,101               7,097
         Notes: Robust standard errors in parenthesis, clustered at group level. Level of
         significance: *** p<0.01, ** p<0.05, * p<0.1. Outcome variable in Column (1)
         ranges from 1 to 4 and in Column (2) from 1 to 10.

                                          Table A2: Informal Lending

                                       Owes to            Owes to            Owed by               Owed by
                                 relative and friends   institutions   relative and friends    business contacts
     Panel A: No Weights
     Account                  -0.063***                    -0.042*             -0.019                -0.014
                                (0.021)                    (0.023)            (0.024)               (0.023)
     Constant                  0.237***                   0.357***           0.377***              0.480***
                                (0.005)                    (0.006)            (0.006)               (0.006)
     Individual Fixed Effects      Yes                        Yes                 Yes                   Yes
     Year Dummy                   Yes                        Yes                 Yes                   Yes
     R2                          0.047                       0.05               0.052                 0.017

     Panel B: With Attrition Weights
     Account                  -0.062***                    -0.040*             -0.019                -0.014
                                (0.021)                    (0.023)            (0.024)               (0.023)

     Constant                         0.237***            0.355***           0.378***              0.480***
                                       (0.005)             (0.006)            (0.006)               (0.006)
     Individual Fixed Effects             Yes                 Yes                Yes                   Yes
     Year Dummy                          Yes                 Yes                Yes                   Yes
     R2                                 0.048                0.05              0.051                 0.017
     Observations                       7086                7087               7094                  7110
       Notes: Robust standard errors in parenthesis, clustered at group level. Level of significance: *** p<0.01, **
       p<0.05, * p<0.1.

To top