Adult Financial Literacy and Households' Financial Assets ...
Transcript of Adult Financial Literacy and Households' Financial Assets ...
Economic Policy 63rd Panel Meeting
Hosted by the De Nederlandsche Bank Amsterdam, 22-23 April 2016
The organisers would like to thank De Nederlandsche Bank for their support. The views expressed in this paper are those of the author(s) and not those of the supporting organization.
Adult Financial Literacy and Households' Financial Assets:
The Role of Bank Information Policies
Margherita Fort (University of Bologna) Francesco Manaresi (Bank of Italy)
Serena Trucchi (University of Bologna)
Adult Financial Literacy and Households’ Financial Assets:
The Role of Bank Information Policies
Margherita Fort∗ Francesco Manaresi† Serena Trucchi‡
March 15, 2016
Abstract
We investigate the role of bank information policies in fostering the accumulation
of financial knowledge. In Italy, banks that belong to the PattiChiari Consortium
implement policies aimed at increasing transparency and procedural simplification,
without offering services at a lower cost with respect to banks outside the Consor-
tium. These policies significantly affect the level of financial literacy of between 5
to 10% of households. We use bank information policies as an instrumental variable
for financial literacy in the financial assets’ equation and show that the positive re-
lationship between the two is significantly underestimated by OLS correlation. The
estimated impact is driven by households whose head is 60+ and low educated.
Key words: Bank Information Policies; Financial Assets; Financial Literacy;
Instrumental Variables
JEL: D14; G11.
∗Universita di Bologna and CESifo and IZA. Email address: [email protected]†Banca d’Italia. Email address: [email protected].‡UCL, Universita di Bologna and CeRP-Collegio Carlo Alberto. Email address: [email protected]
1
1 Introduction
Both in Europe and in the U.S. there is evidence of low levels of financial literacy and
a limited knowledge of basic economic concepts, like inflation or interest compounding
(Jappelli, 2010; Lusardi and Mitchell, 2011c, 2014). Lack of financial literacy and the
mis-perception of crucial economic factors may lead individuals to make suboptimal
investment choices and, thus, reduce individual wealth and welfare. For this reason,
actions to promote financial literacy in the population have been implemented in a
growing number of countries (see OECD (2013a) for a review). These interventions
mainly rely on the role of formal financial education in schools or at the workplace.1
In this paper we study a different channel which may improve financial literacy,
which has been almost neglected by the literature: bank information policies. To assess
their effectiveness, we exploit a peculiar feature of the Italian banking system, namely
the existence of a consortium of banks (PattiChiari), which has the aim of enhancing
banking transparency. Banks that entered the PattiChiari Consortium had to provide to
their customers information materials on basic economic concepts (e.g., about compound
interests, fixed/flexible interest rates, the calculation of debt repayment instalments, etc.)
and more transparent information on current account expenditures. These activities are
in line with the OECD (2013b) suggestions for improving the effectiveness of financial
literacy programs, as the informative material was provided on a regular basis (monthly)
in a format and location that are easy to access (i.e., on the internet and/or at the local
bank branch). We argue that these information policies reduced the cost of acquiring
(basic) financial knowledge for customers without imposing any additional burden in
terms of time, effort, or resources to them. We investigate to which extent information
policies improve financial literacy. Evidence on absence of self-selection of clients into
PattiChiari banks allows us to overcome one of the main issues in the evaluation of
1There is an open debate about the effectiveness of alternative financial education programs onfinancial literacy and financial behaviour (Lusardi and Mitchell, 2014; Willis, 2011). Robust and reliableempirical evidence on the evaluation of their effects is still limited and this motivates some scepticism onthe role played by financial education in improving financial knowledge and household financial welfare(Willis, 2011).
2
voluntary financial education programs (OECD, 2013a) and to interpret our findings as
causal.
Our results suggest that bank information policies do significantly affect the level
of financial literacy. In particular, financial literacy is found to be 10% higher among
PattiChiari clients (about one fourth of a standard deviation of our measure of financial
literacy). Information policies are found to be relevant for a relatively small sample of
the population (around 5-10% of it). They are particularly effective among households
whose head is 60+ and is low educated. This result adds to the literature aimed at
evaluating the effectiveness of alternative interventions on financial literacy (Fernandes
et al., 2014; Lusardi and Mitchell, 2014) by showing that providing bank’s customers
with information about simple economic concepts may represent an alternative way
to improve financial knowledge among adults. Even when formal financial education
programs affect financial literacy, biases, heuristics and non rational influences may
weaken their effectiveness on individual financial behaviour (Hastings et al., 2013). We
find that bank information policies significantly affect financial assets. In this respect,
this stuedy might offer some guidance on how program effectiveness can be improved,
as we discuss in the conclusions.
A second contribution of this paper is to provide a new instrumental variable strat-
egy to assess the effect of financial literacy on financial assets, which exploits bank
information policies for identification. The association between financial literacy and
several economic decisions (risk diversification, debt exposure and retirement planning)
has been widely documented. However, evidence about the strength of the causal re-
lationship between these factors is recent and still limited. Most studies report that
regression correlations provide a lower bound of the true effect, but estimates of this
bias vary substantially. Unobserved heterogeneity, simultaneity (Jappelli and Padula,
2013b) and measurement error (van Rooij et al., 2011b) plague empirical results and
are difficult to address. In line with previous analyses, we show that the causal effect
of financial literacy on financial assets is underestimated by OLS correlation. In par-
ticular, we find that a one standard deviation increase in financial literacy leads to an
3
increase of household financial assets of around 8 thousands euros (35% of a standard
deviation in financial assets). The effect of financial knowledge on financial assets is
mainly driven by elderly and low educated respondents. Moreover, the OLS bias turns
out to be largest precisely for this group of individuals: these are indeed the households
for whom measurement error in financial literacy is likely to be larger. We offer some
insights about the channels through which this effect takes place, showing that financial
literacy promotes the propensity of households to participate to stock market (Christelis
et al., 2010; van Rooij et al., 2011b,c) and, possibly, to hold a more diversified portfolio.
Conversely, we do not find evidence of a higher attitude to plan for retirement (Lusardi
and Mitchell, 2011a) nor of a higher saving rate. Lack of adequate data, prevent us from
examining the effect of financial literacy on the quality of financial decision.
We refer to financial literacy as a broad concept, that concerns basic economic notions
that individuals should have some understanding of when making even simple day-to-
day financial decisions and help them to evaluate future scenarios (Lusardi and Mitchell,
2007b). To measure financial literacy, we follow previous literature (see, for instance,
Fornero and Monticone (2011) and Lusardi and Mitchell (2007a)) and we rely on simple
questions that have the characteristics pointed out by Lusardi and Mitchell (2011c), i.e.
simplicity, relevance, brevity and capacity to differentiate. We show our indicator to
capture one latent construct, which is labelled financial literacy.
The strategy used in this paper hinges on two main assumptions: the absence of
self-selection of, respectively, individuals as clients of a PattiChiari bank and banks
into the PattiChiari Consortium. We extensively discuss them in the paper. Results
are robust to a number of falsification checks regarding potential sorting of customers
or banks. Moreover, first-stage statistics show that the instrumental variable we use
is not weak (Staiger and Stock, 1997) and the availability of multiple instruments for
financial literacy allows us to run an over-identification test that supports the causal
interpretation of our findings.
The paper is organized as follows. We start by briefly reviewing the related liter-
ature, mainly focusing on empirical contributions (Section 2). We then describe the
4
main features of the PattiChiari Consortium and services provided by banks in the
Consortium to their clients, highlighting the channels through which these services may
foster financial literacy (Section 3). We present the data used to perform our analysis
in Section 4. Section 5 illustrates our empirical approach and discusses the hypothesis
at the core of the identification strategy. We present our main findings in Section 6. In
Section, 7 we provide evidence that estimated effects are not driven by self-selection of
clients into PattiChiari banks, while in Section 8 we show that selection of banks into
the Consortium does not explain our results either. Conclusions follow.
2 Related Literature
Our paper relates to the literature that analyzes tools aimed at fostering financial liter-
acy. Lusardi and Mitchell (2014) and Hastings et al. (2013) represent the most recent
reviews on these topics. They cover a) the measurement of financial literacy; b) the
relationship between financial education, financial literacy and financial outcomes; c)
the role of interventions. Both reviews point out the shortage of rigorous evaluations
of the impact of financial education programs, making it difficult to draw inference on
their effectiveness. One of the main issues that motivates the lack of empirical evidence
is the difficulty of proving causal links (OECD, 2013a). Selective participation into pro-
grams and non-random attrition, on the one hand, and measurement issues related to
financial literacy and financial behaviour, on the other hand, make this task empirically
challenging.
Several recent papers aim to assess the impact of a financial education programs on
financial knowledge and behaviours, taking into account these issues. Results are mixed.
The meta-analysis in Fernandes et al. (2014) documents a weak effect of interventions
the improvement of financial literacy. On the contrary, Luhrmann et al. (2015) analyse
a compulsory education program in Germany, showing an effect on both financial liter-
acy and behaviour. Financial education programs targeted to high school students in
Brazil turn out to significantly affect financial knowledge and behaviour, such as sav-
5
ing for purchases, financial planning, participation to economic choices (Bruhn et al.,
2013). Gibson and Zia (2012) use a randomized experiment to assess the impact of
providing financial literacy training on migrants’ financial knowledge and their use of
remittances tools. Cole et al. (2011) find financial education to increase financial liter-
acy in developing countries (India and Indonesia), with small and heterogeneous effect
on economic behaviour (namely the probability of opening a bank account). Similarly,
Mastrobuoni (2011) shows that the availability of information on social security benefits
in the U.S. has significant impact on workers’ knowledge. However, he does not detect
significant changes in financial decisions. With respect to these studies, we show that
bank information policies shape both financial literacy and financial assets, pointing to
the effectiveness of this alternative tool.
The literature on the links between financial literacy and financial decisions in eco-
nomics and behavioural finance (see the recent review by Garcia 2013) suggests that
the difficulty of processing complex information might be relevant for financial decision-
making, and promote “the widest dissemination possible of some suitable financial rules”
(Garcia, 2013). OECD (2013a) offers some guidance on how the effectiveness of financial
education programs can be improved, in line with what is suggested by papers in the
area of behavioural finance (Altman, 2012). According to these remarks, education pro-
grams should: i) offer material in formats and locations that are easy to access; ii) help
consumers to simplify financial decisions, for instance breaking them in intermediate
steps or providing rules-of-thumb or problem-solving strategies; iii) increase saliency by
providing participants with regular reminders or tools to track and visualise individual
progress. The bank information policies we consider in the paper are in line with these
suggestions.
Our paper also relates to the literature that investigates the causal effect of financial
literacy on financial outcomes. Jappelli and Padula (2013b) and Lusardi et al. (2011)
provide a theoretical framework to analyse the investment in financial literacy and its
impact on wealth in the context of inter-temporal utility maximization. These models
highlight that financial literacy and wealth are simultaneously determined and that the
6
incentives to invest in financial literacy may depend on the level of wealth, raising the
issue of the potential endogeneity of financial literacy in the wealth equation.2 Endogene-
ity of financial literacy in the wealth equation may also arise because of omitted variable
bias or substantial measurement error in financial literacy (Lusardi and Mitchell, 2009;
van Rooij et al., 2011b), as we discuss in more detail in Section 5.
Some recent papers have addressed this potential endogeneity problem pursuing instru-
mental variable strategies for identification. However, empirical studies assessing a causal
link between financial literacy and financial outcomes are still rare (Hastings et al., 2013).
Finding an exogenous source of variability in this setting is difficult and most of the iden-
tification strategies adopted in observational studies so far are not free from criticism.3
These papers can be grouped into three broad categories. A first group of papers ex-
ploit as instruments pre-labour market endowment of financial knowledge (Disney and
Gathergood, 2011; Jappelli and Padula, 2013b; van Rooij et al., 2011c). van Rooij et al.
(2011c) are sceptical about the validity of the exclusion restriction for the instrumen-
tal variable they use and discuss the issue at length in the paper, adding a rich set of
controls to their baseline specification. Other studies hinge on the idea that the respon-
dent’s financial literacy is influenced by financial knowledge of peers or reference groups
(Bucher-Koenen and Lusardi, 2011; Fornero and Monticone, 2011; Jappelli and Padula,
2013a; Klapper et al., 2013; Klapper and Panos, 2011; van Rooij et al., 2011a,c). The
assumption that lies behind this identification strategy is that the respondent cannot in-
fluence the peers’ experience significantly, i.e. there is no “reflection problem” (Manski,
1993). An alternative instrumental variable strategy is proposed by Lusardi and Mitchell
(2011b). They take advantage of the fact that several U.S. states mandated high school
financial education in the past and exploit the exogenous variation in financial literacy
induced by the exposure to the mandate to identify the effect of financial literacy on
planning for retirement.
2However Lusardi and Mitchell (2007c) show that the instrumental variable estimate of the impactof wealth on financial literacy is not statistically significant. They interpret this finding to support theabsence of reverse causality in their data.
3We refer to Lusardi and Mitchell (2014) for a detailed discussion of the instruments proposed in theliterature (an insightful synthesis is provided in their Table 4, p.28).
7
Our paper does not belong to any of these categories and proposes a new strategy to
identify the causal effect of financial literacy on household wealth that exploits, for the
first time, the role of bank information policies in fostering household financial knowl-
edge. Moving from the assumption that acquiring financial knowledge is a costly process
(Jappelli and Padula, 2013b), we argue that bank information policies undertaken by
a specific subset of Italian banks may decrease the cost of acquiring financial literacy
without imposing additional burden to the client in terms of time, effort or resources.4
We exploit this exogenous source of variation in the cost of acquiring financial knowledge
to investigate the effect of financial literacy on financial assets. As pointed out Lusardi
and Mitchell (2014), most studies find that the OLS estimate of the impact of financial
literacy is biased downward, but the estimates differ a lot across papers and the mag-
nitude of the bias varies largely in the literature.5 Our findings confirm the downward
bias of OLS estimate, showing that causal effect is four times larger.
Financial literacy might affect financial assets by affecting price perception, expec-
tations, and optimization. As a result of increased financial knowledge individuals: a)
may make more accurate or less biased predictions; b) may be less likely to underesti-
mate compounding effects; c) may better understand risk and this may affect how well
they diversify their portfolios (Jappelli and Padula, 2013b). Moreover, financial literacy
may be related to heterogeneity in assets returns, as shown by Deuflhard et al. (2014)
focusing on returns yielded by savings accounts. Finally, the implementation of less
generous defined-contribution pension schemes shifted the burden of adequate savings
for retirement to workers. In this framework, if financial literacy implies a better ability
to understand current pension rules and to anticipate a reduction in the value of social
security wealth, financial knowledge may promote savings, necessary to sustain an ade-
quate level of consumption during retirement (Lusardi and Mitchell, 2007a). Whatever
is the most relevant channel, all point towards financial literacy being positively corre-
4See Section 3 for a description of the activities of the Consortium, and Section 7 for statisticalevidence against selection of customers into PattiChiari banks.
5The instrumental variable estimates range from three times larger the corresponding OLS estimates(Jappelli and Padula, 2013b; van Rooij et al., 2011c) to thirteen times larger than OLS (Bucher-Koenenand Lusardi, 2011).
8
lated with financial assets.
Assessing the effect of financial literacy on the quality of choices is a less debated and
eventually more demanding question because it is often unclear what a better decision is.
In this respect, Guiso and Viviano (2015) exploit a high frequency dataset with detailed
portfolio management information to investigate whether financial literacy affect the
ability of dismissing a dominated alternative. More precisely, they examine the ability
to avoid financial mistakes and to make autonomous decisions about portfolio manage-
ment. They find statistically significant but economically small effects. Even though
we lack adequate date to make a similar assessment, evidence that literate individuals
hold a more diversified portfolio is consistent with a better portfolio management. Our
evidence is, however, only suggestive.
3 The PattiChiari Consortium
This section describes the main features of the PattiChiari Consortium, how and why
banks join it, and the range of services provided by banks in the Consortium. It highlights
the channels through which these services may foster financial literacy.
The PattiChiari Consortium has been created in 2003 by the Italian Banking As-
sociation (Associazione Bancaria Italiana: ABI, henceforth). By 2010, 98 banks were
members of it, corresponding to around 64% of all bank branches in the Italian territory.
All except one of the 70 banks we can consider in the paper (see Section 8 for more de-
tails) joined the consortium in 2003; one (Cassa di risparmio della provincia di Chieti)
joined in 2004. The geographical distribution is characterized by some regions (such as
Trentino-Alto Adige and Calabria) where there are few PattiChiari branches and others
(e.g., Piedmont and Sardinia) where over 80% of branches belong to a PattiChiari bank.
The share of PattiChiari branches is however homogeneous across broader geographical
areas: about 63% in northern Italy, 67% in central Italy, 64% in southern Italy.
The Consortium, develops actions aimed at fostering bank transparency and improv-
9
ing financial literacy among citizens.6 A bank that joins the Consortium has to introduce
a set of “Quality Commitments” (so called “Impegni per la qualita”). They are aimed to
improving transparency in the relation between bank and clients and, thus, do not have
any goal of advertising specific financial products nor promoting any specific behaviour,
economic choice or investment in any asset. In the period we analyse (year 2010), these
commitments refer to information about mortgage and debt and to comparability of
current accounts. In what follows, we present these two commitments in detail.
Information about mortgage and debt. Every month PattiChiari banks send by
mail a mortgage statement to their clients. This statement includes information on total
and residual amount of debt to be paid and on the residual number of installments, and
reminders information about penalties. In addition, a 24-pages booklet was available
in each PattiChiari branch. The booklet covered several topics related to mortgages
and provided definitions of more general economic concepts.7 Each section included
examples about the calculation of interest rates and installments and provided several
remarks.8 An example of its content is the following remark (our translation from
Italian): “WARNING: With a lower mortgage payment the duration of the mortgage
increases. Therefore, the total amount of interest paid on the mortgage will be higher,
because the number of installments increases”.
Comparability of current accounts. Costs, interest rates and general contract con-
ditions of bank accounts vary both across and within banks. For instance, one bank may
offer up to 12 different bank accounts tailored to a specific type of customer (eg. house-
hold, retired, younger than 18, etc.). The PattiChiari Consortium provides a search en-
6It has implemented a set of formal financial literacy programs in Italian schools and, starting in2012, in workplaces and cultural and charitable associations.
7More specifically it referred to mortgage portability, substitution or renegotiation; mortgage resolu-tion and redemption. It gives a definition for amortization plan, mortgage resolution, Euribor (Euro Inter-bank Offered Rate), IRS (Interest Rate Swap), Eurirs (Euro Interest Rate Swap), fixed-rate, adjustable-rate and mixed-rate mortgages, TAEG (synthetic indicator of the cost of the mortgage), portability,installment, spread, renegotiation, and subrogation.
8A copy of the booklet, available in Italian only, can be downloaded from the URLhttps://web.archive.org/web/20111030200724/http://www.pattichiari.it/dotAsset/12557.pdf
10
gine that allows the direct comparison of specific characteristics of bank accounts across
different banks, as well as the information on the location of bank branches. Thanks to
the provision of detailed information on interest rates, we expect households to become
more aware about how current accounts work and more accustomed to concepts like
interest rate compounding.
The “Quality Commitments” described above are in line with the OECD suggestions
to improve the effectiveness of education programs (OECD, 2013b). Indeed, they pro-
vide informative material on a regular basis (costumers receive the mortgage statement
every month) in format and location that are easy to access (mail, bank’s branch and
websites). These interventions are designed to simplify costumers decision problems,
providing definitions and synthesis of basic and simple information about relevant eco-
nomic concepts, such as amortization plans and interest rates compounding, sometimes
through examples. These concepts are closely related to the standard questions used to
measure financial literacy in the literature and in our paper (see Appendix A for the
exact wording of the questions used to measure financial literacy). At the same time,
the “Quality Commitments” are not directly related to the provision of formal financial
advice and do not typically take this form. However, they can indirectly impact the
probability of consulting an advisor and delegating the portfolio choice management if,
as shown by Calcagno and Monticone (2015), it depends on financial literacy.
Why Banks Join the PattiChiari Consortium? Application to the PattiChiari
Consortium is decided at the bank level, and it involves all the branches on the Italian
territory. Any bank can apply to be a member of the Consortium, provided that it
undertakes the “Quality Commitments”. Applications are evaluated by Patti Chiari
board of directors. If a member bank does not meet the requirements, it can be fined
and excluded from the Consortium.9 The Consortium, based in Rome at the ABI’s
9After considering all merges and acquisitions among banks over the relevant period, we find thatonly 1.5% of households in our baseline sample remained with a bank that changed affiliation with thePattiChiari Consortium. In all these cases these banks exited the Consortium after belonging to it for aperiod between 6 months and 5 years. Thus, we are unable to use variation induced by banks entering
11
headquarters, develops projects that are, then, implemented by each PattiChiari bank.
There is no fee for entering the Consortium, yet the bank has to incur administrative
costs in order to comply to the “Quality Commitments”. Menu costs make the commit-
ment regarding the comparability between current accounts the most demanding task
for member banks. In particular, significant menu costs had to be incurred in order to
adjust the offer of bank accounts to the PattiChiari standards (allowing their compara-
bility across banks). High entry costs may partly explain the large share of PattiChiari
partners among large banks (see Section 8). Concerning the advantages of entering in the
PattiChiari Consortium, banks improve the quality of supplied services and customer
satisfaction, and, in turn, benefit from advertising provided by the Consortium. Informa-
tive advertising can benefit both the firm that makes the investment (in our application a
bank belonging to the PattiChiari consortium) and its competitors, through an increase
in savings, stock market participation and the demand for financial services in general
(Hastings et al., 2013). Free-riding problems may induce underprovision of informative
advertising. In our case, however, the fact that over 60% of all bank branches belong to
the Consortium, according to ABI data, should mitigate the incentive to free-ride.
Descriptive statistics of bank characteristics, and their heterogeneity between PattiChiari
and non-PattiChiari banks are discussed in Section 8, where we provide statistical ev-
idence that our results are not driven by bank observable differences or their potential
self-selection in the Consortium.
4 Data
The empirical analysis is based on the Bank of Italy’s Survey on Household Income
and Wealth (SHIW), a biannual survey that collects detailed information on household
income, savings and portfolios from a nationally representative sample of Italian house-
holds.
Households’ financial assets are defined as the sum of deposits, securities, and commer-
the Consortium in different periods to test the internal validity of our identification strategy.
12
cial credits at the end of the reference year (2010 for the baseline analysis).10 Several
covariates (age, gender, education) refer to the household head, defined as the household
member who is responsible for economic and financial decision. We observe financial
literacy of the household head only. Our baseline empirical analysis is based on the
2010 wave, because it includes both questions about the reasons for choosing the bank
(available only in this year) and questions to measure financial literacy. In Section 9,
we check the validity of the results in a larger sample, that also includes 2006 and 2008
waves: results are fully consistent.
The set of questions on financial literacy are included since 2006 and vary slightly across
waves. In the 2010 wave, respondents are asked three questions related to portfolio
diversification, the risk associated to fixed or adjustable interest rate and the effect of
inflation.11 Questions about diversification and inflation are similar to the ones devised
for the US Health and Retirement Study (HRS henceforth) (Lusardi and Mitchell, 2011a)
while knowledge of mortgage repayment mechanisms is not considered in the HRS. All
these questions follow the principles of simplicity, relevance, brevity and capacity to
differentiate identified by Lusardi and Mitchell (2011c). Table 1 shows that in 2010
almost three out of four respondents answered correctly to the question on inflation;
the percentage of correct answers to the question on mortgages and portfolio diversifica-
tion is, respectively, 64% and 55%. On average, respondents answered correctly to less
than two questions, and one respondent out of three correctly answered to all questions.
We measure financial literacy with the number of questions correctly answered by the
household head. We present evidence corroborating the assumption that our indicators
measure just one latent construct, that we name financial literacy (as in van Rooij et al.
(2011b)), and we show that our baseline results are confirmed when using this indicator
for financial literacy (see Appendix A for more details). Moreover, our results are robust
10More precisely assets include, along with bank accounts, Italian government securities, bonds andItalian investment funds, Italian shares and equities, managed portfolios, foreign securities and a residualcategory. Commercial credits include i) credits of the household’s members with relatives or friendsnot living with the household and ii) trade credits to costumers of respondents who are self-employed,members of a profession, individual entrepreneurs or employed in a family business.
11See Appendix A for the exact wording of the questions and details on differences across waves.
13
to the use of different measures of financial literacy (see Section 6).
We use information on the date of entrance and exit of each Italian banking group
in the PattiChiari Consortium and the answer to questions about the banks used by the
household to build an indicator of whether the household is client of a bank that is part
of the PattiChiari Consortium. Our baseline analysis focuses on households having at
least one bank account. We say that a bank belongs to the Consortium in year 2010 if
the bank was part of it by January 1st of 2010. It is important to highlight that financial
literacy and financial assets are both collected at the interview, and the latter refers to
December 31st 2010. Therefore, we observe financial literacy and financial assets one
year after the moment we measure the PattiChiari treatment and, thus, the effect we
estimate is not contemporaneous but delayed by one year.
Households in the SHIW dataset report their “main” bank -as defined by the household
head- and up to seven additional banks where they hold a current account. Answers
can be chosen from a list of 87 financial institutions (encompassing more than 85% of
total credit granted in Italy). If the bank is not listed among the possible answers,
its name will be missing. For each household, we construct a dummy (Patti) that is
equal to one if the main bank used by the respondent belong to the Consortium at the
beginning of 2010 and zero if it does not; we exclude households who do not have any
bank account or for which the name of the bank is missing. Descriptive statistics show
73% of respondents in our sample to be clients of a bank belonging to the PattiChiari
Consortium (Table 1).12
We include in the sample households whose head is aged 30 or above because we
want to exclude cases in which individuals may be still enrolled in full-time education
(at the university) and thus not financially independent from the family of origin. After
excluding from our sample outliers and respondents who do not report the reason for
choosing the bank, the final sample size in our baseline regression (2010 wave data) is
4865 observations (descriptive statistics are shown in Table 1).13
12Our results are robust to alternative measures for being client of a PattiChiari bank and to differentway of dealing with missing values. This issue is discussed in Section 6.
13We exclude the upper and lower 5% tail of the financial assets distribution. Results are similar with
14
For some complementary analysis we merge this data with data from the 2002 SHIW
wave: due to the rotating panel of the survey, only 1027 households have been interviewed
in 2002 and 2010. Notably, no measure of financial literacy is available in the 2002 SHIW
wave.
We collect information on bank branches active in Italy in 2010 from the SIOTEC
database. This database, administered by the Bank of Italy, is based on compulsory
information provided by each branch to the Supervisory Agency. The only financial
institution excluded by SIOTEC is BancoPosta (the financial arm of the Italian post
office), for which we obtained the number of branches from the notes to its 2010 balance-
sheet.14
We use data from a nationally representative sample of bank accounts (the Survey on
Bank Fees and Expenditures, SBFE hereafter) to check that the fees and costs charged
by PattiChiari and non-PattiChiari banks are on average similar. This dataset, ad-
ministered by the Bank of Italy’s Banking and Financial Supervision Area, is currently
the most comprehensive source of information on bank costs in Italy and it collects in-
formation on fixed bank account costs, debit/credit card fees, costs of bank transfers
and withdrawals. We use the 2010 wave and we rely on data on 6717 current accounts
surveyed in SBFE and belonging to banks present in the SHIW dataset. Note that
the SBFE samples household’s bank accounts rather than banks, bank branches or cus-
tomers. This is a sensible choice because each bank may offer several types of accounts,
with different costs. The data at our hand suggest that SBFE might oversample accounts
of PattiChiari banks.
1% trimming. There are tied values and we discuss this point in more detail in Section B.3.14We were able to retrieve information on total assets in 2010 for 70 out of the 87 bank names presented
in the questionnaire from the Supervisory Register of the Bank of Italy. For the remaining 17 banks,it was not possible to recover such information because of different reasons. In 14 cases banks includedin the questionnaire did not have an autonomous balance-sheet: e.g., Unicredit Banca and UnicreditPrivate Banking share the same balance-sheet data in our data sources; the Banca Popolare di Intra waspart of VenetoBanca. Finally, for 3 banks neither the Supervisory Register nor other sources, such asBankscope, reported any information for year 2010.
15
5 Empirical Strategy
To identify the effect of bank information policies on financial literacy, we consider the
following linear model:
flip = α0 + α1Pattiip +Xipα2 + δp + νip (1)
where flip is financial literacy of the head of household i living in province p in 2010;
Pattiip is a dummy equal to 1 if the household is client of a bank belonging to the
PattiChiari Consortium; Xip is a vector of individual-level observable characteristics.15
δp is a province fixed-effect (province effects are denoted by µp or γp in equations (2)
and (3), respectively).
Similarly, to estimate the effect of bank information policies on financial assets, we
consider the equation:
wip = θ0 + θ1Pattiip +Xipθ2 + µp + ξip (2)
where wip is household’s financial assets in 2010.
The parameters of interest are α1 and θ1, which capture the effect of being client
of a bank in the PattiChiari Consortium on financial literacy and financial assets, re-
spectively. Provided that Pattiip is exogenous in equation (1) and (2), OLS estimates
of these parameters are consistent and have a causal interpretation. We expect α1 and
θ1 to be statistically significant and positive and we argue that the variable Pattiip is
exogenous in both equations. Investing in financial literacy is costly for individuals and
the investment decision is made comparing marginal cost and marginal benefit (Jappelli
and Padula, 2013b), where the latter may depend on financial wealth w and individuals’
characteristics X. We argue that the Quality Commitments undertaken by a PattiChiari
15In the baseline specification, controls include gender of the household head, a second-order polyno-mial in age, years of education of the household head, household current labor income, household size,marital status, size of the municipality (3 categories: 20,000-40,000; 40,000-500,000; more than 500,000inhabitants). We experimented with several subsets of these controls, and the inclusion of additionalcovariates is discussed in Section 6 and Appendix B.
16
bank lower the cost of acquiring financial education without directly affecting the ben-
efits related to additional financial literacy. Thus, other things being equal, we expect
PattiChiari clients to accumulate more financial literacy and end up with a higher stock
of it. For this interpretation to be valid, the following three assumptions must be fulfilled.
First, net of information policies, banks belonging to the PattiChiari Consortium may
not differ from other banks with respect to characteristics that may directly affect house-
hold financial assets (such as costs and fees charged, credit rationing, mortgage policies,
etc.). Second, individuals must not self-select as clients of a specific bank according to
its information policy. In other words, the choice of the bank must not depend on vari-
ables that, conditional on province (or municipality) of residence, directly affect financial
literacy or wealth. Third, banks must not join the Consortium in order to attract clients
that are wealthier or more financially literate. All these assumptions should hold within
province in a given year, because all our regressions include province fixed effects. These
three assumptions are partly testable on the basis of the data at our hands and deserve a
careful discussion. We discuss client selection in Section 7, while we focus on differences
in observable characteristics of banks in Section 8 . The evidence presented supports
the validity of the assumptions above and thus the causal interpretation of our estimates.
We also examine the effect of financial literacy on financial assets by considering the
model:
wip = β0 + β1flip +Xipβ2 + γp + εip. (3)
OLS estimates of β1 in Equation (3) may be biased because of: (i) individual unobserved
heterogeneity (i.e. ability, patience or preferences) correlated with both financial literacy
and the value of financial assets; (ii) reverse causality; (iii) measurement error. In the
first case, the sign of the bias is theoretically ambiguous and depends on the sign of the
correlation between flip and εip in equation (3). The theoretical arguments in Jappelli
and Padula (2013b) suggest that reverse causality should lead to an upward bias in
the estimate of β1, since individual wealth may affect the incentive to increase financial
17
education both through a change in the opportunity-cost of time and through a change
in the relative benefit from knowing how financial tools work.16 Finally, measurement
error in financial literacy may be substantial because respondents may guess the answer
at random or they may misunderstand the question.17 In these cases, OLS estimate
may be biased downward (Jappelli and Padula, 2013b). Overall, these remarks suggest
that: (i) OLS estimates may be biased and (ii) the sign of the bias cannot be assessed
theoretically since it depends on the relevance of each of the channels described above.
To address these issues, we pursue an instrumental variable approach and use equation
(1) as the first stage equation for the financial asset equation (3).18
The instrumental variable strategy relies on the three assumptions illustrated above (and
discussed more in details in Sections 7 and 8), since it rests on consistent estimation of
α1 and θ1 with OLS and an exclusion restriction (Angrist et al., 1996).
6 Findings
Table 2 shows estimate results of the baseline specification, which includes, along with
household covariates and province fixed effects, controls for motivations for choosing
the bank.19 The association between financial literacy and financial assets is positive
(column OLS). One standard deviation increase in financial literacy (i.e. one more correct
answer) is associated with an increase in financial assets by more than 2 thousand euros
16Lusardi and Mitchell (2007c), however, present empirical results that suggest that reverse causalityis not a relevant issue in American data (HRS).
17Evidence from the U.S. (Lusardi and Mitchell, 2009) and the Netherlands (van Rooij et al., 2011b)suggests that survey data on this issue are sensitive to how the question is worded.
18In principle, we might rely on the panel component of our dataset and include household fixed effectsin equation (3). There are three arguments against this choice. First, financial assets and financial literacyare both highly stable over the short time span we consider (2006-2010) in this application. Second,respondents may learn, or at least receive a stimulus to acquire knowledge, from previous interviews:the variation in correct answers may capture a retest effect rather than increases in financial literacy.This would affect estimates delivered by a fixed effect identification approach, while it does not drive ourresults, as we show in Appendix B. Third, using the longitudinal panel would cut sample size by halfbecause of attrition and survey design. Furthermore, in case of reverse causality or measurement error,coefficients in equation (3) are biased also when including fixed effects.
19Unless otherwise stated, estimated standard errors are robust to both heteroskedasticity and serialcorrelation at the province level, the largest level at which we have enough clusters (103) to obtainconsistent estimates of the second moment of the estimator distribution.
18
(about 12% of the average financial assets holdings in our baseline sample). As discussed
in Section 5, this estimate is unlikely to reflect a causal relationship but represent an
important benchmark for our analysis.
Moving to the first stage results (column FS in Table 2), being a client of a PattiChiari
bank leads an individual to give 0.24 additional correct answers, that is approximately
one fourth of a standard deviation of financial literacy, nearly 12% of the average num-
ber of correct answers given in the sample. The effect of bank information policies on
financial literacy is sizeable compared to other explanatory variables in our regressions:
the impact of being client of a bank that belong to the PattiChiari Consortium is com-
parable to the effect of increasing education by five years.20 The F-test on excluded
instruments is about 44, showing that, when we interpret this equation as the first step
of our IV strategy, the instrument is not weak.21
The instrumental variable estimates we present in the paper apply to the sub-
population of compliers (Angrist et al., 1996), i.e. individuals forces to change their
treatment status because of the assignment to the treatment. In our case the compliers
are the individuals whose knowledge of financial instruments increases because of the
information policies of the PattiChiari banks. When both the instrumental variable and
the treatment variable are binary, the proportion of compliers can be estimated and cor-
responds to the first stage estimate (Angrist et al., 1996), i.e. the effect of the instrument
on the endogenous regressor. In our application, the instrumental variable is binary, a
dummy variable indicator of being PattiChiari client. In Table 3 we consider four al-
ternative measures of financial literacy and we estimate the effect of being PattiChiari
20As discussed in Section 2, evidence on the causal impact of interventions on financial literacy is stilllimited. Assessing the effect of bank information policies relative to other interventions is, therefore, notstraightforward. Klapper and Panos (2011) consider the role of newspapers in Russia, and find that a 1%increase in their number raise average financial literacy in the population by nearly 4%. Comparatively,increasing the number of universities by 1% would increase financial literacy by 0.15%. Carpena et al.(2011) assess the effect of a video-based financial literacy program in India, finding it increases financialawareness by about 8%. The relative size of the effect we find seems, thus, slightly larger than the one ofprevious information/training instruments in Russia and India. More recently, Luhrmann et al. (2015)studies the effect of a financial education program for teenagers in Germany, finding it both significantlyincreases their financial knowledge and reduce their propensity to buy on impulse.
21Staiger and Stock (1997) indicate a rule of thumb suggesting that the F-statistic should be greaterthan 10 to rule out weak identification problems.
19
client on each of these binary indicators: these are also alternative estimates of the pro-
portion of compliers in the population. We consider four definitions: a dummy equal to
1 if all three questions have been answered correctly and zero otherwise or we consider
the correct answers of each of these three questions separately. Being a PattiChiari
client is strongly correlated with all these definitions and leads to about 13% increase
in the probability of answering all questions correctly, 14% increase in the probability
of answering correctly the question on inflation and an 11% increase in the probability
of answering the loan or portfolio question correctly, with respect to the sample average
(see Table 1). These estimates suggest that the compliers represent a fraction of our
sample ranging between 5% and 10% of the population.
Financial assets of households who have the main current account in a PattiChiari
are 1.9 thousand euros higher than clients of other banks, roughly 10% of the average
value of assets in our sample (column ITT in Table 2). The resulting instrumental vari-
able estimates (column IV of Table 2) are larger than what simple association suggests:
on average, a one standard deviation increase in financial literacy (i.e. an additional
correct answer provided, about a 50% increase with respect to the average financial
literacy), conditional on other covariates, leads to an increase of household assets of 8
thousand euros, that roughly correspond to 35% of a standard deviation in financial
assets.22 Thus, in line with findings by Lusardi and Mitchell (2011b), the causal effect
of financial literacy is around four times higher than suggested by the corresponding
association. The downward bias in OLS estimate can be explained by a negative corre-
lation between financial literacy and the error term in the financial assets equation or
by the presence of substantial measurement error in financial literacy. First, individual
specific unobservables may be negatively correlated with financial literacy and positively
associated with financial assets (or viceversa). For instance, wealthy individuals may
have a higher opportunity cost of time and/or expect (more efficient) professional port-
folio management to yield a greater return, in absolute terms. For these reasons, they
22This effect apply to the sub-population of individuals whose knowledge of financial instrumentsincreases because of the information policies of the PattiChiari banks and cannot be generalized to theaverage sample household without further assumptions (Angrist, 2004).
20
may decide not to directly invest in their own financial literacy but, on the contrary,
to delegate portfolio management decisions to a professional advisor. This channel is
consistent with downward bias of the OLS estimate: if professional advisors are indeed
able to generate larger returns the observed association between financial literacy and
financial assets will be smaller in absolute terms than the true causal effect. Second, the
indicator(s) for financial literacy may suffer of measurement error, since the respondent
may misunderstand the survey questions or pick the answer randomly. This argument
is consistent with evidence on the sensitivity of answers to financial literacy questions
documented by Lusardi and Mitchell (2009) for the US and by van Rooij et al. (2011b)
for the Netherlands.
We also consider alternative measures of being a PattiChiari client, as reported in
Table 20 in Appendix C. Along with the indicator based on the main bank account we
use in the baseline specification (first panel), we rely on a dummy variable that is equal
to one whether at least one of the banks of the respondent belong to the PattiChiari
Consortium (second panel). Finally, we include records for which the name of the bank
is missing and we consider banks with missing names to signal a non-PattiChiari bank.23
Then, we create a dummy taking the value one if at least one of the accounts recorded in
the survey is with a PattiChiari bank, and zero otherwise. We expect the latter indicator
to have some degree of attenuation bias with respect to our main indicator, because some
of the banks for which the name is unavailable may belong to the Consortium. All results
are confirmed using these alternative indicators.
In the baseline regression we control for province fixed effect. But municipality-
specific factors may affect both the level of financial assets and the probability of being
client of a PattiChiari bank (e.g. different banks may decide to open branches in a
23The SHIW questionnaire includes a list of names from which the respondent may pick her bank.This list seems to include a slightly larger share of PattiChiari banks with respect to the overall universeof Italian banks. Our baseline analysis exclude records for which the name of the banks is not listedand might, in principle, be biased towards PattiChiari clients. Because of this we replicate our results:a) changing the sample and including households whose main bank has a missing name; or b) using analternative indicator of being PattiChiari client that exploits information on all the accounts held bythe household; or c) including controls for the number of current accounts held by the household; or d)restricting the attention to large or small banks only. Baseline results are confirmed.
21
particular type of municipality). We include municipality-fixed effects and replicate our
baseline results: previous findings are confirmed, as shown in Table 21 in Appendix C.
We also the robustness our findings to several falsification checks, which are reported
ans discussed in Appendix B. In particular, we explore: i) the role of retest effects and
the timing of the effect of bank information policies and financial literacy; ii) the role
of trust in institutions; iii) the sensitivity of our results to sample selection and the
functional form. Our results turn out to be robust to all these checks.
7 Evidence Against Self-Selection of Households into Pat-
tiChiari Banks
In this section we discuss evidence supporting the absence of self-selection of households
into PattiChiari or non-PattiChiari banks, that lies behind the validity of our identifi-
cation strategy.
We first provide evidence on the determinants of households choice of banks. Then we
tackle the potential reverse causality that may explain the estimated effect. We then
show that our results are robust to the use of a propensity-score matching estimator,
and we perform an overidentification test on the validity of the exclusion restriction for
the IV model. In the Appendix B.2, we also show that different measures of trust are
not different between PattiChiari and non-PattiChiari households.
It is worth emphasizing that all the results are net of province (or municipality) fixed
effects: we are, thus, exploiting only the within-province variability of the instrument
for identification (see Table 21 in Appendix C).
Why do households become customers of PattiChiari (or non-PattiChiari)
banks?
Bank clients may self-select into PattiChiari banks according to their financial knowl-
edge. Hence, respondents with higher financial literacy may value more financial con-
ditions and products supplied by the bank and, in turn, they may decide to become
22
customer of a specific bank according to its financial offerings. Even though the ho-
mogeneity between PattiChiari and non-PattiChiari banks concerning cost and credit
strategies, documented in Section 8, weakens this self-selection issue, we explore further
the determinants of household choices by exploiting information collected by the SHIW.
We rely on answers to the following question “Why did you choose [the bank you use
more often] when you and your household first began using it?”. The set of available
alternatives can be naturally grouped in the following three categories: 1) convenience,
i.e. convenience to home/work, respondent’s employer’s bank (or respondent’s business’s
bank); 2) financial/economic reasons, i.e. favourable interest rates, speed of transaction
execution, range of services, low fees for services, possibility of online banking; 3) bank
type, i.e. it is a well-known/ important bank, staff courtesy. The main reason for choos-
ing a bank turns out to be related to convenience, that is the only determinant for 63%
of PattiChiari clients and for 67% of other respondents. Financial conditions are the
only determinant for less than 15% of the respondents in both groups. We interpret
this evidence as follows: the preferred financial offerings may be readily available and
consumers typically decide their bank depending on the one that is more conveniently
located to them. In other words, the constraint on financial services is not binding
and it is satisfied by the abundance of options, hence individuals choose on the basis
of their second (binding) constraint, namely convenience. As we shown in Table 22 in
Appendix C, our results are confirmed when we omit these controls. We include them
in our baseline regression to improve the precision of our estimates.
Transparency or, more generally, information acquisition is not mentioned among
the alternative reasons for choosing the bank. For this reason, one may argue that we
cannot rule out the possibility that clients join a PattiChiari bank to get more infor-
mation. Households that are wealthy or willing to use sophisticated financial products
may prefer PattiChiari banks because of the information policies they provide. This
potential problem of self-selection is particularly hard to test, as it entails gouging
unobserved characteristics (propensity to invest) and potentially simultaneous choices
(information seeking and bank choice). Nonetheless, some indirect evidence can be ob-
23
tained by exploiting data on the source of financial information used by respondents to
make investment decisions. In the 2010 SHIW survey, household heads are asked about
the information source they have been consulted before the current investments were
performed. They can be grouped in three broad categories: formal sources (financial
intermediaries, experts and advisors), and two categories of informal ones, namely press
or websites and the residual (friends, others, do not know). Descriptive statistics of the
resulting indicators are reported in Table 1. Since this information is collected only for
respondents who hold financial assets other than bank accounts (“investor households”
hereafter), the sample size reduces to 1492 households, who are, on average, wealthier
and more financially sophisticated than the rest of the sample.24 We estimate the base-
line regressions on the subsample of “investor households”, with and without controlling
for sources of financial information. Particularly, we control for an indicator that takes
value 1 if the respondent relies only on that specific source of financial information and
0 otherwise (descriptive statistics of these variables are shown in Table 1). Results are
reported in Table 4. Conditioning on sources of information does not alter the effect
of the variables of interest. Moreover, the coefficients associated to sources of financial
information are not statistically significant. The reduced form (first-stage and intention-
to-treat) and instrumental variable estimates are shown not to be confounded by the
direct effect of sources of financial information. It is worth noting that, when we restrict
our analysis on the sample of “investor households”, both the intention-to-treat and the
effect of financial literacy on financial assets become larger in magnitude, although less
precisely estimated with respect to the entire sample. This is informative of the potential
heterogeneity of the effects in the population.
The reasons why banks join the PattiChiari Consortium relate to the advertisement
of the bank commitment to transparency (see Section 3). We claim that the effectiveness
of this advertising does not depend on wealth or financial literacy, thus entering into the
Consortium does not have the effect of attracting wealthier or more financially literate
24Average financial wealth of “investor households” is 38.22 thousand euros, compared to 19.38 thou-sand euros for the entire sample. The average number of correct answers on the questions measuringfinancial literacy is 2.2 among “investor households”, 1.9 in the broad sample.
24
clients. In other words, our identification strategy hinges on the assumption that the
effectiveness of advertisement policies implemented by the PattiChiari Consortium does
not depend on financial literacy nor wealth. While we cannot test this assumption di-
rectly due to lack of relevant data, we collect complementary evidence documenting that
individuals do not self-select as client of a bank belonging to the PattiChiari Consortium
according to their wealth, as illustrated in the next paragraph.
Reverse causality: Are wealthier clients more likely to become PattiChiari
clients?
In order to document absence of self-selection of clients according to their wealth, we
exploit the panel component of the SHIW dataset to retrieve information on household
financial assets in 2002, i.e. before the PattiChiari Consortium was created. First, we
perform a“placebo”test, regressing 2002 assets on being a PattiChiari client in 2010. We
fail to find a significant correlation between being such a client in 2010 and 2002 assets.25
Second, in order to control for “pre-treatment” differences in the asset endowment, we
estimate the baseline specification by including financial assets in 2002 among the control
variables. Results are reported in Table 23 in Appendix C. The upper panel shows the
results with the additional controls, while the lower panel reports the baseline results
estimated on the sample for which we can recover information on financial assets in
2002. The main coefficients in the two panels are similar both in terms of size, while
their statistical significance is hampered by the huge cut in sample size. Also the first
stage result (column 2) remains stable when we include financial assets in 2002 as a
regressor.
In order to rule out self-selection of wealthiest clients into PattiChiari banks as the
driver of our findings, we compare households with the same assets before the creation
of the Patti-Chiari Consortium and examine the growth in their assets between 2002
and 2010 or 2002 and 2008 or 2002 and 2006. The exercise is performed on the sample
of individuals surveyed in 2002 and at least in another year (2006, 2008 and 2010).
25The coefficient associated to the PattiChiari dummy in this test regression is 12.663 with a standarderror of 13.847.
25
We compare households with the same assets before the creation of the PattiChiari
Consortium and examine the growth in their assets between 2002 and 2010 or 2002 and
2008 or 2002 and 2006, controlling also for a vector of household characteristics and year
specific province fixed effects. We standardized growth to have zero mean and unitary
variance so that coefficients can be easily interpreted as effects in units of standard
deviation. Results are reported in Table 5. Figures confirm our results: they suggest
that OLS underestimates the effect of financial literacy on growth of financial assets and
that the causal effect of financial assets is as large as about 15% of a standard deviation
and positive. As a robustness check, we estimate a model that allows for heterogeneity
across years in the effects of financial literacy on financial assets growth (Table 24 in
Appendix C). The effect of financial literacy is positive and significant in all years and
about 4% of standard deviation smaller in 2008 (the year after the fall in stock market
prices documented by Bottazzi et al. (2013)), with respect to 2006 and 2010.
We also explore whether there are differences in pre-treatment (2002) financial assets
among individuals who switch between PattiChiari and non-PattiChiari banks between
2006 and 2008 or between 2008 and 2010. We focus on households sampled for at least
two consecutive waves over the period 2006-2010 and we group them in four categories
of households: households who remained clients of PattiChiari over the two waves;
households who remained clients of a non-PattiChiari bank; those who moved from a
PattiChiari to a non-PattiChiari bank; and those that made the opposite move. We
then merge to these data, data on financial assets in 2002. Although for this exercise we
can only use individuals interviewed in 2002, the distribution of the types of switch is
fairly similar to the one in the full sample: 77.9% remained clients of PattiChiari banks,
13.6% remained clients of a non-PattiChiari bank, 5.4% switched to a non-PattiChiari
bank, 3.1% switched to a PattiChiari bank. In almost all these cases (99.4%) the switch
has been voluntary, i.e. due to switch of the household’s current account from one
bank to another, rather than because the bank entered or exited the Consortium. We
then test whether there are significant between group differences in financial assets in
2002, i.e. before the PattiChiari consortium was created, using our baseline regression
26
(yet, controlling for labor income in 2002, rather than in 2010). Table 6 presents the
results for three different regression models: column (1) does not control for reasons for
choosing the main bank; column (2) controls for reasons for choosing the main bank
(baseline); column (3) controls for reasons for choosing the main bank, costs charged
by the bank and bank characteristics (i.e. bank size). We cannot detect significant
differences in financial assets between groups in any specification: clients who remained
with PattiChiari banks between 2006 and 2008 or 2008 and 2010 are not wealthier nor
poorer than those who switched to Non-PattiChiari banks, nor the two types of “stayers”
or “switchers” differ. We interpret this as evidence that there is no selection of clients
into PattiChiari banks.
Households who hold more than one bank account may be wealthier than the average
and may already diversify their portfolios using services from different banks. As having
more bank accounts increases the chances of being a PattiChiari client, this may generate
spurious correlation. To control for this possibility, we include household’s number of
bank accounts in our baseline specification. Results, presented in the third panel of
Table 18 in Appendix B, confirm our baseline findings.
Financial assets and financial literacy may correlated to the tenure status of the
respondent. To examine this issue, we include among the regressors in the baseline
specification an indicator for whether the family owns the dwelling where they live (Table
25 in Appendix C). Homeownership turns out to be positively associated with the level
of both financial literacy and financial assets, but our findings are robust to the inclusion
of this additional regressor.
A matching estimator
We estimate the effects of being client of a PattiChiari bank on financial literacy and
financial assets using a matching estimator, based on propensity score matching, and we
replicate the baseline regression analysis of the paper on the sample where we observe
clients of PattiChiari and non-PattiChiari banks that have similar probability to be a
PattiChiari client based on observables. This exercise allows us to match almost all Pat-
27
tiChiari clients with similar non-PattiChiari clients in terms of demographics, reasons
for choosing the main bank, province of residence and household labour income. After
excluding 36 households we cannot match, we replicate our baseline estimate and find
essentially the same results.
We start estimating the propensity score through a logit model, where the dependent
variable is the probability of being PattiChiari client in 2010. The balancing property is
satisfied once we include among the controls demographics of the household head (age,
age squared, gender, years of education, marital status) and household characteristics
(number of household members, household labour income), municipality population size
(inhabitants), province fixed effects, reasons for choosing the main bank and some in-
teractions between those controls.26 We perform radius matching with caliper 0.005,
with replacement, matching each treated individual with two controls: observations of
PattiChiari clients are used only once, while observations in the control group (non-
PattiChiari clients) may be used in several matches. Table 26 in Appendix C assesses
whether the matching procedure is able to balance the distribution of covariates among
PattiChiari clients and non-PattiChiari clients. For each variable, we show the mean
value in the group of clients of PattiChiari and non-PattiChiari banks in, respectively,
column 2 and 3. The last two columns report the standardized percentage bias across
covariates after matching and the p-value of a t-test of equality between the means. We
do not reject that covariates are balanced at all conventionally used levels of significance,
with few exceptions (for which we do not reject the null at 1% significance level) The
last line in the table shows that the joint test does not reject with a p-value of 0.103.
Figure 1 displays a graphical representation of the common support. Notably, most ob-
servation are on the common support (99.3%) and we are not able to match only 0.007%
of the households (36 out of 4865) with a very high propensity score. We estimate the
effect of being PattiChiari clients on financial literacy (first stage effect) and financial
26More precisely, in addition to the levels of the listed variables, we include interactions between:household head gender and marital status; household head marital status and municipality size; familysize and and municipality size; household head gender and municipality size; household head age andmunicipality size; family size and household labour income; household head marital status and householdlabour income.
28
assets (intention-to-treat effect) for the treated using a matching estimator: 2.38 with a
standard error of 1.03 for financial assets; 0.17 with a standard error of 0.06 for financial
literacy. The intention-to-treat effect is statistically significant at 5%; the first stage
estimate is statistically significant at 1%. Both estimates in line with those based on the
OLS estimator reported in Table 2 (see FS and ITT column). Table 7 reports results of
instrumental variable regression on the common support with the specification of Table
2 and on the common support with the specification that achieves covariate balancing.
An overidentification test
We exploit the panel dimension of the SHIW to perform an over-identification test. We
focus on households sampled for at least two consecutive waves over the period 2006-
2010 (i.e. either in 2006 and 2008 or in 2008 and in 2010; 3196 observations) and
we group them in the four categories of households discussed above: households who
remained clients of PattiChiari over the two waves; households who remained clients
of a non-PattiChiari bank; those who moved from a PattiChiari to a non-PattiChiari
bank; and those that made the opposite move. As discussed above, these four groups
identify three instrumental variables (plus a baseline group). All these variables can be
used to predict financial literacy in 2010. Results are summarized in the first panel of
Table 27 in Appendix C. The lower panel of Table 27 replicates the exercise focusing
on bank-switchers over the 2006-2010 period (two waves). Despite the sample size is
sensibly reduced, the main results of Table 2 are confirmed: one standard deviation rise
in financial literacy causes an increase in financial assets by 12.7 and 9.6 thousand euros
in, respectively, the upper and lower panels. Looking at the over-identification test, the
Hansen J statistic does not reject the null hypothesis of exogeneity of the instruments
at all conventional levels of significance: the p-value associated to the test is 0.43 in the
top panel of Table 27 and 0.75 in the bottom panel. In short, this evidence supports the
exogeneity of our instrument (i.e. being client of a PattiChiari bank) in the financial
assets equation.
29
8 Evidence Against Self-Selection of Banks into the Pat-
tiChiari Consortium
In this Section, we present empirical evidence supporting the absence of self-selection of
banks into the PattiChiari Consortium, one of the pillars on which the causal interpre-
tation of our findings stands. To this scope, we compare observable characteristics of
banks belonging or not to the consortium and we check the robustness of our findings
across relevant sub-samples and to the inclusion of bank-specific controls.
We start by analysing to which extent banks differ according to costs and credit
policies. Descriptive statistics on costs of bank accounts charged by banks belonging
and not belonging to the PattiChiari Consortium are reported in Table 8, together with
a Welch test on equality between means of the two groups. Differences in costs and fees
are generally not statistically significant, the only exception being average debit card fees
that are slightly larger (less than 3 euros per year) for non-PattiChiari banks. Focusing
on ATM withdrawal costs, the Quality Commitments of banks joining the PattiChiari
banks did not prescribe anything related to free access to ATM machines to customers
of other banks of the Consortium and, on average, ATM withdrawal costs are 0.31 for
PattiChiari banks and 0.29 for non-PattiChiari banks, not statistically different from
each other. Our main results are unaffected by the inclusion of these costs among the
control variables (see the upper panel of Table 30 in Appendix C).
We check whether banks that belong to the PattiChiari Consortium differ from other
banks with respect to credit rationing, exploiting information about liquidity constraints
provided by SHIW. We do not find evidence of any correlation between being PattiChiari
client and being liquidity constrained, that may signal differences in credit rationing, at
standard significance levels.27
27Following Jappelli et al. (1998), we consider a household to be “liquidity constrained” if it either (a)applied to a bank or a financial company to ask for loan or mortgages and the application was rejected,or (b) considered the possibility to apply but did not, thinking that the application would have beenrejected. At average values of the covariates, the marginal effect of being PattiChiari client on theprobability of being liquidity constrained is -0.005 with a standard error of 0.007. The model controls forprovince fixed effects, municipality size, and individual observable characteristics (quadratic polynomialin age, gender, marital status, household size, education, family labor income and reason for choosing
30
In principle, large and small banks may differ with respect to their policies about
investment or client relations, to their ability to attract clients, and to other unobserv-
ables that may affect customers wealth or financial literacy. Therefore, we examine to
which extent banks belonging or not to the PattiChiari Consortium are different accord-
ing to their size. To this purpose, Table 9 ranks banks according to their total assets,
and provides their number of branches and PattiChiari status: 49 out of the 70 banks
listed (70%) belong to the PattiChiari Consortium. Names of PattiChiari banks are
underlined, names of non-PattiChiari banks are in blue. The following two columns in
Table 9 report the decile of the empirical distribution of total assets and of the number
of branches to which each bank belongs to. Among the top-10 banks in terms of assets
(reporting more than 40,000 millions euros of assets) there are 9 PattiChiari banks and
1 non-PattiChiari bank, while all top-30 banks in terms of number of branches are part
of PattiChiari. Yet, apart from the top-sized banks, the rest of the distribution is much
more mixed. Within the sub-sample of small banks, PattiChiari and non-PattiChiari
banks are evenly distributed over both measures: among banks with below-median as-
sets, there are 20 PattiChiari and 15 non-PattiChiari banks, among bank with below-
median number of branches there are 17 non-PattiChiari and 18 PattiChiari banks.28
We test for statistical differences in the mean of these variables by PattiChiari status in
Table 10. The difference is not statistically significant.
To check whether different characteristics between PattiChiari and non-PattiChiari
banks drive our results, we perform three different exercises: 1) we replicate our exercise
excluding the ten largest banks (in terms of assets or number of branches); 2) we replicate
our baseline results using data of clients of small or large banks only; 3) we add the control
variables for bank characteristics to our baseline regression.
First, we run our baseline regression on a sample that excludes the top 10 banks (where
the incidence of PattiChiari banks is higher) in terms of, alternatively, level of total
the main bank).28The Wilcoxon rank-sum test of assets and number of branches byPattiChiari status among small
banks largely fails to reject the null of different distributions (p-values equal 0.79 and 0.82, respectively).
31
assets, or number of branches.29 Results are reported in Table 28 in Appendix C and
confirm the baseline results in Table 2.30
Second, we look for a sample of banks where the overall size distribution for PattiChiari
and non-PattiChiari banks is similar. To this purpose, we estimate the baseline model
among households having their current account in a “small bank”, which is defined,
alternatively, as having lower-than-median assets or number of branches.31 Within the
sub-sample of small banks, PattiChiari and non-PattiChiari banks are evenly distributed
over both dimension: among banks with below-median assets, there are 20 PattiChiari
and 15 non-PattiChiari banks, among bank with below-median number of branches
there are 17 non-PattiChiari and 18 PattiChiari banks.32 Results are reported in Table
29, together with results obtained in the complement samples of “large banks”. Point
estimates of the effects are generally consistent with our baseline in both sub-samples,
although the statistical significance of the results on small banks is severely hampered by
the large reduction in sample size: larger banks tend to have more customers and when
we focus on clients of small banks we are left with around 500 households. This is also
consistent with the fact that customers tend to choose their main bank for convenience
to home or work and in this exercise large banks are those who have more branches by
definition.
Finally, we test for the robustness of our results by including assets and number of
branches among the covariates. The bottom panel of Table 30 in Appendix C shows
29These banks are, respectively: UniCredit, Intesa Sanpaolo, Banca Monte dei Paschi di Siena, BancaNazionale del Lavoro, Banca Popolare del Mezzogiorno Banca popolare dell’Emilia Romagna, BancaPopolare di Milano, Cassa di Risparmio di Parma e Piacenza, BancoPosta, Banca Carige (for totalassets) and UniCredit, BancoPosta, Banca Monte dei Paschi di Siena, Intesa Sanpaolo, Banco di Napoli,Banca Popolare di Verona-S. Giminiano e Prospero, Banca Nazionale del Lavoro, Banca Popolare diNovara, Banca Popolare di Lodi, Cassa di Risparmio di Parma e Piacenza (for the number of branches).
30The descriptive statistics on financial assets and financial literacy in these sub-samples are similarto the one reported in Table 1. The share of PattiChiari clients in these sub-samples is around 67%compared to 73% in Table 1.
31The median of total assets and number of branches two variables on the sample of 70 banks for whichthe information is available- see Table 9- are, respectively, 9328.5 and 307. Table 29 reports descriptivestatistics on key variables in the sub-samples (share of PattiChiari clients, financial literacy and financialassets): these are similar to those reported in Table 2 for all sub-samples.
32The Wilcoxon rank-sum test of assets and number of branches byPattiChiari status among smallbanks largely fails to reject the null of different distributions (p-values equal 0.79 and 0.82, respectively).
32
that results are not affected by these additional controls. To sum up, the distribution
ofPattiChiari and non-PattiChiari banks in terms of assets or number of branches differ
mostly in the upper tail, with PattiChiari banks being more frequent among banks with
higher total assets and larger number of branches (see Table 9). Yet, the two groups of
banks do not differ in size on average and, most importantly, our main results seem not
to be driven by bank size.
9 Mechanisms and Heterogeneity
Financial literacy can affect wealth through many different channels.33 With the data
at our hand, we can analyse three potentially relevant mechanisms that may drive the
positive effect measured in the aggregate. First, we investigate to which extent portfolio
composition depends on financial literacy of the respondent (Christelis et al., 2010; van
Rooij et al., 2011b,c). Second, we examine whether literate respondents are more prone
to plan for retirement (Lusardi and Mitchell, 2008; van Rooij et al., 2011a), that, in
turn, may enhance saving. Finally, we focus on the relation between financial literacy
and saving behaviour.
Looking at portfolio composition, we exploit information about assets held by SHIW
respondents and we examine to which extent the probability of holding equities, mutual
funds and government bonds differs according to financial literacy. We also consider a bi-
nary indicator of household participation to the stock market, either directly or through
a mutual fund. To analyze the role of retirement planning, we use answers to the follow-
ing question (not asked to retirees): “Have you ever thought about how to arrange for
your household’s support when you retire?”. Saving rate is defined as the ratio between
savings and income.
Results reported in Table 11 suggest that financial literacy leads to an increase in stock
33We refer to Jappelli and Padula (2013b) (note 5) and Lusardi et al. (2011) for a discussion of thelinks between financial literacy and financial wealth. Relevant channels relate to expectations (liter-ate individuals may be more accurate and/or less biased), preferences (financial literacy may ease theunderstanding of risk and reducing ’direct risk aversion’), perceptions (financial literacy may reduceunderestimation of compounding effects), diversification, and the cost charged on loans or mutual funds.
33
market participation and to a more diversified portfolio. While lack of availability of
more detailed data on portfolio management prevent us to examine the effect of finan-
cial literacy on the quality of financial decisions, as in Guiso and Viviano (2015), our
results are consistent with financial literacy to promote a better portfolio management,
in terms of diversification and stock market participation. Financially literate respon-
dents holding a more efficient and diversified portfolio may contribute to explain the
positive effect of financial knowledge on financial assets. This channel may be partic-
ularly relevant in Italy, a country characterized by relatively low participation rates in
the stock market (Bottazzi et al., 2013).
Turning to planning for retirement and saving rate, while the estimated coefficient is pos-
itive, we fail to detect a significant causal effect of financial literacy on both outcomes.
This finding may reflect the fact that Italy is characterized by a generous pension sys-
tem, compared to other countries, and relatively high saving rates. Undersaving to
meet retirement target in Italy might thus be a less relevant issue with respect to other
countries.
To shed more light on the heterogeneity of the impact of bank information policies and
on the effect of financial literacy on financial assets, we estimate our baseline model across
age and education groups. The results are shown in Table 12. The first and the second
panels report estimate for, respectively, low and high-educated respondent; the third and
fourth ones consider, respectively, the young and the elderly and the bottom panel refers
to the subgroup of low-educated individuals aged 60+. The effect of financial literacy
on financial assets is the largest for low-educated and old respondents (the IV coefficient
is 12.6 in the bottom panel). This finding is consistent with return of financial literacy
that are decreasing in education.34 In addition, individuals aged 60+ are expected to
be wealthier than the young and, in turn, to gain more from their financial knowledge.
The bias in the OLS estimate is the highest when 60+ and low-educated respondents are
considered. The determinants of the bias described above are, thus, more relevant in this
34This latter result is in line with the results by Cole et al. (2011) showing, in a different context, asignificant impact of financial literacy (on the demand for current accounts in India and Indonesia) onlyfor those with limited education or financial literacy.
34
sub-sample. First, education is expected to be negatively correlated with the probability
of choosing the answer at random or not understanding the questions. Moreover, the
elderly may be less concentrated when answering the questionnaire and, thus, more
subject to measurement error. Second, low-educated elderly may be more likely to
delegate their portfolio’s management, rather than investing in their financial knowledge,
further increasing the bias in the OLS estimate of the effect of financial literacy on
assets. Turning to the effectiveness of bank information policies, they significantly foster
financial knowledge in all the subgroups we consider, but the magnitude of their effect
is heterogeneous. Results in Table 12 illustrate that the marginal effect is substantially
greater for low educated respondents and for those aged 60+.
10 Concluding remarks
The interests by scholars and policy-makers, in Europe and in the U.S., on the deter-
minants of financial literacy and on the link between financial literacy and wealth has
been constantly increasing in the last years. Institutions, such as the OECD, the U.S.
Treasury Department and the European Commission, have expressed the need for im-
proved financial knowledge, emphasizing the role of formal financial education in schools
or at the workplace. However, whether these programs affect financial knowledge and
financial decisions, and who are the individuals reacting are still controversial issues.
This research contributes to the discussion focusing on a new and possibly comple-
mentary policy instrument. We start recognizing the potential role of bank information
policies in reducing the cost of acquiring financial literacy and we posit a novel question:
do bank information policies actually increase financial literacy and, in turn, household
financial wealth? To answer this question, we identify a group of Italian banks, the Pat-
tiChiari Consortium, that implements active policies aimed at increasing transparency.
Requirements to enter into the Consortium translate into the provision of information
about simple economic concepts by banks to their customers.
The fraction of individuals whose financial literacy is affected by these information
35
policies ranges between 5% to 10% while being client of a PattiChiari bank translates
into a 12% increase in financial literacy on average. We use the exogenous variation in
financial literacy induced by bank information policies to assess the effect of financial
literacy on financial assets. Our baseline estimate of the effect of about one standard
deviation increase in financial literacy is approximately 8 thousand euros in 2010 and
reflects 35% of standard deviation of financial assets in the sample. The effects of
financial literacy on financial assets growth - over a time horizon of between eight and
four years - are as large as 15% of a standard deviation in growth over the same time-
span.
We find that both the improvement in financial literacy and the increase in financial
assets are driven by a specific subgroup of households: those whose head is older and
less educated. This result is reasonable, given the type of intervention we studied: the
provision of simple and basic economic concepts, which are likely to be more relevant
for individuals who are less exposed to other sources of financial information, such as
economic newspapers or the web. Therefore, even though the effectiveness of information
policies, in terms of compliance, is limited, their impact on welfare may not be negligible
as they affect financial knowledge and behaviour of costumers who may profit more from
them. Moreover, our findings may stimulate further research broader interventions,
aimed to convey more sophisticated information.
We find that the effect of financial literacy on financial assets is mainly driven by
portfolio management and diversification. While we cannot assess whether the quality of
investment decisions improves, increased portfolio diversification in our case-study may
be possibly interpreted as intrinsically better, given Italy’s historically low rates of stock
market participation (particularly among the elderly).35
The impact on financial assets is remarkable, if compared with other studies of fi-
nancial education programs that typically find small or no effects on financial decisions.
We argue that two features, besides the afore-mentioned impact on the elderly, may ex-
35In a different setting, where households already diversify portfolios, the improvement in the qualityof financial decisions related to an increase of financial knowledge may not lead to such large effects (seeGuiso and Viviano (2015)).
36
plain this result. First, the fact that the less educated and elderly are those who benefit
from the policy. Indeed, (i) returns to financial education are likely to be decreasing
in education, and (ii) older households may under-invest their financial endowments ac-
cumulated over their lifetime. Second, individuals who receive information from banks
are treated for a rather long period of time (i.e. about 9 years, the average lenght of
the relation between client and bank), while financial education programs last at most
for some months. Third, individuals receive financial education in a setting where this
information is relevant. Indeed, it is mostly through banks that households make their
financial decisions, such as saving and diversifying their portfolio of investments.
These results show how financial education may be taught in different settings, which
may be potentially more salient than schools or workplaces. They also point to the im-
portance of targeting specific types of individuals whose marginal benefits from financial
literacy may be larger, and who are tipically neglected by standard financial education
programs.36
Finally, our analysis shows that the relationship between financial literacy and fi-
nancial assets is largely underestimated by simple OLS correlation and this is mainly
due to measurement error: this suggests that there is room to improve the way financial
literacy is measured.
Acknowledgements
We acknowledge the financial support of MIUR- FIRB 2008 project RBFR089QQC-003-
J31J10000060001. We thank E. Battistin, R. Bottazzi, T. Bucher-Koenen, M. Cervellati,
M. Giofre, M. Hurd, T. Kuchler, P. Legrenzi, B. Luppi, M. Padula, G. Pasini, E. Ret-
tore, E. Sette, M. Wakefield, Y. Winter and participants to the BoMoPaV Economics
Meeting in Padua, the Netspar International Pension Workshop in Paris, the SAVE-
PHF Conference in Munich, the CERP Annual Conference on Financial Literacy, Saving
and Retirement in an Ageing Society, the European Association of Labour Economists
36Notice, however, that these individuals may be those that have higher levels of financial assets,relative to the average individual in the population. As such, these interventions will not necessarilyreduce inequality in wealth.
37
(EALE) Annual conference in Bonn, Munich Center for the Economics of Aging (MEA)
and seminars at the University of Bologna and at the Johannes Kepler University of
Linz for useful comments on an earlier version of this work. We thank the Bank of Italy
Banking and Financial Supervision Area (notably, P. Franchini and A. Scognamiglio)
for kindly providing data on bank account costs. We are indebted with the PattiChiari
Consortium (notably, V. Panna and L. Napoleoni) for the data provided and for intrigu-
ing discussions about their initiatives. The views here expressed are those of the authors
and do not necessarily reflect those of the Bank of Italy. The usual disclaimer applies.
38
References
Altman, M., 2012. Implications of Behavioural Economics For Financial Literacy and
Public Policy. Journal of Socio-Economics 41, 677–690.
Angrist, J., 2004. Treatment Effect Heterogeneity in Theory and Practice. The Economic
Journal 114, C52–C83, issue 494.
Angrist, J. D., Imbens, G. W., Rubin, D. B., 1996. Identification of Causal Effects Using
Instrumental Variables. Journal of the American Statistical Association 91, 444–455.
Biagi, F., Giraldo, A., Rettore, E., 2009. Gli Effetti dell’Attrito sulla Stima della Dis-
uguaglianza in Italia. In: Brandolini A., C. Saraceno, A. S. (Ed.), Dimensioni della
Disuguaglianza in Italia: Poverta , Salute, Abitazione. Il Mulino.
Bottazzi, R., Trucchi, S., Wakefield, M., 2013. Wealth effects and the consumption of
italian households in the great recession. Working Papers 13/21, IFS.
Bruhn, M., de Souza Leao, L., Legovini, A., Marchetti, R., Zia, B., 2013. The impact of
high school financial education : experimental evidence from Brazil. Policy Research
Working Paper Series 6723, The World Bank.
Bucher-Koenen, T., Lusardi, A., 2011. Financial Literacy and Retirement Planning in
Germany. Journal of Pension Economics and Finance 10 (4).
Bucher-Koenen, T., Ziegelmeyer, M., 2014. Once Burned, Twice Shy? Financial Literacy
and Wealth Losses during the Financial Crisis. Review of Financee 18, 2215–2246.
Calcagno, R., Monticone, C., 2015. Financial Literacy and the Demand for Financial
Advice. Journal of Banking and Finance 50.
Carpena, F., Cole, S., Shapiro, J., Zia, B., 2011. Unpacking the Causal Chain of Financial
Literacy. Policy Research Working Paper 5798, World Bank.
Christelis, D., Jappelli, T., Padula, M., 2010. Cognitive Abilities and Portfolio Choice.
European Economic Review 54 (1).
39
Cole, S., Sampson, T., Zia, B., 2011. Prices or knowledge? what drives demand for
financial services in emerging markets? The Journal of Finance 66 (6), 1933–1967.
Deuflhard, F., Georgarakos, D., Inderst, R., 2014. Financial literacy and savings ac-
count returnsn. Tech. rep., available at SSRN: http://ssrn.com/abstract=2358564 or
http://dx.doi.org/10.2139/ssrn.2358564vo.
Disney, R., Gathergood, J., 2011. Financial Literacy and Indebtedness: New Evidence
for UK consumers. Discussion papers, University of Nottingham, Centre for Finance,
Credit and Macroeconomics (CFCM).
Fernandes, D., Lynch Jr, J. G., Netemeyer, R. G., 2014. Financial literacy, financial
education, and downstream financial behaviors. Management Science 60 (8), 1861–
1883.
Fornero, E., Monticone, C., 2011. Financial Literacy and Pension Plan Participation in
Italy. Journal of Pension Economics and Finance 10 (4).
Garcia, M., 2013. Financial Education and Behavioral Finance: New Insights Into The
Role of Information in Financial Decisions. Journal of Economic Surveys 27 (2), 297–
315.
Georgarakos, D., Inderst, R., 2011. Financial Advice and Stock Market Participation.
Working Paper 1296, European Central Bank.
Gibson, J. McKenzie, D., Zia, B., 2012. The impact of financial literacy training for
migrants. World Bank Economic Review, 130–161.
Guiso, L., Viviano, E., 2015. How Much Can Financial Literacy Help? Review of Finance
19, 1347–1382.
Hastings, J. S., Madrian, B. C., Skimmyhorn, W. L., 2013. Financial Literacy, Financial
Education, and Economic Outcomes. Annual Review of Economics, Annual Reviews
5 (1), 347–373.
40
Jappelli, T., 2010. Economic Literacy: An International Comparison. The Economic
Journal 120 (548).
Jappelli, T., Padula, M., 2013a. Consumption Growth, the Interest Rate, and Financial
Literacy. CSEF Working Papers 329, Centre for Studies in Economics and Finance
(CSEF), University of Naples, Italy.
Jappelli, T., Padula, M., 2013b. Investment in Financial Literacy and Saving Decisions.
Journal of Banking and Finance 27 (08), 2779–2792.
Jappelli, T., Pischke, J., Souleles, N., 1998. Testing for Liquidity Constraints in Euler
Equations With Complementary Data Sources. Review of Economics and Statistics
80 (2).
Klapper, L., Lusardi, A., Panos, G. A., 2013. Financial Literacy and its Consequences:
Evidence from Russia During the Financial Crisis. Journal of Banking and Finance
37.
Klapper, L., Panos, G. A., 2011. Financial Literacy and Retirement Planning: the Rus-
sian case. Journal of Pension Economics and Finance 10.
Luhrmann, M., Serra-Garcia, M., Winter, J., 2015. Teaching teenagers in finance: Does
it work? Journal of Banking and Finance 54, 160 – 174.
Lusardi, A., Michaud, P.-C., Mitchell, O. S., Sep. 2011. Optimal Financial Literacy and
Saving for Retirement. Working papers, RAND Corporation Publications Department.
Lusardi, A., Mitchell, O. S., 2007a. Baby Boomer Retirement Security: The Roles of
Planning, Financial Literacy, and Housing Wealth. Journal of Monetary Economics
54 (1).
Lusardi, A., Mitchell, O. S., 2007b. Financial Literacy and Retirement Preparedness:
Evidence and Implications for Financial Education. Business Economics 42 (1).
41
Lusardi, A., Mitchell, O. S., 2007c. Financial Literacy and Retirement Preparedness:
Evidence and Implications for Financial Education. Business Economics 42 (1).
Lusardi, A., Mitchell, O. S., 2008. Planning and Financial Literacy: How Do Women
Fare? American Economic Review 98 (2).
Lusardi, A., Mitchell, O. S., 2009. How Ordinary Consumers Make Complex Economic
Decisions: Financial Literacy and Retirement Readiness. NBER Working Papers
15350, National Bureau of Economic Research, Inc.
Lusardi, A., Mitchell, O. S., 2011a. Financial Literacy and Planning: Implications for
Retirement Wellbeing. In: Lusardi, A., Mitchell, O. S. (Eds.), Financial Literacy:
Implications for Retirement Security and the Financial Marketplace. Oxford: Oxford
University Press.
Lusardi, A., Mitchell, O. S., 2011b. Financial Literacy and Retirement Planning in the
United States. Journal of Pension Economics and Finance 10 (04).
Lusardi, A., Mitchell, O. S., 2011c. Financial Literacy Around the World: An Overview.
Journal of Pension Economics and Finance 10.
Lusardi, A., Mitchell, O. S., 2014. The Economic Importance of Financial Literacy:
Theory and Evidence. Journal of Economic Literature 52 (1).
Manski, C., 1993. Identification of Endogenous Social Effects: the Reflection Problem.
The Review of Economic Studies 60 (3), 531–542.
Mastrobuoni, G., 2011. The Role of Information for Retirement Behavior: Evidence
Based on the Stepwise Introdution of the Social Security Statement. Journal of Public
Economics 95, 913–925.
OECD, September 2013a. Evaluating Financial Education Programmes: Survey, Evi-
dence, Policy Instruments and Guidance.
42
OECD, June 2013b. Improving Financial Education Effectiveness Through Behavioural
Economics.
Staiger, D., Stock, J., 1997. Instrumental Variables Regression With Weak Instruments.
Econometrica 65, 557–586.
van Rooij, M., Lusardi, A., Alessie, R., 2011a. Financial Literacy and Retirement Prepa-
ration in the Netherlands. Journal of Pension Economics and Finance 10.
van Rooij, M., Lusardi, A., Alessie, R., 2011b. Financial Literacy and Stock Market
Participation. Journal of Financial Economics 101 (2).
van Rooij, M., Lusardi, A., Alessie, R., 2011c. Financial Literacy, Retirement Planning,
and Household Wealth. Economic Journal.
Willis, L. E., 2011. The Financial Education Fallancy. American Economic Review:
Papers and Proceedings, 429–434.
43
Tables
Table 1: Descriptive Statistics.
Variable Mean Std. Dev. NFinancial literacy (number of correct answers out of three) 1.93 0.98 4865Financial literacy, binary indicator 1(three out of three answers correct) 0.34 0.47 4865Financial literacy, binary indicator 1(question on inflation correct) 0.74 0.44 4865Financial literacy, binary indicator 1(question on loan correct) 0.64 0.48 4865Financial literacy, binary indicator 1(question on portfolio diversification correct) 0.55 0.5 4865Financial literacy (number of correct answers out of two present in 2006 & 2008) 1.38 0.71 4865Instrumental variable, binary indicator 1(Household is PattiChiari client) (patti) 0.73 0.45 4865Instrumental variable, binary indicator 1(Household is PattiChiari client) (patti atl) 0.75 0.43 4865Financial assets (thousands) 19.4 22.4 4865Savings rate+ 0.17 0.27 4865Planning for retirement 0.46 0.50 2653Ownership of equities and mutual funds 0.12 0.32 4865Ownership of equities 0.06 0.23 4865Ownership of mutual funds 0.08 0.27 4865Ownership of government bonds 0.11 0.32 4865Male household head 0.56 0.50 4865Married 0.64 0.48 4865Nb. hh components 2.47 1.21 4865Years of education of the household head 9.81 4.46 4865Household labor income 26.3 16.3 4865Municipality 20.000-40.000 inh. 0.26 0.44 4865Municipality 40.000-500.000 inh. 0.44 0.50 4865Municipality 500.000+ inh. 0.10 0.30 4865High trust∗ 0.56 0.50 4865Length of relationship with the bank: less 2 years 0.04 0.18 4757Length of relationship with the bank: 2-4 years 0.07 0.26 4757Length of relationship with the bank: 5-10 years 0.16 0.37 4757Length of relationship with the bank: more than 10 years 0.73 0.45 4757Length of relationship with the bank∗∗: years 9.03 2.30 4757Number of current accounts 1.2 0.46 4865Homeownership 0.74 0.44 4865Motivations for choosing the main bank∗∗∗
Only one reason: related to convenience 0.54 0.50 4865Only one reason: related to financial/economic reason 0.07 0.25 4865Only one reason: related to bank characteristics 0.02 0.15 4865Two reasons: both related to convenience 0.12 0.32 4865Two reasons: related to convenience & financial/economic reasons 0.10 0.30 4865Two reasons: related to convenience & bank characteristics 0.10 0.31 4865Two reasons: both related to financial/economic reasons 0.03 0.16 4865Two reasons: financial/economic reasons & bank characteristics 0.02 0.15 4865Two reasons: both related to bank characteristics 0.00 0.05 4865Sources of information∗∗∗∗
Only intermediaries or experts 0.657 0.475 1492Only press or websites 0.016 0.126 1492Only friends, others, do not know 0.26 0.439 1492
Notes: Unless otherwise stated all descriptive statistics refer to the 2010 estimation sample.+ Calculated as the ratio between winsorized saving and income.∗ Our measure of trust relies on the answer to the question “Do you trust your principal bank? Please assign a score of 1 to 10,where 1 means ’I don’t trust it at all’ and 10 means ’I trust it completely’ and the intermediate scores serve to graduate yourresponse”. Respondents who trust their main bank are those who choose a value above the median of the distribution of answers(8).∗∗ Recoded from the categorical variable collected in the survey: for each category we impute the mid-point.∗∗∗ Each respondent can give at most two answers to the question on why the main bank is chosen. The 13 alternatives among whichthe respondent can choose are grouped into 3 broad categories: convenience (convenience to home/work, respondent’s employer’sbank (or respondent’s business’s bank)), financial/economic reasons (favourable interest rates, speed of transaction execution,range of services, low fees for services, possibility of online banking) and bank type (it is a well-known, important bank, staffcourtesy). The figures show the percentage of respondents who choose only one alternative, two alternatives in the same group ortwo alternatives in different categories.∗∗∗∗ Based on the question about the information source consulted before the current investments were performed. Available onlyfor households who hold financial assets other than bank accounts. The figures show the average number of respondents who relyonly on that specific source of financial information.
Table 2: Effect of Financial Literacy on Financial Assets. Financial literacy is thenumber of correct answers out of the three questions asked in 2010. 2010 SHIW Data.
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.231*** 8.273***(0.630) (2.979)
Client of PattiChiari 0.235*** 1.941***(0.035) (0.723)
Age 0.832*** 0.054*** 0.939*** 0.494**(0.136) (0.006) (0.134) (0.197)
Age squared -0.005*** -0.000*** -0.006*** -0.001(0.001) (0.000) (0.001) (0.002)
Male 1.017 0.170*** 1.347** -0.063(0.642) (0.025) (0.652) (0.705)
Married 2.479*** 0.024 2.497*** 2.295***(0.625) (0.037) (0.645) (0.596)
Nb. hh components -1.501*** 0.051*** -1.391*** -1.812***(0.292) (0.017) (0.292) (0.342)
Years education 0.918*** 0.045*** 0.999*** 0.626***(0.095) (0.004) (0.091) (0.178)
Municipality 20.000-40.000 inhabitants 1.040 0.071 1.256 0.673(1.415) (0.070) (1.456) (1.388)
Municipality 40.000-500.000 inhabitants 0.212 -0.009 0.157 0.235(1.086) (0.088) (1.114) (1.173)
Municipality 500.000+ inhabitants -0.308 0.151 -0.027 -1.275(2.061) (0.165) (2.024) (2.679)
Household labor income 0.414*** 0.000 0.413*** 0.409***(0.033) (0.001) (0.034) (0.032)
Only one motivation: services 1.126 0.092 1.647 0.887(1.207) (0.060) (1.204) (1.310)
Only one motivation: bank characteristics 1.804 -0.108 1.489 2.382(1.870) (0.067) (1.847) (1.970)
Two motivations: two convenience 4.626*** 0.056 4.774*** 4.314***(1.258) (0.052) (1.292) (1.165)
Two motivations: convenience & financial conditions 1.312 0.132* 1.786 0.698(1.155) (0.069) (1.173) (1.208)
Two motivations: convenience & bank characteristics 4.595*** 0.067 4.665*** 4.113***(1.306) (0.057) (1.303) (1.261)
Two motivations: two financial conditions 2.818 0.184** 3.470* 1.949(1.741) (0.077) (1.832) (1.608)
Two motivations: financial conditions 1.644 0.139 2.013 0.862& bank characteristics (2.232) (0.087) (2.242) (2.228)
Two motivations: two bank characteristics 4.698 0.037 4.701 4.391(3.831) (0.172) (3.869) (3.844)
Province FE Yes Yes Yes YesN. of observations 4865 4865 4865 4865
F-test on the excluded instruments 43.849
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ+λp+εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of thehousehold head, Xip is a vector of household controls (age, age squared, gender, marital status, years of education of the householdhead and number of household components, household labour income in 2010) and λp is a province fixed effect. Financial literacyis instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p <0.01.
Table 3: Robustness: Different measures for financial literacy.
OLS FS ITT IVDependent variable: Fin. assets Fin. literacy Fin. assets Fin. assets
Number of answers correct out of 3Financial literacy (nb/3) 2.231*** 8.273***
(0.630) (2.979)Client of PattiChiari 0.235*** 1.941***
(0.035) (0.723)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Binary indicator for 3 answers correct out of 3
Financial literacy (3 correct) 4.512*** 43.378**(1.086) (20.156)
Client of PattiChiari 0.045*** 1.941***(0.016) (0.723)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 8.152
Binary indicator for inflation question correct
Inflation correct 1.921 19.096***(1.204) (7.315)
Client of PattiChiari 0.102*** 1.941***(0.019) (0.723)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 29.128
Binary indicator for loan question correct
Loan correct 1.604** 27.207**(0.798) (10.831)
Client of PattiChiari 0.071*** 1.941***(0.013) (0.723)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 28.297
Binary indicator for portfolio question correct
Portfolio correct 5.287*** 31.491**(1.136) (12.929)
Client of PattiChiari 0.062*** 1.941***(0.019) (0.723)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 10.715
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ+λp+εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of thehousehold head, Xip is a vector of household controls (age, age squared, gender, marital status, years of education of the householdhead and number of household components, household labour income in 2010) and λp is a province fixed effect. Financial literacyis instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p <0.01.
Table 4: Effect of Financial Literacy on Financial Assets - Sample of “Investor House-holds”, i.e. of households who invested some savings in financial markets (bonds, equities,investment funds, etc.)
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Panel A: “Investor Households”, Baseline Specification as in Table 2
Financial literacy 2.473** 38.036*(1.022) (19.587)
Client of PattiChiari 0.134** 5.082***(0.062) (1.306)
N. of observations 1492 1492 1492 1492F-test on the excluded instr. 4.710
Panel B: “Investor Households”, Controlling for Sources of Financial Information
Financial literacy 2.314* 39.034*(1.022) (21.977)
Client of PattiChiari 0.121** 4.717***(0.060) (1.279)
Source of InformationOnly Intermediaries and Experts -0.430 0.069 -0.374 -3.052
(2.256) (0.118) (2.119) (5.463)Only Press and Websites -1.183 0.208 -0.364 -8.493
(5.231) (0.136) (5.278) (7.867)Only Friends and Other -4.245 -0.077 -4.161 -1.172
(2.771) (0.120) (2.675) (5.044)
N. of observations 1492 1492 1492 1492F-test on the excluded instr. 3.992
Notes: “Investor households”: household that invested some savings in financial markets (bonds, equities, investment funds, etc.).
Each panel of the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α+βflip+Xipγ + λp + εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy ofthe household head, Xip is a vector of household controls (as in the baseline specification of Table 2), and λp is a province fixedeffect. Financial literacy is instrumented with a dummy taking the value 1 if the household’s main bank belongs to the PattiChiariConsortium and 0 otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 5: Effect of financial literacy on financial assets growth between 2002 and 2006 or2008 or 2010. SHIW repeated cross-section 2006-2010. Financial literacy is the numberof correct answers out of the two common across waves.
OLS FS ITT IVDependent variable: Fin. ass. growth Fin. lit. Fin. ass. growth Fin. ass. growth
Financial literacy 0.014*** 0.153***(0.005) (0.052)
Client of PattiChiari 0.179*** 0.027***(0.040) (0.008)
N. of observations 3181 3181 3181 3181F-test of excluded instruments 19.755
Notes: Each column reports estimates of a separate regression. OLS, ITT and IV columns are regression where the dependent
variable is(yitp−yi2002p−E[yitp−yi2002p])
V ar[yitp−yi2002p], i.e. standardized growth in financial assets between 2002 and time t for household i
in province p. FS columns are regressions where the dependent variable is financial literacy. All models include the same vector ofhousehold controls: age, age squared, gender, marital status and years of education of the household head; number of householdcomponents, household labour income, reasons for choosing the main bank and size of the municipality. All models include andyear-specific province fixed effects. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if thehousehold’s main bank belongs to the PattiChiari Consortium and 0 otherwise. Heteroskedasticity-robust standard errors clusteredat the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 6: Differences Among Bank Switchers in Household Financial Assets in 2002,before the PattiChiari Consortium was created. Data: panel data SHIW, Bank Switchersbetween 2006 and 2008 and between 2008 and 2010.
(1) (2) (3)Dependent variable: Fin. ass. in 2002 Fin. ass. in 2002 Fin. ass. in 2002
Switchs to Client of PattiChiari -10.787 -9.917 -7.642(6.927) (6.188) (5.109)
Stays Client of Non-PattiChiari -12.536 -8.936 -7.722(13.066) (16.437) (17.156)
Switchs to Client of Non-PattiChiari -3.632 -15.928 5.209(19.331) (24.442) (21.381)
Baseline Controls Y Y YMotivation for Choosing Bank N Y YBank Costs and Characteristics N N YN. of observations 1666 833 812
Null Hypothesis: No Difference Tests p-values
Stays PattiChiari Client vsSwitchs to Non-PattiChiari 0.851 0.516 0.808
Stays Non-PattiChiari Client vsSwitchs to PattiChiari 0.884 0.956 0.997
Stays Non-PattiChiari vs Stays PattiChiari 0.340 0.588 0.654
Switchs to Non-PattiChiari vsSwitchs to PattiChiari 0.704 0.813 0.592
Notes: the table reports OLS regression results of the model yip = α + βZip + Xipγ + λp + εip where yip is financial assetsof household i, living in province p, in year 2002; Zip is a vector of dummies indicating whether the household switched mainbank between waves or not and the type of switch, Xip is a vector of household controls and λp is a province fixed effect. Morespecifically, all regressions include controls for characteristics of the household head (age, age squared, gender, marital status,years of education) and the household (number of household components, household labour income in 2002), in analogy with thespecification in Table 2. Column (2) adds motivations for choosing the main bank to the list of controls; column (3) adds motivationsfor choosing the main bank, bank costs and bank characteristics to the list of controls. Column (1) uses data on all switchers2006-2008 and 2008-2010; column (2) and (3) use records with no missing data on all the added controls. Heteroskedasticity-robuststandard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 7: Instrumental Variable estimates of the effect of financial literacy on financialassets. SHIW 2010.
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Panel A : Baseline model on common support
Financial literacy 2.289*** 8.550***(0.629) (3.005)
Client of PattiChiari 0.233*** 1.996***(0.036) (0.724)
F-test excluded instr. 43.216N. of observations 4829
Panel B: Propensity score model controls on common support
Financial literacy 2.180*** 7.933***(0.629) (3.044)
Client of PattiChiari 0.221*** 1.751**(0.035) (0.695)
F-test excluded instr. 38.917N. of observations 4829
Notes: Each column reports estimates of a separate regression. OLS, ITT and IV columns are regression where the dependentvariable is financial assest. FS columns are regressions where the dependent variable is financial literacy. All models includethe same vector of household controls: age, age squared, gender, marital status and years of education of the household head;number of household components, household labour income, reasons for choosing the main bank and size of the municipality. Allmodels include province fixed effects. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if thehousehold’s main bank belongs to the PattiChiari Consortium and 0 otherwise. Heteroskedasticity-robust standard errors clusteredat the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 8: Bank Costs by PattiChiari Status - Year 2010 (euro)
Type Basic Debit Credit Avg. Cost Avg. Cost Avg. Cost Avg. Cost Obsof Bank Account Card Card Bank Transf. Bank Transf. of Withdrawal of Withdrawal
Fees Fees Fees (desk) (online) (desk) (ATM)
non-Patti-Chiari 34.9 7.09 15.67 1.88 0.35 0.05 0.31 917(30.3) (6.28) (20.5) (5.26) (4.03) (0.90) (1.93)
Patti-Chiari 36.3 4.23 14.06 2.05 0.52 0.07 0.29 5800(38.6) (5.20) (12.93) (6.09) (4.66) (1.62) (1.82)
Welch t-test stat. 1.25 13.1 2.32 0.84 1.17 0.66 0.34(p-value) (0.18) (0.00) (0.28) (0.20) (0.19) (0.32) (0.38)
Source: Bank of Italy Survey on Bank Fees and Expenditures. Notes: an observation is an individual bank account. Due to confidentiality reasons, for each bank, only means and
standard deviations of each variable has been provided, together with the number of bank accounts surveyed. The means and standard deviations provided in this table are, thus,
combined assuming observations are independent between banks. The Welch t-test for equality of means assumes unequal standard deviations between the two groups.
Table 9: Bank Total Assets, Number of Branches, and PattiChiari status in 2010
Bank Name PattiChiari Total Assets No. of Decile Decile(mln) Branches by Assets by Branches
UniCredit 1 929488 11266 10 10Intesa Sanpaolo 1 658757 4060 10 10
Banca Monte dei Paschi di Siena 1 244279 4677 10 10Banca Nazionale del Lavoro 1 98022 1847 10 10Banca Popolare del Mezzogiorno 1 60515.92 174 10 4
Banca popolare dell’Emilia Romagna 1 58498 429 10 6
Banca Popolare di Milano 1 54053 703 10 8
Cassa di Risparmio di Parma e Piacenza 1 46339 1082 9 9
BancoPosta 0 43223.27 7700 9 10Banca Carige 1 40010 807 9 9
Banca Popolare di Vicenza 1 35533 520 9 7
Veneto Banca 0 33057 335 9 6Cassa di Risparmio di Firenze 1 31677 637 9 8
Credito Emiliano 1 29998 976 9 9Banca Popolare di Verona-S. Geminiano e S. Prospero 1 29689 2153 8 10
Banco di Napoli 1 28153 2424 8 10
Credito Valtellinese 1 26761 164 8 3Banca Popolare di Sondrio 0 26282 336 8 6
Deutsche Bank 1 24859 358 8 6Banca Popolare di Bergamo 1 24456 957 8 9
Banca delle Marche 1 21486 501 8 7Cassa di Risparmio del Veneto 1 19625 1000 7 9
Banca Popolare di Lodi 1 19556 1175 7 9
Banco di Brescia 1 17622 738 7 8Banca Popolare di Novara 1 17075 1205 7 9
Banco di Sardegna 1 13930 550 7 7
Unipol Banca 0 12052 332 7 6Banca Mediolanum 1 11622 6 7 1FinecoBank Banca 1 11250 3 6 1Banca popolare dell’Etruria e del Lazio 0 10903 304 6 5Cassa di risparmio in Bologna 1 10161 596 6 8
Banca Popolare Commercio e Industria 1 10130 598 6 8
Aletti and Co. 0 9940 42 6 1Banca Carime 1 9784 700 6 8Banca Fideuram 1 9556 108 6 2Banca Popolare di Ancona 1 9101 310 5 6
Banca Popolare FriulAdria 1 8501 313 5 6
Cassa di Risparmio di Bolzano 0 8210 198 5 5Banco di Desio e della Brianza 0 8163 166 5 4Banca Regionale Europea 1 8132 569 5 8
Banca Sella 1 7979 528 5 7Cassa di risparmio di Ferrara 0 7573 170 5 4Banca Popolare di Bari 0 7286 221 4 5Banca di Credito Cooperativo di Roma 0 7259 152 4 3Cassa di risparmio di Asti 0 6118 210 4 5Banca dell’ Adriatico 1 5613 463 4 7Cassa di risparmio della provincia di Teramo 0 5489 187 4 4Banca Di Credito Sardo 1 5401 117 4 2Banca Popolare dell’Alto Adige 0 5248 140 4 3Banca della Campania 1 5155 210 3 5
Cassa dei Risparmi di Forli e della Romagna 1 4769 171 3 4
Banca Nuova 1 4711 135 3 3Cassa di risparmio di Venezia 1 4650 434 3 7
Cassa di Risparimio di Rimini 0 4551.51 179 3 4Cassa di Risparmio di Biella e Vercelli 1 4523 211 3 5
Banca Popolare di Puglia e Basilicata 1 4460 191 3 5
Banca Agricola Popolare di Ragusa 0 4388 101 2 2Cassa di Risparmio del Friuli Venezia Giulia 1 4300 447 2 7
Credito Siciliano 1 4059 157 2 3SpA-Generbanca 1 3808 70 2 1
Cassa di risparmio della provincia di Chieti 1 3574 116 2 2
Banca Popolare Pugliese 1 3263 103 2 2
Banca Popolare di Spoleto 1 3029 165 2 4
Banca Monte Parma 0 3005 91 1 2Cassa di risparmio di San Miniato 0 2986 132 1 2IW Bank 0 2874 2 1 1Banca di Piacenza 0 2793 62 1 1Allianz Bank 0 2769 19 1 1Cassa di risparmio di Pistoia e della Lucchesia 1 2515 145 1 3
Cassa di risparmio della Spezia 1 2055 152 1 3
Notes: data from Bankscope, Bank of Italy - SIOTEC database, and PattiChiari Consortium.
Table 10: Average Total Assets and Number of Branches by PattiChiari status in 2010
Variable PattiChiari non-PattiChiari p-value No. of Obs.
Total Assets (bns.) 54948.631 10198.561 0.209 70(160923.038) (10734.01)
No. of Branches 922.417 433.704 0.197 87(1689.874) (1455.516)
Notes: Data: SIOTEC, 2010; Banks Listed in SHIW 2010. Sample size for total assets and number of branches differs because forsome banks the information on total assets in 2010 is missing.
Table 11: Effect of Financial Literacy on Other Outcomes
OLS FS ITT IVDependent variable: Fin. assets Fin. literacy Fin. assets Fin. assets
Ownership of stock and fundsFinancial literacy 0.019*** 0.134***
(0.006) (0.047)Client of PattiChiari 0.235*** 0.031***
(0.035) (0.010)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Ownership of equitiesFinancial literacy 0.006 0.026
(0.004) (0.028)Client of PattiChiari 0.235*** 0.006
(0.035) (0.007)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Ownership of mutual fundsFinancial literacy 0.016*** 0.091*
(0.005) (0.047)Client of PattiChiari 0.235*** 0.021**
(0.035) (0.011)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Ownership of government bondsFinancial literacy 0.028*** 0.300***
(0.007) (0.070)Client of PattiChiari 0.235*** 0.070***
(0.035) (0.015)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Retirement PlanningFinancial literacy 0.039*** 0.197
(0.014) (0.122)Client of PattiChiari 0.213*** 0.042
(0.048) (0.026)N. of observations 2653 2653 2653 2653F-test on the excluded instr. 19.967
Saving rateFinancial literacy 0.002 0.023
(0.007) (0.043)Client of PattiChiari 0.235*** 0.005
(0.035) (0.010)N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Notes: The table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model DVip = α + βflip +Xipγ + λp + εipwhere DVip is a dependent variable measured for household i, living in province p, in year 2010; flip is financialliteracy of the household head, Xip is a vector of household controls (as in the baseline specification of Table 2), and λp is aprovince fixed effect. Financial literacy is instrumented with a dummy taking the value 1 if the household’s main bank belongs tothe PattiChiari Consortium and 0 otherwise. In the top four panels the dependent variable is a dummy equal to 1 if the householdowns equities, mutual funds or government bonds; in the fifth panel is a dummy equal to 1 if the household head reports to haveplanned for retirement; in the bottom panel is the winsorized saving rate (ratio between savings and income). Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 12: Heterogeneity in the Effect of Financial Literacy on Financial Assets. Financialliteracy is the number of correct answers out of the three questions asked in 2010. 2010SHIW Data.
OLS FS ITT IVDependent variable: Fin. assets Fin. literacy Fin. assets Fin. assets
Education <= mediana
Financial literacy 1.985*** 9.426**(0.628) (3.747)
Client of PattiChiari 0.256*** 2.412**(0.044) (0.998)
N. of observations 2563 2563 2563 2563F-test on the excluded instr. 33.677Mean 15.354 St. dev 18.971
Education > mediana
Financial literacy 3.034*** 7.333Client of PattiChiari 0.127*** 0.928
(0.046) (1.110)N. of observations 2122 2122 2122 2122F-test on the excluded instr. 7.542
Age < 60b
Financial literacy 1.544** 5.154(0.769) (6.261)
Client of PattiChiari 0.167*** 0.860(0.044) (1.096)
N. of observations 2419 2419 2419 2419F-test on the excluded instr. 14.507
Age 60+b
Financial literacy 2.665*** 10.309**(0.709) (4.032)
Client of PattiChiari 0.256*** 2.635**(0.048) (1.121)
N. of observations 2367 2367 2367 2367F-test on the excluded instr. 28.385
Low education, age 60+Financial literacy 1.722** 12.595**
(0.808) (5.153)Client of PattiChiari 0.232*** 2.918**
(0.052) (1.204)N. of observations 1647 1647 1647 1647F-test on the excluded instr. 19.857
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ + λp + εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacyof the household head, Xip is a vector of household controls, and λp is a province fixed effect. Financial literacy is instrumentedwith a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0 otherwise.Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.a: This group includes respondents with compulsory schooling (years of education lower or equal to 8).b: 60 is the median age in our baseline sample.
Figure 1: Propensity Score Matching: Distribution of estimated propensity score. SHIW2010 data.
0 .2 .4 .6 .8 1Propensity Score
Untreated Treated: On supportTreated: Off support
A Measuring Financial Literacy
The three questions available in the 2010 wave for eliciting financial literacy are: i) un-
derstanding inflation “Imagine leaving 1,000 euros in a current account that pays 1%
interest and has no charges. Imagine also that inflation is running at 2%. Do you think
that if you withdraw the money in a year’s time you will be able to buy the same amount
of goods as if you spent the 1,000 euros today? Yes/No, I will be able to buy less (correct
answer)/No, I will be able to buy more/Do not know”; ii) understanding mortgages
“Which of the following types of mortgage do you think will allow you from the very start
to fix the maximum amount and number of installments to be paid before the debt is ex-
tinguished? Floating rate mortgage/Fixed rate mortgage (correct answer)/Floating rate
mortgage with fixed installments/Do not know”; iii) portfolio diversification“ Which
of the following investment strategies do you think entails the greatest risk of losing your
capital? Investing in the shares of a single company (correct answer) / Investing in the
shares of more than one company/Do not know/No answer”.
Questions (i) and (iii) are similar to the ones devised for the US Health and Retire-
ment Study (HRS) (Lusardi and Mitchell, 2011a) while knowledge of mortgage repay-
ment mechanisms is not considered in the HRS. All these questions follow the principles
of simplicity, relevance, brevity and capacity to differentiate identified by Lusardi and
Mitchell (2011c). Questions (i) and (ii) are common to all three waves of SHIW used in
this paper (2006, 2008, 2010). In 2006, unfortunately, financial literacy questions were
asked to only half of the sample, i.e. to household heads born in even years.
We claim that the answers to the questions about inflation, mortgages and portfolio
diversification in the SHIW reflect a broader knowledge, namely financial literacy. To
corroborate this statement, we perform a factor analysis. This exercise aims at checking
how many latent factors are captured by correct answers to those questions, moving
from the presumption that all questions do not only measure knowledge of a specific
economic concept but rather reflect the same common latent factor. For each question
related to financial knowledge asked in 2010, we construct a binary indicator for answer-
ing correctly, and, then, we perform a factor analysis on these indicators. The analysis
unambiguously suggest that only one factor should be retained and we name this factor
financial literacy. We replicate the same exercise using the repeated cross-section from
2006 and 2010 and the two indicators available across all three waves (the estimated fac-
tor loadings for both exercises are shown in Table 14). Then we obtained factor scores
using the regression method and perform our analysis also using this indicator: Table 13
reports the results using the same indicator used in the baseline regression and the esti-
mated latent factor. In Table 13 we standardized both indicators of financial literacy to
have zero mean and unitary variance so that coefficients can be interpreted as the effect
of a one-standard deviation change in financial literacy of financial assets for both indi-
cators and the comparison is more straightforward: our results are confirmed when using
standardized factor scores to measure financial literacy. This exercise further supports
our definition of financial literacy as a broad concept, that embeds different aspects of
financial knowledge. Table 14 shows that answering correctly the question on inflation
is positively related to the latent factor and this is true also for the other indicators
(answering the loan or portfolio question correctly). Therefore, since all indicators turn
out to capture the same latent factor, when we measure financial literacy by using each
of these questions in turn we get similar results (see Table 3).
Table 13: Effect of Financial Literacy on Financial Assets. Financial literacy is eitherthe number of correct answers out of the three questions asked in 2010 or the estimatedfactor score; both indicators are standardized to have zero mean and unitary variancein sample. 2010 SHIW Data.
Financial literacy measured with standardized number of correct answers
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.193*** 8.131***(0.619) (2.928)
Client of PattiChiari 0.239*** 1.941***(0.036) (0.723)
F-test on the excluded instr. 43.849N. of observations 4865 4865 4865 4865
Financial literacy measured with standardized factor score
Financial literacy 2.196*** 7.953***(0.631) (2.864)
Client of PattiChiari 0.244*** 1.941***(0.037) (0.723)
F-test on the excluded instr. 43.273N. of observations 4865 4865 4865 4865
Notes: All models include the same vector of household controls: age, age squared, gender, marital status and years of educationof the household head; number of household components, household labour income, reasons for choosing the main bank and size ofthe municipality. All models include province fixed effects. Financial literacy is measured as the number of correct question out ofthose asked in 2010. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if the household’s mainbank belongs to the PattiChiari Consortium and 0 otherwise. Heteroskedasticity-robust standard errors clustered at the provincelevel in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 14: Factor loadings corresponding to the three financial literacy questions.Thedata are from SHIW.
Indicator Factor loading2010 2006-2010∗
Inflation question correct 0.4487 0.4116Loan question correct 0.3745 0.4116Portfolio question correct 0.4353 not availableN. of observations 4865 11812
Notes: ∗ repeated cross-section.
B Robustness and Sensitivity Analysis
In this section, we check the robustness our findings to several falsification checks. In
particular, we explore: i) the role of retest effects and the timing of the effect of bank
information policies and financial literacy; ii) the role of trust in institutions; iii) the
sensitivity of our results to sample selection and the functional form. Our results are
robust to all these checks.
B.1 The Role of Retest Effects and the Timing of the Effect of BankInformation Policies and Financial Literacy
The SHIW dataset has a rotating panel sub-sample. Households that have been in-
terviewed in previous surveys may have looked for the correct answer to that specific
question (retest effect). In addition, attrition in the panel sub-sample tends to be higher
among worse-off households (Biagi et al., 2009). The simultaneous presence of retest
effects and non-random attrition may upwardly bias our IV results. To test whether this
bias is present, we check whether the effect of financial literacy is different for house-
holds who participated to the survey for the second time. Hence, we interact both the
instrument and financial literacy variables with a dummy equal to 1 if the household was
interviewed before 2010. Results, reported in Table 15, do not provide evidence of retest
effects in the sample. Controlling for this channel does not alter the ITT and IV results:
the interaction between being PattiChiari client and being interviewed more than once
is not statistically significant. The estimated effect of financial literacy for respondents
interviewed more than once is not statistically different from the one estimated for the
rest of the sample.
A different, albeit related, concern refers to the timing of the effect of the treatment
and of the instrumental variable. We examine two issues. First, year 2010 may not be a
representative year to assess the role of financial literacy. Household wealth in 2010 may
be affected by the financial turmoil due to the Great Recession and, particularly, by the
fall in the value of stock prices observed in 2007-2008 (Bottazzi et al., 2013). However
whether and to which extent the effect of financial literacy on financial assets depend on
the recession in not straightforward (see Guiso and Viviano (2015), Bucher-Koenen and
Ziegelmeyer (2014)). Hence, financial literate respondents are more likely to be exposed
to the stock market, and, therefore, to experience financial losses in late 2000s but, at
the same time, they may be better endowed to cope with the fall in stock prices. Our
data cover three years - 2006 (pre-crisis), 2008 (the Great Recession, after Lehman’s de-
fault), and 2010 (the short recovery period that was followed, from mid-2011, by the debt
crisis). Two questions concerning financial literacy, about inflation and loan repayment
schemes, are common to all three waves.37 We restrict our analysis to these common
items and we interact year dummies (for 2008 and 2010) with both the instrumental and
the financial literacy variables. Results, reported in Table 16. The impact of financial
literacy (and of the PattiChiari dummy) on financial assets turns out not to be statis-
tically different in 2006, 2008 and 2010. The model allows for the the level of financial
assets being affected by the Great Recession because we include year specific province
specific fixed effects, but the impact of financial literacy on financial assets being not
substantially different before and after the crisis. We fail to detect heterogeneity in the
estimated effect of financial literacy on financial assets over this short time span: this
may signal that financial literacy have positive effect on long-term financial accumula-
tion, while it does not play a specific role in the presence of large shocks, such as the
financial crisis. This would be consistent with the recent findings by Guiso and Viviano
(2015), who show that financial literacy had a positive but economically small impact
on the way in which households responded to the financial crisis.
Second, we exploit the panel component of the sample to estimate the effect of being
a client of a PattiChiari bank from, respectively, 1 3 or 5 years (Table 17).38 We do
not detect statistically significant differences in the impact of information policies on
financial literacy (first stage) nor financial assets (intention-to-treat) according to the
length of the exposure of clients to the treatment (test on the null assumption of equal
treatment does not allow to reject it at standard levels of significance). However, the
37More details are provided in Appendix A.38It is worth noting that being a client of a PattiChiari bank is measured at the beginning of the year,
while financial literacy and assets are measured at the end of the year.
number of respondent who become or cease to be client of a PattiChiari bank is limited,
because the client relation with the bank is typically long lasting and because entry/exit
of banks from the PattiChiari Consortium is limited. Indeed, all banks who were part
of the Consortium in 2010 entered in 2003 (except the Bank of Chieti that entered in
2004). This may prevent us from precisely estimating the effect of the duration of the
treatment.
B.2 The Role of Trust in Institutions
As discussed in Section 5, our identification strategy is based on the assumption that the
choice of the bank is not correlated with factors that may affect wealth. Georgarakos
and Inderst (2011) find that the level of trust in institutions and financial advisors may
affect individual investment decision. We check whether including proxies for trust as
additional controls alters our results. In the first panel of Table 18, we add a dummy
capturing whether the respondent trusts his/her main bank.39 The effect of this variable
is not significant and the key results remain comparable to the baseline specification. In
the second panel, we control for the length of the relationship between the client and the
bank: we expect a longer relationship to reflect higher trust in the bank. This variable
is correlated with level of financial assets (ITT and IV) but controlling for it does not
affect the impact of the main variables of interest. The two proxies for trust in banks
are not significantly different from zero in the first stage equation (FS columns in the
top panels of Table 18) at conventional levels of significance. This evidence suggests
that the effects of bank information policies (proxied by our binary indicator of being a
PattiChiari client) on financial literacy are not mediated by trust.
39Our measure of trust relies on the answer to the question “Do you trust your principal bank? Pleaseassign a score of 1 to 10, where 1 means “I don’t trust it at all” and 10 means “I trust it completely” andthe intermediate scores serve to graduate your response”. Respondents who trust their main bank arethose who choose a value above the median of the distribution of answers (8); results are confirmed ifwe define the we set this binary indicator equal to one if the answer is above 5, 0 otherwise.
B.3 Sensitivity Analysis: Sample Selection and Functional Form
Our estimation sample excludes the upper and lower 5% tails of the financial assets
distribution. However, in the bottom part of the distribution there are tied values: about
11% of the households do not have any financial asset. One may question whether our
results are driven by sample selection, i.e. by the fact that we focus on households with
positive financial assets. Panel A in Table 19 presents estimates of our baseline model on
a sample that includes households with zero assets. In Panel B, we use a Tobit model to
distinguish the effect of financial literacy on the probability of having any financial asset
from its effect on the amount of assets owned. In both cases, we contrast the results of the
instrumental variable approach with those obtained when we consider financial literacy
as an exogenous variable. Results broadly confirm the findings illustrated in Section 6.
When endogeneity is not addressed, the effect of financial literacy on financial assets
is underestimated. The magnitude of the instrumental variable estimates on average
assets is fairly similar to the one discussed in Section 6. Some interesting new findings
emerge: instrumental variable estimates from the Tobit model suggest that financial
literacy affects both the decision to hold financial assets as well as their amount, so that
the marginal effect of one additional correct answer to the financial literacy questions
is lower for those who hold financial assets (+7 thousand euros) than for the average
individual (+ nearly 10 thousand euros).40
40Notice, however, that this model has two drawbacks: first, it imposes restrictions on the effect offinancial literacy over the intensive and the extensive margins; second, it is well suited for continuousendogenous variables. For these reasons, we choose as baseline the results delivered by linear models.
Table 15: Robustness: Retest effect.
OLS ITT IVDependent variable: Fin. assets Fin. assets Fin. assets
Financial literacy 2.003*** 7.715***(0.703) (2.954)
Financial literacy at 2nd/3rd interviews 0.489 1.366(0.578) (3.061)
Client of PattiChiari at 1st Interview 1.831**(0.920)
Client of PattiChiari at 2nd/3rd Interviews 0.153(1.162)
N. of observations 4865 4865F-test on the excluded instr. 22.827
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ+λp+εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of thehousehold head, Xip is a vector of household controls (age, age squared, gender, marital status, years of education of the householdhead and number of household components, household labour income in 2010) and λp is a province fixed effect. Financial literacyis instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p <0.01.
Table 16: Robustness: Heterogeneity of the Effect of Financial Literacy on FinancialAssets across years, 2006-2010 SHIW Data (repeated cross-section). Financial literacyis the number of correct answers out of the two common across waves.
OLS ITT IVDependent variable: Fin. assets Fin. assets Fin. assets
Financial literacy 0.731 16.404***(0.734) (4.973)
Financial literacy · 1(year=2008) 0.602 -0.418(0.704) (3.693)
Financial literacy · 1(year=2010) 0.898 -0.372(0.848) (4.037)
Client of PattiChiari 2.621***(0.822)
Client of PattiChiari · 1(year=2008) -0.127(0.983)
Client of PattiChiari · 1(year=2010) -0.236(1.098)
N. of observations 11812 11812 11812
Descriptive statistics -Avg (se)- on this sampleFinancial Assets : 18.45 (21.35); Financial Literacy : 1.37 (0.74)
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wipt = α + β1flipt +
β2flipt · 1(t = 2008) + β3flipt · 1(t = 2010) + Xiptγ + λpt + εipt where wipt is financial assets owned by household i, livingin province p, in year t; flipt is financial literacy of the household head measured as the number of correct answers over the twoquestions asked in 2006, 2008 and 2010; 1(·) is the indicator function, Xipt is a vector of household controls (as in the baselinespecification of Table 2, excluding motivations for choosing the bank), and λpt is a province-year fixed effect. Financial literacyand its interaction with year dummies are instrumented with a dummy that takes the value 1 if the household’s main bank belongsto the PattiChiari Consortium in year t, interacted with year dummies, and 0 otherwise. Heteroskedasticity-robust standard errorsclustered at the province level in parentheses.
Table 17: Robustness: Timing of the Effect of Being PattiChiari Client. Measure ofFinancial Literacy: Number of Correct Answer over the Three Questions (in 2010).SHIW panel individuals over 2006-2010.
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.274** 9.158(0.919) (8.000)
Client of PattiChiari in 2010 0.179*** 1.635(0.058) (1.516)
N. of observations 1662 1662 1662 1662F-test on the excluded instruments 9.376
Financial literacy 2.274** 7.997(0.919) (6.177)
Client of PattiChiari since 2008 0.215*** 1.720(0.060) (1.357)
N. of observations 1662 1662 1662 1662F-test on the excluded instruments 12.664
Financial literacy 2.274** 8.997(0.919) (6.646)
Client of PattiChiari since 2006 0.195*** 1.751(0.058) (1.292)
N. of observations 1662 1662 1662 1662F-test on the excluded instruments 11.124
Notes: Elaborations based on the 2010 wave data on financial assets and financial literacy and lagged values of the instrumentalvariable as indicated in the tables; elaborations restricted to the households in the panel sub-sample. Measure of financial literacy:number of correct answers over the three questions.All regressions include the controls in the baseline specification (Table 2).Errors robust to heteroskedasticity and clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 18: Effect of Financial Literacy on Financial Assets, Adding Additional Controlsto the Baseline Specification: Trust, Length of Relation with Bank or Number of BankAccounts Held by the Household
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Additional control: Trust
Financial literacy 2.230*** 8.258***(0.627) (3.058)
Client of PattiChiari 0.240*** 1.981***(0.036) (0.741)
High trust 0.150 0.049 0.373 -0.032(0.627) (0.044) (0.645) (0.746)
F-test on the excluded instruments 44.023N. of observations 4865 4865 4865 4865
Additional control: Length of Relation with Bank
Financial literacy 2.099*** 6.239**(0.656) (3.089)
Client of PattiChiari 0.223*** 1.393*(0.037) (0.713)
Length relation bank: more than 10 years 3.618*** 0.069 0.3620*** 3.191***(0.895) (0.044) (0.884) (0.947)
F-test on the excluded instruments 35.721
N. of observations 4757 4757 4757 4757
Additional control: Number of Bank Accounts Held by the Household
Financial literacy 2.195*** 8.123***(0.633) (2.997)
Client of PattiChiari 0.234*** 1.900***(0.035) (0.727)
Number of bank accounts 2.730*** 0.051 2.824*** 2.407***(0.750) (0.033) (0.744) (0.784)
F-test on the excluded instruments 43.479
N. of observations 4865 4865 4865 4865
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ + λp + εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy ofthe household head, Xip is a vector of household controls (as in the baseline specification of Table 2 plus the additional controlsspecified in each panel), and λp is a province fixed effect. Financial literacy is instrumented with a dummy taking the value 1 ifthe household’s main bank belongs to the PattiChiari Consortium and 0 otherwise.Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 19: Robustness: Sample Selection and Functional Form
Panel A: Baseline model including households with no financial asset
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.106*** 6.342**(0.539) (2.664)
Client of PattiChiari 0.238*** 1.509**(0.034) (0.629)
F-test of excluded instruments 49.278
N. of observations 5451
Panel B: Tobit regression model results
Coefficient of Average Partial Effects onFinancial Literacy E[Y ] E[Y |Y > 0] Prob[Y > 0]2.593∗∗∗ 2.005∗∗∗ 1.442∗∗∗ 0.038∗∗∗
(0.621) (0.475) (0.341) (0.009)
Instrumental variable Tobit regression model results
Coefficient of Average Partial Effects onFinancial Literacy E[Y ] E[Y |Y > 0] Prob[Y > 0]13.094∗∗∗ 9.881∗∗∗ 7.056∗∗∗ 0.183∗∗∗
(3.425) (2.583) (1.845) (0.048)
N. of observations 5451
Notes: Models include the same vector of household controls (as in the baseline specification of Table 2) and province fixed effects.Financial literacy is instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiariConsortium and 0 otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
C Additional Tables
Table 20: Robustness: Alternative measure of PattiChiari
OLS FS ITT IVDependent variable: Fin. assets Fin. literacy Fin. assets Fin. assets
Financial literacy 2.231*** 8.273***(0.630) (2.979)
Client of PattiChiari (main bank) 0.235*** 1.941***(0.035) (0.723)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 43.849
Financial literacy 2.231*** 8.886***(0.630) (2.811)
Client of at least one PattiChiari 0.258*** 2.289***(0.038) (0.742)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 46.655
Financial literacy 2.191*** 11.238***(0.582) (4.078)
Client of at least one 0.157*** 1.760***PattiChiari, imputed zero if missing (0.031) (0.613)
N. of observations 5756 5756 5756 5756F-test on the excluded instr. 26.128
Notes: Each column reports estimates of a separate regression. All models include the same vector of household controls: age,age squared, gender, marital status and years of education of the household head; number of household components, householdlabour income, reasons for choosing the main bank and size of the municipality. All models include province fixed effects. Financialliteracy is measured as the number of correct question out of those asked in 2010. In IV columns, financial literacy is instrumentedwith one of this indicators; i) a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortiumand 0 otherwise (top panel); ii) a dummy that takes the value 1 if at least one of the household’s bank account is held in aPattiChiari bank and 0 otherwise (middle panel); iii) a dummy that takes the value 1 if at all the bank accounts of the householdare held in a PattiChiari bank and 0 otherwise (bottom panel) Heteroskedasticity-robust standard errors clustered at the provincelevel in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 21: Robustness: Controlling for municipality fixed effect
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.331*** 12.051***(0.510) (4.275)
Client of PattiChiari 0.192*** 2.354***(0.037) (0.732)
N. of observations 4843 4843 4843 4843F-test on the excluded instr. 26.852
Notes: Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α +
βflip + Xipγ + λp + εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financialliteracy of the household head, Xip is a vector of household controls (age, age squared, gender, marital status, years of educationof the household head and number of household components, household labour income in 2010) and λp is a province fixed effect.Financial literacy is instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiariConsortium and 0 otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p <0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 22: Robustness: Not controlling for reasons for choosing the main bank
OLS FS ITT IVDependent variable: Fin. assets Fin. literacy Fin. assets Fin. assets
Financial literacy 2.293*** 8.802***(0.637) (3.243)
Client of PattiChiari 0.219*** 1.925**(0.034) (0.752)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 41.751
Notes: The table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ+λp+εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of thehousehold head, Xip is a vector of household controls (age, age squared, gender, marital status, years of education of the householdhead and number of household components, household labour income in 2010) and λp is a province fixed effect. Financial literacyis instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p <0.01.
Table 23: Robustness: Controlling financial assets in 2002
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Controlling for financial assets in 2002Financial literacy 1.491 5.199
(0.949) (8.168)Client of PattiChiari 0.184*** 0.957
(0.056) (1.582)Financial assets in 2002 0.021* 0.000 0.021* 0.019*
(0.011) (0.000) (0.012) (0.011)N. of observations 1027 1027 1027 1027F-test on the excluded instr. 10.679
BaselineFinancial literacy 1.605* 6.487
(0.961) (8.089)Client of PattiChiari 0.188*** 1.222
(0.056) (1.598)N. of observations 1027 1027 1027 1027F-test on the excluded instr. 11.194
Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α + βflip +Xipγ+λp+εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of thehousehold head, Xip is a vector of household controls (age, age squared, gender, marital status, years of education of the householdhead and number of household components, household labour income in 2010) and λp is a province fixed effect. Financial literacyis instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses. ∗p < 0.1,∗∗ p < 0.05,∗∗∗ p <0.01.
Table 24: Effect of financial literacy on financial assets growth between 2002 and 2006or 2008 or 2010, allowing for heterogeneous effects across years. SHIW repeated cross-section 2006-2010. Financial literacy is the number of correct answers out of the twocommon across waves.
OLS ITT IVDependent variable: Fin. ass. growth Fin. ass. growth Fin. ass. growth
Financial literacy 0.018** 0.234***(0.009) (0.082)
Financial literacy · 1(year=2008) -0.001 -0.038*(0.008) (0.022)
Financial literacy · 1(year=2010) 0.003 -0.018(0.009) (0.022)
Client of PattiChiari 0.028**(0.013)
Client of PattiChiari · 1(year=2008) -0.007(0.014)
Client of PattiChiari · 1(year=2010) 0.008(0.020)
N. of observations 3181 3181 3181
Notes: Each column reports estimates of a separate regression. OLS, ITT and IV columns are regression where the dependent
variable is(yitp−yi2002p−E[yitp−yi2002p])
V ar[yitp−yi2002p], i.e. standardized growth in financial assets between 2002 and time t for household i
in province p. FS columns are regressions where the dependent variable is financial literacy. All models include the same vector ofhousehold controls: age, age squared, gender, marital status and years of education of the household head; number of householdcomponents, household labour income, reasons for choosing the main bank and size of the municipality. All models include andyear-specific province fixed effects. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if thehousehold’s main bank belongs to the PattiChiari Consortium and 0 otherwise and interactions of this indicator with year dummies.Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 25: Robustness: Controlling for homeownership
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy 2.131*** 6.975**(0.629) (2.975)
Client of PattiChiari 0.230*** 1.603**(0.036) (0.713)
Homeowner 4.906*** 0.070** 4.987*** 4.497***(0.764) (0.027) (0.783) (0.792)
N. of observations 4865 4865 4865 4865F-test on the excluded instr. 41.490
Notes: Observations: 4865. Notes: the table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the
model wip = α+ βflip +Xipγ+λp +εip where wip is financial assets owned by household i, living in province p, in year 2010; flipis financial literacy of the household head, Xip is a vector of household controls (age, age squared, gender, marital status, yearsof education of the household head and number of household components, household labour income in 2010) and λp is a provincefixed effect. Financial literacy is instrumented with a dummy that takes the value 1 if the household’s main bank belongs to thePattiChiari Consortium and 0 otherwise. Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 26: Propensity Score Matching: Test of the Balancing Property on the commonsupport. SHIW 2010 data.
Mean t-testVariable Patti-Chiari non-Patti-Chiari % bias p-value
clients clients
Age 57.7 57.4 1.6 0.51Male 0.58 0.55 5.3 0.03Married 0.67 0.66 1.9 0.42N. hh. components 2.54 2.56 -2.1 0.39Years of education 10.4 10.3 2.9 0.23Household labour income 27.9 27.9 0.4 0.87Province∗ 43.7 43.3 1.4 0.59Municipality type∗ 1.43 1.42 0.5 0.81Reasons for choosing the main bank∗ 5.8 5.5 5.0 0.03Male · Married 0.46 0.45 2.8 0.24Small municipality · married 0.17 0.17 0.1 0.98Small municipality · n. hh. components 0.65 0.67 -1.4 0.56Large municipality · male 0.06 0.06 2.5 0.36Large municipality · age 6.3 6.03 1.8 0.47N. hh. components · hh. labour income 79.0 80.7 -2.5 0.34Married · hh. labour income 21.1 21.0 0.6 0.82
Joint test 0.103∗ For these variables we include in the pscore a list of K-1 binary indicators, where K is the number of
possible values of the variable.
Table 27: Association and Causal Effect of Financial Literacy on Financial Assets withInstruments that Use Information on“Bank-Switchers”Between Two years (i.e. BetweenSurvey Waves) or Between Four Years (Between Two Survey Waves).
HHs observed in 2006-2008 or in 2008-2010Descriptive statistics -Avg (se)- on this sample
Financial Assets: 22.9 (29.9); Financial Literacy : 1.46 (0.68)
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy∗ 2.009** 18.673**(1.008) (8.538)
Switchs to Client of PattiChiari -0.108 -0.989(0.067) (2.317)
Stays Client of Non-PattiChiari -0.133*** -3.660**(0.048) (1.448)
Switchs to Client of Non-PattiChiari -0.183*** -1.980(0.048) (1.884)
N. of observations 3196 3196 3196 3196
Hansen J p-value 0.4269
HHs observed in 2006 and 2010Descriptive statistics -Avg (se)- on this sample
Financial Assets : 21.9 (23.8); Financial Literacy: 1.43 (0.68)
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Financial literacy∗ 1.245 14.081*(1.091) (7.529)
Switchs to Client of PattiChiari -0.009 0.405(0.088) (2.991)
Stays Client of Non-PattiChiari -0.197*** -3.072**(0.064) (1.299)
Switchs to Client of Non-PattiChiari -0.087 0.638(0.071) (2.759)
N. of observations 1560 1560 1560 1560
Hansen J p-value 0.7535
Notes: ∗ We consider the two questions common across waves. The baseline analysis on year 2010 considers three questions. Seeappendix A for more details.
The table reports OLS, first-stage, intention-to-treat, and instrumental variable results of the model wip = α+ βflip +Xipγ+λp +εip where wip is financial assets owned by household i, living in province p, in year 2010; flip is financial literacy of the householdhead, Xip is a vector of household controls (as in the baseline specification of Table 2, excluding motivations for choosing the bank),and λp is a province fixed effect. Financial literacy is instrumented with three dummies identifying whether household switchedfrom being a PattiChiari to a non-PattiChiari bank (or vice-versa) or remained client of non-PattiChiari banks between 2008 and2010 (in the top panel) or between 2006 and 2010 (in the bottom one). Heteroskedasticity-robust standard errors clustered at theprovince level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.
Table 28: Effect of Financial Literacy on Financial Assets by Size of the Main Bank (Measured as the Level of Total Assetsin 2010 or the Number of Branches in 2010). Financial literacy is the Number of Correct Answers out of the Three QuestionsAsked in 2010. 2010 SHIW Data.
OLS FS ITT IV OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets Fin. assets Fin. lit. Fin. assets Fin. assets
Excluding top 10 largest banksfor nr. branches for total assets
Share of PattiChiari clients: 0.67 Share of PattiChiari clients: 0.66Avg. financial assets (sd): 18.26 (21.40) Avg. financial assets (sd): 18.40 (21.32)Avg. financial literacy (sd): 1.86 (1.00) Avg. financial literacy (sd): 1.86 (1.00)
Financial Literacy 2.268*** 7.710** 2.288*** 9.369**(0.733) (3.401) (0.765) (3.977)
Client of PattiChiari 0.276*** 2.127** 0.253*** 2.371**(0.048) (0.934) (0.050) (0.990)
N 3,084 3,084 3,084 3,084 3,002 3,002 3,002 3,002
F-stat excluded instrum. 33.47 25.87
Notes: Each column reports estimates of a separate regression. OLS, ITT and IV columns are regression where the dependent variable is financial assest. FS columns are regressionswhere the dependent variable is financial literacy. All models include the same vector of household controls: age, age squared, gender, marital status and years of education of thehousehold head; number of household components, household labour income, reasons for choosing the main bank and size of the municipality. All models include province fixedeffects. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0 otherwise.Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.+ Medians are computed on the sample of 70 banks listed in Table 9. The median of the number of bank branches is 307; the median of total assets is 9328.5.
Table 29: Effect of Financial Literacy on Financial Assets by Size of the Main Bank (Measured as the Level of Total Assetsin 2010 or the Number of Branches in 2010). Financial literacy is the Number of Correct Answers out of the Three QuestionsAsked in 2010. 2010 SHIW Data.
OLS FS ITT IV OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets Fin. assets Fin. lit. Fin. assets Fin. assets
Main bank: nr. branches ≤ median+ Main bank: nr. branches > median+
Share of PattiChiari clients: 0.55 Share of PattiChiari clients: 0.73Avg. financial assets (sd): 20.85 (20.38) Avg. financial assets (sd): 19.09 (22.09)Avg. financial literacy (sd): 2.07 (0.93) Avg. financial literacy (sd): 1.90 (0.99)
Financial Literacy 3.509*** 12.051 2.395*** 7.857**(1.294) (9.794) (0.701) (3.052)
Client of PattiChiari 0.292** 3.515 0.275*** 2.158**(0.116) (3.246) (0.043) (0.850)
F-stat excluded instrum. 6.35 40.89
N. of observations 471 471 471 471 4020 4020 4020 4020
Main bank: assets ≤ median+ Main bank: assets > median+
Share of PattiChiari clients: 0.65 Share of PattiChiari clients: 0.72Avg. financial assets (sd): 21.25 (23.16) Avg. financial assets (sd): 19.01 (22.09)Avg. financial literacy (sd): 2.00 (0.94) Avg. financial literacy (sd): 1.91 (0.99)
Financial Literacy -0.181 19.829 2.631*** 8.249***(1.485) (17.981) (0.708) (2.875)
Client of PattiChiari 0.246 4.882 0.273*** 2.253***(0.174) (3.437) (0.040) (0.785)
F-stat excluded instrum. 2.002 46.17
N. of observations 525 525 525 525 3966 3966 3966 3966
Notes: Each column reports estimates of a separate regression. OLS, ITT and IV columns are regression where the dependent variable is financial assest. FS columns are regressionswhere the dependent variable is financial literacy. All models include the same vector of household controls: age, age squared, gender, marital status and years of education of thehousehold head; number of household components, household labour income, reasons for choosing the main bank and size of the municipality. All models include province fixedeffects.. In IV columns, financial literacy is instrumented with a dummy that takes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0 otherwise.Heteroskedasticity-robust standard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.+ Medians are computed on the sample of 70 banks listed in Table 9. The median of the number of bank branches is 307; the median of total assets is 9328.5.
Table 30: Robustness: Including Main Bank Assets and Number of Branches - Year2010
OLS FS ITT IVDependent variable: Fin. assets Fin. lit. Fin. assets Fin. assets
Controlling for costs
Financial literacy 2.344*** 7.549**(0.687) (3.277)
Client of PattiChiari 0.236*** 1.785**(0.042) (0.769)
Main bank: transfer (desk) -0.440 -0.013 -0.468 -0.367(0.630) (0.028) (0.639) (0.591)
Main bank: transfer (online) 0.109 0.022 -0.261 -0.429(0.894) (0.051) (0.912) (0.918)
Main bank: withdrawal (desk) 1.104 0.185 -0.035 -1.434(3.270) (0.175) (3.179) (3.208)
Main bank: withdrawal (ATM) 4.808 0.194 4.892* 3.425(2.970) (0.127) (2.914) (3.243)
N. of observations 4256 4256 4256 4256F-test on the excluded instr. 32.391
Controlling for bank characteristics
Financial literacy 2.301*** 8.614*(0.648) (4.626)
Client of PattiChiari 0.223*** 1.918*(0.043) (1.023)
Log(Assets) -Main Bank- 0.175 0.005 -0.114 -0.158(0.316) (0.015) (0.445) (0.440)
Log(Branch) -Main Bank- -0.540* -0.016 -0.294 -0.160N. of observations 4568 4568 4568 4568F-test on the excluded instr. 26.749
Notes: All models include the same vector of household controls: age, age squared, gender, marital status and years of educationof the household head; number of household components, household labour income, reasons for choosing the main bank and size ofthe municipality. All models include province fixed effects. In IV columns, financial literacy is instrumented with a dummy thattakes the value 1 if the household’s main bank belongs to the PattiChiari Consortium and 0 otherwise. Heteroskedasticity-robuststandard errors clustered at the province level in parentheses.∗p < 0.1,∗∗ p < 0.05,∗∗∗ p < 0.01.