3.1 Average associations between financial literacy and preferences

Table 2 presents the results of a set of OLS regressions of country-level financial literacy on country-level preferences. The left-hand side variable is the percentage of adults correctly answering 3 out of 4 financial literacy questions in the S&P survey in a given country. Each panel presents results using a different measure of preference as key right-hand side variable and the average association between the S&P FL index and the preference across the 74 countries is estimated holding constant the country log GDP per capita and sequentially adding other covariates controlling for human capital development at the country level.

Table 2. Financial literacy and preferences

Notes: The dependent variable is the country-level percentage of individuals who have correctly answered 3 out of 4 financial literacy questions in the S&P survey. All regressions include a constant, a control for log of GDP pc in PPP and robust standard errors (in parenthesis). Column (4) reports average marginal effects from a logit model.

* p < 0.1, **p < 0.05, ***p < 0.01

Column 1 of Table 2 reveals that patience is positively associated with financial literacy in our sample, whereas trust and altruism are negatively related to financial literacy after controlling for the country’s log GDP per capita. As financial literacy has been found in the literature to be closely related to education and human capital development (Jappelli, Reference Jappelli2010; Cupák et al., Reference Cupák, Fessler, Silgoner and Ulbrich2021), in column 2 we further control for the average schooling of the adult population and the expected schooling for children in each country. Following the findings that financial literacy is related to math ability (measured with numeracy by Skagerlund et al., Reference Skagerlund, Lind, Strömbäck, Tinghög and Västfjäll2018), in column 3, we further control for a subjective measure of math skills at the country level, which proxies for math cognitive ability. While the association between financial literacy and patience or altruism remains unchanged, the association between financial literacy and trust becomes weaker and is no longer statistically significant once country differences in education are accounted for. Interestingly, after controlling for differences in education across countries, financial literacy is positively associated with risk-taking (albeit only marginally at the 10 percent level) and negatively associated with positive reciprocity.

Based on our baseline estimates, shown in column 3 of Table 2, a one standard deviation increase in patience is associated with 7.05 percentage points increase in financial literacy,Footnote ⁷ the equivalent of an increase of 18.75 percent in financial literacy given the S&P FL index averages 37.60 percent in our sample.Footnote ⁸ This association is statistically significant at the 1 percent level. Estimates in column 3 also reveal that a one standard deviation increase in risk-taking is associated with 2.26 percentage points increase in financial literacy,Footnote ⁹ the equivalent of an increase of 6 percent in financial literacy.Footnote ¹⁰ However, this association is only statistically significant at the 10 percent level. In contrast, a one standard deviation increase in altruism or positive reciprocity is associated with a decrease of 7.87 and 6.85 percent, respectively, in financial literacy.Footnote ¹¹ Both coefficients are statistically significant at the 5 percent or lower. These results hold when using a Logit instead of a linear regression as shown in column 4 of Table 2. The Logit regression model allows us to better accommodate the fractional nature of the dependent variable, a proportion bounded between 0 and 1.

3.2 Heterogeneity by Economic Development

Average associations between financial literacy and preferences in our sample conceal notable differences by whether the country is below or above the median GDP per capita. Table 3 reports estimates from regressions of the S&P FL index on the different preferences by level of economic development using the following model:

\begin{align*}{\text{S}}\& {\text{PFL}}{{\text{I}}_{\text{j}}} & = {{\beta}_0} + {{\beta}_1}{\text{Preferenc}}{{\text{e}}_{\text{j}}} + {{\beta }_2}{\text{AboveMedianGD}}{{\text{P}}_{{\text{pc}}}}_j\\ &\qquad+\,{{\beta}_3}\left( {{\text{AboveMedianGD}}{{\text{P}}_{{\text{pc}}}}_j{\text{x Preferenc}}{{\text{e}}_{\text{j}}}} \right) + Z{'_j}{\beta _4} + {{\varepsilon}_{\text{j}}}\end{align*}

Table 3. Financial literacy and preferences by level of economic development

Notes: The dependent variable is the country-level percentage of individuals who have correctly answered 3 out of 4 financial literacy questions in the S&P survey. All regressions include a constant and robust standard error (in parenthesis).”Above median log(GDP)” is a dummy for countries with GDP pc above the sample median. Column (4) reports average marginal effects from a logit model.

* p < 0.1, **p < 0.05, ***p < 0.01

where $S\& PFL{I_j}$ is the S&P FL index, which measures financial literacy at country $j$ level; ${\text{Preferenc}}{{\text{e}}_{\text{j}}}$ is the GPS preference measured at the country $j$; ${\text{AboveMedianGD}}{{\text{P}}_{{\text{pc}}}}_j$ is a dummy variable indicating whether country j’s GDP per capita is above the median GDP per capita in our sample; and $({\text{AboveMedianGD}}{{\text{P}}_{{\text{pc}}}}_j{\text{x Preferenc}}{{\text{e}}_{\text{j}}}$) is the interaction between the two aforementioned variables. The vector ${Z_j}$ includes country-level covariates that will vary with the estimated specification. Standard errors are heteroscedasticity robust.

In the equation above, ${\hat \beta _1}$ measures the average association between the country-level S&P FL index and country-level preference, say patience, in countries below the median GDP per capita. The coefficient, ${\hat \beta _2}$, measures the average difference in financial literacy between countries with a GDP per capita above the median (wealthier countries) and those below (impoverished countries). The coefficient, ${\hat \beta _3}$, measures the differential association between financial literacy and preference in wealthier relative to impoverished countries. Finally, the sum of the coefficients ${\hat \beta _{1{\text{ }}}},{\text{ }}{\hat \beta _2},{\text{ }}and{\text{ }}{\hat \beta _3}$, namely $\left( {{{\hat \beta }_1} + {{\hat \beta }_2} + {{\hat \beta }_3}} \right)$, measures the overall association between the country-level S&P FL index and country-level preference in countries with GDP per capita above the median.

Panel A of Table 3 reveals that patience is only salient in wealthier countries. The positive association between financial literacy and patience observed in Table 2 is solely driven by countries whose GDP per capita is above the sample median. Based on estimates of column 3, a one standard deviation increase in patience is associated with an increase of about 30 percent in financial literacy in countries above the median GDP per capita.Footnote ¹² This effect is statistically significant at the 1 percent level. In contrast, there is no association between patience and financial literacy in countries below the median. This difference by countries’ economic development is statistically significantly different from zero at the 1 percent level (see ${\hat \beta _3})$.

Moving to trust, we find a diverging association with financial literacy for countries above and below the median GDP per capita. A one standard deviation increase in trust is associated with an increase of 7.47 percent in financial literacy in wealthier countries, but with a decrease of 12.63 percent in impoverished countries.Footnote ¹³ These effects are statistically significant at the 5 percent level. While the coefficients ${\hat \beta _1}$ and ${\hat \beta _3}$ have opposite signs for risk-taking and altruism, we do not have enough precision to reject the null hypothesis ${H_0}:{\hat \beta _3} = 0$.

Furthermore, the observed negative association between positive reciprocity and financial literacy when using the whole sample of 74 countries is driven solely by countries with a GDP per capita below the median. In these countries, a one standard deviation increase in positive reciprocity is associated with a decrease of 11.45 percent in financial literacy.Footnote ¹⁴ It is important to underscore that such association is inexistent in countries above the median GDP per capita as $\left( {{{\hat \beta }_1} + {{\hat \beta }_2} + {{\hat \beta }_3}} \right)$ = 0.01 with a standard error of 0.076.

These findings are robust to different validation tests. First, in column (4) of Table 3, we estimate the parameters of our preferred specification through a fractional Logit model. Doing so, does not affect the main results. In Appendix Table B.1, we estimate a fully interacted model with all covariates interacted with the above the median GDP per capita dummy. While some of the estimates lose precision, the direction of the associations remains stable. Finally, in Appendix Table B.2, we consider different definitions of country economic development by estimating our baseline specificationFootnote ¹⁵ using GDP quartiles (Panel A), income groups defined by the World Bank (Panel B), and an OECD-countries indicator (Panel C). Our main results are mostly robust to these alternative measures of economic development. The major difference is observed for OECD-countries where we observe higher financial literacy for those with higher levels of risk-taking. This may suggest that the differential effect of risk-taking is not exclusively driven by economic development but by an overall improved policy and legal framework.

3.3 Institutional and Cultural Drivers

To explore which factors may be driving these correlations, we introduce in our OLS model other relevant controls capturing institutional and cultural factors which previous studies have deemed to be relevant in explaining financial literacy. As the culture of a country may influence the degree of interest and investment in financial knowledge, we introduce proxies for cultural institutions.

Results are presented in Table 4.Footnote ¹⁶ Column 1 presents our baseline model for comparison purposes. Column 2 adds to the baseline model the Gini index. This index measures the extent to which the income distribution among individuals within a country deviates from a perfectly equal distribution (with an index close to 1 being very unequal and an index close to 0 being very equal). As conjectured by Ahunov & van Hove (Ahunov and Van Hove, Reference Ahunov and Van Hove2020) in their analysis of national culture and financial literacy, in highly unequal countries less people may have the material means to invest in their financial education. Column 3 adds to our baseline model the legal rights index, which measures the degree to which collateral and bankruptcy laws protect the rights of borrowers and lenders and thus facilitate lending, and column 4 adds to the baseline model the legal origin of the country, which is a strong determinant of the depth of a country’s financial markets (Rafael et al., Reference Rafael, Florencio, Shleifer and Vishny1997, Reference Rafael, Florencio, Shleifer, Vishny, Constantinides, Harris and Stulz2013). Both covariates aim at capturing institutional differences across countries that are likely to be associated with financial knowledge. Column 5 controls for the share of the population of each country that belongs to the three most widely spread religions in the world, as countries in which the majority of population follow a particular religion may have different incentives to invest in financial literacy (Grohmann et al., Reference Grohmann, Klühs and Menkhoff2018). Finally, columns 7 and 8 include controls for geographical characteristics of the country and for the share of population in the country older than 65 years. Following Falk et al (Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018), we control for a set of geographic variables that have been proposed as potential determinants of preferences in the literature. To the extent that these geographical factors also shape human capital accumulation in the country (and hence, financial literacy), not including such controls may lead to omitted variable bias. A similar motive is also behind the choice of including country-level age structure as an additional control variable. In fact, recent evidence has underscored systematic variation in preferences over the life cycle (Sunde, Reference Sunde2023), and the link between age-related cognitive and physical decline, financial literacy, and intertemporal decision-making (Bialowolski et al., Reference Bialowolski, Cwynar and Weziak-Bialowolska2022; Huffman et al., Reference Huffman, Maurer and Mitchell2019; James et al., Reference James, Boyle, Yu, Han and Bennett2015).

Table 4. Financial literacy and preferences: additional controls

Notes: The dependent variable is the country-level percentage of individuals who have correctly answered 3 out of 4 financial literacy questions in the S&P survey. All regressions include a constant, controls for education and math skills and robust standard errors (in parenthesis).”Above median log(GDP)” is a dummy for countries with GDP pc above the sample median. Geographical controls are: average precipitations, average temperature, distance to the nearest waterway, latitude, rugged terrain, distance to Equator (as in Falk et al, Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018). Age structure is the share of population being 65 years old or older.

* p < 0.1, **p < 0.05, ***p < 0.01

As a first result, it is worth noting how preferences on patience remains practically unaffected by the introduction of new controls. To put it differently, the relationship underlined in Table 3 between patience and financial literacy in countries above the median GDP per capita appears very robust and not channeled through the different institutional controls we have added to the specification.

As far as concerns risk-taking, trust, altruism and positive reciprocity, we observe a noticeable change, in magnitude and significance of the correlations, following the introduction in the model of legal origin, and religion dummies. Legal origin absorbs all of the association as far as concerns risk-taking and about half of the association with trust and positive reciprocity in countries below the median GDP per capita; and all of the association with trust in countries above the median. At the same time, the association between altruism and financial literacy vanishes and that of trust is cut in half in countries below the median GDP per capita when introducing controls for the countries’ main religious groups. Appendix Table B.3, which reports the full set of estimates, suggests that emphasizing traits common among Protestantism would increase financial literacy in countries below the median GDP per capita. Similarly, emphasizing traits common among Scandinavian legal origin would also benefit financial literacy in impoverished countries. Column (6) in Table 4 reveals that continent dummies absorb most of the correlation between financial literacy and risk-taking and positive reciprocity.

3.4 Subcomponents of Financial Literacy

As a last step, we try to disentangle which field of knowledge is mostly associated to preferences and analyze the different subcomponents of the financial literacy index. The S&P FL index is composed of four different questions on risk diversification, inflation, numeracy, and interest compounding, with quite different response rates across the countries. For example, the questions on inflation and numeracy are the ones that exhibit the highest response rate, with more than half of the sample answering them correctly, whereas, on average, 60 percent of people exhibit illiteracy as far as concern the risk diversification concept (see Table 1). While being correlated with each other (Appendix Table A.4), these questions capture different aspects of financial literacy, and hence may exhibit heterogeneous relationship with preferences. Table 5 estimates our baseline specification, replacing the overall financial literacy index with the four different sub-components of the S&P FL index.

Table 5. Financial literacy sub-questions and preferences

Notes: The dependent variable is the country-level percentage of individuals who have correctly answered 3 out of 4 financial literacy questions (column (1)-(2)) or the specific questions (column (3)-(10)). All regressions include a constant, education and math skills regressors as in Table 3, column (6) and robust standard errors.

Odd columns control also for log of GDP per capita.”Above median log(GDP)” is a dummy for countries with GDP per capita above the sample median.

* p < 0.1, **p < 0.05, ***p < 0.01

Interestingly, inflation does not appear to correlate with any of the GPS preferences. As for the share of population correctly answering the numeracy question, we observe an association with trust, regardless of the country’s economic development measured by the GDP per capita.

The subcomponents of the S&P FL index that truly correlate with preferences, however, are the ones related to risk-diversification and interest compounding. Based on estimates from column 10 of Table 5, a one standard deviation increase in the country-level measure of patience is associated with an increase of about 41 percent in the rate of people correctly answering the risk-diversification question in countries with GDP per capita above the median.Footnote ¹⁷ Furthermore, the difference by countries’ economic development is statistically different from zero at the 1 percent significant level. In contrast, risk-taking, altruism, and positive reciprocity are associated with the rate of people correctly answering the risk-diversification question only in countries below the GDP sample median, as the effect in countries above the median is not precisely estimated – that is, the estimated $\left( {{{\hat \beta }_1} + {{\hat \beta }_2} + {{\hat \beta }_3}} \right)$ is not significantly different from zero. In fact, a one standard deviation increase in risk-taking is associated with an 18 percent increase in the rate of correct responses to the risk diversification question. A one standard deviation increase in altruism and positive reciprocity corresponds to a 17 percent and a 21 percent decrease in the rate of correct responses to the risk diversification question.Footnote ¹⁸

Trust has, like with the general index of financial literacy, a diverging effect according to the economic development: a one standard deviation increase in trust is associated with an increase of about 11 percent in the rate of correct responses to the risk diversification question in wealthier countries (significant at the 10 per cent), and with a decrease of 18 percent in impoverished countries (significant at the 1 per cent).Footnote ¹⁹

Similarly, based on the estimates of column 8 we observe that, for countries above the median GDP per capita, a one standard deviation increase in patience (trust) is associated with a statistically significant increase of around 22 percent (8 percent) in the rate of respondents answering correctly the question on interest compounding.Footnote ²⁰ In contrast, in countries below the median, a one standard deviation increase in trust and altruism diminish the rate of correct response by around 8 percent. At the same time, a one standard deviation increase in positive reciprocity decreases financial literacy by about 10 percent in countries below the median.Footnote ²¹

4.1 Discussion of the results

While the literature observes large differences in average financial literacy across countries (Annamaria and Mitchell, Reference Annamaria and Mitchell2014), meaningful cross-country evidence on the topic is scarce. Preferences and financial literacy both are considered to be important correlates of general well-being in the population, but the interplay between the two has not been widely studied. Merging two global, representative, and relatively novel datasets, the Global Preference Survey and the Standard & Poor’s Ratings Services Global Financial Literacy Survey, we uncover important differences in the association between financial literacy and preferences by the level of economic development.

In countries below the median GDP per capita, we document a higher level of financial literacy in countries with higher levels of risk-taking, but with lower levels of trust, positive reciprocity, and altruism. Countries’ legal origin drives all of the association with risk-taking and about half of the relationship with trust and positive reciprocity. The countries’ distribution of major religions drives the association between altruism and financial literacy. Ultimately our results document that the association between these preferences and financial literacy in countries below the median GDP per capita appears to be driven by institutional factors, suggesting that training program and policies emphasizing certain traits associated with Scandinavian legal origin or Protestantism would help improve financial literacy.

Interestingly, patience is only salient in countries above the median GDP per capita, with higher financial literacy in countries with higher level of patience. Importantly, this association is very robust, and it is not driven by a multitude of institutional or cultural factors known to be related to financial literacy. This positive association between financial literacy and patience is somewhat in line with the work of Hanushek et al (Reference Hanushek, Kinne, Lergetporer and Woessmann2022) and with the idea that human capital investment decisions are, by nature, intertemporal decisions, and hence inextricably linked to cultural traits, such as time preferences, related to the valuation of present versus future payoffs. Furthermore, as pointed out by Lusardi et al. (Annamaria et al., Reference Annamaria, Michaud and Mitchell2017), if more patient households would be more inclined to invest in human capital and, more specifically, in financial knowledge, education and wealth levels are likely to be important factors of heterogeneity in the patience-knowledge association. The lack of relationship between patience and financial literacy in impoverished countries, and the generally lower level of patience, might be related to the more uncertain horizon that populations in those countries may face.

The last interesting result of our analysis relates to the differential relationship between preferences and the four sub-components of the financial literacy index. The associations outlined above only emerge when knowledge on risk diversification and interest compounding are taken into account, the two financial literacy questions people struggle the most with (Annamaria and Mitchell, Reference Annamaria and Mitchell2014). In contrast to other studies, we analyze the link between financial literacy and preferences in a unified framework, by looking at both social and economic preferences as possible correlates of financial literacy. This is important because preferences do not typically affect decisions and life outcomes in an isolated manner bur rather jointly, hence it makes sense to consider them in a unified framework.Footnote ²² Our analysis is, to some extent, similar to the ones performed by Ahunov & van Hove (Ahunov and Van Hove, Reference Ahunov and Van Hove2020) and De Beckker et al (Reference Beckker, Kenneth and Van Campenhout2020). The first employ the Standard and Poor’s Financial Literacy Survey while the latter the OECD/INFE financial literacy surveys to uncover correlations between financial literacy and national cultures measured by mean of Hoefstede’s cultural dimensions, such as individualism, power distance and uncertainty avoidance. However, the findings from these two studies diverge, possibly due to differences in sample sizes or the non-representativeness of Hofstede’s cultural proxies. In fact, differently from the GPS, Hofstede’s measures of national culture are the result of interviews to non-representative samples of IBM employees, and, while being still in use, the validity of these measures have been largely criticized (see, e.g., Shaiq et al., Reference Shaiq, Khalid, Akram and Ali2011). Hence, our study contributes to this cross-country framework by using comparable and standardized measures of both financial literacy and preferences for a large set of countries. Moreover, we identify a differential pattern by economic development allowing us to identify relevant institutional factors specific for impoverished countries.

4.2 Limitations and endogeneity

This study explores associations between financial literacy, risk preferences, and cultural traits across countries, making no causality claims. While our cross-country data and aggregated preference measures uncover novel correlations that provide valuable insights for future research, several limitations must be considered when interpreting our results.

First, when analyzing a complex construct like financial literacy, confounding factors present a significant challenge, as it is difficult to control for all relevant variables. Preferences are not inherently bestowed on certain countries; institutional and societal characteristics contribute to their development while also influencing country-level financial literacy accumulation.

Second, bidirectional relationships may exist between financial literacy and preferences. Countries with more developed financial markets might provide better opportunities for individuals to access financial information, potentially strengthening preferences such as patience and risk-taking. More patient individuals may self-select into activities that promote financial knowledge, anticipating long-term returns on such human capital investment (Meier and Sprenger, Reference Meier and Sprenger2013). Simultaneously, gaining financial knowledge could affect individuals’ deep time preferences (Lührmann et al., Reference Lührmann, Serra-Garcia and Winter2018). Different studies demonstrate that time preferences correlate with financial information acquisition and better financial outcomes (Lührmann et al., Reference Lührmann, Serra-Garcia and Winter2018; Meier and Sprenger, Reference Meier and Sprenger2012, Reference Meier and Sprenger2013), while financial literacy interventions can alter individuals’ subjective discount rates and time preferences (Lahav et al., Reference Lahav, Rosenboim and Shavit2015; Lührmann et al., Reference Lührmann, Serra-Garcia and Winter2018). Thus, the directionality between time preferences and financial literacy remains ambiguous.

Third, we cannot control for factors such as present bias and exponential growth bias, both linked with financial behavior and patience. The GPS patience measure captures a concept close to present bias, as respondents indicate their willingness to delay gratification. Since both present-bias and exponential-growth bias appear negatively correlated with financial literacy (Almenberg and Gerdes, Reference Almenberg and Gerdes2012; Wang and St John, Reference Wang and St John2025), omitting these measures likely creates a downward bias in our estimates of the patience-financial literacy relationship.

While it is challenging to address the endogeneity of preferences, evidence suggests their origins lie deeper than economic development. Becker et al (Reference Becker, Enke and Falk2020), utilizing GPS data, show that preference differences between populations are largely explained by distant migration patterns of early ancestors, somewhat confirmed by Galor & Özark (Reference Galor and Özak2016). Falk et al (Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018) and Wang et al (Reference Wang, Rieger and Hens2016) provide evidence that aggregate preferences strongly relate to cultural and geographic conditions. It is reasonable to consider preferences as deep determinants affecting financial literacy, possibly through proximate determinants such as GDP or country wealth. However, distinguishing preference effects on financial literacy from other factors possibly determined by preferences remains econometrically challenging, which is why our results should be interpreted as robust correlations rather than evidence of causal relationships.

Though some instrumental variables have been used to address financial literacy endogeneity in other contexts (Behrman et al., Reference Behrman, Mitchell, Soo and Bravo2012; Grohmann et al., Reference Grohmann, Klühs and Menkhoff2018; Ricci and Carratelli, Reference Ricci and Carratelli2017), there is, to our knowledge, no paper successfully instrumenting preferences at the country level. As there is no suitable instrumental variable for all our preference measures, we cannot use an IV estimation strategy.

While it should be underlined that our results do not imply causality, the robust correlations found in our analysis stress the importance of further research into the causal link between financial literacy and preferences. Overall, our study underlines the need for additional studies and theories to explain a possible link between financial literacy and different preferences in the context of varying levels of economic development. Furthermore, it suggests that designing policies to improve financial literacy requires a multi-faceted approach that considers not only the formal institutional framework where initiatives are promoted but also the societal and cultural context.

Despite these limitations, our study advances the understanding of financial literacy’s relationship with cultural and behavioral attributes across national contexts. Future research employing longitudinal designs or natural experiments could further illuminate the mechanisms underlying these associations and potentially address questions of causality beyond the scope of this investigation.