A. Outcome Variables

Our outcome variables are various measures of asset allocation of the population of second-generation migrants in Sweden. These data come from the Swedish Wealth Registry (Förmögenhetsregistret) and were collected by Statistics Sweden (the government’s statistical agency) for tax purposes. The data include all financial assets held outside retirement accounts at the end of a tax year, Dec. 31, reported by a variety of different sources, including the Swedish Tax Agency, welfare agencies, and financial institutions. Importantly, nontaxable securities and securities owned by investors below the wealth tax threshold were included in the reports (Calvet, Campbell, and Sodini (Reference Calvet, Campbell and Sodini2007)). With information based on statements from financial institutions and the full coverage of the population, issues of measurement error and selection bias (which are frequently substantial concerns) are negligible in our setting. We have data on assets (linked with the holders’ country of birth) from 1999 to 2006.

In our analysis of second-generation migrants, we focus on wealth in the year 2006. Between 1999 and 2005, banks were not required to report small bank accounts to the Swedish Tax Agency unless the account accrued more than 100 SEK (about 11 USD) in interest during the year. From 2006 onward, banks were required to report all bank accounts above 10,000 SEK.Footnote ¹⁰ Also, focusing on 2006 allows us to have more second-generation migrant children to be old enough to participate in the stock market than earlier in the sample.Footnote ¹¹

We look at five outcome variables in total. The first is an indicator of participation in the risky asset market, through stocks or mutual funds, and the second is the share of risky assets in financial assets conditional on participation. Since market participation may be influenced by both preferences and other individual characteristics, such as wealth and cognitive ability, which can help overcome entry barriers into the markets, we are most interested in analyzing portfolio composition conditional on participation that could best reveal the role of culturally transmitted preferences in investment behavior.Footnote ¹²

Outcomes that we use to analyze portfolio composition conditional on risky asset market participation include: i) an indicator for whether the individual owns stocks directly, which we refer to as stock market participation; ii) the share of directly held stocks out of the total value of mutual funds and directly held stocks, which we refer to as stock share; iii) a proxy for portfolio volatility.

Investment data from the wealth register that is linked with country of birth for the population of Sweden is only available to us at the aggregated level, meaning that we observe the total value of individuals’ directly held stocks, fixed-income funds, and equity funds but not the composition of stocks and fund portfolios.Footnote ¹³ Therefore, we cannot precisely measure portfolio volatility and instead follow the approach of Calvet et al. (Reference Calvet, Campbell and Sodini2007), where they devised an approach to proxy portfolio diversification with more limited data.Footnote ¹⁴

B. Variable of Interest

Our variable of interest is a measure of risk-taking culture captured by the risk preferences associated with second-generation migrants’ ancestral countries (i.e., the country of origin of their parents). These data come from the GPS; an experimentally validated survey data set of the global variation in preferences (Falk et al. (Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018)).Footnote ¹⁵ GPS provides us with measures specifically designed to capture economic and social preferences from 80,000 people in 76 countries that represent approximately 90% of the world population.Footnote ¹⁶ The surveys are carried out on representative samples within each country and exhibit substantial heterogeneity in preferences across countries.Footnote ¹⁷

Risk preferences (risk taking) were elicited through a series of related quantitative questions as well as one qualitative question (see Falk et al. (Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018) for details). The quantitative survey measure consists of a series of five interdependent hypothetical binary choices, a format commonly referred to as a “staircase” (or “unfolding brackets”) procedure (Cornsweet (Reference Cornsweet1962)). Choices were between a fixed lottery, in which the individual could win x or zero, and varying sure payments, y.Footnote ¹⁸

The qualitative item asks for the respondents’ self-assessment of their willingness to take risks on an 11-point Likert scale, “In general, how willing are you to take risks?” This qualitative subjective self-assessment has previously been shown to be predictive of risk-taking behavior in the field in a representative sample (Dohmen, Falk, Huffman, Sunde, Schupp, and Wagner (Reference Dohmen, Falk, Huffman, Sunde, Schupp and Wagner2011)) as well as of incentivized experimental risk taking across countries in student samples (Vieider, Chmura, Fisher, Kusakawa, Martinsson, Thompson, and Sunday (Reference Vieider, Chmura, Fisher, Kusakawa, Martinsson, Thompson and Sunday2015)). The qualitative item and the outcome of the quantitative staircase measure were combined through roughly equal weights.

Figure 1 shows the distribution of risk taking measure by quartiles across countries in our sample. Risk taking measure ranges between −0.79 and 0.97. For a complete list of countries and their risk taking scores, see Supplementary Material Table A.1. Risk preferences vary substantially geographically, as well as within a set of countries with similar levels of development. For example, within Europe, while the Netherlands are in the top risk-taking quartile, Spain and Germany are in lower quartiles.Footnote ¹⁹

Figure 1Risk Taking Across Countries

The data in Figure 1 come from Falk et al. (Reference Falk, Becker, Dohmen, Enke, Huffman and Sunde2018)

Figure 1 Long description

A choropleth map of the world where countries are shaded in four gradients of red based on risk-taking data.

* High Risk-Taking (Darkest Red, > 0.14): Concentrated in Canada, Peru, Algeria, Nigeria, South Africa, Saudi Arabia, Iran, and Afghanistan.

* Moderate-High Risk-Taking (Medium Red, -0.02 to 0.14): Visible in the United States, Mexico, Argentina, Australia, China, and parts of Western Europe.

* Moderate-Low Risk-Taking (Light Red, -0.15 to -0.02): Found in Brazil, Russia, and several Central African nations.

* Low Risk-Taking (Lightest Pink, < -0.15): Observed in India, parts of Southeast Asia, and Eastern Europe.

* No Data (Grey): Greenland and various countries in Central Africa and Southeast Asia are shaded grey.

A legend at the bottom center provides the numerical scale for the four color categories from left to right: less than -0.15 (lightest), -0.15 to -0.02, -0.02 to 0.14, and greater than 0.14 (darkest).

C. Controls

In our baseline regressions, we control for gender, year of birth, and time preferences from GPS. In our robustness exercises, we go further by taking into account parents’ years of birth, education, income ranks and parental financial wealth ranks (both within parental birth cohort).Footnote ²⁰ We also investigate the mediating role of individuals’ education level, income rank, and financial wealth rank. All of the variables are provided by Statistics Sweden and are derived from administrative records, primarily from the Swedish tax authority.Footnote ²¹

Our final sample for the main analysis conditional on equity participation contains 38,702 observations (64,466 observations for unconditional outcomes). Table 1 provides summary statistics. Conditional on participation in equity markets, 45 percent directly hold stocks.

Table 1Summary Statistics

Table 1 Long description

The table contains six columns: Variable Name, Mean, Standard Deviation, Minimum, Maximum, and Number of Observations. It is divided into five thematic sections.

1. Outcomes, 2006: Risky asset participation has a mean of 0.60. Conditional risky asset share, stock-market participation, and stock share of risky assets have means of 0.48, 0.45, and 0.29 respectively. Portfolio volatility proxy has a mean of 0.23.

2. Cultural variable: Risk-taking culture mean is 0.002. Ancestral risk-taking chance games mean is 0.47. Ancestral climatic instability mean is 0.27. Patience mean is 0.043.

3. Individual characteristics: Female mean is 0.47. Age mean is 35.47. Years of education mean is 12.66. Labor income mean is 233. Financial wealth mean is 225.

4. Parental characteristics, 1999: Father's age mean is 59.26 and mother's is 55.89. Father's education mean is 10.71 and mother's is 10.52. Father's labor income mean is 114.92 and mother's is 92.03. Financial wealth of parents mean is 416.74. Municipal share of same origin region mean is 0.05. Parents from same country mean is 0.76.

5. Country-level characteristics: Log G D P mean is 9.61. Life expectancy mean is 70.18. Trust mean is 0.0025. Rule of law mean is 0.18.

In the same sample, stock holdings make up 29 percent of risky assets (stocks and mutual funds).Footnote ²²

Identifying Assumption

The key assumption of our empirical strategy is that, by including the risk preference measure in parents’ countries of origin, we capture the effect of cultural preferences and not that of potentially omitted variables. By observing second-generation immigrants in a common environment, we are able to distinguish cultural factors from institutional and economic ones, as these latter ones do not vary, while cultural heritage does. The assumption will be violated if cultural preferences are systematically correlated with other factors that affect financial behavior. One such example is if migrants from relatively risk-loving countries are wealthier (for other reasons than their high risk tolerance) and children of wealthier parents are more likely to participate in equity markets. The fact that we can observe and control for parental characteristics greatly mitigates these concerns. In Sections IV and V, we address the issue of confounding variables in detail and perform a number of robustness analyses.

A. Risky Asset Market

We start our analysis by showing the effects of country of origin risk preferences on risky asset market participation, regardless of whether participation is through mutual funds or directly owned stocks, in Panel A of Table 2, and the share of financial wealth invested in risky assets conditional on participation in Panel B. In all specifications, we control for year-of-birth and gender fixed effects. We do this because the previous literature has documented that the life cycle has important implications for equity-market participation (Fagereng, Gottlieb, and Guiso (Reference Fagereng, Gottlieb and Guiso2017)), and men and women could behave differently in their investment behavior. As explained before, we also control for the country of origin time preferences, although the results are very similar without those.

Table 2Risky Asset Participation, Conditional Risky Asset Share, and Cultural Risk Preferences

Table 2 Long description

The table is organized into eight numbered columns representing different regression models.

Panel A: Risky Asset Participation.

- Risk-taking culture coefficients range from 0.0779 in column 1 to 0.0239 in column 8. Significance levels vary from p < 0.05 to p < 0.01.

- Number of observations is constant at 64,466.

- R-squared values increase from 0.021 in column 1 to 0.113 in column 8.

Panel B: Conditional Risky Asset Share.

- Risk-taking culture coefficients range from 0.0491 in column 1 to 0.0296 in column 8. Most are significant at the p < 0.01 level.

- Number of observations is constant at 38,702.

- R-squared values increase from 0.013 in column 1 to 0.038 in column 8.

Control Variables (Bottom Section):

- Baseline controls are ‘Yes’ for all columns.

- Parental education F E and Parental cohort F E are ‘No’ in column 1 and ‘Yes’ in columns 2 through 8.

- Income rank for father and mother are ‘No’ in columns 1 and 2, and ‘Yes’ in columns 3 through 8.

- Parental wealth rank is ‘No’ in columns 1 through 3, and ‘Yes’ in columns 4 through 8.

- Individual education F E is ‘Yes’ only in columns 5 and 8.

- Individual income rank is ‘Yes’ only in columns 6 and 8.

- Individual wealth rank is ‘Yes’ only in columns 7 and 8.

While in the literature the decision to participate in risky asset markets is partly ascribed to overcoming fixed costs, we would expect those with higher risk preferences to be more likely to own equities. Column 1 in Panel A of Table 2 shows that culturally transmitted preferences for risk increase the likelihood of participation in risky asset markets. The coefficient estimate indicates that a 1-standard-deviation increase in ancestral risk preferences (0.29) increases the probability of risky-asset-market participation by 2.2 percentage points, compared to the mean of 60%.

Similarly in column 1 in Panel B of Table 2, we observe that, for those who participate in the risky asset market, individuals who come from more risk-loving cultural backgrounds designate more of their financial wealth to risky assets. A 1-standard-deviation increase in the culture of risk-taking (0.29) leads to an increase of 1.4 percentage points in the risky asset share, compared to the mean of 48%.

3. Further Robustness to Alternative Country of Origin Characteristics

A possible scenario is that countries with higher measures of cultural risk-taking might be different in other ways that affect the investment behavior of children of immigrants from those countries. One should note that, for this to be a threat to our identification, these potential effects should be in addition to their impact on the socioeconomic characteristics of the first-generation immigrants themselves, which we account for.

To investigate this, in Table 3, in addition to baseline controls and parental controls (as in column 4 of Table 2), we add further controls to proxy for many aspects of heterogeneity across countries from which first generation migrants descend. In column 1, we add continent fixed effects, allowing us to use within continent variation. This has no discernible effect on either participation (Panel A) or conditional risky share (Panel B). These suggest that a few country clusters do not drive the results.Footnote ³¹

Table 3Risky Asset Participation, Conditional Risky Asset Share, and Cultural Risk Preferences $\mid$ $ \mid $ vertical bar Robustness to Other Cross-Country Controls

Table 3 Long description

The table is organized into six numbered columns.

Panel A: Risky Asset Participation.

Risk-taking culture coefficients are 0.0538 (standard error 0.0268) in column 1, 0.0600 (0.0188) in column 2, 0.0585 (0.0202) in column 3, 0.0624 (0.0198) in column 4, 0.0225 (0.0268) in column 5, and 0.0227 (0.0251) in column 6. Number of observations is approximately 64,400 with R-squared values around 0.064 to 0.065.

Panel B: Conditional Risky Asset Share.

Risk-taking culture coefficients are 0.0487 (0.0179) in column 1, 0.0432 (0.0125) in column 2, 0.0452 (0.0133) in column 3, 0.0427 (0.0151) in column 4, 0.0300 (0.0159) in column 5, and 0.0456 (0.0154) in column 6. Number of observations is approximately 38,700 with R-squared values around 0.025 to 0.026.

Control Variables:

Continent F E is ‘Yes’ in columns 1 and 6.

Log G D P per Cap. is ‘Yes’ in columns 2 and 6.

Life expectancy is ‘Yes’ in columns 3 and 6.

Trust is ‘Yes’ in columns 4 and 6.

Rule of law is ‘Yes’ in columns 5 and 6.

Baseline controls, Parental education F E, Income rank (father and mother), Parental wealth rank, and Parental cohort F E are ‘Yes’ across all six columns.

Next, we control for GDP per capita and life expectancy of the source countries in columns 2 and 3. Data on GDP per capita are from the Penn World Tables, measured in 1995, and data on Life Expectancy are from the World Bank, WDI, measured in 2016. Estimates of the coefficients of interest in the following 2 columns are very similar to the previous estimates. GDP per capita and life expectancy coefficient estimates are both economically and statistically insignificant. This suggests that the level of development of the source countries is unlikely to drive our findings and it is not confounding the cultural variable of interest.

Alternatively, one could argue that the selection of immigrants from countries with differential levels of development is not what we are picking up in our regressions, but, rather, economic preferences could correlate with social preferences affecting financial behavior. This is not an argument against the role of culturally transmissible traits in general, but the coefficient estimates on our variable of interest could be biased. More specifically, Guiso et al. (Reference Guiso, Sapienza and Zingales2004), (Reference Guiso, Sapienza and Zingales2008) suggest that trust (or social capital in general) is a cultural factor shaping financial behavior. To address this, in column 4, we account for the trust measure from the GPS, which could potentially affect our outcomes independently. Results suggest that controlling for trust has no effect on the coefficients of interest.

Lastly, one may be concerned about institutional characteristics of the origin countries confounding the influence of culture. It is worth noting that institutions are arguably endogenous to culture, and therefore, controlling for their quality would make for adding bad controls. Nevertheless, in column 5, we control for the rule of law as a measure of institutional quality in the origin country, extracted from the Worldwide Governance Indicators (Kaufmann, Kraay, and Mastruzzi (Reference Kaufmann, Kraay and Mastruzzi2010)).Footnote ³² While the estimate for market participation becomes statistically insignificant, the estimate of the effect of risk-taking culture on risky shares remains meaningful and statistically significant, indicating that culture plays a distinct role not explained by institutional differences. In column 6, we control for all the characteristics of countries of origin discussed before. The estimates are very close to the ones in column 5, and even stronger in the risky asset share regression of Panel B.

B. Compositional Effects

So far, we have documented that those from more risk-loving cultures are more likely to participate in risky asset markets and, conditional on that, designate a larger share of their financial wealth to equities. While participation is not necessarily attributed to risk-taking preferences, as in classical models everyone is expected to participate and nonparticipation is usually attributed to an inability to overcome fixed costs, it is predicted that a more risk-loving individual chooses a higher risky share.

However, our findings do not indicate whether more risk-taking in financial markets induced by cultural background is manifested in a mean–variance efficient portfolio with the best possible Sharpe ratio, as predicted in the canonical model of investment behavior, or a portfolio with high idiosyncratic risk and high volatility, where the expected return might be low. In other words, it is not clear if there are compositional effects whereby cultural background induces taking on more idiosyncratic risk or if it is just about a larger mean–variance efficient portfolio.

Compared to investments in mutual funds, direct investments in stocks typically exhibit greater volatility with potentially more extreme returns, characteristics particularly appealing to investors with an appetite for idiosyncratic risk. Previous research suggests that individuals who invest in stocks rather than mutual funds often perceive trading as a form of gambling and may substitute between the two activities (Barsky et al. (Reference Barsky, Juster, Kimball and Shapiro1997)). Additionally, survey respondents frequently equate stock investing with gambling, buying lottery tickets, or visiting a casino (Chopra and Haaland (Reference Chopra and Haaland2023), Henkel and Zimpelmann (Reference Henkel and Zimpelmann2023)).Footnote ³³

Dorn et al. (Reference Dorn, Dorn and Sengmueller2015) show that fluctuations in lottery prizes in Germany influence trading behavior in individual stocks and options, whereas mutual fund trading does not provide the same gambling experience for such gambling-motivated investors. Similarly, Gao and Lin (Reference Gao and Lin2015) find that larger lottery jackpots in Taiwan lead to a decline in trading volume for stocks favored by individual investors. Further supporting this connection, Kumar (Reference Kumar2009) finds that individual investors are drawn to stocks with lottery-like characteristics, such as low prices, high idiosyncratic volatility, and high idiosyncratic skewness.

In Table 4, we investigate how culturally inherited risk preferences affect the composition of risky portfolios. The sample is restricted to risky asset market participants. Specifically, we try to understand if those from more risk-loving cultures take on more idiosyncratic risk by investing (relatively) more in directly held stocks as opposed to investing in mutual funds and if they form risky portfolios with higher volatility.

Table 4Stock Participation, Stock Share, Portfolio Volatility Proxy, and Cultural Risk Preferences

Table 4 Long description

The table is organized into eight columns representing different regression models.

Panel A, Stock Participation: The coefficient for Risk-taking culture ranges from 0.212 in column 1 to 0.141 in column 8, all significant at the 1 percent level. R-squared values increase from 0.064 to 0.149 across the models.

Panel B, Stock Share: The coefficient for Risk-taking culture ranges from 0.165 in column 1 to 0.140 in column 8, all significant at the 1 percent level. R-squared values increase from 0.056 to 0.072.

Panel C, Portfolio Volatility Proxy: The coefficient for Risk-taking culture ranges from 0.0139 in column 1 to 0.0170 in column 8, significant at the 5 percent level in column 1 and the 1 percent level in columns 2 through 8. R-squared values increase from 0.033 to 0.054.

Control Variables: All models include Baseline controls. Parental education F E and Parental cohort F E are included in columns 2 through 8. Father and mother income ranks are included in columns 3 through 8. Parental wealth rank is included in columns 4 through 8. Individual education F E is included in columns 5 and 8. Individual income rank is included in columns 6 and 8. Individual wealth rank is included in columns 7 and 8.

The number of observations is constant at 38,702 for all models.

The outcome of interest in Panel A of Table 4 is an indicator of whether participants directly own stocks. The baseline estimate in column 1 indicates that, conditional on participation, those from more risk-loving cultures are more likely to directly hold stocks. A 1-standard-deviation increase in ancestral risk-taking preferences leads to a 6.1 percentage point increase in the likelihood of owning stocks (compared to a mean of 45%). An alternative way to get a sense of the quantitative significance of the effect is to compare individuals from countries in the top and bottom quartiles of risk-taking. For example, if a market participant with a Portuguese heritage (−0.79) had the risk-taking preferences of someone with an Algerian heritage (0.39), her probability of stock-market participation would go up by 25 percentage points.

As in Table 2, columns 2–4 of Table 4 control for parental characteristics. None of these has any discernible effect on the coefficient estimate. Comparing columns 1 and 4 reveals an Altonji ratio of 13.1, which rules out selection on unobservables to a great extent.Footnote ³⁴

Again, as previously, in columns 5–8, we add individual characteristics. The coefficient estimate survives even controlling for these, suggesting that the effect of cultural heritage might work beyond the mediating effect of other personal outcomes.

These findings suggest that culturally transmitted risk-taking preferences have a compositional effect on individuals’ portfolios by inducing people to hold stocks directly and take more idiosyncratic risk.

In Panel B of Table 4, we investigate the intensive margin by exploring whether those who descend from more risk-loving cultures would also invest (relatively) more in directly held stocks than mutual funds, which presumably would also lead to the formation of less diversified portfolios. The outcome is the share of directly held stocks when the denominator is the total value of funds and stocks. Estimates suggest that those from more risk-loving cultures assign a larger share of their portfolio to directly held stocks. A 1-standard-deviation increase in ancestral risk-taking preferences leads to a 4.8 percentage point increase in the share of stocks among risky assets (compared to a mean of 29%). Also, comparing the baseline coefficient in column 1 with that in column 4 reveals that taking into account parental characteristics barely changes anything, with a high Altonji ratio of 26.5.Footnote ³⁵

Instead, the rest of the table suggests that individual characteristics are not strong enough mediating factors to make a difference in the direct effect of cultural risk preferences (columns 5–8).

While these exercises provide suggestive evidence, they do not, on their own, establish that individuals with a more risk-tolerant cultural background take on greater idiosyncratic risk. To strengthen our evidence, we attempt to approximate individuals’ portfolio volatility. Because we do not have information on the exact portfolio of assets held by individuals, we create a portfolio volatility proxy using the methodology detailed in Appendix 2.12 of Calvet et al. (Reference Calvet, Campbell and Sodini2007) and explained in Section II. The results are presented in Panel C of Table 4. The findings in column 1 indicate that a standard deviation increase in cultural risk-taking raises portfolio volatility by almost 7% of a standard deviation. Adding parental background characteristics in columns 2–4 and individual characteristics in columns 5–8 does not have any recognizable effect.

Importantly, in additional exercises in Supplementary Material Tables A.3 and A.4, we observe that the findings in all 3 panels of Table 4 are robust to taking into account parental investment behavior and individual cognitive ability. This rules out the possibility that the influence of cultural risk preferences on portfolio composition is driven by parental influence (e.g., due to differential parental financial sophistication) or individuals’ ability (e.g., better understanding of financial markets).

Finally, we further check the sensitivity of our findings in Table 4 to previously mentioned country-level controls. Table 5 shows that the results are robust and that the previous conclusions carry over.

Table 5Stock Participation, Stock Share, Portfolio Volatility Proxy, and Cultural Risk Preferences $\mid$ $ \mid $ vertical bar Robustness to Other Cross-Country Controls

Table 5 Long description

The table is organized into six columns representing different econometric models.

Panel A: Stock Participation. Risk-taking culture coefficients range from 0.137 to 0.228, all statistically significant. R-squared values are approximately 0.085 to 0.088.

Panel B: Stock Share. Risk-taking culture coefficients range from 0.129 to 0.218, all statistically significant. R-squared values are approximately 0.062 to 0.065.

Panel C: Portfolio Volatility Proxy. Risk-taking culture coefficients range from 0.0113 to 0.0252, all statistically significant. R-squared values are approximately 0.039 to 0.041.

Below the panels, the table indicates which controls are included in each model.

Model 1 includes Continent F E and Baseline controls.

Model 2 includes Log G D P per Capita and Baseline controls.

Model 3 includes Life expectancy and Baseline controls.

Model 4 includes Trust and Baseline controls.

Model 5 includes Rule of law and Baseline controls.

Model 6 includes all aforementioned controls.

All models include Parental education F E, Income rank for father and mother, Parental wealth rank, and Parental cohort F E. The number of observations is 38,669 for models 1 and 6, and 38,702 for models 2 through 5.

Overall, the findings suggest a large effect of cultural heritage on portfolio composition by increasing the level of idiosyncratic risk, which indicates a departure from mean–variance efficiency for those born into more risk-loving cultures. While this may suggest that the cultural component of risk-taking, at least as measured by GPS, induces both greater aggregate and idiosyncratic risk, it is also consistent with the possibility that individuals from more risk-loving cultural backgrounds are less sophisticated investors. However, the latter possibility is very unlikely since one would expect investor sophistication to be highly correlated with individual and parental characteristics, and we have shown that the effects survive not only controlling for (arguably) influential parental background, but also individual characteristics (such as cognitive ability).

All estimates shown in Tables 4 and 5 are economically meaningful and statistically significant, indicating that the largest effects of cultural heritage seem to be on portfolio composition and taking idiosyncratic risk. With this in mind, in the rest of the article, we focus our attention on portfolio composition and the mechanisms through which culture influences portfolio composition.Footnote ³⁶

A. Ancestral Risk-Taking Culture Proxied by Ethnographic Chance Games in Parental Birth Country

Part of the literature that studies the impact of cultural values on economic outcomes has focused on cultural variables that are measured before modernization and industrialization and that predate economic outcomes by a very long time (Alesina et al. (Reference Alesina, Giuliano and Nunn2013), Giuliano and Nunn (Reference Giuliano and Nunn2013), Michalopoulos (Reference Michalopoulos2012), and Nunn and Wantchekon (Reference Nunn and Wantchekon2011)). The advantages of using cultural variables measured very far back in time are twofold. First, it rules out reverse causality (e.g., gender norms today cannot have caused plough usage centuries ago (Alesina et al. (Reference Alesina, Giuliano and Nunn2013))). Second, it provides an intuitive understanding of where the differing cultural norms come from, as these measures capture characteristics of ancestral tribes or communities before any modernization and industrialization took place.

In our setup, with the spatial separation that our identification strategy relies on, reverse causality is already ruled out; there is no plausible mechanism by which cross-sectional variance in financial decision-making in Sweden has a material impact on measured average risk preferences across countries. Furthermore, as we are mainly concerned with the impact of cultural values on financial decision-making, and not how those cultural values are formed, we prefer using a direct measure of risk preferences as our baseline. Nevertheless, using a “deeper” measure of cultural risk-taking provides an intuitive justification for where these differences may come from. Also, this addresses the unlikely concern that our main cultural preferences, obtained from the GPS, might have been formed (even partly) by contemporaneous institutional and economic policies that could have also affected parents of immigrants in ways not reflected in their wealth, income, and education.Footnote ³⁷

We draw on the Ethnographic Atlas from Murdock (Reference Murdock1965) which allows us to approximate ancestral risk-taking culture.Footnote ³⁸ It further buttresses the interpretation of the GPS measure of risk-taking as capturing deeper cultural differences with an actual bearing on economic decisions, as opposed to solely reflecting some economic or institutional differences across countries that induce differences in survey respondents’ lottery certainty-equivalence.

The Ethnographic Atlas includes information gathered by ethnographers reflecting various cultural and socio-economic characteristics of pre-modern societies before industrialization and European contact.Footnote ³⁹ The Atlas provides us with information on what types of games a given society had in their cultures. It classifies societies’ games when any combination of the following three elements were present: i) chance, ii) physical skills, and iii) strategy.Footnote ⁴⁰ We proxy the ancestral culture of risk taking in the parents’ country of origin with the share of people whose ancestors played chance games.Footnote ⁴¹

In Panel A of Table 6, we present our findings on the effects of the long-term ancestral cultural measure of risk-taking described above on portfolio composition. Namely, we investigate to what extent children from cultures in which their ancestors’ games were more heavily based on chance, rather than strategy or physical activities, are more likely to take more idiosyncratic risks in their portfolio, keeping the institutional setting constant.

Table 6Mechanisms of Cultural Persistence

Table 6 Long description

The table presents regression results across six columns for three dependent variables: Stock Participation (columns 1 and 4), Stock Share (columns 2 and 5), and Portfolio Volatility Proxy (columns 3 and 6).

Panel A: Ancestral Risk-Taking Culture Captured by Ethnographic Chance Games. Shows positive coefficients for Ancestral risk taking (chance games) across columns 1 to 3, with values of 0.120, 0.0923, and 0.00865 respectively, all statistically significant. Number of observations is 38,702.

Panel B: Cultural Persistence. Examines the interaction between Ancestral risk taking and ancestral climatic instability. Coefficients for the interaction term are negative and significant: -0.542 (column 1), -0.445 (column 2), and -0.0531 (column 3). Columns 4 to 6 show results for Risk-taking culture and its interaction with climatic instability. Number of observations is 32,572.

Panel C: Parents from the Same Country (Vertical Transmission). Analyzes the interaction between risk culture and having parents from the same country. The interaction term for Risk-taking culture times same country parents is positive and significant: 0.143 (column 4), 0.100 (column 5), and 0.0133 (column 6). Number of observations is 38,702.

Panel D: Municipal Share of Locals from the Same Region as Parents (Horizontal Transmission). Shows interaction terms for risk culture and the share of locals from the same region. The interaction with Ancestral risk taking is positive and significant: 0.997 (column 1), 1.106 (column 2), and 0.128 (column 3). Number of observations is 38,521.

Bottom Rows: All models include Baseline controls, Parental education F E, Income rank for father and mother, Parental wealth rank, and Parental cohort F E, all marked as Yes.

Consistent with the baseline findings, in columns 1 to 3 we find that, conditional on equity market participation, children with an ancestral culture of risk taking are more likely to participate in the stock market, have a greater share of their financial wealth directly in stocks, and form portfolios with higher volatility.

This analysis reassures that cultural traits, defined in this analysis based on those passed down from centuries ago, influence individuals’ financial behavior today and the effect is similar to traits drawn from contemporary societies.

B. Cultural Persistence

The next analysis is based on the idea that cultural transmission is stronger for individuals whose parents come from countries in which maintaining cultural norms has been more beneficial across generations. Giuliano and Nunn (Reference Giuliano and Nunn2021) study cultural persistence and change and argue that the similarity of the environment across generations matters for cultural transmission. When the environment is more stable across generations, traits that have evolved up to the previous generation are more likely to be beneficial for the current one, and hence, the more beneficial it is to maintain existing customs. They empirically show that populations whose ancestors lived in environments with more cross-generational instability exhibit less cultural persistence.

Based on this, we expect the relations we find between culturally transmitted preferences and financial behavior to be weaker for those from more unstable places, as cultural persistence is weaker. We test this idea by interacting two measures of risk-taking culture—i) ancestral risk-taking based on the ethnographic data and ii) modern risk-taking from the GPS—with an exogenous measure of cross-generational climatic variability of the environment built by Giuliano and Nunn (Reference Giuliano and Nunn2021). Giuliano and Nunn (Reference Giuliano and Nunn2021) calculate cross-generational climatic instability of the ancestors of individuals living in each country today by measuring the average temperature variation over 70 generations (20 years to a generation) from years 500 to 1900 using sources of paleoclimatic data.Footnote ⁴²

Panel B of Table 6 depicts a general pattern for individuals whose ancestors come from more unstable countries. The effects of culturally transmitted preferences on financial decision-making are mitigated. In the first 3 columns, we use the ancestral culture of risk-taking extracted from the Ethnographic Atlas and, in columns 4–6, we use the measure of cultural risk-taking from the GPS. The estimates on the interaction terms are more precise in the case of ancestral risk-taking. This is expected as the measure of environmental instability is built for the ancestors of current generations.

Overall, the results are consistent with the idea that our variables of interest indeed capture the cultural transmission of economic preferences. For instance, if we look at the share of directly held stocks as the outcome of analysis in column 5, the net effect of risk-taking for those from highly unstable countries (90th percentile = 0.41) is low at 3.7 percentage points. Whereas, the net effect of risk-taking for those from highly stable countries (10th percentile of instability measure = 0.12) is quite substantial at 21.4 percentage points (compared to the mean share of 29%). This reiterates the importance of cultural persistence in the transmission of cultural traits.

C. The Role of Socialization

Bisin and Verdier (Reference Bisin and Verdier2000) argue that socialization is one of the main channels of cultural transmission. In this section, we explore additional channels of cultural transmission via socialization, namely, vertical and horizontal transmission.

Dohmen et al. (Reference Dohmen, Falk, Huffman and Sunde2012) find that the correlation between parents’ and children’s cultural attitudes is stronger when parents are from similar cultures (i.e., the vertical transmission of culture is stronger). To test this idea, we create a Same Country indicator—a binary variable equal to 1 if both parents originate from the same country—and interact it with our risk-taking measure. If cultural transmission is stronger when parents are from the same country, the relationship between risk preferences and investment behavior should be accentuated. The results in Panel C of Table 6 are consistent with this idea—the association between risk-taking and outcome variables are much stronger for those with both parents from the same country. For instance, column 4 indicates that the positive effect of risk-taking culture on the likelihood of participating in the stock market is 14.3 percentage points higher for those with both parents from the same country.

At the same time, parents’ cultural transmission might be weaker or stronger depending on the region they live (Dohmen et al. (Reference Dohmen, Falk, Huffman and Sunde2012)). For example, peers at school or neighbors in the same community may also play a role in cultural transmission (horizontal transmission). Therefore, to capture the horizontal transmission of culture, we also look at the shares of people from the same region as parents who live within the same municipality as the individuals and interact this with the risk-taking variable.Footnote ⁴³ If cultural transmission is stronger when individuals have more opportunities to interact with others from a similar cultural background, then, the relationship between risk preferences and investment behavior should be even more pronounced. The evidence in Panel D of Table 6 aligns with this idea. For example, if we increase the share of people in a municipality with a similar culture from 0 to 10 percent, the positive effect of risk-taking on the share of stocks more than doubles and goes from 3.1 to 8.5 percentage points (see column 5).

In sum, this section provides evidence on the role of socialization in cultural transmission and both vertical and horizontal transmissions of culture seem to matter.

A. Excluding Finnish Parents

As mentioned in Section II, since we need our sample of analysis to be adults when we observe them in the wealth register and we do not have financial outcomes after 2006, parents of the individuals in our sample must have migrated to Sweden far back in time for their children to have been born in Sweden early enough to be in our sample. Also, we require parents to be in the data set in 1999 (the earliest year of the wealth register), so that we have information on their wealth and financial behavior.

Historically, Finland was part of the Swedish Kingdom and has maintained a significant Swedish-speaking population, leading to substantial migration and exchange between the two countries. Consequently, because migration of non-Nordics to Sweden was not a widespread phenomenon before the 1990s, parents from Finland are over-represented in our sample. While it is clear from Supplementary Material Table A.1 and Figure 1 that variation in risk preferences across European countries is almost as large as variation in the whole sample, we formally investigate how our findings change if we limit our sample to second-generation immigrants whose parents come from outside Finland. Coefficient estimates in Supplementary Material Table A.6 are very similar to our baseline findings, indicating that the large set of immigrants from Finland are not driving our results.

B. Outcome Data from an Earlier Year

For our outcome variables so far, we have used data from the year 2006. As discussed before in Section II, there are two main reasons for that. One is that we want our sample of second-generation immigrants to be as old as possible when we observe them and make decisions about financial market behavior during their adult lives. The second is that, between 1999 and 2005, banks were not required to report small bank accounts to the Swedish Tax Agency unless the account accrued more than 100 SEK (about 11 USD) in interest during the year. For 2006, banks were required to report all bank accounts above 10,000 SEK and, as a result, we have more complete information on financial wealth when we study risky share and stock share in our analysis. Nevertheless, in Supplementary Material Table A.7, we present the findings when we instead use data from the year 2000. Note that, because we have more people with reported financial wealth of zero, our sample of analysis is smaller. However, we get very similar results to our baseline, indicating that our analysis is not sensitive to the sample year.