Portfolio Allocation, Risk Aversion, and Digital Literacy Among the European Elderly

1 Introduction

The last two decades was marked by breakthroughs in information communication technologies (ICTs), which provided new opportunities for economic development all over the world, but especially in developing and developed countries. Despite the improvements in connectivity and enhanced productivity generated by technological improvements, persons who cannot utilize digital tools and platforms may be marginalized, suffering from different forms of the digital divide. The digital divide is prominent between younger and older persons since younger individuals are more experienced with technology and have grown up with more exposure to technology. Nevertheless, the digital divide can impede older individuals from undergoing important activities and benefiting from opportunities, including those related to personal finances and investments.

Retirement represents a phase of life when people often (re)consider investing in financial assets as an opportunity to simply increase the income replacement rate. However, the lack of digital skills may represent a burden, especially for elder persons, when purchasing financial assets and absorbing financial information. Consequently, the portfolio choices of individuals can be influenced in various manners by digital literacy, including the use of online trading platforms for processing buy and sell orders, the access and analysis of financial data using the internet for making informed decisions, as well as using the newly emerged robo-advisors.

The literature shows that the digital divide continues to represent a persistent issue in the European Union (EU), even though several efforts have been made to reduce the digital gap. The first-order digital divide is concerned with the access of individuals to ICT. The uneven access to ICT could determine the marginalization of individuals and communities (Elena-Bucea et al., 2021). However, the focus in the literature has shifted from the simple access to ICT to the skills of individuals and communities in using ICT, namely, the second-order digital divide, and the tangible outcomes of using technology (i.e., third-level digital divide). Several categories of citizens are more prone to lack digital skills and more vulnerable to being affected by the digital divide in the EU: older persons living in sparsely populated areas, especially older women, or elder individuals with a low level of education, little internet use, and a relatively low level of income (Botrić & Božić, 2021; Vasilescu et al., 2020).

The portfolio allocation of elder individuals was previously analyzed by various studies. The following determinants were considered to influence the propensity to own stocks, either directly or indirectly through mutual funds or retirement accounts: psychological factors (Angelini & Cavapozzi, 2017; Bucciol & Zarri, 2017; Georgarakos & Pasini, 2011), cognitive abilities (Christelis et al., 2010), and the social network (Ostrovsky-Berman & Litwin, 2019). Other predictors that could influence the investment behavior of elder persons were considered, including the health status and the coverage of the healthcare system (Atella et al., 2012; Banks et al., 2020), as well as the existence of a life insurance policy (Cavapozzi et al., 2013).

Our research makes three distinct contributions to the literature. First, we examine the influence of digital skills at the individual level and country level on the portfolio riskiness of senior persons. To the best of our knowledge, the relationship between digital skills and portfolio choices of senior persons was not examined so far in the literature. Elderly persons represent a vulnerable category for ICT marginalization. Our empirical results suggest the access of older individuals to riskier financial assets might be improved by reducing the digital divide, which might subsequently create investment opportunities. Second, this article introduces novel variables in the empirical analysis at both individual and country levels, providing a starting point for policy formulation to reduce the digital divide. Finally, we combine data obtained from wave eight of Survey of Health, Ageing and Retirement in Europe (SHARE) with data from international databases, which enables us to identify the influence of digital literacy at the individual and country levels on portfolio riskiness.

The empirical analysis is conducted using multilevel models, more specifically two-level random intercept models with covariates, as well as a two-level random slope model. For the models, the likelihood ratio (LR) test establishes that the models are adequate compared to single-level models and models with fewer predictors. A similar result was reiterated by the Information Criteria (i.e., Akaike information criterion [AIC] and Bayesian information criterion [BIC]). The data from our empirical analysis are clustered at two different levels: individuals (Level 1) are nested in 25 EU countries (Level 2). The predictors utilized in the empirical analysis were based on theoretical considerations, including common socio-demographic control variables and variables indicating more recent forms of digital literacy.

The multilevel logistical models that we have applied enable us to depict the within- and between- country variation across the EU, as to finally identify heterogeneous interest in financial assets. Furthermore, we investigate the impact of age on portfolio riskiness across different countries by using random slopes for age in a multilevel logistic regression. The individual- and country-level data were retrieved from three datasets: (1) wave eight of the SHARE, which was collected in 2019/2020 (Börsch-Supan, 2022; Börsch-Supan et al., 2013); (2) Eurostat (Eurostat, 2023a,b); and (3) Digital Economy and Society Index (European Commission, 2022).

The rest of this article is organized as follows. Section 2 is dedicated to the analysis of the specialized literature. Section 3 introduces the databases and the methods that were used. Section 4 is dedicated to results and interpretations, while Section 5 is dedicated to conclusions and discussions.

2 Literature Review

3 Data and Methods

4 Results and Interpretations

The empirical analysis aims to explain older people’s portfolio riskiness upon a set of individual- and country-level predictor variables, and to also find out which of the two types of variables has a higher explanatory power. To accommodate both individual- and country-level variables, the multilevel analysis is first considered to be used here. However, we first run a series of tests to check the appropriateness of this methodology.

The variance partition coefficient (VPC) is used in multilevel modeling to partition the variation in a dataset within groups or between groups and to suggest whether a multilevel model is adequate, as opposed to a single-level model. The VPC for this regression is 0.122, which indicates that 12.2% of the variance in capital allocation can be attributed to differences between countries. The proportion of the total (residual) variance in the underlying propensity to have high interest in risky assets is attributable to differences between EU states. Aside from the VPC, we have applied LR tests, which established that a random intercept model fits better the data compared to a single-level model.

Table 2 reveals the empirical results we have obtained by applying a variance components model compared to two random intercept models. The first random intercept model includes only individual-level variables, while the second random intercept model includes both individual- and country-level variables. For the random intercept models, portfolio riskiness was regressed on the set of variables that we have explained, and the coefficients of both models are similar and the direction of relationships between the predictors and the response variable is the same.

In Model 3 of Table 2, the following country-level explanatory variables are added: national economic growth (variable real GDP/capita growth rate) and above the basic level of digital skills (for persons aged 55–74). By running LR tests, we have established that adding the extra parameters for random intercepts was not worthwhile and we present therein the results without random intercepts for those variables.

The relationship between portfolio riskiness and life satisfaction is found to be statistically significant and positive, which indicates that the propensity of individuals to invest in the stock market is higher if the persons are satisfied with their lives. There are various channels through which life satisfaction might affect financial risk tolerance and the willingness of individuals to invest in risky assets (stocks). First, this relationship may be attributable to the fact that individuals with high life satisfaction might have a good financial situation, which could be measured by income/pension or wealth. This result is consistent with previous studies from the specialized literature, which argued that the impact of economic circumstances was underestimated and economic factors significantly impact happiness (Headey et al., 2008; Powdthavee, 2010). Therefore, a lack of life satisfaction might correlate with a lack of financial resources, implying that individuals who are not satisfied might not have the disposable funds to invest in risky assets.

Second, persons who are not satisfied and happy with their lives may exhibit regret, including financial regret. Previous studies from the specialized literature have identified that the inclination for regret and happiness in life has been identified to influence financial risk tolerance. If one individual experiences regret with a particular type of investment, then the tendency to make a similar investment is reduced (Bailey & Kinerson, 2005).

Risk aversion had a negative impact on portfolio riskiness, which is an anticipated and logical result, indicating that individuals with a higher risk aversion are less likely to invest in risky portfolios, which include directly or indirectly owned stocks. General risk aversion was previously identified to negatively affect risky investment preferences (Dinç Aydemir & Aren, 2017).

More importantly, computer skills and use of the internet in the last 7 days were both positively correlated with portfolio riskiness and the relationships were statistically significant, indicating that individuals with better digital skills were more inclined to include stocks directly in their portfolios or indirectly through mutual funds or individual retirement accounts than individuals who lacked digital skills and did not use the internet. This may be attributable to the necessary digital skills required to directly invest in stocks, since currently most of the trading is conducted online, either via the website of the broker or by using a mobile app. The buy/sell orders are processed online, which may impede older persons from trading if they do not possess digital skills or lack an internet connection. As far as indirect stock ownership is concerned, computer skills are required to invest in mutual funds too.

The impact of computer skills was further analyzed at the country level. The variable denoting the above basic digital skills (for persons aged 55–74 old) has a positive impact on the portfolio riskiness and is also significant at the 0.01 threshold. Individuals who have above basic digital skills are more prone to own risky assets (stocks) compared to individuals who lack such skills. This result provides further evidence of the presence of a relationship between digital skills and portfolio allocation among older individuals. Opening a brokerage account, using mobile banking, and performing stock market orders require at least basic digital skills; therefore, the variable utilized in this analysis is based on theoretical foundations.

The standard socio-demographic indicators are included in the list of control variables: age, gender, and education, following the existing practices from the literature. The relationship between portfolio riskiness and education is positive and statistically significant. Several studies support a positive correlation between financial risk tolerance and the level of education (Faff et al., 2009; Grable, 1997; Grable & Joo, 2004). Individuals with a higher level of education may be more prone to taking financial risks and investing in riskier assets. One channel that may explain this relationship might be the level of earnings, which is positively correlated with education (Tamborini et al., 2015). Variables indicating the age and gender of persons were not statistically significant.

The national economic growth (variable real GDP/capita growth rate) is found to have a non-significant relationship, which is also the case for the following country-level variables: ordering online and severely materially and socially deprived. Ordering online was considered in the analysis since it reflects a more recent form of the digital divide that concerns the results of using ICT. The country-level variable of severely materially and socially deprived was included in the analysis from theoretical considerations. As expected, a negative relationship was identified between material and social deprivation and portfolio riskiness.

We have increased the number of covariates for robustness testing using variables based on theoretical considerations and variables used in previous studies in the literature. Therefore, we included in the models’ specifications variables related to income (i.e., payments from old age, early retirement, survivor pensions, total household income) and cognitive abilities (i.e., self-rated reading skills; self-rated writing skills, score of verbal fluency test). The digital variables are still statistically significant after the inclusion of the income and cognitive variables. The results of the robustness testing are included in Table 3. As expected, individuals with larger payments from old age, early retirement, and survivor pensions are more likely to invest in riskier assets at the 0.01 threshold, which is also the case for the total household income variable. The relationship is also positive and statistically significant between portfolio riskiness and the following cognitive-related variables: self-rated reading skills and scores on verbal fluency tests. The relationship between cognitive abilities and stockholding was investigated in the literature. Christelis et al. (2010) provided evidence for a strong association between cognitive abilities and stockholding using the SHARE database. The mechanism through which reading skills and verbal fluency influence portfolio riskiness may be related to the absorption of financial information or the spread of information in the social network. The size of the social network was identified to be a relevant predictor for stockholding among the European elderly (Ostrovsky-Berman & Litwin, 2019).

To further examine the differences between countries regarding the interest of individuals in financial assets, we have analyzed the residuals of the variance components model. Figure 1 shows the estimated residuals for the EU states based on the variance components model. For a substantial number of states, the 95% confidence interval does not overlap zero, indicating that interest in financial assets is above average (above the zero line) or below average (below the zero line) at the 5% level for these states.

Figure 1

Residual estimates based on two-level variance components ordinal logistic model. [Note: The country effects are in rank order together with a 95% confidence interval. The horizontal red line (at 0) represents the average country in the data].

The list of the residual rank for each country from the EU is illustrated in Table 4, indicating how countries are situated compared to the baseline level regarding the risk aversion for financial assets. The most risk-averse countries are Latvia, Romania, and Bulgaria, indicating that the interest for financial assets is considerably below average. Those countries exhibit financial risk aversion, and individuals from those countries are more prone to invest in safer assets (saving accounts) rather than to invest in stocks directly or indirectly. Nordic countries such as Sweden, Denmark, and Finland are the countries that are the least financially risk-averse, and individuals are willing to invest in financial assets such as stocks. Other studies have identified significant stock ownership differences between stock market participation across EU countries. It has been argued that stock market participation is relatively high in Nordic countries since financial markets and institutions are more developed (Angelini & Cavapozzi, 2017). Countries within the EU exhibit a stark contrast in their capital allocation. Microdata analysis showed that the stock market participation of individuals with a low wealth in Sweden and Denmark is twice those of wealthy individuals from Austria, Spain, and Italy (Georgarakos & Pasini, 2011).

We further examined the age effect on portfolio riskiness across the EU countries. To establish that, we have fitted a random slope two-level ordered logistic regression where the portfolio riskiness was the response variable, as in the previous regressions, where we have introduced a random slope for age. The sole purpose of the model was to analyze whether the impact of age is different across the sample countries; therefore, we have reduced the number of predictors to achieve model convergence and avoid overparametrization. The working variable was included to identify the influence of employment on the investment behavior of elder persons. The results of the model with a random slope for age are illustrated in Table 5.

We have maintained a variable related to digital skills since we were interested in determining the influence of country-level digital literacy on investment riskiness. Maximum likelihood was used as an estimation method for the coefficients and the cut points.

The results of the LR test indicate that the addition of a random slope for age is justified at the 0.01 threshold and the age effect varies across countries, and this specification captures the relationship better than a model without the random slope for age.

The model indicates a negative and statistically significant relationship between age and portfolio riskiness, indicating that older individuals are less prone to invest in risky assets compared to younger persons. This is consistent with previous findings in the literature, and age may influence the propensity to own stocks through multiple channels. Elder persons normally encounter higher health risks compared to younger individuals, which may determine them to be more risk-averse. The current health status and the expected health status negatively influence the proportion of risky assets (Edwards, 2008). The increased health risks of elder persons may cause them to reduce their allocation of capital toward risky assets. Older persons are likely to face higher future medical expenditures compared to younger persons. Poor health and risky asset ownership were found to be negatively related (Rosen & Wu, 2004). Another plausible explanation for the risk aversion of older persons is that their investment horizon is shorter compared to younger investors, who have a longer timespan to recover losses.

The relationship between life satisfaction and the response variable is positive and statistically significant in the random slope model for age and in the previous models. The positive correlation and the statistical significance are maintained also for the variable denoting the above basic level of digital skills. The variable of material and social deprivation maintains a negative correlation to portfolio riskiness, as in the previous models.

The influence of age on the investment behavior of older individuals between the countries from the sample was examined. We have introduced this analysis based on theoretical considerations considering expected heterogeneity across the EU states, namely, that the varying economic conditions, cultural norms, and regulatory frameworks can influence retirement investment behavior. From a practical perspective, understanding the manner age influences the investment behavior during the retirement phase across the EU states is necessary for establishing adequate policies by policymakers. The results of the models can serve as information for tailoring retirement planning policies or financial education programs to specific age groups and regions, to promote healthy financial behavior during retirement.

The age slopes were plotted against the country intercepts in Figure 2, based on the ordered logistic regression with a random slope for age. The coefficients of age in the random slope model indicate that age impacts differently the investment behavior across the EU countries. The countries in the top-left quadrant are countries that had lower-than-average portfolio riskiness but better-than-average participation as age increases. For instance, the most prominent differences in investment behavior when age is considered are present in France, indicating that if age increases, individuals are more willing to own risky assets, namely, stocks. Other countries such as Malta, Hungary, Luxembourg, Cyprus, Greece, Germany, and Finland are all in the same group. For the aforementioned countries, age has a positive impact on the propensity to own risky assets.

Figure 2

Age slopes versus country intercepts.

In the bottom-left quadrant are present countries with below-average mean portfolio riskiness with below-average slopes. This category includes Romania, the Netherlands, and Latvia. The interest for risky assets in these countries was lower than the mean and the inclination to own shares decreased as individuals aged.

For all the other countries in the right quadrants, the portfolio riskiness was higher than the average, with age having a positive impact on portfolio riskiness for the countries situated in the top-right quadrant, and a negative impact for the ones in the bottom-right quadrant.

5 Discussion and Conclusions

This study examined the propensity of older persons to invest in risky assets considering their digital skills, using individual- and country-level data from SHARE, Eurostat, and the Digital Economy and Society Index.

Our contribution to the literature is threefold. First, it extends the literature on portfolio allocation among elder individuals by investigating the influence of digital skills at individual and country levels on portfolio riskiness. To our knowledge, the impact of digital skills and the propensity to own risky assets have not been considered yet by research articles. The empirical findings suggest that reducing the digital gap might improve the access of older individuals to riskier financial assets, which may, in turn, create investment opportunities. Second, the introduction of novel variables into the empirical analysis at the individual and country levels might serve as a starting point in policy formulation for interventions to reduce the digital gap. Third, we have combined data retrieved from wave eight of SHARE with data from international databases, which permitted us to portray the impact of digital skills at the individual level, as well as the country level, on portfolio preferences.

Our study is particularly relevant since it explores the relationship between the digital skills of senior citizens and portfolio riskiness in the EU, using a methodological framework that permits us to analyze the differences across countries, which is necessary for policy interventions. Furthermore, the global context in which the population ages and the life expectancy increases implies that senior citizens comprise a growing percentage of the total population and the potential users of ICT. Consequently, persons spend a longer period of their lives as senior citizens, which indicates the necessity of acquiring digital skills and updating them, to bridge digital gaps and ensure well-being, including financial well-being.

The appropriateness of the methodological framework was considered after running LR tests and examining the VPC, which indicates clustering effects across the EU countries. The tests were in favor of multilevel models and suggested the specifications with a random intercept for the country and a random slope for age. The presence of covariates in our model was indicated by the LR tests compared to a null multilevel model.

Our results suggest the presence of positive and statistically significant relationships between digital skills and portfolio riskiness, indicating that individuals with digital skills are more prone to invest in risky assets, namely, stocks, either directly or indirectly. Both individual-level variables that considered digital skills were statistically significant: computer skills and the use of the internet in the past 7 days. A similar result was obtained for the country-level variable denoting the above digital skills, which had a positive impact on portfolio riskiness. The results are consistent with our expectations since digital skills are required to absorb financial information and participate in the stock market.

Moreover, the interest in risky financial assets in the EU was heterogeneous. The lowest interest for risky financial assets was observed in Latvia, Romania, and Bulgaria, while Nordic countries such as Finland, Denmark, and Sweden were the least financially risk-averse and more likely to participate in the stock market. Previous studies from the literature have identified high participation in the stock market in Nordic countries (Angelini & Cavapozzi, 2017; Georgarakos & Pasini, 2011).

Another empirical finding suggests that age influences differently the investment behavior of individuals across the EU countries, based on the results of the regression with a random slope for age. Even though age has a heterogeneous impact on portfolio riskiness across the EU sample countries, the overall coefficient for age is negative, namely, that as individuals age, their risk appetite is decreased, and they are less willing to invest in the stock market. This behavior can reinforce the assumptions of the life cycle hypothesis of Modigliani and the permanent income hypothesis of Friedman. The life cycle hypothesis assumes that individuals make consumption and saving decisions based on their expected lifetime income, namely, they consume less when they earn more (i.e., during active years) to have more savings to maintain their standard of living during retirement when the income is reduced (Modigliani & Brumberg, 1954). A smoothing of the consumption behavior throughout life is also suggested by the permanent income hypothesis, which implies that during the decumulation phase, individuals are prone to enhanced spending (Friedman, 1957).

The results from this article may serve as policy recommendations. The reduction of the digital divide should represent a policy debate among the EU member states for subsequent interventions to reduce the gap, especially among vulnerable categories such as elder persons. The lack of addressing this issue might exacerbate existing divides and possibly result in further segregation of individuals in society based on their digital literacy. The digitalization of financial services will hinder vulnerable categories from investing in financial assets (i.e., elder persons). Assuring assistance regarding the use and benefits of digital services to deprived people must be considered since such persons may have integration difficulties.

Furthermore, the globalization of capital markets can potentially explain the lack of statistical significance of the real GDP/capita variable in our models, suggesting that the evolution of external markets has a greater influence on the investment behavior of older individuals, as opposed to the origin’s country’s economic conditions.

The main limitation identified in this study concerns the variables used in the empirical analysis, namely, the response variable is a categorical variable proxy for portfolio riskiness. Even though it is based on theoretical considerations from prior studies in the literature, it led to an exclusion of individuals from the sample that could be allocated to two risk groups at the same time. A more precise measurement of portfolio riskiness may be considered, which makes a distinction between directly holding stocks or indirectly owning stocks. Since our contribution is novel regarding the relationship between digital skills and portfolio riskiness, upon our knowledge, a drawback is related to the lack of the possibility of confronting our results with prior studies from the literature.

Digitalization may serve as a barrier for elder persons in investing in stocks since they may not have a mobile bank app/digital brokerage account or are not willing to trust such apps. Therefore, elder persons may lose significant investment opportunities and may not allocate their wealth adequately, especially in periods with high inflationary pressure, in which their wealth will diminish. Moreover, the lack of digital skills will not only refrain elder persons from participating in the stock market but will affect other aspects of their lives as well since digitalization will be implemented in other sectors and governance.