A within and across country analysis on the relationship between cognitive
ability and individual retirement account ownership when approaching
retirement age
Author: F.C. van Nieuwpoort1 Supervisor: C. Laureti Master’s Thesis Economics
University of Groningen June 2018
Abstract
This paper used a large and representative dataset of 18 European countries to investigate a possible link between cognitive ability and the ownership of an individual retirements account among individuals approaching the retirement age. Five measures of cognitive function were used to derive a compound binary score of good cognitive ability, which, along with a series of control variables, was used to find the possible effect of cognitive ability on the probability of holding an individual retirement account. The results from the present analyses do not suggest that cognitive ability has a significant impact on the probability of individual retirement account ownership in the countries under consideration.
Keywords: Cognitive ability, Personal Finance, Investment Decisions JEL Codes: D91, D14, G11
2 1. Introduction
In a world with a continuously evolving financial system and declining state pension provisions, household financial planning is a matter of concern to more individuals around the world than ever before. Continually rising life expectancy and a relatively aging population, especially in developed countries, have increased the burden on social welfare systems in many countries. Solutions to this often include increasing retirement age or reducing social security generosity (OECD, 2002; Coppola and Wilke, 2010). Demographic trends will likely continue affecting and burdening social welfare programs and reliance on private financial planning for retirement is likely to increase in the future (Banks and Oldfield, 2007). A number of European nations have sought to stimulate the development of a ‘third pillar’ of pension funding, by incentivizing private retirement savings arrangements (OECD, 2002; Galasso and Profeta, 2007). Banks and Oldfield (2007) for example discuss the situation in Great Britain regarding public and private pensions: over the past decades, a number of reforms have been implemented to cope with demographic and market changes. Due to political pressures, these changes tend to distribute costs over longer time horizons to avoid creating immediate losers, at the expense of the individuals in the system in the future. The well-documented consequence of this string of reforms however, is a more convoluted and less transparent system, and an increased reliance on private pension savings for the United Kingdom (Banks and Blundell, 2005). Other European countries also express a private savings rate higher than that in the United States, partly associated with institutional factors, specifically social security and pension schemes as well as expectations regarding future economic situation. (Leetmaa et al., 2009).
3 Approaches to improving retirement income security related to the first two factors have been amply discussed in both economic and financial publications as well as in politics, as retirement reforms are a recurring topic in many political systems. Frequent changes, transitional periods, and a variety of other social security regulations tend to contribute to a more complicated and unstable environment, further obscuring the bottom-line for many individuals. Some research suggests that using the third and fourth of the factors mentioned above might offer some outcomes (Banks and Oldfield, 2007). A recent paper by Dolls et al. (2016) investigated a German policy of providing simple information on pension payments to citizens and found that this increased retirement savings, among other outcomes.
Lack of knowledge regarding pension savings and payments is widespread in Europe, and cognitive ability has been linked to financial decision making in several ways. A positive relation between cognitive ability and stock ownership is well documented (e.g. Christelis et al., 2010), and cognitive ability has been associated with retirement savings by Benartzi and Thaler (2007).
LeBlanc et al. (2011) suggest that individual retirement account ownership varies based on both policy and personal factors like cognitive ability and education. This paper uses survey data from individuals approaching retirement age to see if there is a relation between cognitive ability and holding an individual retirement account. Pension systems and regulations vary significantly between countries, both in generosity and in complexity, so this paper also discusses the effect of country policy on the relation between cognitive ability and individual retirement account ownership. To gain insight into the relation between cognitive ability and the ownership of an individual retirement account in the demographic of people approaching retirement age, the research question of this paper is as follows:
Is cognitive ability related to individual retirement account ownership for those individuals approaching retirement age and does this relationship vary by country?
4 to gain insight into the difference in dynamics between each country. The results of this analysis will be discussed and placed into context to assess if possible recommendations for action arise.
5 2. Literature review
Personal finance and planning for the future are part of the lives of more people than ever before. Developed financial and economic markets offer people a plethora of ways and means to manage their wealth and maximize purchasing power. How and when to save, invest, consume and spend money in order to maximize utility over a lifetime is however a task about as difficult as it is important, and the body of work studying financial planning has been growing for a number of decades (Bodie, 2002). In addition to the necessary allocation of income to consumption or saving, households are increasingly required to personally build up a financial buffer to supplement state-funded pensions if they would like to retain their standard of living throughout retirement (Wallmeier and Zainhofer, 2006).
In general, pension systems in European countries are based on three institutions: public schemes, or social security programs regulated and supplied by the government, occupational schemes, pension income arranged through employers, and individual pension savings (Hershey et al., 2009). The first pillar, public schemes, is the primary domain of government policy here and has contributed the largest share of retirement income in past decades in most countries (Commission of the European Communities, 2000). Due to demographic aging, the number of retirees in Europe is growing both absolutely and relatively compared to the number of employed individuals. Time spent in retirement has also been on the rise, as life expectancy has risen faster than retirement age (European Commission Report, 2015). This ever-increasing burden on social security necessitates more or less constant reform of regulations, such as increasing the retirement age or decreasing monthly benefits, to maintain a stable and equitable system of pensions.
6 An increased reliance on income from the third pillar, private savings, has been documented for Europe in recent years. LeBlanc (2011) states that due to expectations of future retirement income and experience with occupational schemes, in addition to regulatory differences between countries and personal characteristics of households, participation in this third pillar has increased in recent years. While private savings is not within the direct purview of government policy, regulations can be implemented to stimulate private retirement savings.
2.1 Individual retirement accounts
One way for individuals to supplement the pension income from the state and their employers is to invest in individual retirement accounts. These are institutions that allow individuals to privately invest part of their income now, to receive this back when they are retired. Stimulating individuals to invest in individual retirement accounts by letting them subtract (part of) these investments from income for tax purposes is an attractive option for governments to help individuals increase retirement income, without increasing the cost of pension schemes directly.
The decision to invest in an individual retirement account for retirement income in addition to government and employer funded retirement income depends on the generosity of these pensions and the incentives to invest, as well as on individual characteristics. A less generous state pension would logically motivate individuals to seek additional retirement income, as would a large tax incentive for individual retirement account investment. Individual retirement account ownership varies greatly between countries and demographics, but the extent to which differences in pension schemes and individual retirement account tax benefits interact with individual characteristics is less well-documented. This research will delve into some facets of this relation with the aim to improve understanding and find clues for effective and fair policy making.
7 factors that affect this decision. The following section will discuss the impact of cognitive function on financial decision making, and relate it to individual retirement account ownership.
2.2 The impact of cognitive function on financial decision making
One possible reason that policies affect individuals in different ways comes from differences in cognition, both in term of preferences, and of ability to process and use information to make well-reasoned decisions. The link between cognitive ability and financial decision making has been studied to some extent, resulting in a number of interesting findings.
Agarwal and Mazumder (2010) explored the link between cognitive ability and financial decision making using data from the United States, and found that individuals with higher overall test scores and specifically mathematics test scores were less likely to make financial mistakes. Mcardle and Smith (2009) use different measures of cognitive function, e.g. numeracy and word recall, combined with economic outcome measures related to wealth, accumulation, and portfolio composition to find a similar relation. Their findings corroborate the importance of numeracy and memory for financial decision making. Burks et al. (2009) also relate cognitive ability to financial decision making in a 2009 publication. They use three measures of cognitive skill, a non-verbal IQ test, a numeracy test, and a planning test, and study its relation with economic preferences and financial decisions. Cognitive ability is strongly correlated with economic preferences: high cognitive ability is correlated with better patience both in the long - and short-run, and negatively correlated with a willingness to take calculated risks.
8 2.3 Hypotheses
As mentioned before, state-provided pension schemes are declining in many countries and financial markets are becoming more complicated and differentiated. With people relying more on their own preferences and decisions for their long-term financial wellbeing, it is important that individuals understand the options they have and the consequences of saving and investing money in various ways, thus making cognitive ability and financial literacy more important to the public today (Banks and Oldfield, 2007). Cognitive ability has been linked to anomalous preferences, e.g. being positively related to risk seeking in small takes games, and negatively with short-run discounting (e.g. Benjamin et al., 2006). Burks et al. (2009) agree that higher cognitive skills are related to higher willingness to take some financial risk, and find that high cognitive skill predicts higher patience in both the short- and long run.
Cognitive ability can thus affect financial decision making in a variety of ways, and could influence the decision to invest in an individual retirement account through the following channels: higher cognitive ability allows individuals to better understand and compare their financial options, increases likelihood of financial market participation, and contributes to more patience in both the long- and short-run. Because saving for retirement is particularly important for individuals approaching retirement age, it is useful to see how this dynamic affects the decision to invest in an individual retirement account in those individuals aged 50-59. This paper will investigate this by testing the following hypothesis:
9 As mentioned before, the decision to invest in an individual retirement account is also dependent on the culture and regulatory environment in a country. Less generous retirement incomes from the first pillar, government pensions, would for example logically drive individuals to engage in more voluntary retirement savings. The degree to which individuals understand the pension system and regulations in place, also affects their decision making: clearer and more certain understanding allows for more robust decisions. To see how between -country differences affect individual retirement account ownership, the following hypothesis is tested:
H2: The relation between cognitive ability and individual retirement ownership probability in
10 3. Data and Methodology
3.1 Data
The Survey of Health, Ageing and Retirement in Europe (SHARE) was selected as the source of data for the research described in this paper. SHARE is a multidisciplinary and cross-national panel dataset that, to date, has selected micro data on health, socio-economic status and social networks of more than 120,000 individuals aged 50 and over from numerous European countries and Israel. Given that all questions proposed by SHARE are standardized across the countries the dataset allows for consistent international comparisons. To date, SHARE has collected six panel waves (2004, 2006, 2010, 2013 and 2015) of current living circumstances and individual’s life histories including all important areas of respondents’ lives (SHARELIFE, 2008). Since the dataset includes respondents ranging from Scandinavia through Central Europe to the Mediterranean it is representative of the European region. In addition to the numerous European countries included in the dataset, Israel has joined the SHARE effort, being the first country in the Middle East to initiate a systematic study on its ageing population. Each wave of SHARE includes participants from former waves as well as a number of new participants in order to maintain a rich sample and improve the data in later waves, resulting in a unique longitudinal dataset. The survey consists of 20 modules and is collected using face-to-face computer-aided interviews combined with self-completion papers and pen questionnaires2.
Given the unique dataset that SHARE offers, it has been used in many fields of research including, among others, (behavioral) economics and finance, political science, and health care (see e.g. Alessie et al., 2013; Christelis et al., 2012; Bressan et al., 2014). See Börsch-Supan et al. (2013) for methodological details.
3.2 Dataset
SHARE has thus far released six panel waves. The research described in this paper is mainly based on the sixth wave of SHARE, which took place in 2015 in eighteen countries (Austria, Belgium, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Italy, Israel, Luxembourg, Poland, Portugal, Slovenia, Spain, Sweden, and Switzerland). Each wave of SHARE is divided into several modules, which are provided in separate datasets. Each
11 participant is matched to their responses throughout the survey using a unique identification number. This identification number is used in merging the different modules into a single dataset, yielding a total of 68,231 individuals. One caveat here is the fact that, even though most questions posed by SHARE refer to the individual, some questions are only posed to the financial respondent of a household. Questions relating to e.g. an individual’s cognitive ability, job status, educational attainment and health status are unique to the individual, while questions on the financial portfolio composition refer to the household (e.g. income, savings and other financial asset holdings). The questions related to the ownership of an individual retirement account and the characteristics of such an account are only posed to the financial respondent of the household, who answers the questions for both themselves and their partner. Answering the question referring to the ownership of an individual retirement account, the financial respondent should choose from the following options: 1) respondent only, 2) husband/wife/partner only and 3) both. Following the approach used by Christelis et al. (2010), answers to questions that are posed to the financial respondent only are aggregated over the two partners in a couple using the unique household identification number. Please refer to Section 3.2.1 for a detailed description on the aggregation of the individual retirement account holding variable, and to Appendix A for an overview of the variables used.
Consistent with the approach of e.g. Banks and Oldfield (2007) and Banks and Blundell (2005), the dataset has been narrowed down to individuals aged 50 up to and including 59, given that this is an age range when relevant decisions about retirement savings are finalized. Additionally, individuals who have indicated that they are retired have been excluded from the dataset as they are no longer in their preretirement years. Some questions in the dataset permitted answers such as ‘Don’t know’ or ‘Refusal’ in case respondents would rather not share information. These answers are treated as missing values, as are other responses that are invalid or genuinely missing. Finally, all individuals with missing values for the dependent or independent variables are dropped from the dataset, resulting in a dataset of 4,222 individuals.
3.3 Variable selection and construction
12 3.3.1 Construction of dependent variables
The main goal of this research is to examine a possible connection between an individual’s cognitive ability and the ownership of an individual retirement account. As such, the dependent variable is a binary variable, indicating one in the case that an individual holds an individual retirement account and zero otherwise. As previously discussed, the information on the ownership of an individual retirement account is aggregated over the household using the household identifier. The financial respondent of the household is asked who in the household owns an individual retirement account and can choose from the following options: 1) respondent only, 2) husband/wife/partner only and 3) both. The variable displaying this information is as020_. As a first step in aggregating the information, a variable (hh_has_iras) is created displaying the maximum value of as020_ within the household using the household identifier (hhid6). As a second step, the combination of hh_has_iras and whether an individual is a financial respondent (fin_resp) is used to assess whether an individual holds an individual retirement account (“IRA”) or not. Please refer to Table 1 for details on the derivation of the individual retirement account holding variable.
Table 1
Derivation of the individual retirement account holding variable
3.3.2 Construction of independent variables
For testing the hypothesis on the relationship between an individual’s cognitive ability and the ownership of an individual retirement account, an independent variable has been created measuring an individual’s cognitive ability. Measuring cognitive ability is a science in itself and as such, no standard or certain way exists to do so. Most definitions of cognitive ability include a
Maximum value who has iras?
(hh_has_iras)
Financial respondent?
(fin_resp)
Holds an IRA?
1 - Respondent only Yes Yes
2 - Husband/wife/partner only Yes No
3 - Both Yes Yes
1 - Respondent only No No
2 - Husband/wife/partner only No Yes
13 variation on “the ability of the brain to process, retrieve, and store information” (Taber's Cyclopedic Medical Dictionary, 2009). The ability to learn, understand and use relevant knowledge, and solve problems quickly are also elements associated with cognitive ability (e.g. Rindermann et al., 2010; Nelson and Phelps, 1966). Within the concept of cognitive ability, a distinction between fluid intelligence and crystallized intelligence is recognized, where fluid intelligence relates to abstract reasoning and crystallized intelligence relates more to depth and breadth of knowledge (Cattell and Horn, 1967). Fluid intelligence testing commonly includes solving numerical or logical problems, whereas crystallized intelligence can be assessed by for instance questions on general knowledge or vocabulary (Grabner and Stern, 2011). Both of these domains are relevant to financial decision making, as this consists of constantly gathering and storing information (crystallized intelligence) and comparing these options to optimally allocate resources (fluid intelligence).
14 used in the analysis. In order to verify this assumption a separate analysis has been performed on all individuals with numeracy information from the sixth waves, leading to the same results.
Following Doblhammer-Reiter et al. (2011) a cognitive score is computed by assigning scores to different ranges of points for each question and adding these scores up. Memory (recall) is tested simply by reading a list of ten short words to the respondent and asking them to repeat as many of the words back to the interviewer as possible right after. For either no or a single word, no points were assigned. For two correct words, one point was assigned, three words meant two points, four words meant three points and five or more words were assigned a score of four. For delayed recall, testing memory on a longer term, respondents were asked to repeat the same ten words again, after the interview. Because this is generally more difficult, one fewer correct answer was needed for each number of points, so that four correct words got four points, three words got three points, and so on.
According to Christelis et al. (2010) numeracy directly relates to financial decision making and understanding of finances. Additionally, Banks and Oldfield (2007) find a direct correlation between numeracy and retirement savings. Therefore, numeracy is considered especially relevant in the context of this research. SHARE uses a series of problems to be solved by respondents to test their numeracy. These problems were 1) to find ten percent of a number, 2) find half of a number, 3) use a fraction of a number to compute the whole number and 4) find the value of an amount in two years. Solving none of the problems correctly resulted in a score of one, with a point added for each correct solution, so that solving each problem correctly resulted in five points. For consistency, the scores were brought down by one point as the other cognitive ability related variables are also on a zero to four scale.
15 Lastly, orientation in time was also included in the dataset. Respondents were asked about their orientation in time by simply asking questions about the current year, month, day of the month (as in e.g. the 23rd) and the day of the week (as in e.g. Monday). As with numeracy, each correct answer is assigned one point, for a minimum of zero and a maximum of four.
Now each of these five measures have each been assigned zero to four points, they can be compiled into a single cognitive ability score by adding the points for each part. This results in an indicator with a minimum score of zero points and a maximum score of twenty points. Figure 1 to Figure 6 show the distributions of the components underlying the cognitive ability score as well as the computed cognitive ability score itself. To be able to draw more general conclusions from the analyses, the 20-point scale is taken as a binary variable of cognitive ability: individuals who have scored above the median of 18 are considered to have ‘good cognitive ability’ whereas individuals who scored below the median are assigned to the group with ‘poor cognitive ability’. Resulting in 2,645 (62.65%) respondents with ‘good cognitive ability’ and 1,577 (37.35%) with ‘bad cognitive ability’. To determine whether the median was a reasonable cutoff point for good cognitive ability, cutoff points between 16 and 20 were also tested, but results were mostly stable and the impact of the dummy did not change significantly, suggesting that the median is indeed an acceptable cutoff.
Figure 1
The distribution of the recall scores
Figure 2
The distribution of the delayed recall scores
0 .2 .4 .6 .8 1 D en si ty 0 2 4 6 8 10 Recall score 0 .2 .4 .6 .8 D en si ty 0 2 4 6 8 10
16 Figure 3
The distribution of the numeracy scores
Figure 4
The distribution of the fluency scores
Figure 5
The distribution of the orientation scores
Figure 6
The distribution of the cognitive ability scores
3.3.3 Construction of control variables
In order to maintain the validity of the model and results, a number of variables are to be controlled for. These include variables that can be expected or have shown to have an influence
0 1 2 3 4 D en si ty 1 2 3 4 5 Numeracy score 0 .02 .04 .06 .08 D en si ty 0 20 40 60 80 100 Fluency score 0 2 4 6 8 D en si ty 0 1 2 3 4 Orientation score 0 .1 .2 .3 .4 .5 D en si ty 0 5 10 15 20
17 on the outcomes of interest, but are not directly related to the study at hand. First, a number of basic, socio-economic factors are considered. Age, gender, employment status, education, family situation, income and statistical dispersion in an individual’s country of residence are included.
SHARE focuses in part on the dynamics of ageing and consists of data gathered from individuals aged 50 and over. Age is accepted to have an influence on life-cycle financial behavior (e.g. Berg, 1996) as the time horizon of planning is more uncertain for the elderly as mortality risk rises, which tends to shift the distribution of consumption closer to the present day. Additionally, consumption patterns change with age, for example through children moving out or increasing physical constraints to consumption (Skinner, 2007; Börsch-Supan and Stahl, 1991). Literature suggests that age affects financial decision making in other ways as well: in general, older individuals tend to be more risk averse than younger individuals, even when controlling for cohort effects (e.g. Bonsang and Dohmen, 2015; Bodie et al., 1992).
Gender is a standard control variable in most statistical analysis of individual-level data. In addition to being of general interest, gender has been shown to interact with financial decision making in distinct ways. Men are overall more risk-seeking than women, which is also expressed in financial decision making, but not necessarily in outcomes (e.g. Powell and Ansic, 1997; Almenberg and Dreber, 2015). Barber and Odean (2001) cite overconfidence in men as a cause of the difference in risk attitude between genders and found that gender does interact with risk attitude. In the dataset used for this research, gender is included as a straightforward binary variable where zero indicates female and one indicates male.
18 employment status on the probability of owning an individual retirement account a binary variable is included with a value of one if the individual is employed or self-employed and zero otherwise.
For education, the international standard classification of education (ISCED 1997) was used by SHARE. This standard assigns a value from zero to six, whereby zero indicates ‘pre-primary education’ and six indicates ‘second stage or tertiary education’, e.g. a PhD research. Education has an obvious relation to finances through the positive relation between education and income, but education has also been found to independently affect savings rate (Solmon, 1975). Education was included as a binary variable, where having attained an ISCED 4 or higher education indicates “highly educated”.
Also an individual’s marital status has been controlled for. Research finds that being in a couple affects financial decision making both in terms of spending and saving, as individuals in couples have for instance been shown to accumulate more wealth in two cohorts of 51-56 year olds (Drolet and Schwarz, 2010; Lusardi and Mitchell, 2007). Additionally, Wilmoth and Koso (2002) find that marital history affects wealth outcomes in preretirement adults; individuals who have continuously married have higher wealth outcomes than divorcees, but “remarriage offsets the negative effect of a marital dissolution”. As such, two binary variables related two marital status have been included: one for being in a couple and one for being divorced.
In addition to the two marital status related variables, having children is taken into account as a standard control variable. Having children is associated with different consumption and savings outcomes. Skinner (2007) suggests that couples with children find it easier to save sufficiently for retirement and consume at a comfortable level once children had left the home. Lusardi and Mitchell (2007) find that baby boomer couples with children do indeed accumulate more wealth.
19 Health status is highly relevant to individual financial outcomes. Bad health is associated with health expenditure risk and lower wages, earnings, and retirement savings (Hurd, 1990), whereas good health is correlated with higher wealth (Lusardi, 2003). Considering the dataset used, including individuals aged 50 up to and including 59, a binary variable for self-perceived health has been included. The variable indicates one if the individual assesses its health as “good”, “very good” or “excellent” and zero otherwise.
As discussed before, the dataset used in this research consists of individuals from eighteen countries. Recent research (i.e. Christelis et al., 2010) shows that significant within and across country differences exist in the holding probabilities of an individual retirement account. As such, country dummies have been included to control for this effect. To avoid multi-collinearity of the data one country dummy has been omitted from the analyses, being Croatia. Three more countries, Poland, Israel, and Greece, did not have sufficient respondents in the dataset to be used in marginal effects analysis, and were thus dropped for this part of the research. Following the across country analysis with the country dummies included, separate analyses are employed for each of the countries. Please refer to Section 3.2.4 for more details on the analyses employed.
3.2.4 Methodology
The main goal of this research is to examine a possible relationship between an individual’s cognitive ability and the ownership of an individual retirement account. As previously discussed, given that previous research suggests significant within and across country differences in the holding probabilities of an individual retirement account (e.g. Christelis et al., 2010; Le Blanc, 2011), the research will be split up to across and within country analyses. For the research employed in this paper, the holding probability of an individual retirement account is examined using a logistic regression model.
In order to test the first hypothesis, the relation between investing in an individual retirement account and cognitive ability is analyzed. Because the dependent variable, the holding of an individual retirement account, is binary and there are several binary and continuous independent variables, a logistic regression model is used. For the across country analysis, the following specification has been used:
𝐼𝑅𝐴𝑖 = {1 𝐼𝑅𝐴𝑖 = 𝛼 + 𝛽𝐶𝑖 + 𝑋𝑖′𝛿 + 𝛾1𝐷1+ ⋯ + 𝛾𝑁𝐷𝑁+ 𝜀𝑖 0
20 Where 𝐼𝑅𝐴𝑖 is the dependent holding variable (i.e. holding an individual retirement), 𝐶𝑖 represents the good cognitive ability dummy, 𝑋𝑖 is the vector of control variables, 𝐷𝑖 represents the dummy variables controlling for country effects and lastly 𝜀𝑖 is the zero-mean residual.
As described earlier, in addition to analyzing the relationship between an individual’s cognitive ability and the ownership of an individual retirement account for the entire dataset, a separate analysis on this relationship is performed for each country. For the within country analyses, the following specification has been used:
𝐼𝑅𝐴𝑖 = {1 𝐼𝑅𝐴𝑖 = 𝛼 + 𝛽𝐶𝑖+ 𝑋𝑖′𝛿 + 𝜀𝑖 0
21 4. Results and analysis
4.1 Descriptive statistics
In this section, descriptive statistics on the dependent, independent, and control variables are reported and discussed. Table 2 reports the mean and standard deviation of the probability that an individual owns an individual retirement account for each country as well as for the total sample of 4,222 individuals. The mean probability for the entire sample is 0.8344, with the highest probability in the Czech Republic (0.9186) and lowest in Italy (0.6912).
Table 2
Descriptive statistics of holding an individual retirement account
The descriptive statistics of the good cognitive ability score dummy, based on survey questions testing numeracy, fluency, orientation, and memory, can be found in Table 3. The threshold for “good cognitive ability” has been set at the median score, as described in Section 3.3.2, and 62.7% of the total sample scores the median or higher and is regarded as having good cognitive ability. This number varies strongly from country to country, the highest being in Switzerland (75.1%) and lowest in Greece (13%). It should be noted that only 23 individuals from Greece were included, such that this number is not particularly robust compared to most other countries. A North-South
Mean
Standard
deviation Minimum Maximum Observations
Austria 0.7927 0.4079 0 1 82 Germany 0.8244 0.3810 0 1 410 Sweden 0.8708 0.3360 0 1 271 Spain 0.8268 0.3795 0 1 179 Italy 0.6912 0.4654 0 1 68 France 0.7470 0.4356 0 1 253 Denmark 0.8134 0.3899 0 1 686 Greece 0.7826 0.4217 0 1 23 Switzerland 0.8472 0.3604 0 1 301 Belgium 0.9149 0.2792 0 1 752 Israel 0.8649 0.3466 0 1 37 Czech Republic 0.9186 0.2738 0 1 430 Poland 0.8421 0.3746 0 1 19 Luxembourg 0.8349 0.3730 0 1 109 Portugal 0.8125 0.3934 0 1 64 Slovenia 0.7423 0.4387 0 1 163 Estonia 0.7232 0.4482 0 1 271 Croatia 0.7500 0.4351 0 1 104 Total 0.8344 0.3717 0 1 4,222
22 gradient seems to exist to some extent, with the Northern European countries scoring generally higher than Mediterranean countries, consistent with finding by e.g. Hanushek and Woessmann (2009).
Table 3
Descriptive statistics of having good cognitive ability
The descriptive statistics for the independent control variables are reported in Table 4. The average age of the sample is near the midpoint of the age range at 55.4 years of age. This is also reflected in the other variables, for example the large share of the sample that is in a couple (91.2%) and has children (94.3%). 4.6% of the individuals in the sample has been through a divorce and has not remarried since, and 81.6% reports that they are in good health, defined as having answered that they were in either “good”, “very good”, or “excellent” health. 42.3% of the individuals is highly educated, defined as having attained an ISCED education score of 4 or higher, and the majority of the sample, 86.4%, was employed at the time of survey. Because of the sample selection of pre-retirement individuals, those not employed were also not retired, but homemaker, permanently ill, or otherwise not working. The average Gini-coefficient of the countries included
Mean
Standard
deviation Minimum Maximum Observations
Austria 0.7195 0.4520 0 1 82 Germany 0.7098 0.4544 0 1 410 Sweden 0.6605 0.4744 0 1 271 Spain 0.3743 0.4853 0 1 179 Italy 0.3971 0.4929 0 1 68 France 0.4585 0.4993 0 1 253 Denmark 0.7187 0.4500 0 1 686 Greece 0.1304 0.3444 0 1 23 Switzerland 0.7508 0.4333 0 1 301 Belgium 0.6104 0.4880 0 1 752 Israel 0.4865 0.5067 0 1 37 Czech Republic 0.7000 0.4588 0 1 430 Poland 0.6316 0.4956 0 1 19 Luxembourg 0.5688 0.4975 0 1 109 Portugal 0.2031 0.4055 0 1 64 Slovenia 0.6442 0.4802 0 1 163 Estonia 0.6421 0.4803 0 1 271 Croatia 0.3846 0.4889 0 1 104 Total 0.6265 0.4838 0 1 4,222
23 in the sample is 0.283, in line with expectations for Euro countries. Table 5 shows the Gini-coefficient for each country separately.
Table 4
Descriptive statistics of the control variables
Table 5
Gini coefficients for the countries included in the dataset
4.2 Results
This paragraph lays out the results of the empirical analyses based on the data and methodology as described in the previous chapter. Table 6 shows the results of the logistic regression analyses
Mean
Standard
deviation Minimum Maximum Observations
Age 55.3683 2.5429 50 59 4,222 Male 0.4278 0.4948 0 1 4,222 Couple 0.9119 0.2835 0 1 4,222 Divorced 0.0462 0.2099 0 1 4,222 Having children 0.9432 0.2316 0 1 4,222 Employed 0.8636 0.3433 0 1 4,222 High education 0.4327 0.4955 0 1 4,222
Self-reported good health 0.8160 0.3876 0 1 4,222
Household income 11.1964 1.2899 0 14.3714 4,222
Household gross financial assets 11.5589 1.6105 0 15.9621 4,222
Household real financial assets 12.3588 3.1205 -13.9669 16.6112 4,222
24 employed to find the relationship between the dependent variable (individual retirement account holding probability in pre-retirement individuals) and the independent variables, consisting of both binary and continuous variables. For logistic regression models the regression coefficients indicate whether the relationship between the binary dependent variable and each of the independent variables included in the model is positive or negative. In order to derive the absolute effect that an independent variable has on the holding probability of the dependent variable, one should calculate the marginal effects. These marginal effects are reported in Table 6 and reflect the change in the relevant holding probability of the dependent variable when the variables of interest change from zero to one for binary variables and by one unit for continuous variables. The second part of the analysis concerns the regression results for the relation between individual retirement account ownership and the dependent variables within each country (Table 7).
4.2.1 Relation between cognitive analysis and the probability of owning an individual retirement account.
-25 standing and more favorable institutions regarding individual retirement accounts have higher ownership probabilities.
Table 6
Marginal effects of the logistic regression on holding an individual retirement account
M.E. Std. Dev.
Good cognitive abilities -0.0024 0.0119
Age 0.0056 0.0023 ** Male 0.0261 0.0123 ** Couple -0.0446 0.0286 Divorced -0.1101 0.0360 *** Having children -0.0289 0.0256 Employed 0.1053 0.0148 *** High education 0.0342 0.0120 ***
Self-reported good health 0.0253 0.0139 *
Household income 0.0051 0.0038
Household gross financial assets 0.0290 0.0043 *** Household real financial assets -0.0005 0.0017 Gini coefficient -1.1919 0.7802 Austria -0.1091 0.0604 * Germany -0.0600 0.0440 Sweden -0.0548 0.0569 Spain 0.0398 0.0296 Italy -0.0901 0.0343 *** France -0.0802 0.0386 ** Denmark -0.1407 0.0665 ** Greece 0.0555 0.0711 Switzerland -0.0774 0.0429 * Belgium 0.0218 0.0593 Israel 0.0508 0.0593 Czech Republic 0.0856 0.0664 Poland 0.0266 0.0665 Luxembourg -0.0579 0.0567 Portugal 0.1012 0.0394 *** Slovenia -0.1063 0.0701 Estonia 0.0000 (omitted) Observations 4,222
26 4.2.2 Between country analysis
The variable for good cognitive ability is insignificant when taking the countries together, and this is also the case for the countries individually, with the exception of Sweden (please refer to Table 7). Here, cognitive ability was associated with an 8.5% drop in individual retirement account holding probability. Overall, there is considerable variation in the significance and effect of variables compared between the different countries. Employment status is the only variable that is significant in the majority of countries, eight out of fifteen, and is positively related to individual retirement account holding probability, which is unsurprising considering the nature of individual retirement savings and the implications mentioned in Section 3. Age and gender are significant in respectively five and six countries, both with varyingly positive and negative effects. The effect of divorce is significant in six countries as well, and is negative for each country in which it is significant, in line with findings by Wilmoth and Koso (2002).
27 Table 7 (1/2)
Marginal effects of the logistic regression on holding an individual retirement account by country
M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev.
Good cognitive abilities -0.1016 0.1172 -0.0429 0.0442 -0.0854 0.0456 * 0.0870 0.0563 -0.0026 0.0933
Age 0.0239 0.0223 0.0081 0.0088 0.0150 0.0081 * 0.0103 0.0114 -0.0306 0.0169 * Male -0.1340 0.0971 -0.0454 0.0405 -0.0692 0.0406 * 0.1429 0.0624 ** 0.2052 0.1285 Couple -1.9909 0.4249 *** 0.0486 0.0697 -0.1779 0.1080 -1.6681 0.3011 *** 0.0000 (omitted) Divorced -1.8627 0.4225 *** -0.0534 0.1205 -0.1149 0.1372 -1.9371 0.3322 *** 0.0000 (omitted) Having children 0.2100 0.1541 0.0643 0.0608 0.1712 0.0972 * -0.1476 0.1135 0.0299 0.1601 Employed 0.0668 0.1445 0.0085 0.0625 0.0148 0.0944 0.1499 0.0477 *** 0.3594 0.1092 *** High education 0.1122 0.0957 0.0766 0.0422 * 0.0249 0.0438 0.0555 0.0657 -0.0298 0.1021
Self-reported good health -0.0350 0.1350 0.0399 0.0444 0.0877 0.0573 -0.0457 0.0650 -0.2940 0.1569 *
Household income 0.0927 0.1228 -0.0164 0.0326 0.0066 0.0308 -0.0108 0.0202 0.0401 0.0354
Household gross financial assets -0.0038 0.0576 0.0284 0.0161 * 0.0442 0.0166 *** 0.0083 0.0270 0.0245 0.0162 Household real financial assets 0.0016 0.0147 -0.0034 0.0048 0.0006 0.0050 0.0596 0.0462 0.0443 0.0098 ***
Observations 82 410 271 179 65
Note: M arginal effects indicate the change in the relevant choice probability when the binary variable changes from zero to one and the continuous variable by one unit. ***, **, * denote significance at 1%, 5% and 10%, respectively. Greece, Israel and Poland are not included as their datasets include too few individuals.
Austria Germany Sweden Spain Italy
M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev.
Good cognitive abilities -0.0587 0.0600 0.0458 0.0313 -0.0644 0.0531 -0.0287 0.0213 0.0178 0.0246
Age 0.0126 0.0102 0.0089 0.0053 * 0.0041 0.0094 0.0036 0.0038 0.0092 0.0049 *
Male 0.1483 0.0586 ** 0.1472 0.0327 *** 0.0776 0.0503 0.0214 0.0219 -0.0019 0.0250
Couple 0.0898 0.0977 -0.0897 0.0787 0.0269 0.1049 0.0013 0.0551 -0.8138 0.1174 ***
Divorced -0.0986 0.1462 -0.1637 0.0936 * 0.0691 0.1326 -0.0291 0.0659 -0.8842 0.1252 ***
Having children 0.0000 (omitted) -0.1832 0.1058 * -0.1202 0.0900 0.0428 0.0290 0.2591 0.0613 ***
Employed -0.0060 0.0790 0.1175 0.0484 ** 0.1423 0.0548 *** 0.0427 0.0248 * 0.1237 0.0281 ***
High education 0.0704 0.0626 -0.0178 0.0302 0.0549 0.0441 0.0090 0.0224 0.1064 0.0607 *
Self-reported good health 0.0118 0.0597 -0.0084 0.0442 0.0615 0.0600 0.0876 0.0212 *** -0.0206 0.0288
Household income 0.0000 0.0468 0.0006 0.0182 -0.0214 0.0377 -0.0128 0.0154 -0.0017 0.0073
Household gross financial assets 0.0915 0.0182 *** 0.0311 0.0118 *** 0.0280 0.0234 0.0159 0.0093 * 0.0514 0.0119 *** Household real financial assets -0.0064 0.0107 -0.0053 0.0056 0.0000 0.0035 0.0059 0.0025 ** -0.0058 0.0045
Observations 244 686 301 752 430
Note: M arginal effects indicate the change in the relevant choice probability when the binary variable changes from zero to one and the continuous variable by one unit. ***, **, * denote significance at 1%, 5% and 10%, respectively. Greece, Israel and Poland are not included as their datasets include too few individuals.
Czech Republic
28 Table 7 (2/2)
Marginal effects of the logistic regression on holding an individual retirement account by country
M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev. M.E. Std. Dev.
Good cognitive abilities -0.0279 0.0863 -0.0595 0.1237 0.0830 0.0625 0.0370 0.0498 0.0338 0.0716
Age -0.0133 0.0173 0.0448 0.0281 0.0065 0.0132 -0.0053 0.0097 -0.0307 0.0151 **
Male 0.1647 0.1089 -0.2614 0.1008 ** 0.0053 0.0722 -0.1487 0.0590 ** -0.0378 0.1068
Couple -1.7778 0.3622 *** 0.0000 (omitted) 0.0000 (omitted) 0.0040 0.0918 0.0000 (omitted)
Divorced -2.0751 0.4797 *** 0.0000 (omitted) ** 0.0000 (omitted) -0.1901 0.1162 0.0000 (omitted)
Having children 0.0329 0.1233 0.1309 0.1404 -0.1047 0.1202 -0.1800 0.1570 0.0000 (omitted)
Employed 0.0820 0.0968 0.1986 0.0924 0.2967 0.0668 *** 0.0912 0.0932 0.3652 0.0528 ***
High education 0.1164 0.0926 -0.1771 0.1419 0.0772 0.0575 0.0255 0.0527 0.1040 0.1270
Self-reported good health 0.0000 (omitted) 0.1464 0.0959 -0.0021 0.0794 0.0009 0.0579 -0.0151 0.0835
Household income -0.0349 0.0818 -0.0136 0.0286 0.0014 0.0114 0.0283 0.0279 0.0161 0.0079 **
Household gross financial assets 0.0208 0.0376 0.0518 0.0198 *** 0.0198 0.0207 0.0165 0.0216 0.0102 0.0356 Household real financial assets 0.0128 0.0402 -0.0494 0.0183 *** 0.0060 0.0057 -0.0143 0.0089 -0.0073 0.0151
Observations 96 62 163 271 98
Note: M arginal effects indicate the change in the relevant choice probability when the binary variable changes from zero to one and the continuous variable by one unit. ***, **, * denote significance at 1%, 5% and 10%, respectively. Greece, Israel and Poland are not included as their datasets include too few individuals.
Croatia
Table 8
Marginal effects of the logistic regression on holding an individual retirement account for individuals having good cognitive ability
M.E. Std. Dev. Age 0.0067 0.0029 ** Male 0.0346 0.0149 *** Couple -0.0517 0.0361 Divorced -0.1025 0.0464 ** Having children -0.0311 0.0330 Employed 0.0930 0.0201 *** High education 0.0282 0.0139 **
Self-reported good health 0.0271 0.0187 Household income -0.0008 0.0071 Household gross financial assets 0.0301 0.0057 *** Household real financial assets -0.0014 0.0023 Gini coefficient -1.3104 1.2978 Austria -0.1364 0.0953 Germany -0.0875 0.0710 Sweden -0.0967 0.0909 Spain 0.0988 0.0578 * Italy -0.0764 0.0554 France -0.1006 0.0651 Denmark -0.1477 0.1104 Greece 0.0262 0.0878 Switzerland -0.1028 0.0651 Belgium -0.0141 0.0985 Israel 0.0000 (omitted) Czech Republic 0.0728 0.1116 Poland 0.0776 0.1385 Luxembourg -0.1000 0.0866 Portugal 0.0189 0.0715 Slovenia -0.1011 0.1184 Estonia 0.0000 (omitted) Observations 2,627
5. Conclusion and recommendations
The purpose of this research was to gain insight into the relation between cognitive ability and the ownership of an individual retirement account. The sustainability of pension systems have been on the political agenda in many European countries for some years, and calls for reform in a variety of ways are heard. In addition to the common and more direct regulatory outcome of raising the retirement age and lowering net benefits, a shift in responsibility from the government to the individual has been noted as well. Governments increasingly look towards stimulating private retirement savings as a way to improve prospects for retirement income in the future. A common method to do this is to create tax incentives over sums invested in private retirement accounts, to make doing so more attractive to individuals, without directly increasing the budget for the public pension scheme. This allows individuals to affect a relatively larger increase in retirement income, while maintaining freedom of choice compared to raising taxes or contributions.
As private saving for retirement is highly dependent on individual preferences and circumstances, it is important to understand how these might affect financial decision making. While literature suggests that cognitive ability might affect life-cycle financial planning through planning horizon, information cost and risk attitude, it is useful relation between cognitive ability and voluntary pre-retirement saving. A number of logistic regressions were done with a compound variable for good cognitive ability as the independent variable and the probability of owning an individual retirement account as the independent variable, including a set of control variables relating to the country and personal characteristics of respondents (age, gender, family status, employment and financial status, education, and health). The relation between cognitive ability and individual retirement account ownership probability was analyzed and compared between different countries. Based on this regression, there is no evidence to suggest a difference in the relation between individual retirement account ownership and cognitive ability between countries.
ownership among a restricted demographic subset. The heterogeneity in the economic circumstances and institutions of different countries, as well as the long time horizons inherent to social security policy, imply that more concentrated and longer-term analysis will likely reveal more insights into the relation between country, individual factors, and third pillar participation.
6. References
Agarwal, S., and Mazumder, B. (2010). Cognitive Abilities and Household Financial Decision Making. American Economic Journal 5(1), 193-207.
Alessie, R., Angelini, V., van Santen, P. (2013). Pension Wealth and Household Saving in Europe: Evidence from SHARELIFE. European Economic Review 63, 308-328.
Almenberg, J., Dreber, A. (2015). Gender, stock market participation and financial literacy. Economics Letters 137, 140–142.
Banks, J. and Blundell, R. (2005), Private Pension Arrangements and Retirement in Britain. Fiscal Studies, 26: 35-53.
Banks, J., Oldfield, Z. (2007). Understanding Pensions: Cognitive Function, Numerical Ability and Retirement Saving. Fiscal Studies 28, 143-170.
Barber, B., Odean, T. (2001). Boys Will be Boys: Gender, Overconfidence, and Common Stock Investment. The Quarterly Journal of Economics 116(1), 261-292
Benartzi, S., and Thaler, R. (2007). "Heuristics and Biases in Retirement Savings Behavior." Journal of Economic Perspectives, 21 (3): 81-104.
Benjamin, D., Brown, S., Shapiro, J. (2006). Who is 'Behavioral'? Cognitive Ability and Anomalous Preferences. Journal of the European Economic Association 11(6), 1231-1255. Berg, L. (1996). Age Distribution, Saving and Consumption in Sweden. Uppsala University Working Paper 1996:22.
Bodie, Z., Merton, R., Samuelson, W. (1992). Labor supply flexibility and portfolio choice in a life cycle model. Journal of Economic Dynamics and Control 16, 427-449.
Bodie, Z. (2002). Life-Cycle Finance in Theory and in Practice. Boston University School of Management Working Paper No. 2002-02.
Bolton, Brian J. (2009) The U.S. Financial Crisis: A Summary of Causes & Consequences Bonsang, E., Dohmen, T. (2015). Risk attitude and cognitive aging. Journal of Economic Behavior and Organization 112, 112–126.
Börsch-Supan, A., & Stahl, K. (1991). Life cycle savings and consumption constraints. Journal of population economics, 4(3), 233-255.
Bressan, S., Pace, N., Pelizzon, L. (2014). Health status and portfolio choice: is their relationship economically relevant? International Review of Financial Analysis 32, 109-122
Burks, S., Carpenter, J., Goette, L., Rustichini, A. (2009). Cognitive skills affect economic preferences, strategic behavior, and job attachment. Proceedings of the National Academy of Sciences 106(19), 7745–7750.
Christelis, D., Jappelli, T., Padula, M. (2010). Cognitive Abilities and Portfolio. European Economic Review 54, 18-38.
Christelis, D., Dobrescu, L., Motta, A. (2012). Early Life Conditions and Financial Risk-Taking in Older Age. Unpublished working paper. ARC Centre of Excellence in Population Ageing Research.
Cole, S., Shastry, G. (2009). Smart money: The effect of education, cognitive ability, and financial literacy on financial market participation. Harvard Business School Finance Working Paper No. 09-071.
Commission of the European Communities (2000). The future evolution of social protection from a long-term point of view: Safe and sustainable pensions.
Coppola, M., (2014), Eliciting risk-preferences in socio-economic surveys: How do different measures perform?, Journal of Behavioral and Experimental Economics (formerly The Journal of Socio-Economics), 48, issue C, p. 1-10
Coppola, Michela and Wilke, Christina B. (2010) How Sensitive are Subjective Retirement Expectations to Increases in the Statutory Retirement Age? The German Case. MEA Discussion Paper No. 207-10.
Doblhammer-Reiter, G., van den Berg, G.,. Fritze, T. (2011). Economic Conditions at the Time of Birth and Cognitive Abilities Late in Life: Evidence from Eleven European Countries. IZA Discussion Paper No. 5940
Dolls, M., Doerrenberg, P., Peichl, A., & Stichnoth, H. (2016). Do savings increase in response to salient information about retirement and expected pensions? (No. w22684). National Bureau of Economic Research.
Drolet, Aimee & Schwarz, Norbert & Yoon, Carolyn. (2010). The Aging Consumer: Perspectives from Psychology and Economics.
Duflo, E., and E. Saez (2002), “Participation and Investment Decisions in a Retirement Plan: the Influence of Colleagues’ Choices”. Journal of Public Economics, 85, 121-48.
Galasso, V., & Profeta, P. (2007). How does ageing affect the welfare state?. European Journal of Political Economy, 23(2), 554-563.
Hurd, M. D. (1990). Research on the elderly: Economic status, retirement, and consumption and saving. Journal of economic literature, 28(2), 565-637.
Lachowska, M. (2017). A Note on the Approximate Interpretation of Dummy Variable Coefficients in Inverse Hyperbolic Sine Regression.
Le Blanc, J. (2011). The third pillar in Europe: institutional factors and individual decisions. Leetmaa, P., Rennie, H., & Thiry, B. (2009). Household saving rate higher in the EU than in the USA despite lower income. Eurostat Statistics in Focus, 29, 1-11.
Lusardi, Annamaria (2003). Planning and saving for retirement. Working paper. Dartmouth College, 2003
Lusardi, A., and Mitchell, O., (2007), Baby Boomer retirement security: The roles of planning, financial literacy, and housing wealth, Journal of Monetary Economics, 54, issue 1, p. 205-224 Lusardi, A. (2008). Household saving behavior: the role financial literacy, information and financial education programs. NBER Working Paper Series No. 13824.
Mazzonna, Fabrizio and Peracchi, Franco, (2012), Ageing, cognitive abilities and retirement, European Economic Review, 56, issue 4, p. 691-710
McArdle, J., Smith, J., Willis, R. (2009). Cognition and Economic Outcomes in the Health and Retirement Survey. NBER Working Papers No. 15266.
Pence, K. (2006). The Role of Wealth Transformations: An Application to Estimating the Effect of Tax Incentives on Saving. The B.E. Journal of Economic Analysis & Policy 5(1), 1-26.
Phelps, E. and Pollak, R. A., (1968), On Second-Best National Saving and Game-Equilibrium Growth, Review of Economic Studies, 35, issue 2, p. 185-199,
Pezzuto, I. (2008) Miraculous Financial Engineering or Toxic Finance? The Genesis of the U.S. Subprime Mortgage Loans Crisis and its Consequences on the Global Financial Markets and Real Economy.
Powell, M., Ansic, D. (1997). Gender differences in risk behaviour in financial decision making: An experimental analysis. Journal of Economic Psychology 18(6), 605-628.
Schneider, O. (2009). Reforming pensions in Europe: Economic fundamentals and political factors.
Solmon, Lewis C., (1975), The Relation between Schooling and Savings Behavior: An Example of the Indirect Effects of Education, p. 253-294, Education, Income, and Human Behavior, National Bureau of Economic Research, Inc.
Tumino, Alberto, (2016), Retirement and cognitive abilities, No 2016-06, ISER Working Paper Series, Institute for Social and Economic Research
UNESCO, 2006. International Standard Classification of Education: ISCED 1997. Venes, D. (2009). Taber's cyclopedic medical dictionary. FA Davis.
7. Appendix Appendix A (1/2)
Overview of variables used in the analysis
Mnemonic SHARE variable
name Variable description
Variables used for constructing the dataset
mergeid mergeid Unique identification number per individual.
hhid6 hhid6 Unique identification number per household.
financial_respondent fin_resp Dummy variable indicating one if individual is financial respondent (fin_resp = 1).
retired cjs Dummy variable indicating one if individual is retired (cjs = 1).
Dependent variable
who_has_iras as020_
A variable on the ownership of an individual retirement account ("IRA") answered by the financial respondent of the household, yield the following options: 1) respondent only, 2) husband/wife/partner only and 3) both.
hh_has_iras - Aggregated variable (by using hhid6 ) indicating the maximum of the
who_has_iras variable.
iras Dummy variable indicating one if the individual has an individual retirement
account (please refer to Table 1 for construction details).
Independent variable
recall_score cf008tot Variable indicating four if cf008tot > 4, three if cf008tot = 4, two if cf008tot = 3, one if cf008tot = 2 and zero otherwise.
delayed_recall_score cf016tot Variable indicating four if cf016tot > 3, three if cf016tot = 3, two if cf016tot = 2, one if cf016tot = 1 and zero otherwise.
numeracy_score numeracy Variable indicating four if numeracy = 5, three if numeracy = 4, two if numeracy = 3, one if numeracy = 2 and zero otherwise.
fluency_score cf010_ Variable indicating four if cf010_ > 23, three if cf010_ > 18 and < 23, two if cf010_ > 15 and < 18, one if cf010_ > 11 and < 15 and zero otherwise. orientation_score orienti Variable indicating four if orienti = 4, three if orienti = 3, two if orienti =2, one if
orienti = 1 and zero otherwise.
cognitive_score recall_score, delayed_recall_score , numeracy_score, fluency_score, orientation_score
Sum of the five cognitive abilities components: recall_score +
delayed_recall_score + numeracy_score + fluency_score + orientation_score.
good_cognitive_ability cognitive_score Dummy variable indicating one if the individual has good cognitive abilities (cognitive_score ≥ 18).
Control variables
Age age Continuous variable indicating the age of an individual.
Gender gender Dummy variable indicating one if the individual is male (gender=1)
Couple mstat Dummy variable indicating one if the individual is in a couple (mstat = 1,2 or 3)
Divorced mstat Dummy variable indicating one if the individual is divorced (mstat =
Having_children nchild Continuous variable indicating the number of children an individual has.
Employed cjs Dummy variable indicating one if the individual is employed or self-employed (cjs = 2).
High_education isced Dummy variable indicating one if the individual has taken post-secondary education (isced = 4,5 or 6).
Self_reported_health_good sphus Dummy variable indicating one if the individual's self-assessed health is at least "good" (sphus = 1,2 or 3).
Hyper_total_hh_income thinc Inverse hyperbolic sine transformation of the total household income. Hyper_gross_assets hgfass Inverse hyperbolic sine transformation of the household gross financial assets. Hyper_real_assets hrass Inverse hyperbolic sine transformation of the household real financial assets.
Appendix A (2/2)
Overview of variables used in the analysis
Mnemonic SHARE variable
name Variable description
Variables used for constructing the dataset
Austria country Dummy variable indicating one if individual's country of residence is Austria (country = 11)
Germany country Dummy variable indicating one if individual's country of residence is Germany
(country = 12)
Sweden country Dummy variable indicating one if individual's country of residence is Sweden
(country = 13)
Spain country Dummy variable indicating one if individual's country of residence is Spain (country = 15)
Italy country Dummy variable indicating one if individual's country of residence is Italy (country = 16)
France country Dummy variable indicating one if individual's country of residence is France
(country = 17)
Denmark country Dummy variable indicating one if individual's country of residence is Denmark
(country = 18)
Greece country Dummy variable indicating one if individual's country of residence is Greece
(country = 19)
Switzerland country Dummy variable indicating one if individual's country of residence is Switzerland (country = 20)
Belgium country Dummy variable indicating one if individual's country of residence is Belgium (country = 23)
Israel country Dummy variable indicating one if individual's country of residence is Israel (country = 25)
Czech Republic country Dummy variable indicating one if individual's country of residence is Czech Republic (country = 28)
Poland country Dummy variable indicating one if individual's country of residence is Poland
(country = 29)
Luxembourg country Dummy variable indicating one if individual's country of residence is Luxembourg (country = 31)
Portugal country Dummy variable indicating one if individual's country of residence is Portugal (country = 33)
Slovenia country Dummy variable indicating one if individual's country of residence is Slovenia (country = 34)
Estonia country Dummy variable indicating one if individual's country of residence is Estonia (country = 35)
