THE INFLUENCE OF FINANCIAL KNOWLEDGE AND SELF- PERCEPTION ON THE PROPENSITY TO SAVE FOR RETIRE- MENT.

(1)

1

THE INFLUENCE OF FINANCIAL KNOWLEDGE AND

SELF-PERCEPTION ON THE PROPENSITY TO SAVE FOR

RETIRE-MENT.

Master Thesis Finance

Ana García Moreno

S3490742

June 2018 Msc Finance

(2)

2 ABSTRACT

This study investigates the influence of the two dimensions of financial knowledge and self-esteem on the decision of saving for retirement. This study uses the LISS panel ad-ministered by CentERdata. The final sample consists of 1,872 respondents on a Dutch representative dataset. This thesis distinguishes from other literature by including the two dimensions of financial knowledge instead of just objective knowledge. An additional meaningful new feature is the analysis of the role of personality traits, specifically the effect of self-esteem on saving for retirement. The results of the multivariate regressions and probit models indicate that both financial knowledge dimensions have a positive and significant effect on savings for retirement. However, when high self-esteem is included in the model, it exerts the strongest influence on the decision of saving for retirement. The results indicate that financial education programs need also to take into account re-spondents self-esteem and self-perception of their financial literacy. In conclusion, this thesis contributes to a broader understanding of the concept of financial knowledge and the influence of personality traits on the propensity to save for retirement.

Keywords: Objective financial knowledge, self-perceived knowledge, self-esteem, sav-ing for retirement

(3)

3

3.3. Explanatory variables. ... 13 3.3.1. Financial knowledge ... 13 3.3.2. Self-esteem ... 14 3.4. Control variables.... 14 3.5. Summary statistics. ... 15 4. METHODOLOGY. ... 18 4.1. Hypotheses. ... 18 4.2. Robustness checks. ... 21 5. RESULTS. ... 22 5.1. Correlations ... 22 5.2. Regression analysis ... 24

5.2.1. Financial knowledge and saving for retirement. ... 24

5.2.2. Interaction between actual and perceived knowledge.... 28

5.2.3. Self- esteem and saving for retirement. ... 31

6. DISCUSSION OF THE RESULTS AND LIMITATIONS. ... 34

6.1. Discussion of the results. ... 34

6.1.1. Influence of actual knowledge on saving for retirement. ... 34

6.1.2. The effect of self-perceived knowledge on saving for retirement. ... 34

6.1.3. Interaction between actual and self-perceived knowledge. ... 35

6.1.4. The effect of high self-esteem on saving for retirement. ... 36

6.1.5. Socioeconomic control variables. ... 36

6.2. Limitations. ... 37

7. CONCLUSIONS AND IMPLICATIONS FOR THE FUTURE... 37

8. REFERENCES. ... 39

(4)

4 1. INTRODUCTION

Over the last decades, a good number of countries have started reforming their retirement systems. These transformations range from the traditional defined benefit pensions to

in-dividual account defined contribution schemes. Individuals have to determine the amount

saved for retirement and the allocation of the pension wealth (Lusardi, 2008). As a con-sequence, a substantial part of the risk and responsibility for the retirement plans have been allocated to individuals and households (Prast & Van Soest, 2016). These reforms are partially beneficial, due to the flexibility that defined contribution schemes provide. Nonetheless, this flexibility might lead individuals to bear greater responsibility and to make serious mistakes, such as undersaving or failing to invest wisely (Lusardi & Mitch-ell, 2011). Moreover, individuals face a complex catalogue of financial products they do not understand, and they are poorly informed about them.

This changing environment has caused an increase of the studies related to the extent of financial illiteracy around the world, the scope of the definition of financial knowledge and its association with financial behaviour and specifically, with the propensity to save and the amount saved for retirement.

With regards to the extent of financial illiteracy around the world, research proves low financial literacy and lack of information to be widespread among the general population (Bucher-Koenen et al., 2011; Lusardi & Mitchell, 2011; Lusardi, 2008). In practice, even the most basic financial concepts are unknown for many households (Lusardi and Mitch-ell, 2007). These studies found evidence that the majority of people cannot answer simple questions regarding compound interest, inflation and risk diversification.

(5)

5 Furthermore, in response to these problems, financial education courses have been im-plemented by governments, employers and financial institutions. Nevertheless, studies concerning the effectiveness of such programs have shown varied outcomes. Some of them have concluded that although an individual might have sufficient objective financial knowledge, he or she will not engage in efficient financial behaviour because low self-perceived financial literacy (Hadar, Sood and Fox, 2013; Assad, 2015). Therefore, im-proved financial behaviour does not automatically follow from a higher financial literacy. With respect to the scope of the definition of financial knowledge, the approach solely targeting objective financial literacy might not be sufficient. Consequently, we might take into consideration that financial literacy can be divided into two components: objective and subjective. Subjective literacy is defined as what the person thinks he knows whereas objective refers to actual knowledge (Hadar, Sood and Fox, 2013).

General literature has acknowledged the impact of actual financial knowledge on plan-ning or saving for retirement. However, the effect of self-perceived financial knowledge on financial behaviour, specifically, on planning for retirement and how subjective knowledge interacts with objective knowledge is still relatively unstudied. This is relevant

because most of investment decisions are made under uncertainty.Some studies

docu-ment that holding objective financial knowledge constant, subjective knowledge pro-motes the choice of risky and complex investment options (Asaad, 2015). By contrast, other authors state that even if the respondent knows about an investment, if he feels that he does not possess sufficiently knowledge, his choice will not be one that suits his needs the best, but the safest or the most familiar (Hadar, Sood and Fox, 2013).

Part of the literature has demonstrated that perceived knowledge exerts an influence on financial behaviour (Allgood & Walstad, 2013,2; Anderson, Baker, & Robinson, 2017; Asaad, 2015; Croy, Gerrans, & Speelman, 2010). This raises questions related to the scope of the definition of self-perceived knowledge. Therefore, determining which factors can affect subjective knowledge is also fundamental. The role of personality traits on financial decisions can be relevant and, in some cases, they can affect financial knowledge (Brown & Taylor, 2014). In particular, self-esteem is the level of confidence in one’s own worth or abilities (Tang & Baker, 2016). Early findings have encouraged the belief that self-esteem can enhance a positive development of academic performance due to the pos-itive correlation between self-esteem and educational achievement. (Baumeister et al.,

2003).Regarding financial behaviour, Tang and Baker (2016) study the association

(6)

6 responsible financial behaviour. Hence, financial education programs must cover both financial literacy dimensions and the development of psychological traits.

This paper contributes empirically to the existing literature in two ways: first, studying the association between self-perceived and saving for retirement, and its interaction with actual knowledge; secondly, linking self-esteem to households’ propensity to save for retirement controlling for other sociodemographic factors in the Netherlands.

The remainder of this study is organized as follows. Section 2 discusses a review of the current literature on this subject, leading to my research hypotheses. Section 3 describes the data retrieved from LISS panel administered by CentERdata and the construction of the variables. Section 4 provides the design of the research hypotheses and describes the methodology used. Section 5 introduces results of my analysis. Section 6 provides a dis-cussion of the results and limitations. Section 7 gives the conclusions and implications for future research.

2. LITERATURE REVIEW.

The role of financial literacy is in debate and a part of this debate may be related to the variation in how researchers establish and measure financial knowledge. Prior research definitions of financial literacy vary quite substantially and so do the methods used to measure financial literacy. Nonetheless, the most common principle for the definition is knowledge (Hung, Parker, & Yoong, 2009; Stolper & Walter, 2017). The majority of studies have documented the association between financial knowledge and planning for retirement, defining financial literacy as a specific form of actual knowledge (Alessie, Rooij, & Lusardi, 2011; Lusardi, 2008; Lusardi & Mitchell, 2007). However, actual knowledge is important but insufficient. Hence, the role of perceived knowledge might be significant in order to find if there is a positive relationship between financial knowledge and planning for retirement (Agnew and Skyzman, 2005; Hadar et al., 2013; Hung, Parker, & Yoong, 2009; Parker et al.,2012 )

In the next subsections, first I will present an overview of the actual Dutch pension system and then I will review current literature related to objective financial literacy, subjective financial literacy and self-esteem and their link in the existing literature to financial be-haviour and specifically to saving for retirement.

2.1.Dutch pension system

Three pillars compose the Dutch pension system: The General Old Age Pensions (AOW), occupational benefits and private pension schemes (Van Els, Van den End, & Van Rooij, 2004). Nevertheless, two of them are fundamental.

(7)

7

ensures that a lifelong basic income is provided to persons aged 65 and over, based on

the years resided in the Netherlands between 15 to 65 years old.

Apart from the AOW, an employee might receive an occupational benefit from his or her employer. Solidarity is fundamental in the Dutch pension system; a member pays a high percent of his/her wage for the future pension rights each year. Contributions are made jointly from employer and employee. Historically, little freedom of choice has been for company retirement plan because the level of contribution is established by trade unions and the decision of the investment policy by pension funds. Approximately, 90 percent of the employees is member of a pension fund in the Netherlands.

Despite the strong development of the actual pension system, there are two main chal-lenges.

Firstly, the effects of the financial crisis. During the last financial crisis, pension funds incurred in huge investment losses. A committee of pension experts appointed by the government concluded that current pension system is not sustainable anymore (Van Rooij, Lusardi, & Alessie, 2011).

The second one is related to a demographic change. As ageing of the population increases, also does the retirees to workers ratio. This causes a rise in the cost of pensions and con-tributions are significantly higher than before (Van Der Smitte, 2013).

In order to know if it is necessary to take additional measures to maintain their way of living when retirement age is reached, understanding the amount and the adequacy of their pension income is fundamental (Van Der Smitte, 2013). Therefore, there is an in-creasing necessity for households to prepare for retirement and thus acquire the necessary financial skills to do so.

2.2.Objective financial knowledge and financial behaviour

The relationship between financial literacy and household financial decisions is steadily becoming critical, individuals are increasingly required to take financial responsibility for their well-being and retirement plans. However, financial illiteracy is widespread around the world and households are not prepared to make financial decisions regarding retire-ment (as stated by Bucher-koenen et al., 2011; Lusardi & Mitchell, 2007; Lusardi & Mitchell,2011).

In the case of the Netherlands, participation in retirement plans is compulsory and Dutch pension system is starting to shift further responsibility towards employees for retirement financial decisions (Alessie, Van Rooij & Lusardi, 2011). In general, existing retirement plans are inadequate to fulfil the desired income for retirement. In addition, most employ-ees reveal a substantial lack of knowledge and interest in financial issues (confirmed by other studies such as Agnew & Skyzman, 2005; Van Els, Van den End & Van Rooij, 2004).

(8)

8 relationship (as discussed in studies such as Bucher-Koenen et al., 2011, Chan and Stevens, 2008; Lusardi and Mitchel 2006, 2007, 2011). This implies that as financial knowledge increases, the likelihood of saving for retirement raises. “Well- informed in-dividuals are five times more responsible to pension incentives than the average individ-ual when knowledge is ignored” (Chan and Stevens 2008, p.1).

Due to the change to defined contribution pension schemes in the United States (US), the majority of research on these topics to date has focused on the level of financial knowledge in the US (Hilgert, Hogarth, & Beverly, 2003; Lusardi, 2008; Lusardi & Mitchell, 2007).

By using a similar approach, studies such as Van Rooij, Lusardi, & Alessie, 2009, 2011 confirmed the evidence from US studies about the positive relationship between financial literacy and planning for the retirement in the Netherlands. They found that more financial

literate households are more prone to make arrangements for retirement. Moreover,

re-spondents with high actual financial literacy have a higher propensity to invest in stocks and to plan for retirement. Van Rooij et al. (2009, 2011) measure retirement planning by including a question in their survey on how much the respondents have thought about retirement. In order to construct a financial literacy index, they designed a set of questions to evaluate financial literacy and skills of the respondents.

Although the topic of financial literacy has become fundamental, there is no universally accepted definition (as discussed in some studies like Stolper and Walter, 2017). Lusardi & Mitchell (2006) constructed three questions “to identify numeracy levels of households and understanding of basic economic concepts, namely inflation, interest compounding and risk diversification”(Van Rooij et al., 2009, p.1). These questions have been widely adopted to design a financial literacy index and are known as the Big Three (Bucher-Koenen et al., 2011; Hung et al., 2009; Stolper & Walter, 2017). However, authors criti-cise that respondents are not encouraged to respond reflecting their actual knowledge. Moreover, the aim of these three questions is to extract basic objective knowledge of the economic concepts mentioned above (Lusardi, Mitchell, & Curto, 2014). Hence, the scope of the definition of financial literacy is currently under discussion.

Regarding objective financial literacy, several studies have revealed its influence on fi-nancial decisions, suggesting that fifi-nancial education programs have a positive and sig-nificant influence on financial behaviour (Lusardi, 2008; Lusardi & Mitchell, 2007). However, other studies now argue that implementing education programs related to

fi-nance has not been a successful strategy(Fernandes, Lynch, & Netemeyer, 2014; Hadar,

Sood, & Fox, 2013). Therefore, an increase in objective financial knowledge does not directly result in an improvement of financial behaviour (Alessie et al., 2011).

(9)

9 have not dealt with other factors related to financial knowledge that can influence finan-cial behaviour (Alessie et al., 2011; Hilgert, Hogarth, & Beverly, 2003; Lusardi, 2008; Lusardi & Mitchell, 2007; Van Rooij et al.,2009). Therefore, it is fundamental to detect which other factors can influence financial behaviour.

2.3.Perceived financial knowledge and financial behaviour

An alternative measure that is increasingly relevant in the literature is self-assessed finan-cial knowledge (Stolper & Walter, 2017). Self-assessed finanfinan-cial literacy is also known as subjective knowledge, self-perceived knowledge or financial confidence. There are mixed results regarding the relationships between actual and self-perceived knowledge, and the last one with financial behaviour. This measure implies asking individuals for a self- assessment of their financial expertise.

The extent of actual knowledge of respondents is unknown and self-assessment is funda-mental when they confront a decision (Allgood & Walstad, 2013, 2016). Similarly, Asaad (2015) is one of the authors that demonstrates that financial behaviour is influenced by actual knowledge and financial confidence. He challenges the widely held view that ob-jective financial literacy is the main driver of financial behaviour, by demonstrating that the subjective knowledge is just as large as the effect of actual knowledge

On the one hand, part of the literature compares both index and they report that self-assessed knowledge is larger than actual knowledge. Therefore, there is overconfidence and this may be problematic (Stolper & Walter, 2017). As a consequence, individuals with high subjective literacy but low objective knowledge; are more prone to make risky financial decisions (Asaad, 2015).

Moreover, overconfident investors underestimate financial risks and trade more, leading to a negative association between financial confidence and stockholding (Christelis, Jappelli, & Padula, 2010).

Contrary to the view that overconfidence can engage in riskier financial decisions, Parker et al. (2012) demonstrate that higher confidence is correlated with more prudent behav-iour. In addition, they find that both perceived and actual financial knowledge are predic-tive for retirement planning. Using Dutch Household Survey (DHS), Alessie et al. (2011) document that respondents with high self-perceived financial literacy are more likely to plan for retirement.

In a related way, studies such as Anderson, Baker, & Robinson (2017) found evidence that household financial decisions are not just driven by financial knowledge itself. Spe-cially, they notice that much of the retirement planning is promoted by subjective finan-cial knowledge and not by actual finanfinan-cial knowledge itself. They conclude that “savers and planners are those who believed that they are informed, but they are not necessarily those who are informed” (Anderson, Baker, & Robinson, 2017, p.398).

(10)

10 change investment allocations. This study reveals that the perception of planning im-portance exerts a crucial influence on planning preparedness. Therefore, future research might benefit from studying the underlying perceptions that might influence financial behaviour.

Moreover, other studies have demonstrated that there is not always strong correlation between perceived financial literacy and objective knowledge (Agnew and Skyzman, 2005; Hadar et al., 2013; Hung, Parker, & Yoong, 2009; Parker et al., 2012). For instance, Agnew & Skyzman (2005) find a median correlation of 0.49 between objective and sub-jective financial literacy index, varying depending on the characteristics of the respond-ent. Hence, there is a mismatch between actual and self-assessed knowledge. Conse-quently, using objective knowledge as a proxy for both dimensions of knowledge might not be suitable.

Developing an index of financial knowledge compounded by the combination of actual financial knowledge with self-perceived knowledge can influence more significantly fi-nancial behaviour than objective knowledge alone (Allgood & Walstad, 2013, 2016). For these reasons, I have expanded my investigation into the influence of subjective fi-nancial knowledge on the likelihood of saving and the amount saved for retirement and the influence of the interaction between both knowledge dimensions on retirement sav-ings.

2.4.Self-esteem and financial behaviour

The importance of understanding what influences financial decisions of households has been highlighted during the recent financial crisis, when difficulties for many retirement savers appeared. In an effort to find a remedy, the influence of psychological factors on these decisions is becoming relevant amongst academic and policymakers (Asaad, 2015; Brown & Taylor, 2014; Fernandes et al., 2014; Hadar et al., 2013).

Studies such as Fernandes et al. (2014) reported that the effects of financially oriented

education programs decrease dramatically when controlling for psychological traits.

Many personality traits have been associated with financial behaviour. As revealed by Brown & Taylor (2014), some personality aspects can influence significantly the amount of unsecured debt and financial assets held by British households. Their findings suggest that certain psychological traits are related to financial decision-making.

In this study, after testing if subjective financial knowledge makes a meaningful contri-bution to the decision of saving or not for retirement. We are going to focus on one per-sonality trait, specifically, studying if self-esteem contributes to this decision. This asso-ciation might be relevant, as self-perceived knowledge reflects the part of individual’s level of self- confidence (Tang & Baker, 2016).

(11)

11 Using a sample of undergraduates, their results document a negative relation between compulsive buying behaviour and self-esteem.

Likewise, Tang and Baker (2016) emphasize the relevance of psychological aspects, spe-cifically, the association between self-esteem and financial behaviour. They argue that individuals who possess high self-esteem achieve a higher amount of goals. Their study analyses the direct and indirect influence of self-esteem in financial behaviour, consider-ing the double dimension of knowledge (objective and subjective). Usconsider-ing a representative dataset of US and selecting eight measures reflecting individuals ‘financial behaviour, they state that self-esteem is significantly and positively linked to financial decisions. Hence, high levels of self-esteem can be critical to make a financial decision.

These findings have important implications for enhancing financial educations programs. Financial education courses could be improved by including information about psycho-logical traits.

With this paper I also want to contribute to current literature analysing the implications of personality traits for financial decisions. In particular, I examine the role of self-esteem in relation to financial behaviour due to the possible association between self-perceived knowledge and self-esteem.

2.5.Socioeconomic control variables.

Furthermore, “particular sub-groups of the population have even greater objective finan-cial literacy deficits” (as quoted in Lusardi, Mitchell, & Curto, 2014, p. 349). Women possess greater financially illiteracy compared to men (Agnew & Szykman, 2005; Agnew et al., 2013; Alessie et al., 2011, Lusardi & Mitchell, 2011). Additionally, education is statistically significant when it is included in the multivariate regression, thus, the least educated are also the less financially knowledgeable (Bucher-Koenen et al, 2011; Lusardi & Mitchell, 2011). Moreover, high income households have better financial knowledge than low income households (Agnew & Skyzman, 2005; Alessie et al, 2011; Van Rooij et al., 2009). Finally, as revealed by studies such as Agnew et al (2013), Alessie et al.(2011), Lusardi & Mitchell (2011), Van Rooij et al. (2009), as age increases, financial knowledge raises. Retirement is a distant concept for young respondents, it is not until they approach retirement age that they start worrying about it

3. DATA

3.1.Sample selection

(12)

12 starting from 2007 until 2017, containing a large variety of fields or variables as work, education, income, housing, personality or politics. The LISS data website provides a SPSS or STATA dataset for each topic of the Core Study, a codebook containing the survey questions, a description of the methodology, and some summary statistics. More-over, a variety of other single wave survey modules regarding different topics are availa-ble, which can be combined with the Core Study.

For this study several datasets have been selected. First of all,for measuring the

depend-ent variable saving for retiremdepend-ent, the fifth wave of the study economic situation: assets, has been used. The questionnaire was presented to 6,557 panel members in July 2016 and it was completed by 5,447 respondents (response rate 83.1%).

There are no questions concerning financial literacy in the LISS Core Study. However, in August 2011, LISS panel members were presented questions about the extent to which they possess financial literacy. The full questionnaire consisted of two parts; the first part was about the self-reported score of the respondent’s financial knowledge, and the second part presented the financial questions. The questionnaire was presented to 6,778 panel members, of which 4,857 fully completed the questionnaire (response percentage 71.7%). To capture a measure of self-esteem, the ninth wave of the LISS Core Study of personality characteristics has been selected. The questionnaire was filed in May 2017 and was re-peated for non-respondents in June 2017. It was presented to 7,167 panel members and 6,010 respondents completed the questionnaire (response rate 83.9%).

Regarding control variables, age control variable is already incorporated in thecore study

of economic situation: assets. The remainder control variables such as level of education,

gender and monthly net income are included in single datasets called background

varia-bles. These variables are collected every month using a separate questionnaire. After

merging the different datasets, the final sample consists of 1,872 respondents.

3.2.Dependent variables.

3.2.1. Saving for retirement.

(13)

13 different proxies for savings for retirement. First, I constructed a variable for the natural

logarithm of total amount saved inendowment insurance policies, single-premium

insur-ance policies and life-annuity insurinsur-ance. Secondly, in order to test the robustness of the model, a dummy variable is created, with a value equal to one if the respondent has sup-plemented their defined-benefit pensions with at least one of the products mentioned above and equal to zero if he has not.

3.3.Explanatory variables. 3.3.1. Financial knowledge

As a difference from previous studies, I have taken under consideration the bi-dimen-sional nature of financial knowledge. These two dimensions are objective and subjective financial knowledge. The expertise is related to objective knowledge, whereas the expe-rience and the ability to make decisions is associated to subjective knowledge (Tang & Baker, 2016). This implies that comparing two individuals with the same level of objec-tive financial knowledge but different level of self-perceived knowledge level, it might lead to different financial behaviour outcomes (Hadar, Sood & Fox, 2013).

i) Objective financial knowledge

On the one hand, individual’s actual knowledge, what is called objective financial liter-acy. Previous studies measure objective financial knowledge constructing the basic finan-cial literacy index of Lusardi and Mitchel. This index is normally constructed out of 5 questions. However, the single wave study I used, only provided 4 questions for LISS panel members. These questions covered the following topics: numeracy, inflation, risk and diversification, and the relationship between bond prices and interest rates. In other studies, such as Van Rooij et al. (2011), the last two topics are classified as being indica-tors for advanced financial literacy. Therefore, the financial literacy index I have designed is a mixed of basic and advanced financial knowledge.

Objective financial literacy variable is constructed counting the number of correct an-swers. Then, in order to test interaction effects, two dummy variables are created. One it is equal to one if the respondent has high level of actual knowledge and zero otherwise. The second one is equal to one if the respondent has low actual knowledge and zero oth-erwise. I categorized individuals as having high level of financial knowledge if their over-all score is above the overover-all mean, otherwise they are considered as respondents with low level of financial knowledge (Allgood & Walstad, 2013).

(14)

14

ii) Perceived financial knowledge

On the other hand, what he believes he knows, subjective financial literacy. It is also known as self-reported, self-assessed or perceived financial literacy. The respondents were asked to score their understanding of financial matters, ranging from 1; very poor, to 7; very good. This self-assessed financial literacy score will be served as an index for the subjective financial literacy of the respondent. I propose that respondents who possess higher levels of self-perceived financial literacy will save more for retirement, even after controlling for their actual financial knowledge and socio-economic variables.

Following again, Allgood & Walstad (2013) methodology, in order to check for interac-tion effects, two categorical variables are constructed. If respondents overall rating is above the mean, they are placed as high perceived financial literacy individuals, other-wise, they are classified as low perceived literacy individuals.

We can expect that objective and subjective knowledge would be strongly and positively correlated. However, empirical research has found that these dimensions do not always

coincide (Asaad, 2015; Hadar et al., 2013).For instance, when consumers face complex

product information, their actual knowledge will strengthen meanwhile their perceived knowledge will decrease (Hadar, Sood and Fox, 2013).

3.3.2. Self-esteem

Following Tang & Baker (2016) line of reasoning, “self-esteem not only related to cial behaviour as a positive emotional resource, it also alters the way one views his finan-cial sophistication” (p. 166). To determine individuals level of self-esteem I will rely on LISS survey which perform a Rosenberg self-esteem scale. Respondents have to indicate in what extent they agree or disagree with thirteen statements, where one is equal to totally disagree and seven equals to totally agree. I have reversed the scores of negatively charged statements. Firstly, by adding the scores of all statements, self-esteem variable is created. Then, two more variables are created: If respondents overall rating is above the mean, they are classified as high self-esteem individuals otherwise, they are considered as low self-esteem individuals. High scores indicate a high level of self-esteem of the respondent. In this research, due to the positive effect of high self-esteem on financial behaviour, I will incorporate the variable high level of self-esteem. For the exact wording of self-esteem statements, the appendix can be checked.

3.4.Control variables.

(15)

15 2005; Agnew et al., 2013; Alessie et al., 2011; Bucher-Koenen et al, 2011; Lusardi & Mitchell, 2011; Van Rooij et al., 2009). Younger respondents, women and respondents with less level of education exhibit superior financial illiteracy and they save less for retirement. For the construction of socio-economic control variables, the following vari-ables are established:

i. Age. The age of the respondent is included as a control variable. According to Van Rooij et al. (2009), there are three age groups in which respondents are

ex-pected to make the greatest financial decisions. For this reason, these three age

categories are constructed: 30-40 years, 40-50 years and 50-60 year, and they are incorporated as control variables.

ii. Education is included as a control variable due to the positive influence on finan-cial behaviour demonstrated by general literature. In the Netherlands, the highest levels of educations are HBO (vocational college) and WO (university). The in-fluence of these two categories and the lowest level of education (primary school) are examined. A dummy variable is created for each category.

iii. One dummy variable is constructed in order to take into account gender effect, one is defined for female and zero for male.

iv. Income.A variable for the natural logarithm of the net monthly income on a per-sonal level is created.

3.5.Summary statistics.

The descriptive statistics on the dependent variables and on the independent variables (including the control variables) are reported in Table 1. Out of the 1,872 respondents, only 129 (6.94%) have supplemented their defined-benefit pensions with endowment in-surance policies, single-premium inin-surance policies or life-annuity inin-surance. This comes into agreement with Van Rooij et al. (2009) results, where only 12.9% of the respondents thought a lot about retirement in the Netherlands.

The objective financial literacy measure averaged at 2.39 correct answers out of 4. 46.85 percent of the respondents were classified as having high objective financial literacy and 53.15% as low objective financial literacy. Meanwhile, the mean subjective financial lit-eracy score is 4.94 out of 7. The percentage of respondents with high perceived knowledge is equal to 69.28%, while with low perceived knowledge is equal to 30.72%. Therefore, the percentage of respondents with high perceived knowledge is relatively

high compared with the one with high actual knowledge.This comes in agreement with

part of the literature such as Lusardi & Mitchell (2014), who state that respondents tend

(16)

16 out of 84. Out of 1,872 respondents, 54.91% possess high self-esteem, whilst 45,09% have low self-esteem.

The final sample includes slightly more females (50.53%) than males (49.47%). The av-erage age of the participants is 57.66, ranging from 20 to 91 years old. The avav-erage monthly net income of the respondents is € 1,613.681

(17)

17 Table 1: Summary statistics

This table reports the descriptive statistics of the variables, which include the number of observations, the mean values, standard deviation, the maximum and minimum values.

Nº observations Mean Std. Dev. Min Max

Dependent varia-bles

Save for re-tirement

129 6.94% 0.2542 0 1

(18)

18 4. METHODOLOGY.

The aim of this study is to relate an individual’s financial knowledge and self-esteem with their amount saved and their propensity to save for retirement. This study tests four hy-potheses.

4.1.Hypotheses.

i) Hypothesis 1 (H1): Objective financial literacy affects positively and

signifi-cantly individuals’ propensity to save and amount saved for retirement. In this section an ordinary least squared (OLS) regression is estimated, setting the natural logarithm of the total amount saved in retirement products as dependent variable. An OLS is a proper method in this case because the dependent variable is a continuous variable. In order to test the influence of actual financial knowledge on savings for retirement, a simple OLS regression of objective financial literacy and socioeconomic controls is im-plemented. Following the line of reasoning provided by current literature, more finan-cially knowledgeable respondents are more likely to save for retirement. In order to con-firm this, I start performing a multivariate analysis of the relationship between saving for retirement and objective financial knowledge. Consistent with prior literature, each equa-tion control for socio-economic variables: age, educaequa-tion, gender and income. These con-trol variables are included in order “to take into account individual heterogeneity that might affect the literacy-planning relationship” (Van Rooij et al., 2009, p.13).

The baseline model is presented below:

𝑆𝑎𝑣𝑖𝑛𝑔 𝑓𝑜𝑟 𝑟𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡𝑖 = 𝛽1+ 𝛽2𝐴𝑐𝑡𝑢𝑎𝑙 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 + 𝛽3𝐼𝑛𝑐𝑜𝑚𝑒 +

𝛽₄𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛽₅𝐴𝑔𝑒3040 + 𝛽₆𝐴𝑔𝑒4050 + 𝛽₇𝐴𝑔𝑒5060 + 𝛽₈𝑃𝑟𝑖𝑚𝑎𝑟𝑦𝑠𝑐ℎ𝑜𝑜𝑙 +

𝛽₉𝐻𝐵𝑂 + 𝛽₁₀𝑊𝑂 + 𝜀_𝑖 (1)

Where the dependent variable is the natural logarithm of the total amount saved in retire-ment products (endowretire-ment insurance policies, single-premium insurance policies and life-annuity insurance). 𝛽1 is the constant term, followed by the independent variable: actual financial knowledge. Then, eight control variables are included: natural logarithm of monthly income, gender, dummy age between 30 and 40 years, dummy for age be-tween 40 and 50, dummy for age bebe-tween 50 and 60, primary education, higher vocational

education and university education, respectively and εi is the error term. This model will

(19)

19

ii) Hypothesis 2 (H2): Subjective financial literacy predicts individual’s

propen-sity to save for retirement.

Then, it is fundamental to understand the extension of the concept, financial literacy. In order to do so, another multivariate regression is estimated, in this case using self-per-ceived financial knowledge as explanatory variable.

𝑆𝑎𝑣𝑖𝑛𝑔 𝑓𝑜𝑟 𝑟𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡_𝑖 = 𝛽₁+ 𝛽₂𝑃𝑒𝑟𝑐𝑒𝑖𝑣𝑒𝑑 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 + 𝛽₃𝐼𝑛𝑐𝑜𝑚𝑒 +

𝛽9𝐻𝐵𝑂 + 𝛽10𝑊𝑂 + 𝜀𝑖 (2)

On the left side of the equation the dependent variable is presented: natural logarithm of the total amount saved in retirement products (endowment insurance policies, single-pre-mium insurance policies and life-annuity insurance). The right side of the equation starts with the constant term, followed by the explanatory variable: perceived financial knowledge. After that, the eight control variables mentioned in the baseline model are included. The final term is the error term.

Finally, both types of knowledge are included as explanatory variables, controlling for socioeconomic variables in the last multivariate regression.

𝑆𝑎𝑣𝑖𝑛𝑔 𝑓𝑜𝑟 𝑟𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡𝑖

= 𝛽₁ + 𝛽₂𝑃𝑒𝑟𝑐𝑒𝑖𝑣𝑒𝑑 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 + 𝛽₃𝐴𝑐𝑡𝑢𝑎𝑙 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒

+ 𝛽4𝐼𝑛𝑐𝑜𝑚𝑒 + 𝛽5𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛽6𝐴𝑔𝑒3040 + 𝛽7𝐴𝑔𝑒4050 + 𝛽8𝐴𝑔𝑒5060

+ 𝛽₉𝑃𝑟𝑖𝑚𝑎𝑟𝑦𝑠𝑐ℎ𝑜𝑜𝑙 + 𝛽₁₀𝐻𝐵𝑂 + 𝛽₁₁𝑊𝑂 + 𝜀_𝑖

iii) Hypothesis 3 (H3): Financial confidence interacting with objective financial

literacy exerts a positive influence on retirement savings.

We are going to explore if subjective financial knowledge interacting with actual knowledge affects individual propensity to save for retirement. Two different models are estimated with its respective probit model.

As a first step, I construct an OLS model that only incorporates as explanatory variable the simple interaction term between both knowledge dimensions and controls for socio-economic factors. The respective probit model is also estimated in order to check the robustness of the model.

(20)

20 literacy and low actual financial literacy with the last combination serving as omitted term.

According to Allgood & Walstad (2016) “if perceived and actual financial literacy are useful and valid measure, then by extension the interaction between the two is especially worthy of study for capturing the wide range of individual differences”, p.5. Following this line of reasoning, in order to understand better the contribution of financial knowledge on the propensity to save for retirement, the interaction between the different levels of objective and subjective financial literacy is analysed.

As a first step, the overall effect of the interaction of both knowledge is investigated. The comparison between both dimensions high and both low provides the estimate of the combined effects of both dimensions on savings for retirement as they change in one direction.

Then, in order to assess the separate impact of perceived or actual knowledge in explain-ing the amount saved for retirement, two comparisons are made. First, to evaluate the relative contribution of actual knowledge on the amount saved for retirement, we compare group one (both knowledge high) with group two (just perceived knowledge high). To test if the coefficients are equal for both groups a Wald test is performed.

In the case of perceived financial knowledge, we compare group one (both knowledge high) with group three (just actual knowledge high), objective financial knowledge is held at a high level meanwhile perceived knowledge varies from high to low. Another Wald test is implemented.

iv) Hypothesis 4 (H4): High self-esteem influences positively on individual’s

re-tirement savings.

In this hypothesis, I will evaluate the influence of self-esteem on planning for retirement behaviour. I propose that high self-esteem is positively related with how much an indi-vidual save for retirement.

As already done during the first and second hypotheses, I perform a simple multivariate analysis, in this case the explanatory variable is high self-esteem, conditional on the eight socioeconomic control variables.

𝑆𝑎𝑣𝑖𝑛𝑔 𝑓𝑜𝑟 𝑟𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡_𝑖 = 𝛽₁+ 𝛽₂𝐻𝑖𝑔ℎ 𝑠𝑒𝑙𝑓𝑒𝑠𝑡𝑒𝑒𝑚 + 𝛽₃𝐼𝑛𝑐𝑜𝑚𝑒 +

𝛽9𝐻𝐵𝑂 + 𝛽10𝑊𝑂 + 𝜀𝑖 (1)

(21)

21 𝑆𝑎𝑣𝑖𝑛𝑔 𝑓𝑜𝑟 𝑟𝑒𝑡𝑖𝑟𝑒𝑚𝑒𝑛𝑡_𝑖 = 𝛽₁+ 𝛽₂𝑆𝑒𝑙𝑓𝑒𝑠𝑡𝑒𝑒𝑚 + 𝛽₃𝑃𝑒𝑟𝑐𝑒𝑖𝑣𝑒𝑑 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 + 𝛽₄𝐴𝑐𝑡𝑢𝑎𝑙 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 + 𝛽₅𝐼𝑛𝑐𝑜𝑚𝑒 + 𝛽₆𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛽₇𝐴𝑔𝑒3040 + 𝛽8𝐴𝑔𝑒4050 + 𝛽9𝐴𝑔𝑒5060 + 𝛽10𝑃𝑟𝑖𝑚𝑎𝑟𝑦𝑠𝑐ℎ𝑜𝑜𝑙 + 𝛽11𝐻𝐵𝑂 + 𝛽₁₂𝑊𝑂 + 𝜀_𝑖 4.2.Robustness checks.

After having performed all these regressions, in order to verify the influence of financial

literacy on savings for retirement, we should carry out some robustness tests.We build a

model that allows us to evaluate the impact of the different dimensions of financial knowledge on a binary outcome, such as saving or not for retirement (whether a house-hold supplement or not its retirement plan with one of the products mentioned above). A binary model is more suitable. “Logit and probit models are able to overcome the limita-tion of the limited dependent variable that can produce estimated probabilities that are negative or greater than one” (Brooks, 2004, p.514). Both models will provide compara-ble characterisations of the data, thus, the election of the model is arbitrage. This thesis uses a probit model in order to test the robustness of each model.

The dependent variable is a dummy variable equal to one if the household supplements his defined-benefit pension with endowment insurance policies, single-premium insur-ance policies or life-annuity insurinsur-ance, otherwise is equal to 0. In each model the eight control variables are included: natural logarithm of monthly income, gender, dummy age between 30 and 40 years, dummy for age between 40 and 50, dummy for age between 50 and 60, primary education, higher vocational education and university education.

For the first hypothesis, the independent variable is actual knowledge index. Then, we change the explanatory variable for perceived knowledge and eventually, we incorporate both dimensions as I have already done in the OLS models. The necessary adjustments in the explanatory variables will be implemented to estimate the correspondent probit model to check robustness in the rest of the models.

Subsequently, in order to analysis the robustness of the models constructed in the third hypothesis, we shift the explanatory variables for the simple interaction term between actual and self-perceived knowledge and then for the four combined financial knowledge variables.

(22)

22 Finally, the correspondent probit models are estimated to test the consistency of self-es-teem models.

The results of this study are reported in next section. 5. RESULTS.

5.1.Correlations

Table 2 shows the correlation matrix. For each variable the Pearson correlation coefficient is displayed. It provides some initial insight about the relationship between the variables. Correlations between the dependent variable and the explanatory variables are positive and significant. Thus, more financially knowledgeable (actual and perceived) respondents and respondents with high self-esteem save more for retirement However, all correlations seem to be weak with coefficients not larger than 0.3.

Correlations between socioeconomic controls and the dependent variable are significant, except from age. Therefore, the covariates seem to be of influence in explaining financial behaviour. All of them are positive correlations, except from the negative association be-tween female variable and saving for retirement. Gender correlates negatively and signif-icantly with savings for retirement, objective knowledge and subjective knowledge (-0.0908, -0.2802, -0.1722). In line with the results of other studies such as Alessie et al. (2011) & Lusardi & Mitchell (2011) women display much lower knowledge than men, and their propensity to save for retirement is lower.

Regarding the correlation between the independent variables, compared to the other co-efficients, the association between self-esteem and subjective knowledge and objective and subjective knowledge are significant and the strongest. So, at first glance, although weak, there seems to be a positive association between self-esteem and subjective finan-cial knowledge (0.2075). Furthermore, correlation between subjective knowledge and ob-jective knowledge is significant and positive but not strong (0.2369), as already shown in Agnew and Skyzkman (2005). Contrary to the strong correlations presented in Lusardi & Mitchell (2007) & Van Rooij et al. (2007). As a consequence, actual financial might not

be an adequate proxy for self-perceived knowledge.There are no correlations higher than

(23)

23 Table 2: Correlation matrix between variables in the regression

*, represents significance level at 5 percent

Variable Saving for

retirement Objective financial knowledge Perceived knowledge

Self-esteem Age Income Female College

educa-tion

Saving for retirement 1 0.1672* 0.0695* 0.0745* 0.008 0.1742* -0.0908* 0.1050*

(24)

24

5.2.Regression analysis

Tables 3,4,5 and 6 present the results of the probit and OLS regression for the explanatory variables.

5.2.1. Financial knowledge and saving for retirement.

In this section, I examine the relationship between financial knowledge dimensions and savings for retirement. Two models are used: an OLS model displayed in the first column,

and a probit model presented in the second column. The results are reportedin the

fol-lowing paragraphs.

Firstly, Table 3 presents the results from the OLS and probit analysis made in order to estimate the separate effect of self-perceived knowledge and actual knowledge respec-tively, controlling for socioeconomic variables on the savings for retirement of Dutch households. As can be seen in Panel 3A, objective financial knowledge has a strongly positive influence on saving for retirement. More financially knowledgeable respondents are more likely to save for retirement and the amount saved is higher. These results sup-port our first hypothesis; thus, objective financial literacy affects positively and signifi-cantly individuals’ propensity to save and amount saved for retirement.

Meanwhile, panel 3B presents the results related to perceived knowledge on saving for retirement. Both regressions also show a positive effect for self-perceived knowledge. These findings confirm our second hypothesis, which states that subjective financial lit-eracy predicts individual’s propensity to save for retirement.

After that, we estimate a new model including both financial knowledge dimensions. The results are displayed in table 4. Table 4 reveals that the inclusion of self-assessed financial knowledge does not make actual knowledge insignificant. In both models, actual knowledge is still significantly related to saving for retirement, thus, a higher level of financial knowledge will lead to an increase the amount saving for retirement. One stand-ard deviation higher of actual financial literacy is associated with 30.06 percentage points higher probability of saving for retirement.

(25)

25 the likelihood of saving for retirement. At a first sight, actual and perceived financial knowledge appear to explain the propensity to save for retirement.

Although perceived knowledge and actual knowledge show great influence on saving for retirement, as revealed in table 4, there are some control variables that influence the amount saved and the propensity to save for retirement. The effect of gender on savings for retirement is significant and negative in the OLS analysis and in the probit regression including only perceived knowledge. This implies that women are less likely to save for retirement than men. However, this effect is insignificant in the remainder probit models. With regards to age, according to OLS results, young people (aged between 30 to 40) are less likely to be savers. Nevertheless, marginal effects are insignificant in all probit mod-els. The range of age between 40 to 50 years old is insignificant. Meanwhile, older re-spondents (aged between 50 to 60) are more likely to increase their savings for retirement. Furthermore, concerning the level of education, primary level is insignificant. However, higher vocational education and university education have a positive and significant in-fluence on savings for retirement. This implies that highly educated respondents are more likely to save for retirement. The role of having university education exerts the greatest influence on the dependent variable. Regarding income, it has a positive impact, but it is insignificant.

(26)

26 Table 3: Baseline regression results of each financial knowledge with saving for retirement.

NOTE: Panel 3A displays the results for the relationship between objective knowledge and savings for retirement. While, panel 3B presents the results for the relationship between self-perceived knowledge and savings for retirement. This table presents four models. First, in column one and three, an OLS model is presented for the dependent variable: natural logarithm of the total amount saved in retirement products. In the second and fourth column, a probit model is presented with the marginal effects for a binary dependent variable of saving for retirement or not. Each model is calculated on the set of socioeconomic control variables. *** indicates statistical significance at 1% level, ** at 5% level and * at 10% level. Standard errors are presented in parentheses.

Panel 3A Panel 3B

OLS Probit model OLS Probit model

Variables Natural logarithm of the total amount saved in retirement products

Dummy variable of sav-ing for retirement

(27)

27 Table 4. Relationship between financial knowledge and saving for retirement.

This table presents two models. First, in column one, the OLS model is presented for the dependent variable: natural logarithm of the total amount saved in retirement products. In the second column, a probit model is presented for a binary dependent variable of saving for retirement or not. Each model is calculated on the set of socioeconomic control variables.

OLS Probit model

Variables Natural logarithm of the

total amount saved in re-tirement products

Dummy variable of sav-ing for retirement

Objective knowledge 0.3006*** (0.0646) 0.0322*** (0.0053) Perceived knowledge 0.1187** (0.0490) 0.0109** (0.0072) Gender effect -0.2757** (0.1283) -0.0192 (0.0131) Ln income 0.0304 (0.0309) 0.0050 (0.0038) Primary school -0.0325 (0.3426) -0.0265 (0.0500) Higher vocational

educa-tion 0.2692* (0.1375) 0.0234* (0.0132) University 0.5531*** (0.1890) 0.0398** (0.0165) Dummy age 30-40 -0.3684* (0.2170) -0.0354 (0.0248) Dummy age 40-50 0.1968 (0.1736) 0.0133 (0.0168) Dummy age 50-60 0.5396*** (0.1528) 0.0417*** (0.0139) Constant -0.9237 (0.3513) Observations (N) 1,855 1,856 Adjusted R-Squared 0.0462 0.0987

(28)

28

5.2.2. Interaction between actual and perceived knowledge.

In the previous section we have examined separately and then together, the relationship between both financial knowledge dimensions and saving for retirement. In this section we observe the association between the interaction terms of financial knowledge dimen-sions and savings for retirement. Table 5 displays the interaction effects of knowledge dimensions on the amount saved and the propensity to save for retirement. Panel 5A pre-sents the results for the OLS and probit regressions of the interaction effects between the different levels (high or low) of self-perceived financial knowledge with the different levels (high or low) of actual financial knowledge on the amount saved for retirement and the probability that a person saves for retirement, respectively.

The comparison between group one (both dimensions high) and four (both dimensions low) presents the total effect of both dimensions on saving for retirement (Allgood & Walstad, 2013). Panel 5A reports that respondents with both knowledge dimensions high are more likely than respondents with both knowledge low to save for retirement. To analyse the effect of a change in perceived knowledge, actual knowledge is held at a high level and the level of self-perceived knowledge varies from high to low (group one and three). The comparison between this two groups is significant, and it reveals that financial confidence can enhance objective knowledge. Therefore, a change in perceived knowledge has a significant influence on the amount and the probability to save for re-tirement.

Regarding the influence of a variation in the level of actual knowledge, we compare group one and two, which keeps perceived at high level and lets actual knowledge changes from high to low. An alteration in the level of actual financial knowledge has a significant effect on the propensity that the respondent saves and the amount saved for retirement. On the whole, changing the level of any knowledge dimensions will affect the likelihood of saving and the amount saved for retirement.

Although the interaction between both financial knowledge dimensions at a high level displays the greatest effect on savings for retirement, socioeconomic control variables also affect this behaviour in the same way that in the previous section. The role of gender is significant and negative, women are less likely to save for retirement compared to men. The results also show that primary school, the lowest education category is not significant on saving for retirement. However, being a higher vocational graduated or university graduated increases the likelihood of saving for retirement. With regards to age, as is to be expected, adults age 50 to 60 are more likely to save for retirement.

(29)

29 knowledge dimensions and controls for socioeconomic factors. As panel 5B reveals, the simple interaction term between both knowledge is positive and significant. Second col-umn of this panel reports the marginal effects from the probit regression. The marginal effect for the interaction term is significant but modest (0.0055). If we compare the mar-ginal effect of the simple interaction term (0.0055) with the marmar-ginal effect of the com-posite variable of both knowledge high (0.0563), we can see that the last one is much larger. Therefore, this implies that when both knowledge are high, the probability to save for retirement increases.

(30)

30 Table 5: Interaction effects of knowledge on the propensity to save for retirement.

This table presents the results for the interaction effects of financial knowledge dimensions on the propensity to save for retirement. Panel 5A displays the results for an OLS model and probit model for savings for retirement. It reports the results for the interaction between the different levels and dimensions of knowledge on the propensity to save for retirement. Mean-while, panel 5B presents the results for an OLS and a probit model of the relationship between the simple interaction knowledge term and saving for retirement, controlling for socioeconomic variables.

OLS Probit model

Variables Natural logarithm of the total amount saved in retirement prod-ucts Dummy variable of saving for retirement

Variables Natural logarithm of the total amount saved in retirement products Dummy variable of saving for retirement (1) Both knowledge high 0.6421*** (0.1692) 0.0563*** (0.0181) Interaction between both knowledge 0.0617*** (0.0099) 0.0055*** (0.0010) (2) Perceived high

& Actual low

0.0692 (0.1648)

0.0072 (0.0194) (3) Perceived low &

Actual high 0.0935 (0.2303) 0.0135 (0.0251) Ln income 0.0409 (0.188) 0.0056 (0.0386) Ln income 0.0306 (0.0310) 0.0051 (0.0038) Gender -0.3034** (0.1291) -0.0269** (0.0131) Gender -0.2521* (0.1284) -0.0192 (0.0131) Primary school -0.8078 (0.3438) -0.0370 (0.0513) Primary school -0.0274 (0.3417) -0.0280 (0.0498) Higher vocational education 0.2888** (0.1380) 0.0268** (0.0133)

Higher vocational edu-cation 0.2598* (0.1369) 0.0238* (0.0131) University 0.6115*** (0.1884) 0.0478*** (0.0165) University 0.5419*** (0.1873) 0.0405** (0.0164) Dummy age 30-40 -0.3975* (0.2180) -0.0393 (0.0251) Dummy age 30-40 -0.3587* (0.2166) -0.0351 (0.0248) Dummy age 40-50 0.1777 (0.1742) 0.0131 (0.0169) Dummy age 40-50 0.1937 (0.1733) 0.0135 (0.0168) Dummy age 50-60 0.5283*** (0.1535) 0.0422*** (0.0140) Dummy age 50-60 0.5351*** (0.1525) 0.0416*** (0.0138) Constant 0.0484 (0.2728) Constant -0.3751 (0.2709) Observations (N) 1,858 1,858 Observations 1,856 1,856 Pseudo R-Squared 0.0392 0.0825 Pseudo R-Squared 0.0494 0.0987

Wald Tests

Groups 1 & 2 14.40*** 10.66*** Groups 1 & 3 6.59** 4.09**

(31)

31

5.2.3. Self- esteem and saving for retirement.

To examine further, the association between self-esteem and saving for retirement is an-alysed. As already done in the preceding sections, two models are conducted. In addition to subjective financial knowledge, self-esteem might assert a direct effect on savings for retirement.

As a first step, the influence of high self-esteem on savings for retirement controlling for socioeconomic factors is presented in panel 6A. High self-esteem has a strongly positive

significant effect on saving for retirement in OLS and probit models (0.4274,0.0403

re-spectively). This means that respondents with high self-esteem are more likely to save for retirement and they save more for retirement.

The sign and significance of all control variables are equivalent in both models. As ex-pected, gender is negatively and significantly related to the decision of saving for retire-ment. The role of monthly net income is insignificant. Concerning the level of education, primary school variable is not significant. However, the highest level of education (higher vocational education and University) exerts a positive and significant influence on the decision of saving for retirement. Finally, older respondents (aged between 50 to 60) are more prone to save for retirement.

Furthermore, Panel 6B reports the results of the influence of both knowledge dimensions and high self-esteem on the amount saved and the propensity to save for retirement. With the inclusion of high self-esteem, high objective knowledge continues affecting signifi-cantly and positively (0.2961) the dependent variable. The second column of panel 6B reports the estimates of the marginal effects and the corresponding standard errors of the probit model. An increase of one in actual knowledge leads to 3.15 percent increase in the probability of saving for retirement. Higher marginal effect in the same direction has also been found for high self-esteem, a unit change in self-esteem leads to 3.25 percent increase in the probability of saving for retirement. Both models show that high self-esteem is significant and slightly positively related to the decision to supplement or not the retirement plan. However, self-perceived knowledge is only significant at 95% confi-dence level in the OLS model and it becomes insignificant in the probit model.

On the whole, the influence of high self-esteem is slightly higher than actual knowledge in both models. Therefore, high self-esteem exerts the greatest influence on savings for retirement and perceived knowledge seems to play a minor or no role when self-esteem is included in the model.

(32)

32 in the probit model. Both models give significant results for the relationship between university education and saving for retirement This implies that people with high educa-tion are more prone to save for retirement. Concerning age, the only significance variable is age between 50 and 60 years, hence, the relation between age and saving for retirement strengthens with age.

Summarizing, the results are in line with the expectations set in the fourth hypothesis. In other words, high self-esteem exerts a positive and significant influence on individual’s retirement savings.

(33)

33 Table 6: Relationship between self-esteem and saving for retirement.

OLS Probit model OLS Probit model

Dummy variable of sav-ing for retirement

Dummy variable of sav-ing for retirement High self-esteem 0.4274*** (0.1193) 0.0403*** (0.0124) Objective knowledge 0.2961*** (0.0645) 0.0315 *** (0.0072) Gender -0.4407*** (0.1253) -0.0391*** (0.0127) Perceived knowledge 0.0929* (0.0497) 0.0087 (0.0072) Ln income 0.0445 (0.0311) 0.0056 (0.0382) High self-esteem 0.3490*** (0.1208) 0.0325*** (0.0125) Primary school -0.1505 (0.3431) -0.0435 (0.0513) Gender -0.2811** (0.1280) -0.0195 (0.0131) Higher vocational education 0.3526** (0.1365) 0.0324** (0.0132) Ln income 0.0252 (0.1374) 0.0040 (0.0038) University 0.7134*** (0.3433) 0.0570*** (0.0163) Primary school 0.0131 (0.3423) -0.0183 (0.0494) Dummy age 30-40 -0.3398 (0.2187) -0.0316 (0.0251) Higher vocational education 0.2528* (0.1374) 0.0217 (0.0132) Dummy age 40-50 0.2194 (0.1745) 0.0191 (0.0169) University 0.5267*** (0.1889) 0.0382** (0.0164) Dummy age 50-60 0.5472*** (0.1535) 0.0455*** (0.0139) Dummy age 30-40 -0.3131 (0.2174) -0.0278 (0.0246) Constant 0.0844 (0.2570) Dummy age 40-50 0.2169 (0.1734) 0.0159 (0.0168) Dummy age 50-60 0.5344*** (0.1525) 0.0420*** (0.0138) Constant -0.9393*** (0.2687) Observations 1,857 1,858 Observations 1,855 1,856 Adjusted R-squared 0.0362 0.0763 Adjusted R-squared 0.0500 0.1062

(34)

34 6. DISCUSSION OF THE RESULTS AND LIMITATIONS.

6.1.Discussion of the results.

6.1.1. Influence of actual knowledge on saving for retirement.

The baseline results indicate an important and significant influence of actual financial knowledge on savings for retirement. These results are aligned with current literature, more financially literate respondents have a higher propensity to plan for retirement (Alessie et al., 2011; Bucher-Koenen et al., 2011, Chan and Stevens, 2008; Lusardi and Mitchel 2006, 2007, 2011; Van Rooij et al., 2009).

However, in contrast to this current literature, this effect is influential but insufficient on the decision of saving for retirement. In order to enhance this influence, self-perceived knowledge must be incorporated in financial knowledge scope. This comes into agree-ment with Allgood & Walstad (2016) results, which states that objective financial knowledge affects financial behaviour, but the effect of the combination of self-perceived with actual knowledge is greater than objective knowledge itself. Moreover, correlation between both knowledge dimensions is positive, significant but not strong (in line with Agnew & Szykman, 2005). In contrast to the strong correlations presented by Lusardi & Mitchell, 2007 & Van Rooij et al. (2007) between both knowledge dimensions. Therefore, actual knowledge might not be an adequate proxy for self-perceived knowledge.

The impact of other variables apart from actual knowledge can be a possible explanation of why part of financial education programs are not effective (Asaad, 2015; Fernandes et al., 2014; Hadar et al., 2013) This paper suggests moving away from just making use of

objective measures to also include self-assessed measures to the analysis.

6.1.2. The effect of self-perceived knowledge on saving for retirement.

First of all, descriptive statistics indicate that the percentage of respondents with high perceived knowledge is relatively high compared to the percentage of high actual knowl-edgeable respondents. Hence, respondents overestimate their financial knowledge (in line with Lusardi & Mitchell, 2014).

Our analysis reports mixed results regarding the association between self-perceived knowledge and savings for retirement.

Before including self-esteem in our models, subjective knowledge is positive and signif-icant in both models. This denotes a signifsignif-icant positive association between self-per-ceived knowledge and saving for retirement. Hence, if the respondent believes that his/her

knowledge is high, the propensity to save for retirement is higher.This in line with Parker

(35)

35 behaviour, in our particular case, saving more for retirement. A possible explanation of these results is that actual financial knowledge does not entirely capture individual´s knowledge or that self-perceived knowledge provides a wider perspective in order to make financial decisions (Allgood & Walstad, 2013). Nevertheless, the effect of self-perceived literacy on savings is not as large as the effect of objective financial literacy, as a matter of fact, it is smaller (it differs from Asaad, 2015 results, which states that self-perceived impact on financial behaviour is a large as actual knowledge impact).

Contrary to our expectations, when self-esteem is included in the model, in the OLS model, subjective knowledge is only significant at 95% confidence level and no effects are found of subjective knowledge on the propensity to save for retirement in the probit model. Therefore, self-esteem might play a critical role in self-perceived knowledge. De-spite the insignificance of self-perceived variable, one needs to keep in mind the possible association between self-perceived knowledge and self-esteem. The significant and pos-itive correlation between self-esteem and perceived knowledge might indicate the pres-ence of a relationship between self-perceived knowledge and self-esteem. One could state that self-esteem is a driving factor of an individual’s self-perception and therefore of one’s perception of his or her subjective financial knowledge (Hadar et al., 2013; Parker et al.,2012; Tang & Baker, 2016).

6.1.3. Interaction between actual and self-perceived knowledge.

In relation to the interaction between objective and subjective knowledge, in order to un-derstand better the contribution of financial knowledge on saving for retirement, the model constructed following Allgood & Walstad (2016) might be more appropriate than the one that only includes the simple interaction term. With this model we can analyse the interaction between different levels of objective and subjective knowledge.

The results reveal that the combined effect of subjective and objective knowledge exerts a positive influence on the propensity to save for retirement. What these results suggest is that when both dimensions are high is more likely that the respondent will save for retirement. Moreover, altering the level of one of the dimensions will affect the likelihood of saving for retirement. Therefore, a financial knowledge index that only takes into con-sideration the objective dimension will leave out the positive effect of subjective knowledge on the propensity to save for retirement.

THE INFLUENCE OF FINANCIAL KNOWLEDGE AND SELF- PERCEPTION ON THE PROPENSITY TO SAVE FOR RETIRE- MENT.