Behavioral biases in individual investment decision making : the role of overconfidence

(1)

Behavioral biases in individual investment decision

making: the role of overconfidence

Jelmer Michiel Thomas Reijerink 11749709

Supervisor: prof. dr. J.H. Sonnemans Second examiner: dr. J.B. Engelmann

University of Amsterdam Amsterdam School of Economics

MSc Economics, Behavioral Economics and Game Theory July 11, 2018

Abstract

The present study examines the variation in investment performance among individuals and elaborates on the role of overconfidence as explaining variable. Data from the DNB Household Survey is used to compute both individual overconfidence and investment performance. For overconfidence, a measure of absolute overconfidence is computed based on individuals’ financial literacy, whereas investment performance is proxied by the variance of the changes of individuals’ level of wealth. Following, a mediation analysis is conducted in order to report whether the differences in performance among individuals can be explained by the corresponding levels of overconfidence. In line with existing literature, it is found that several personal characteristics do significantly influence an individual’s level of overconfidence. Moreover, the results show that overconfidence acts as a mediator, i.e. the differences in investment performance among individuals can be (partially) explained by the overconfidence bias. These findings provide good grounds to tackle the implications of overconfidence as a behavioral bias.

Keywords: Individual decision making, investment behavior, overconfidence JEL Classification: D91, G41

(2)

Statement of Originality

This document is written by Jelmer Michiel Thomas Reijerink who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1 Introduction

In traditional economics, an individual is assumed to make perfectly rational decisions and therefore maximize utility. In both the fields of finance and economics, however, there is evidence of irrationality in individual decision making. For example, in economics, human decision making consistently violates the axioms of decision theory (Gintis,2000). This im-plies that individuals do not always act rationally. In finance, the modern portfolio theory (Markowitz,1952) forms the foundation for asset pricing models. This modern portfolio theory is based on the assumption that, among others, investors are rational. Evidence on the irrationality of investors is reflected in, for example, under-diversification in port-folio decisions (van Horne, Blume, & Friend,1975), which has led to the development of behavioral models. Economists have tried to come up with explanations for these obser-vations in the past decades. One main explanation of irrationality is the overconfidence bias (Kahneman & Tversky,1974). Overconfident investors are investors who believe that their knowledge about the value of a certain financial product is greater than it actually is (Odean,1998). Both in literature and experiments, it is shown that overconfident indi-vidual investors trade more than rationally would be expected. This raises the question whether overconfidence plays a bigger role than has been suggested in current literature and, therefore, more thought should be put into this issue.

Where most studies focus on overconfident behavior of executive directors or behav-ior at firm level, there is little convincing evidence of the individual overconfidence bias leading to worse performance on the financial markets; it appears that the majority of the studies is focused on the effects of overconfidence on investment behavior, but a direct association with regard to actual performance is missing. To the best of my knowledge, there is no study where the direct effects of overconfidence on investing performance are analyzed. This study therefore concentrates on investment performance in terms of ul-timate monetary results, as this perfectly reflects the achievements of individuals that bought risky assets. Furthermore, there is not much known about the characteristics of overconfident individual actors on the financial markets themselves. In the present study study, the goal is to find out whether differences in investment performance can be, at least partially, explained by the extent to which individuals are overconfident. This indirect re-lationship between an individual’s characteristics and the investment performance through

(5)

overconfidence has not been researched before. Hence, this study adds to the literature by analyzing whether overconfidence can explain the differences in investment performance of individuals. This analysis is important, both in terms of academic perspective and from an economic and political point of view. First, it is important to see if and how over-confidence directly affects investing behavior and returns, in order to identify what the economic results of overconfidence are. Additionally, overconfidence can be exploited, as credit card companies may try and exploit consumers who overestimate their ability to pay back on time (Malmendier & Taylor, 2015), and overconfident CEOs are less likely to engage in corporate social responsibility (McCarty, Oliver, & Song, 2017). Moreover, in financial markets overconfident investors might overestimate the value of certain risky assets which in turn might lead to a stock market bubble (Scheinkman & Xiong, 2003). As overconfidence leads to suboptimal decisions, it is therefore interesting to look at the individual characteristics that influence overconfidence. Finally, it is likely that investor behavior affects what happens in markets. That is, it would be inaccurate to model market behavior without taking these psychological factors into account.

In order to research the effects that overconfidence has on investment performance, data from the DNB Household Survey (DHS) is used. This dataset provides the opportunity to look at the effects of individual characteristics on investment performance. By constructing a variable for overconfidence, the direct effects of overconfidence on investment performance will be analyzed. A mediation analysis will be conducted in order to find the extent to which overconfidence plays a role in determining investment performance. The results show that there is no direct effect of both the individual characteristics and overconfidence on investment performance if this performance is defined as the change in wealth of an individual. However, if an additional measure for performance in terms of variance of change in net wealth is analyzed, there appear to be significant effects. Also, it appears that overconfidence in this relation acts as a mediator variable.

The structure of the rest of this paper is as follows. First, an overview of existing liter-ature will be provided, upon which the research question and hypotheses are constructed. Accordingly, the summary statistics are evaluated and necessary modifications with regard to the dataset will be conducted. Following, a research method is established and subse-quently the results will be presented and discussed in detail. Additionally, the limitations

(6)

and caveats of this study are discussed. Finally, the implications of the results of this study will be discussed and proposals for further research are presented.

2 Literature review

In this section, an overview of the literature is given. First, the main literature about personal factors of individual investors will be discussed. Additionally, the most impor-tant literature about overconfidence is covered. After a clear definition of overconfidence according to relevant previous research is stated, the relationships between overconfidence and both personal factors of an individual and his or her investment performance is re-viewed. Subsequently, the aim is to make up expectations of this study and to set up a clear research question; this will be covered in the last part of this chapter.

2.1 Heterogeneity of investment performance

While there is much available research about consumer expenditures, there is little empir-ical research concerning individual investment behavior. Historempir-ical data shows that, for example in the United States, the amount of people owning stocks has been fluctuating a lot. It would be interesting to look at the factors that drive success for individual investors. Research about investment managers who manage funds within an organization shows that, among other factors, education, experience with investing and an individual’s risk attitude partially explain the fund performance (Gallagher, 2003). This suggests that these spe-cific manager characteristics are related to performance, and this also raises the question whether these individual characteristics are also responsible for individual investment per-formance. In previous literature, there has been extensive research about demographic characteristics of individual investors, and it appears that investors can be segmented into a number of groups based on these demographic characteristics (Kiran & Rao, 2005). In the article, Kiran and Rao state that there are several demographic and psychographic characteristics that influence the extent to which individuals invest. For example, age, education, gender, marital status, but also risk attitude and retirement planning, seem to affect investing behavior of individuals. The authors state that these findings can be used by financial service companies to target their financial products to these investor groups

(7)

in order to effectively market them. However, the article is limited to investment decisions and does not cover the returns on these investments. Nevertheless, it does imply that individual characteristics seem to affect the investment behavior and it is therefore likely that also performance is affected by this. More research about this dispute is therefore necessary, and might lead to more conclusive evidence.

It is interesting through which channels the previously mentioned personal factors seem to be affecting investment returns. Literature in behavioral economics and finance suggest that these findings might be attributed to psychological biases (Zaidi & Tauni,2012). One of these psychological biases is the overconfidence bias: individuals may be overconfident, and overestimate their knowledge and abilities when making investment choices (Tapia & Yermo,2007). The research these authors cite appears to be related to individual investors who make a choice to trade, but this choice is restricted to investments in the individual pension account. Still, this suggests that overconfidence affects investment decisions and therefore personal factors might influence investment performance through overconfidence. More research on this matter will therefore clarify this indirect relationship.

2.2 Definition and measurement of overconfidence

Before analyzing the overconfidence bias, a comprehensive review of the definition and measurement of overconfidence itself is indispensable. According to Glaser and Weber

(2007), there is no precise definition of overconfidence. In their research they suggest that overconfidence can be explained as a combination of miscalibration, the better-than-average effect and too tight volatility estimates. This vision is often applied in literature: the distinction between the overestimation of one’s actual performance, the over-placement of one’s performance relative to others and the excessive precision in someone’s beliefs is also evident from experimental research (Moore & Healy, 2008). It is suggested that, for difficult tasks, individuals overestimate their actual performance but also believe they perform better than others; the opposite holds for easy tasks. These findings are in line with other research on this matter: Kruger (1999) showed that for hard tasks, underconfidence is found. This might be explained by the fact that individuals expect their behavior to produce success, so they attribute outcomes to their actions when they succeed, and to bad luck when they do not succeed (Malmendier & Tate,2005). This implies that this so-called

(8)

self-serving attribution bias boosts overconfidence. Svenson explains in his 1981 article that people are overconfident when they rate their relative driving skills. He describes the overconfidence effect that is found as the better-than-average effect, which is in line with the previously discussed research.

It appears that overconfidence leads to faulty assessments; overconfident individuals overestimate their abilities, and the choices these individuals make are therefore too opti-mistic. In the case of overconfident investors, this might for example lead to a too optimistic assessment of future stock prices. Contrarily, underconfidence can cause individuals to be underestimating their abilities, which might cause individuals to not invest in certain risky assets (whereas investing would have been a rational choice). Existing research about con-fidence (both under- and overconcon-fidence) shows that concon-fidence mediates how investment knowledge influences investing self-efficacy (Forbes & Kara, 2010). This means that the assessment of an individual’s abilities are blurred by the level of confidence. Convincingly, this shows that both under- and overconfidence influence the way how individuals make choices.

For this study it is important to narrow down the analysis to the study of overconfidence of investors. Research about overconfidence of investors has led to the composition of the overconfidence hypothesis - in which stock returns, high levels of trading volume, excessive volatility and high risks are explained by the assumption that investor overconfidence can be accounted for these observed anomalies (Chuang & Lee, 2006). It is observed that overconfidence is present among corporate financial officers (Ben-David, Graham, & Harvey,2013) and traders and investment bankers (Glaser, Langer, & Weber,2012). Also, theory shows that optimism about financial prospects directly affects financial decisions (Puri & Robinson,2007), so the presence of overconfidence within financial decision making is not up for discussion in this paper. However, two questions remain unanswered in this debate: how can overconfidence be measured? And how, specifically, does overconfidence affect investment performance? The latter question will be covered in chapter2.4, whereas the remainder of this section will discuss the composition of overconfidence.

Although the psychometric measurement of overconfidence is not the focus of this pa-per, a short overview of empirics is useful in order to discuss this topic. In order to define a variable for overconfidence it is important to distinguish between the two main

(9)

interpreta-tions of overconfidence, specifically between the so-called performance-based (relative) and miscalibration-based measures of overconfidence, because they appear to be only minimally related in empirical research (Glaser & Weber, 2007). First, research shows (Bi, Du, & Li, 2013) that real overconfidence can be significantly detected by using peer-comparison questions. This implies that future studies can employ these peer-comparison questions in order to find overconfidence. Additionally, if research about overconfidence for investors in specific is evaluated, financial literacy is the variable of interest. More explicitly, the method of miscalibration (i.e. the difference between actual knowledge about financial matters and the self-assessed knowledge about these matters) is often used as a proxy for overconfidence for individual investors (Kramer, 2014). However, also the relative mea-surements of overconfidence (i.e. the better than average effect; overplacement) seems to be a good predictor of overconfidence. On average, higher perceptions of ability compared to others predicted larger scales of overconfidence (Larrick, Burson, & Soll, 2007). Con-clusively, research seems to indicate that overconfidence of individual investors manifests itself in many different ways.

2.3 Determinants of overconfidence

The existence of overconfidence is not up for discussion in this study; evolutionary models indicate that there is a biological underpinning as well. That is, research about the evo-lution of overconfidence shows that overconfidence is dominant in an evoevo-lutionary sense (Bernardo & Welch,2001). This is a remarkable finding; as overconfidence can lead to faulty assessments, unrealistic expectations and hazardous decisions, one would expect that over-confidence would not remain stable in a population (Johnson & Fowler, 2011). However, more research of the determinants of overconfidence might lead to a better understanding of this debate. Even though it is known that investors are subject to the bias of overcon-fidence, only limited work has been done on its demographics: empirical research about demographic and personal factors of an individual and the level of overconfidence is mini-mal. An exception is the relationship between gender and overconfidence, which is shown to be significant (Lundeberg, Fox, & Puncochar,1994). However, additional research shows that the observation that men are more inclined to be overconfident than women, is only found in areas that are perceived to be mainly dominated by men, such as finance (Deaux

(10)

& Farris, 1977). The authors report that, overall, men claim more ability than women, but this difference appears to be strongly driven by masculine tasks. Subsequently, this led to the conclusion that in areas such as finance, men are more overconfident than women, which leads to more excessive trades of men and therefore a reduction of returns for men (Barber & Odean,2001).

Additionally, Mishra and Metilda (2015) show that, besides gender, investment ex-perience and education are correlated with higher levels of overconfidence. Data from mutual fund investors is analyzed and shows that overconfidence increases with invest-ment experience and education. Moreover, they found a significant association between self-attribution and overconfidence. If the relationship between age and overconfidence is discussed, research shows that there seems to be an unclear link, as some studies found that overconfidence is increasing with age (Hansson, R¨onnlund, Juslin, & Nilsson,2008), as opposed to others finding contradictory results (Touron & Hertzog,2004). However, these studies do not specifically look at overconfidence in an investment context, and therefore the implications of this paper are limited. Research within the domain of investment be-havior show that age is usually correlated with investment experience, but both might have contradicting effects on behavior. Therefore, it is important to look at both variables when discussing the determinants of overconfidence. Regarding income levels and education, the small body of literature available shows that both of these variables do not have a direct impact but they do correlate with overconfidence (Koellinger, Minniti, & Schade,2004).

Another variable that appears to be related to overconfidence is the family situation of an individual. According to Bhandari and Deaves (2006), in a study they conducted among individuals that could choose their retirement assets, having a partner or having children are both indicators of overconfidence too. Hence, it is likely that also for individual investors, with other intentions than retirement planning, these variables are indicators of overconfidence.

Finally, past success might also attribute to the overconfidence level of an individual. The explanation is that investors who are successful tend to overestimate their abilities in the consequent period(s), as the previously discussed confirmation bias explains that most investors attribute successes to their own competence. Hilary and Menzly (2006) support this reasoning, by testing whether past successes lead analysts to become overconfident,

(11)

and find a significant relationship. Considering the fact that these analysts also make forecasts, fundamentally comparable to individuals that make investment decisions, it is likely that past successes are also indicators of overconfidence if the subject pool is restricted to individual investors only.

Nonetheless, evidently, overconfidence not only depends on these personal and demo-graphic factors. Undeniably, it is clear that multiple factors attribute to an individual’s level of overconfidence. Even though not all factors can be measured, since it is too difficult to find the exact underpinnings of overconfidence, it is still useful to dive into this matter to be able to get an indication or understanding of how overconfidence as a behavioral bias works.

2.4 The overconfidence bias and investment performance

Before looking at the effects the overconfidence bias has on investment performance, it is relevant to underline how investment performance can be measured. Finding and applying the right evaluation methods for investment performance is important in order to draw the correct conclusions.

Literature suggests that there are numerous approaches in evaluating investment perfor-mance. One example uses portfolio decisions as a measurement of investment perforperfor-mance. It suggests that overall performance of the portfolio decision incorporates the difference between the return on the chosen portfolio and the return on the riskless asset, which can be split into two parts: selectivity and risk (Fama, 1972). However, this method does not take into account the fact that individuals tend to under-diversify. Therefore,Calvet, Campbell, and Sodini(2007) argue that investment performance is invoked by the loss on return. This loss measures how much an individual investor is missing by not picking a cer-tain portfolio. Hence, it measures how much an individual loses from under-diversification. Measuring investment performance by under-diversification is reasonable in this particular case, asGoetzmann and Kumar(2005) showed that this under-diversification is a result of overconfidence.

Additionally, the link between overconfidence and high trading levels appears to be leading to lower levels of wealth for individuals (Barber & Odean,2000); people are over-confident and this leads to too much trading. It is argued that trading too much is not

(12)

good for an individual’s level of wealth. For assessing an individual’s level of wealth, the authors look at the household’s total wealth, which can be calculated by subtracting his total liabilities from the total assets. The approach of determining investment performance in terms of wealth is supported byCohn, Lewellen, Lease, and Schlarbaum (1975), where a higher proportion of total wealth invested in risky assets is explained as a measure of overinvestment. Positively, this suggests that multiple methods can be used in order to find an indicator of investment performance.

From existing literature and empirics, it appears that the overconfidence bias does have an effect on investment behavior. Odean (1998) shows that people are overconfident, and explains that this affects financial markets. He finds that overconfidence increases expected trading volume, market depth, and decreases the expected utility of overconfident traders. In his1999paper,Odean tests the hypothesis that overconfidence leads to investors’ will-ingness to trade too much, and finds a significant relationship. Additionally, Glaser and Weber(2007) state in their article that investors who think they are above average trade more. Hence, literature suggests that overconfidence leads to excessive trading. Barber and Odean (2000) argue that this in turn leads to a lower expected utility in terms of returns on investment. Additionally, there are studies that considered the relationship be-tween overconfidence and and under-diversification, which is also shown to be significant (Goetzmann & Kumar, 2005). Conclusively, it appears that overconfident investors are more likely to invest too much in an under-diversified portfolio. A potential reason for this behavior might be that more overconfident investors are less likely to take financial advice (Kramer,2014).

Nonetheless, there is no conclusive evidence of the direct effects of overconfidence on investment performance. Even though the relationship between overconfidence and certain investment behavior has been observed, a link in empirics between overconfidence and performance in terms of net results has not been proven.

(13)

2.5 Research question

Literature suggests that both demographic and personal characteristics of an individual are related to investment behavior: this behavior in turn reflects certain performance. For example, women tend to be better investors than men, and also factors as age, education and income level seem to have a correlation with investment returns. Previously discussed empirical research and literature indicate that overconfidence might be one of the reasons why these relationships are observed. The aim of this study is to find out whether over-confidence can explain, at least part of, the differences in investment performance among individual investors. Expected is that overconfidence acts as a mediator, where there is an indirect effect of demographic and personal factors through overconfidence on investment performance. Therefore, the research question is:

“To what extent can overconfidence explain differences in investment performance among individual investors? ”

3 Theoretical framework

This chapter contains a theoretical framework that focuses on the previously discussed literature review. From the literature, four hypotheses will be derived. These hypotheses will be tested in chapter five.

First, literature suggests that there is a positive relation between an individual’s per-sonal characteristics and the investment performance. It appears that both the perper-sonal and demographic factors influence an individual’s investment performance. As discussed before, gender, age, education and income appear to be positively related with investment returns. Based on this assumption, the first hypothesis is:

H1: Individual characteristics have a negative effect on investment returns and therefore

(14)

Moreover, to find an explanation for differences in investment performance among in-dividuals, an analysis of levels of overconfidence is appropriate. As Barber and Odean

(2001) and Mishra and Metilda (2015) find that certain characteristics are typically re-lated to higher levels of overconfidence, which might imply that these personal and key characteristics are determinants of overconfidence. Based on these findings, the following hypothesis is derived:

H2: Individual characteristics have a positive effect on the level of overconfidence and

therefore lead to higher levels of overconfidence.

Furthermore, existing literature indicates that changes in levels of overconfidence lead to changes in investing behavior and, therefore, it is likely that also investment returns are affected. Based on that discussion, the third hypothesis is:

H3: Overconfidence has a negative effect on investment returns. Higher overconfidence

leads to lower investment returns.

Finally, the above hypotheses hint that the effect that demographic and personal factors have on investment performance are mediated by an individual’s level of overconfidence. Therefore, the last hypothesis is:

H4: Differences in investment performance among individuals can be explained by

overcon-fidence as a mediator.

The hypotheses are visually represented in the following figure:

Individual characteristics Overconfidence Investment performance H3 (-) H2 (+) H4 H1 (-)

(15)

4 Data and methodology

To answer the research question, data from the DNB Household Survey (DHS) is used. The data from this survey is collected through the CentERpanel, and provides data on several topics including general information, data about income, assets and liabilities and psychological factors. The data is collected yearly, from more than 5,000 households and is a close representation of the Dutch population. The dataset allows the study of both psychological and economic aspects of financial behavior, which is the topic of interest for this study. Extensive questions on both psychological and economic factors are given so a broad analysis can be conducted on these matters. Additionally, the datasets offers a large variety of background variables which makes it possible to control for numerous aspects.

First, descriptive and summary statistics are presented (in order to get a general feel for the data). Next, in order to find out whether overconfidence mediates the effect that demographic and personal factors have on investment performance, and because the dataset does not contain off-the-rack variables for both overconfidence and investment performance, both of these variables have to be constructed following the main literature on these themes. This will be covered for in the next section.

Subsequently, it is convenient to do transformations so the data better fits the assump-tions of the proposed model(s). Moreover, in order to make more robust regressions, the fourth section deals with the control variables. The last part of this chapter covers the methodology that is used and discusses the model that is most appropriate to use in this study.

4.1 Data

4.1.1 Dataset

In this study, the focus is on the data from 2005. Datasets about household information, work & pension data, wealth data and economic & psychological concepts are combined in order to get an extensive background knowledge about the individuals that answered the questionnaires. Moreover, the dataset from the 2005 wave includes an additional question-naire about financial literacy, which provides the possibility to extract overconfidence from the data, followingKramer(2014).

(16)

Additionally, for this study it is important to focus on the individuals within a certain household that are most involved with the financial administration of the household, be-cause financial decision making of an individual is the topic of interest. The questionnaire includes a question “Are you the person who is most involved with the financial administra-tion of the household?”, which is used to filter the data in order to take care of this matter. Yet, it is important to note that, within a household, it is likely that decisions are at least partially influenced by other household members. However, each decision made in real life will be influenced by such externalities so it is hard to control for this. Nevertheless, this might have implications for the internal validity of the results in this study.

Since not all respondents participated in every questionnaire, some records have missing values for certain datasets. Furthermore, the additional module was sent out to 2,028 households. Out of these 2,028 households, 1,508 respondents completely filled out the questionnaire. When the additional module is combined with the original dataset from 2005, this results in a merged dataset that contains observations from 1,071 individuals. Out of these 1,071 individuals, roughly 26 percent invested in either mutual funds, stocks or both. A more extensive description of the data is given in the next section.

4.1.2 Descriptive statistics

Extensive descriptives of the full dataset and the subsample of investors can be found in Table1. Additionally, an overview of the variables used is given in Table 5 in Appendix

A, including a description and the type of variable. The data shows that, on average, the subsample of investors contains more males who are older, higher educated and higher paid individuals when compared to the other individuals in the dataset, which might be attributed to a selection effect, a vision that is widely acknowledged in existing literature. If the values for risk attitude for both samples (complete dataset and investors only) are evaluated, the data shows that on average investors are willing to take less risk than others. The opposite holds for overconfidence: investors are relatively more overconfident than others. This might be explained by the fact that, on average, investors have a higher level of actual financial literacy. However, investors also know that they have higher levels of actual financial literacy and therefore overestimate their abilities; this is reflected in a higher level of overconfidence. Moreover, the differences in level of overconfidence for both

(17)

Table 1: Summary statistics by group

Key variables Complete dataset Only investors

Gender 57.52% male 72.92% male

Age (s.d.) 51.23 (14.96) 54.46 (14.17)

Partner 66.85% yes 70.76% yes

Monthly gross income e2415.48 (e5906.36) e2978.03 (e1778.06)

Education level Higher vocational

education (hbo), 27.01%

Higher vocational education (hbo), 36.10%

Occupation Employee, 50.61% Employee, 53.43%

Children Number of children (s.d.) 32.12% yes 0.63 (1.01) 26.35% yes 0.59 (1.06) Other variables Risk attitudea Mean (s.d.) 0 (1.00) -0.55 (0.95) Min -2.91 -2.91 Max 1.73 1.73 Cognitive abilitiesa Mean (s.d.) 0 (1.00) 0.28 (0.75) Min -4.00 -4.00 Max 0.71 0.71

Self-assessed financial literacyb

Mean (s.d.) 4.77 (1.13) 5.16 (1.04)

Min 1 1

Max 7 7

Actual financial literacyb

Mean (s.d.) 0 (0.86) 0.47 (0.59) Min -3.04 -3.04 Max 0.96 0.96 Overconfidenceb Mean (s.d.) 0 (1.09) 0.22 (1.00) Min -3.73 -3.73 Max 2.69 2.11 Investment performancec Mean (s.d.) 0.68 (4.92) 0.18 (1.02) Min -17.96 -0.96 Max 75 10.16

Notes: This table shows the summary statistics for the complete dataset and the investors only. The data from 2005 of the DNB Household Survey is used. The number of observations is 1,071 for the complete dataset and 277 for the investors only. Standard deviations are in parentheses.

a. More about the composition of risk attitude and cognitive abilities in section4.5

(18)

groups can be partially explained by a selection effect: if individuals are underconfident they are, evidently, less interested in investing as they are not confident about their financial knowledge. Finally, the numbers for investment performance show that, on average, both the complete dataset and the investors have a positive investment performance, which means that the wealth of individuals is increasing. However, since the minimum value of performance for the full sample is lower (and the maximum value is higher) when compared to the sample of investors only, the standard deviation is larger for the full sample. Overall, these statistics reveal that, when analyzing investment performance, this is something that needs to be taken into account; it would be inaccurate to find results for the subsample only and apply these findings to the entire dataset as there are significant differences in the composition of these different subsamples.

4.2 Construction of variables

4.2.1 Overconfidence

As discussed before, overconfidence manifests itself in many different ways. Following literature, the variable overconfidence in this paper will be explained within the concept of financial literacy since the overconfidence of individual investors is the field of interest. Therefore, overconfidence will be interpreted as absolute overconfidence. First, discussing the concept of absolute overconfidence, literature suggests that this part of overconfidence can be explained for by comparing an individual’s self-assessed level of financial literacy with this person’s actual level of financial literacy (Kramer,2014). Therefore, it is possible to find the difference between the knowledge an individual thinks he or she has, and the actual knowledge: this very closely mirrors the definition of absolute overconfidence in the light of investment behavior. The dataset used in this study provides the opportunity to compute both self-assessed and actual financial literacy. For self-assessed financial literacy, the answers to the question “How knowledgeable do you consider yourself with respect to financial matters? ” is used. This question was asked to all respondents, where they could answer in between 1 (not knowledgeable at all) and 7 (very knowledgeable). Roughly 24% of the respondents considered themselves averagely (4) skilled when looking at financial matters, whereas 33% thought they were slightly above average (5) and 23% thought their financial knowledge was good (6). This means that, on average the level of self-assessed

(19)

financial literacy is 4.73, which indicates an above average knowledge.

In order to find the level of absolute overconfidence, a comparison with the actual level of financial literacy has to be carried out. The additional module of the DHS dataset contains a number of questions that try to measure the respondents’ knowledge on financial matters1. The answers to these questions illustrate the extent to which the respondents possess financial knowledge. Table 6 (Appendix B) shows extensive information about these questions. On average 68% of the questions are answered correctly, and only about 11% of the respondents answers all questions correctly. Next, a factor analysis needs to be conducted, in order to find the actual level of financial literacy (van Rooij, Lusardi, & Alessie,2011). A factor analysis indicates that there are two main factors with divergent loadings for the two types of questions: the first five questions create an index related to basic knowledge, whereas the second index measures a more advanced financial knowledge. Therefore, two separate factor analyses have to be performed in order to find the factor loadings. In this case, multiple observed variables (all questions testing financial literacy) have similar patterns of responses because they are all associated with financial literacy, which is not directly measured. Therefore a factor analysis provides the factor loadings of the components that express financial literacy. Before conducting the factor analysis, an important step is to analyze whether this analysis is appropriate or not (Bartlett, 1950). First, a Bartlett’s test for sphericity is carried out and these statistics (p-value < 0.01) show that the results are significant. This indicates that there are sufficient intercorrelations to conduct this factor analysis. Subsequently, a Kaiser-Meyer-Olkin measure of sampling adequacy is computed, and the results (KMO > 0.50) show that there is some overlap between the variables but not enough to where it would produce spurious results. Following these statistics, it is correct to conduct a factor analysis for defining the actual level of financial literacy. The results of the factor analysis can be found in Table6(AppendixB), and these indicate that all questions contribute in expressing financial literacy. Therefore, the factors can be weighted according to the factor loadings in order to calculate the actual level of financial literacy, followingAcock(2013).

The scores on actual financial literacy need to be combined with the level of

self-1

Originally, this module contains 16 questions to measure actual financial literacy, but since the last 3 questions are asked both positively and negatively by randomization, these are not incorporated in this study.

(20)

assessed financial literacy in order to compute the variable overconfidence. Specifically, the measure of self-assessed literacy is regressed on the financial literacy score and the residuals are used as overconfidence measure. Hence this overconfidence measure therefore indicates whether confidence, conditional on actual knowledge, is higher or lower than the average individual. Using this method, however, has the implication that there are negative values for overconfidence that express underconfidence. This is something that needs to be acknowledged when interpreting the results; the effects of underconfidence are also incorporated in analyzing the data.

4.2.2 Investment performance

In most researches that cover the concept of investment performance, a popular approach is to look at the return loss, or to evaluate the Sharpe-ratio of the portfolio of this particular household2. Both these calculations incorporate the fact that investors take risk for which they get rewarded. However, these explanations of investment performance rely on the assumption that underdiversification is a result of overconfidence (Calvet et al., 2007and

Goetzmann and Kumar,2005). As the focus of this study is on the results of overconfidence in terms of investment performance, rather than making assumptions about outcomes of overconfidence, a more direct evaluation of investment performance is appropriate: an approach that is supported byBarber and Odean(2000) and Cohn et al.(1975).

A more direct approach of evaluating investment performance is to look at the out-come of investing: the profits and losses an individual acquires. If the concept of investing is defined as the allocation of money in the expectation of a benefit in the future, these outcomes can be analyzed by looking at the assets and liabilities of an individual, because household finances reflect the direct results of investing behavior. The dataset used in this study provides enough information to calculate both the assets and liabilities, from which an individual’s net wealth can be calculated. As the hypotheses describe that overconfi-dence leads to worse investment performance, the present study estimates the relationship between overconfidence and the successive investment performance. Therefore the relative change in level of net wealth is the variable of interest in this particular study. In order

2_{The return loss reflects the expected return a household loses by not choosing a position on the efficient}

frontier with the same level of risk as its portfolio, whereas the Sharpe-ratio measures the risk-adjusted return of a financial portfolio.

(21)

to find this variable, the level of net wealth in 2005 and 2006 is calculated from which the change in net wealth is derived3. Summary statistics of this variable can be found in Table

1.

However, as an individual is overconfident, he or she is be willing to take more risk than rationally would be expected. This mere fact reveals that the above defined performance in terms of increase in net wealth does not necessarily have to decrease when an individual becomes more overconfident: individual investors might also get rewarded for taking high risks, albeit irrational. Therefore, in order to make a better assessment of investment performance, an interesting indicator of performance to look at is the variance of the increase in net wealth of an individual. The variance of return of an overconfident agent is likely to be larger than that of a rational agent, which is why this assessment of performance is a good measure. An extensive analysis of this variance is conducted in section5.1.

4.3 Explanatory variables

The background variables used in this study are the gender of the respondent, age, marital status, number of children, income level and educational level as these appear to be re-lated with both overconfidence and investment performance following the main literature. Moreover, to look at the impact of the work of the respondents on the dependent variables, another variable for occupation is added. An overview of these variables is given in Table

1. A more detailed description of these variables can be found in Table 5in AppendixA.

4.4 Data preparation

Before the data can be used to conduct the statistical analysis, some modifications need to be done. In the first place, the height of the level of income might suffer from measurement error, as there are some outliers and many small values in terms of monthly income. A log transformation makes this positively skewed distribution of income more normal and is therefore an appropriate modification. One drawback of using this modification is that the logarithm of zero is undefined, which causes the number of individuals with a missing value for income to increase. Therefore, the number of observations that can be used for

3

The change in wealth is calculated in percentages as the relative change is the focus in this study rather than absolute differences in wealth.

(22)

this study decreases. Nonetheless, since this only entails 81 individuals, this does not have a large impact on the reliability of the dataset.

Additionally, the dataset contains a variable that reveals the occupation of the respon-dent. This variable is transformed in multiple variables describing retirement, employment and self-employment as it is likely that these factors affect both overconfidence and invest-ment return.

Furthermore, missing values in the dataset were coded so these are excluded from the analyses. Additionally, a graphical analysis shows that there are no errors and large outliers in the dataset that disturb the data4.

4.5 Control variables

In order to enhance the robustness of the models used in this study, control variables are added; there might be potentially confounding variables, and due to the richness of the dataset this can be controlled for.

First, a control variable for risk attitude is added. It might be that an individual’s risk attitude is related to the level of overconfidence of that individual. Therefore, risk attitude is added as a control, so this alternative channel can be excluded. The questionnaire used in this survey contains six statements that concern saving and taking risks. The answers to these statements are used in order to measure risk attitude. These questions have been show to predict the levels of risk aversion and can therefore be used to measure an individual’s risk attitude (Dohmen et al.,2011). Following the same procedure as in calculating actual financial literacy (section 4.2.1), a factor analysis is conducted in order to measure an individual’s risk attitude. Both the statistics from a Bartlett’s test for sphericity (p-value < 0.01) and the Kaiser-Meyer-Olkin measure of sampling adequacy (KMO > 0.50) indicate that also here a factor analysis is appropriate in order to find the factor loadings of the questions defining risk attitude. For a more extensive elaboration of this variable, see Table

7in Appendix C. In this study an extra variable for risk-aversion is created by looking at this constructed risk attitude level, measured on a scale where 1 means risk-averse and 0 means not risk-averse.

4

Several histograms and frequency tables are derived in order to visually show that the data does not contain errors and outliers.

(23)

Furthermore, a control for an individual’s level of cognitive abilities is included, as it is unclear how this might affect both overconfidence and investment performance. It might be that individuals with high cognitive abilities are better in assessing their own level of financial literacy, which is why this needs to be controlled for. The variable of cognitive abilities is constructed by using a factor analysis on the basic questions from the financial literacy module, which is extensively described in Table6, AppendixB5.

Because not all (control) variables that have been discussed are observed for every indi-vidual, the number of complete observations drops to 972 for the individual characteristics including control variables, and 945 when including the level of overconfidence. If the sam-ple is split up into a subsamsam-ple of investors only, there remain 295 comsam-plete observations; this again indicates that roughly one out of three individuals invests in risky assets.

4.6 Methodology and econometric models

4.6.1 Method

Previous literature and empirical evidence indicate that the key variables influence invest-ment performance6. Therefore, and in order to find out whether this effect is present for the data used in this study, an ordinary least squares (OLS) regression will be performed. Significant values from this OLS regression will reveal which key variables are related to in-vestment performance. Nevertheless, this does not explain anything about the mechanism behind this relationship. In order to find out how these variables are related and to discover if there is an underlying factor that can explain these results, a mediation analysis needs to be conducted where overconfidence is acting as a mediator (Hayes,2017). The most basic model of this concept contains two dependent variables (overconfidence and investment performance) and two main independent variables (key variables and overconfidence).

5

Again, the statistics (p-value: 0.00 and KMO: 0.68) indicate that a factor analysis is appropriate.

6_{In this research the key variables used in the regressions are defined as the individual characteristics:}

(24)

From existing literature it appears that a mediation analysis is frequently used in psy-chological research. Since both the fields of behavioral economics and behavioral finance cover extensive topics of psychological relationships, this mediation analysis might also be appropriate to use in this research, as it gives more insight into the underlying mechanisms of investment decision making.

4.6.2 Models

In order to be able to test the above described method, a simple mediation model is used (also followingHayes,2017). Following this model, first of all, the total effect can be found by regressing investment performance (IPh) on the key variables (Xn) for every individual.

The control variables are also incorporated in this model (C).

IPh= α1+ cnXn+ δ1C + ε (1)

Subsequently, it is important to find the effect that the key variables have on the mediator variable (OCh). By doing so it can be shown that the key variables are correlated

with the mediator which is an essential element in a mediation analysis.

OCh= α2+ anXn+ δ2C + ε (2)

Finally, in order to find the indirect effect and following the simple mediation model, a last regression needs to be composed in which investment performance is regressed on both the key variables (direct effect) and the mediator (the indirect effect).

IPh= α3+ bOCh+ c0nXn+ δ3C + ε (3)

Ultimately, by combining the estimates of these models, the amount of mediation can be described, which is visually represented in Figure2. The letters in the figure correspond to the coefficients in the above discussed models.

(25)

OCh

Xn IPh

a b

c’

Total effect = direct effect + indirect effect c = c’ + ab

Figure 2: Mediation model

5 Results

In this section the results of the previously covered models will be discussed. The first part deals with the direct effect of the key variables on investment performance in which, more specifically, an extensive analysis of the determinants of investment performance is performed. Subsequently, the second paragraph focuses on the mediator variable in this study: overconfidence. Next, the impact of both the key variables and the mediator variable on investment performance is examined in order to be able to make conclusions about both the direct and indirect effects. This chapter ends with a discussion on how to further assess the robustness of the results, for which several steps are undertaken.

5.1 Investment performance

First, an analysis of investment performance in terms of change in net wealth of an in-dividual is conducted. An OLS shows that the key variables used in this study do not significantly explain this change in net wealth, neither for the full sample nor the subsam-ple of investors7. This suggests that, surprisingly, other variables than personal factors may be more important in explaining this measure of investment performance. However, as discussed in section 4.2.2, another approach might be more relevant in this particular case: the variance of the growth can be used as a measure of performance. An additional variable is created for both underconfident and overconfident individuals that captures the variance of the change in wealth. Since high fluctuations in changes in net wealth is not

(26)

Table 2: Investment performance measured by variance of change in wealth Variance of ∆wealth (1) (2) Gender -0.756 -0.087 (-1.15) (-0.21) Age -0.058** 0.015 (-2.10) (0.83) Partner 1.976*** -0.467 (3.09) (-1.15)

Logarithm of gross income 0.481 -0.385

(1.06) (-1.37) Education level -0.237 0.137 (-1.18) (1.08) Number of children -0.136 0.217 (-0.42) (1.06) Retired -0.855 -0.277 (-0.83) (-0.43) Self-employed -0.455 1.476 (-0.26) (1.37) Employee -0.206 0.114 (-0.23) (0.21) Risk aversion 0.358 -0.330 (0.63) (-0.92) Cognitive abilities -0.403 0.366* (-1.37) (1.84) Level of overconfidence 6.324*** (39.14) Constant 24.510*** 26.010*** (6.26) (10.66) N 972 945 t statistics in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

(27)

something investors are typically interested in, at least not in the long term, a higher vari-ance of change in wealth is considered as bad investment performvari-ance in this study8. The complete dataset is used here, as the choice to not invest can also be seen as an individual investment decision. Using this measure leads to the results shown in Table2, column (1). In this output the control variables are included, as they do not change the significance of the other variables. These regression outputs show that, first of all, age has a negative im-pact of the measurement of performance used in this study. This negative number implies that older individuals typically have a higher variance of the change in net wealth, which means that older individuals are better investors.

Additionally, column (1) shows that having a partner positively influences the perfor-mance measure which means that individuals with a partner in the household typically have a higher variance of change in wealth. This, in turn, means that these individuals are worse investors.

Remarkably, as opposed to existing literature, there is no significant effect of educa-tion and income level on this measure of investment performance. The former might be explained by the fact that the effects of education on performance are partly captured by the control variable included for cognitive abilities. For the latter, whereas a high in-come does presumably lead to higher levels of overconfidence (see next section), it does not make sense for income to affect investment performance directly when both cognitive abilities and education are included in the model. This might be an explanation to the fact that the statistics show an insignificant relationship between income level and investment performance.

Positively, this implies that the first hypothesis cannot be rejected. More explicitly; the results show that both age and partner significantly influence investment performance.

5.2 Determinants of overconfidence

Table3shows the OLS results of overconfidence. This first column indicates that both age and the level of education are negatively and significantly related to overconfidence, whereas having a partner and the level of income are positively (and also significantly) related to overconfidence. These estimates are robust to the inclusion of multiple control variables in

(28)

column (2) and is in line with existing literature. When the analysis is restricted to the subsample of investors, column (3) and (4) show that additionally gender is significant. The fact that female investors are less overconfident is in line with previous research.

The significance of having a partner, that has a relatively large positive effect on over-confidence, is remarkable, even though it is in line with previous research. As this result indicates that an individual’s level of overconfidence is related to having a partner or not, it may indicate that choices made by this individual are influenced by their partner even though the dataset is restricted to respondents who are most involved with the household’s financial administration. This must be taken into account when drawing conclusions.

Convincingly, the above results point in the same direction as the second hypothesis. It appears that several key factors are significantly related to the level of overconfidence of an individual. More specifically: age, having a partner and income (among others) determine the level of overconfidence for the full sample. For example, having a partner or not has an average difference of 0.41 in an individual’s level of overconfidence.

Additionally, when the sample is restricted to the subsample of investors only, it appears that also gender has a significant impact on overconfidence. This is in line with the findings from previously discussed literature where male investors appear to be more overconfident than female investors. Conclusively, the second hypothesis cannot be rejected.

5.3 Mediation model

As the key factors significantly influence both investment performance and overconfidence, an interesting next step is to analyze the role of overconfidence in this relationship. When an OLS of performance on both the key variables (including controls) and the mediator variable (overconfidence) is performed, following model (3), the results show some inter-esting findings. The output, shown in Table2column (2), reveals that overconfidence can significantly explain changes in investment performance. The positive sign of this coef-ficient shows that an increase of overconfidence leads to a higher variance of change in wealth, which in this study reflects worse investment performance and is therefore in line with the third hypothesis. This finding can be graphically supported by creating a scat-terplot of investment performance and overconfidence (see Figure 3, Appendix E). This figure shows that as the level of overconfidence increases, also the variance of the change

(29)

Table 3: Overconfidence

Overconfidence

Full sample Investors only

(1) (2) (3) (4) Gender -0.105 -0.099 -0.342** -0.332** (-1.28) (-1.18) (-2.25) (-2.15) Age -0.013*** -0.013*** -0.017** -0.016** (-3.64) (-3.55) (-2.58) (-2.47) Partner 0.402*** 0.406*** 0.261* 0.267* (4.93) (4.98) (1.88) (1.91)

Logarithm of gross income 0.136** 0.139** 0.237** 0.235**

(2.39) (2.45) (2.01) (1.98) Education level -0.062** -0.056** -0.025 -0.019 (-2.48) (-2.20) (-0.60) (-0.45) Number of children -0.047 -0.051 -0.023 -0.029 (-1.12) (-1.23) (-0.37) (-0.46) Retired -0.081 -0.083 -0.180 -0.173 (-0.62) (-0.63) (-0.78) (-0.75) Self-employed -0.299 -0.298 -0.076 -0.059 (-1.37) (-1.37) (-0.17) (-0.13) Employee -0.074 -0.073 -0.407* -0.399* (-0.66) (-0.65) (-1.88) (-1.84) Risk aversion 0.124* 0.154 (1.72) (1.25) Cognitive abilities -0.078* -0.084 (-1.93) (-0.97) Constant -0.150 -0.272 -0.118 -0.245 (-0.30) (-0.55) (-0.12) (-0.25) N 945 945 295 295 t statistics in parentheses * p < 0.10, ** p < 0.05, *** p < 0.01

(30)

in wealth increases. The left-hand side of the panel shows the changes of wealth for un-derconfident individuals, where only a small spread is visible. Contrarily, the right-hand side of the panel (observations for overconfident individuals) reveals that the changes in wealth are much more spread: this in turn indicates that overconfident individuals have much more fluctuating changes in wealth. Although the figure shows a large number of points clustered around the x-axis, it does clearly show that the variance of the change in wealth is much higher for overconfident individuals; a finding that matches the statistical analysis given above.

Moreover, if column (1) and (2) are compared in Table2by focusing at the grey shaded rows, the initially significant effects of both age and partner on investment performance appear to be diminished. This does not only indicate that age and partner are insignificant when overconfidence is included, but also shows that the magnitude is different from the initial value which is shown by the change in the coefficient from column (1) to (2). There-fore, it is likely that overconfidence in this model acts as a mediator, as expected and in line with hypothesis 4. Although, in order to make sure that the indirect effect of age and partner through overconfidence is significant and in line with literature, more tests need to be conducted in order to give conclusive evidence about the impact of overconfidence on the relationship of interest in this study. To find the extent to which the key variables are mediated by overconfidence, Structural Equation Modeling (SEM) is used. This model reveals the links between observed variables and latent variables, which perfectly fits the mediation analysis used in this study (Cheung & Lau,2008). Results from this model are shown in Table4; these results confirm the prognosis that overconfidence acts as a mediator for age and having a partner. It shows that the indirect effect of age and having a partner are both significant and the magnitude is relatively large, as for both variables roughly 84 percent of the total effects are mediated through overconfidence. As the direct effects of age and having a partner is insignificant at a 10 percent level, overconfidence appears to be fully mediating this relationship.

(31)

Table 4: Results Structural Equation Model

c’ ab c

Direct effect Indirect effect Total effect Mediation

Age -0.0146 -0.0790 -0.0936 84.39%

*** **

Partner 0.4669 2.5649 3.0318 84.60%

*** ***

Notes: This table shows the disaggregated results of the SEM with overconfidence as a mediator variable and the variance of change in wealth as dependent variable. The results are based on the 2005 and 2006 wave of the DNB Household Survey.

5.4 Bootstrap procedure

In order to verify the above reported results, a bootstrap procedure is used9_{. According}

to Hayes (2009), “Bootstrapping generates an empirical representation of the sampling distribution of the indirect effect by treating the obtained sample as a representation of the population in miniature”. It is therefore a more powerful and valid method for testing intervening variable effects, which makes it an appropriate choice in this study. Moreover, this method can be applied when the assumption of a large sample size and normality may not hold. Therefore, in order to enhance the robustness of the results, a bootstrap procedure with 1000 replications is conducted. The results from this procedure are in line with the results shown in Table4, and hence support the finding that overconfidence acts as a mediator for both age and having a partner.

5.5 Robustness checks and falsification tests

The sample size of the dataset is relatively small compared to other studies. Even though the bootstrap procedure addresses this drawback, it might still be possible that the results are driven by households that performed bad in 2005 or exceptionally good in 2006 in terms of wealth. Therefore, it is relevant to check whether the findings remain intact if the analysis is focused on changes in wealth of different years. The dataset also provides levels of wealth in subsequent years, which creates the opportunity to look at the change of

(32)

wealth in 2006 and 2007. Both the models described in section4.6.2 and the results from the SEM discussed in section5.4 remain intact if the analysis is focused on the growth of wealth in 2005-2007 and 2006-2007 (see Table9 in Appendix F). The results of this OLS confirm that overconfidence significantly influences investment performance even when the analysis is focused on different years.

Furthermore, the regressions used show that there is no multicollinearity present, which means that there is no perfect linear relationship among the independent variables and therefore there are no problems with the coefficients and standard errors of the results.

Additionally, there might be issues regarding reversed causality in this study. Through-out this study it is argued that key factors influence overconfidence, and this in turn in-fluences investment performance. However, it might also be that levels of overconfidence are a result of previous investment performance (Hilary & Menzly, 2006). Therefore, in order to find out whether big changes in wealth in prior years affect an individual’s level of overconfidence, an OLS of change in wealth in 2004-2005 is performed on overconfidence and the key variables plus controls10. The results of this regression (see Table 10 in Ap-pendix G) indicate that the change in wealth of an individual in the prior year does not significantly influence the level of overconfidence. Accordingly, it is highly unlikely that there are problems with reversed causality in this study.

Finally, it would be interesting to look at the effects for the subsample of investors only, as this reveals whether the effects found are mainly driven by investors or not. First, performing an OLS (see Table 11 AppendixH) of performance on the key variables plus controls shows that the determinants of overconfidence are not significant for the subsam-ple of investors: the previously found results appear to be mainly driven by non-investors. However, these insignificant results can also be caused by the small sample size as there are only 302 observations for investors. This results in higher standard errors and therefore a higher chance of type II errors11. However, when overconfidence itself is included as a variable in this OLS, there appears to be a significant relationship between overconfi-dence and investment performance. This supports the finding that, also for investors only, overconfidence influences investment performance.

10

As bad performance in prior years might influence an individual’s level of overconfidence in the subse-quent years, the percentage change from 2004 to 2005 is taken as a measure here.

11

(33)

6 Discussion and conclusion

Literature shows that there appears to be a relationship between individual characteristics and individual investment performance. For example, individual characteristics like age, gender, education level and having a partner all seem to have an effect on how individuals perform when they buy risky assets. Another interesting finding from literature is not only the observation that individuals are overconfident on average; above all the conclusion that the level of overconfidence of an individual is related to his or her individual characteristics is striking. A growing body of literature leads to the belief that overconfidence plays a big role in individual decision making. However, little research has been done on whether and how this overconfidence bias affects individual investment decision making.

This study tries to explain the extent to which overconfidence affects individual invest-ment decision making by finding empirical evidence using data from the DNB Household Survey. First, a variable for overconfidence is constructed that very closely mirrors the definition of absolute overconfidence: it measures the difference between self-assessed fi-nancial literacy and actual fifi-nancial literacy. Second, a measure of investment performance is composed. Since investment performance can be explained using various approaches, this assessment is much more elaborate. As the dataset used is somewhat limited in terms of number of observations, it is interesting to look at a measure that also captures individuals that are not investors; the decision to not invest in risky assets can also be explained as an investment decision. One way to find performance in terms of individual wealth is to look at an individual’s level of assets and liabilities. However, simply evaluating the changes in wealth does not lead to the conclusions that were expected. Both the key factors and overconfidence seem to be unrelated to this change in wealth. This unexpected finding might be explained by the fact that individual investors can also get rewarded for taking exceptionally high risks because they were overconfident; therefore the effects of both the key variables and overconfidence on change in wealth are both insignificant. Nonetheless, this does open a new door for assessing performance: it shows that overconfidence leads to more fluctuating changes in wealth, both positively and negatively. Therefore looking at the variance of the changes in wealth might be a more appropriate evaluation of perfor-mance. This method of evaluating performance is more promising, as the results from this study show some interesting findings.

(34)

First, by regressing this variance of change in wealth on the key factors it is found that both age and having a partner are significantly related to this assessment of performance. Second, as Table 3 shows, age and having a partner are also significantly related to an individual’s level of overconfidence, yet in the opposite direction. These two findings already point in the direction of the fourth hypothesis: it appears that overconfidence plays a role in this relationship. However, to further uncover the role of overconfidence some additional tests are performed. These tests, as discussed in section 5.3 and 5.4, reveal that indeed overconfidence acts as a mediator; the effects that both age and having a partner initially had on investment performance are weakened when overconfidence is included in the model. It appears that roughly 83 percent of the initial effects are mediated by overconfidence; results that remain intact when additional robustness checks and falsification tests are performed. Therefore, for age and having a partner, the hypotheses stated in section3are not rejected and hence the research question of this paper can be answered. Conclusively, for age and having a partner, overconfidence can explain the differences in investment performance of individual investors. Hence, individuals do not perform better because they are older, but because they are less overconfident.

Yet, there are some shortcomings and caveats that need to be addressed. These sug-gestions also provide good directions for further research. First, this study uses absolute overconfidence to assess an individual’s level of overconfidence. However, as extensively discussed before, there are other factors that might also play a role in overconfidence. Un-fortunately, the dataset used in this study does not contain questions or variables from which subjective overconfidence (i.e. the better-than-average effect) can be extracted: this might be an interesting topic for further research.

Second, the measure of performance used in this study has some drawbacks. The aim was to look at the results of investment performance by evaluating changes in net wealth. Even though the variance of change is a good measure of how individuals perform, as high variances are undesirable, still the conclusions that can be drawn from this study are somewhat limited. That is, in terms of net wealth there is no mediation, but in terms of variation of returns there is. Therefore, a good approach in next studies might be to focus on how an individual’s level of wealth changed, i.e. what the drivers are of the changes in wealth. Diving into this matter can perhaps lead to more comprehensive explanations of

(35)

the results that are found.

Additionally, valuable insights can be gained by looking at panel data. As the variable overconfidence in this study is constructed at one point in time (2005), overconfidence is considered as a time-invariant variable. Even though the focus of this study is on the effects of overconfidence at a certain point in time, it would still be interesting to look at how changes in levels of overconfidence would affect performance throughout time. Unfortunately, the data used in this paper does not provide the possibility to look at this. However, additional research can be done to investigate this dynamic relationship. Also, the external validity of this study might be limited as the dataset is from the Netherlands only; research on a similar topic using data from different countries might provide more generalizable results.

Overall, even though the discussed caveats and shortcomings provide good possibilities for further research, some general results can be reported. There is evidence to conclu-sively say that at least the effects of age and having a partner on investment performance are mediated by an individual’s level of overconfidence. One possible explanation of this finding might be that certain people rely less on financial advice (Kramer,2014). As over-confident individuals assess their financial knowledge incorrect, these particular individuals mostly do not obtain financial advise from professionals which might lead to worse results. Moreover, it is likely that overconfident investors are overoptimistic about their prospects and therefore overinvest and underdiversify, as also explained and expected according to previous literature. Additionally, the effects through overconfidence might be explained by an individual’s personality traits. E.g., the level of an individual’s “aggreeableness” might also affect a certain level of overconfidence; a finding that is also reported byBashir, Azam, Butt, Javed, and Tanvir (2013). Therefore, the findings of this study are in line with previous research.

Behavioral biases in individual investment decision making : the role of overconfidence

Behavioral biases in individual investment decision

making: the role of overconfidence

Contents

1

Introduction

2

Literature review

3

Theoretical framework

4

Data and methodology

5

Results

6

Discussion and conclusion