Rijksuniversiteit Groningen
Faculty of Economics and Business
DISPOSITIONAL OPTIMISM AND PORTFOLIO COMPOSITION
Combined Master’s Thesis in Economics and Finance
EBM000A20
LOU HARTMANN
S 2502836
Supervisor: Dr. Viola Angelini
June 2015
Abstract:
In this research project, we analyse the impact of dispositional optimism on the probability of buying risky assets. We examine therefore data of the year 2002 from Germany, using logit and multinomial logit regression models.
At the first view, we find that optimistic people rather tend to hold safe assets only, but the marginal effect at the mean of this finding is close to zero. In a second step, we subdivide the observations into five different groups depending on their level of
optimism. We find then a parabolic relationship between optimism and the probability of holding risky assets. The probability is thus the highest for people that we qualify as
I would like to thank very much my supervisor, Dr. Viola Angelini for all the useful feedback and support throughout the entire process of writing this thesis.
Furthermore, I would like to express my gratitude to Dr. Mark Kramer and Jun. Prof. Israel Waichman whose inspiring lectures on Behavioural Finance at the
Rijksuniversiteit Groningen respectively the Ruprechts-‐Karls-‐Universität in Heidelberg provoked my interest in this topic.
DISPOSITIONAL OPTIMISM AND HOUSEHOLD PORTFOLIO CHOICES
1. INTRODUCTION
In traditional neo-‐classical economics, human beings are usually considered to be fully rational. However, this assumption of completely rational humans was already criticised in the late 19th century (Veblen (1898)). Indeed, when it comes to decision-‐making, including but not limited to financial decision-‐making, human behaviour is often not consistent with the predictions of the models and axioms of the neo-‐classical utility theory (Kahneman and Tversky (1979)). Among the biases that lead humans to behave in a different way than predicted by the neo-‐classical utility theory, the optimism bias plays a role. People tend to be optimistic in an unrealistic way about future life events (Weinstein (1980)).
The goal of this research project is to figure out if and how portfolio decisions are affected by dispositional optimism.
Scheier and Carver (1985) provide a definition for dispositional optimism in terms of expected outcomes. Optimists tend to expect a desirable outcome. Dispositional optimism 1 is thus a trait of personality that leads to expectations of favourable outcomes.
In this paper, we analyse the relationship between dispositional optimism and the probability of buying risky assets. It seems for two reasons intuitive that optimism is one of the driving forces when people decide what assets they want to buy. First of all, optimism is likely to lead directly to higher expected returns on assets, and second via the channel of overconfidence. Dispositional optimism is likely to lead to overconfidence (Nofsinger (2005)).
On an empirical basis, optimistic persons tend to be overconfident (Fabre and François-‐ Heude (2009)) as both biases are closely linked (Taylor and Brown (1988)).
Overconfident persons tend to overestimate their abilities on the stock market and may therefore have a higher probability to buy risky assets. Barber and Odean (2000) find
that the most active investors underperform the market the most and provide overconfidence as explanation for this finding. Thus, one can expect a link between overconfidence and behaviour on the stock market.
Portfolio composition is one of the most complex financial decisions that individuals have to make and is therefore worth being analysed in detail. A big challenge in this research project is to find a valid way of measuring optimism.
Puri and Robinson (2007) analysed the relationship between financial decision-‐making and optimism by constructing an indicator of optimism comparing self-‐reported life expectancy and the life expectancy implied by actuarial life-‐tables. They compared the self-‐reported expectations and the statistical life expectancy of the individuals and interpreted the expectations as optimism if they were higher than the statistical life expectancy issued from the actuarial life tables.
We use a similar approach, however, we do not consider life expectancy but life
satisfaction as relevant variable. In our setting, we compare the expected life satisfaction in five years at a given moment in time to the realized life satisfaction five years later. The idea behind our main measure of optimism is that we interpret the miscalibration between expected life satisfaction and actual life satisfaction in five years from that moment in time as optimism or pessimism.
We have data from the German Socio Economic Panel (SOEP). This data is collected on households in all parts of Germany. We use the year 2002 as it provides all the relevant information we need for this study: we need data on portfolio composition, a bundle of control variables and the data on life satisfaction that we require to construct our optimism measures, i.e. current life satisfaction, expected life satisfaction in five years and the realized life satisfaction five years later.
In the survey, people were asked to self-‐report their life satisfaction in 2002 on a scale from 0 to 10. At the same moment, they were asked to report their expected life satisfaction in five years. Five years later, in 2007 they were asked again about their current life satisfaction.
Our main measure of optimism compares the expected life satisfaction in 2007, as it was estimated in 2002 and the actual life satisfaction in 2007. This variable is created in the same fashion by Abolhassani and Alessie (2013). However, they do not interpret this forecast error explicitly as optimism but as evidence that people are not fully rational when they make their expectations or as evidence that new information may appear during the time period of five years with an impact on the life satisfaction. Our
alternative measure of optimism just compares the difference between the expectation for 2007 and the life satisfaction at the moment when this expectation was reported, in 2002. This measure is used as optimism measure by Abolhassani and Alessie (2013).
This way of measuring optimism using life satisfaction as proxy was, as far as we know, not used before in the literature to assess portfolio composition. We will present the measures of optimism we use more in detail in section (3). Unfortunately, we only have access to data that indicates what types of assets households own and no information about the proportions of the different types of assets in their portfolio. We can thus only evaluate the probability that a household owns a certain type of assets and not draw any conclusions about the share of wealth in each type of assets.
This paper is organized as follows: in section (2), we provide a brief overview of the existing literature on optimism and overconfidence and explain the link between these two concepts. In section (3), we present the data we use.
In section (6), we provide possible interpretations for these at the first view maybe contradictory findings. Finally, we conclude this research project in section (7).
2. LITERATURE REVIEW
In standard portfolio theory, the choice of assets should be made by combining in an efficient way maximum expected return and minimum variance. However, assumptions have to be made in that case about the expected return on assets and covariance
between them (Markowitz (1952)). Markowitz considers the formation of beliefs as „first stage“ and the portfolio composition given the beliefs as the „second stage“ of the process of portfolio selection. In his analysis, he focuses above all on the second stage. He defines portfolios as efficient if it is not possible to get the same expected return with a lower variance and no higher expected return with the same variance. In order to make the beliefs required to end up with expectations, Markowitz suggested to use statistical computations first and to let adjustments of these forecasts be done by experts if they consider this as necessary. He concludes however that his well-‐known paper does not consider in depth the first stage of portfolio composition (p.91): „the formation of the relevant beliefs on the basis of observation.“
In this research project, we would like to figure out if and how optimism impacts portfolio composition. Optimism is likely to have an influence on the formation of beliefs. Formation of beliefs is actually hard, if not impossible, to combine with the concept of the fully rational homo economicus. A behavioural approach is therefore in our opinion required to do so. Behavioural economics is defined by Mullainathan and Thaler (2000) in the abstract of their working paper as „combination of economics and psychology that investigates what happens in markets in which some of the agents display human limitations and complications.“ Optimism can be seen as one of these human complications.
Brissette, Scheier and Carver (2002) find that optimistic people are better able to deal with stressful life events. Rasmussen, Scheier and Greenhouse (2009) conclude, by reviewing 83 studies on the topic, that optimism is a significant predictor for good physical health.
Optimism and overconfidence are two similar and closely related concepts. Fabre and François-‐Heude (2009) provide the following definitions: optimism is (p.80) „the tendency or inclination to perceive an event or an action as more likely to result in a favourable outcome, irrespective of the objective probability of that outcome actually occurring.“ On the other hand, they define overconfidence as (p.80) „the tendency to overestimate the probability of achieving one’s objectives as a result of a presumptuous belief in one’s abilities or attributes as they may be used to bring about a particular outcome.“ Thus, the main difference between both concepts is the role of one’s own impact on the final outcome. Furthermore, as Fabre and François-‐Heude underline, Scheier and Carver (1985) argue that dispositional optimism is stable in nature, whereas McGraw et al. (2004) showed that some information on the overconfidence bias during a five-‐minute-‐break is already enough to significantly reduce
overconfidence.
Weinstein (1980) finds in a study that students tend to suffer from an optimistic bias, especially in situations that seem controllable to them, which shows thus that there is a strong link between the concepts of optimism and overconfidence.
This finding is in line with the one of Langer (1975) who finds that people think they have an influence in situations in which the outcome is basically decided by luck. Also Svenson (1981) shows evidence of overconfidence: he concludes that 80% of the
students in his study would consider themselves as being in the top 30% of car drivers.
with increasing mood like optimism lead to overconfidence. According to Hilton (2001), the idea of being better than the average, which actually is driven by overconfidence, is part of the optimism bias definition.
So, there is evidence from the literature that people tend to suffer from optimism and overconfidence biases and that the concepts are indeed closely related. This is likely to have an impact on financial decisions.
Odean (1998) shows that overconfidence is a potential source of inefficient decision-‐ making. This finding is confirmed by a paper of Barber and Odean (2000): they conclude that active retail investors underperform the market the most and they explain this observation basically by the overconfidence of these active investors. Barber and Odean (2002) drew similar conclusions when analysing the switch to online trading in the 1990s. The retail investors who made strong performances on the financial markets with phone based trading were likely to switch to online trading. Their good earlier performances lead to overconfidence. When trading online, these investors traded much more and showed worse performances.
Puri and Robinson (2007) show an empiric relation between optimism and a broad range of economic and financial variables. Using data from the United States, they find that optimistic individuals are more likely to invest in individual stocks rather than mutual funds or other equity investment vehicles. Furthermore, they test whether moderate and extreme optimists behave in the same fashion and find several differences between these two types of optimists: moderate optimists take rather prudent economic decisions, whereas extreme optimists do not. They finally conclude that moderate
optimism can improve the economic decision-‐making, whereas extreme optimism is clearly bad.
Sengmuller conclude that some investors get some non-‐monetary benefits from trading. What they call „entertainment trading“ can be seen as some kind of leisure activity that increases utility, even though it may even lead to losses. In a similar spirit, Kumar finds that some people’s desire to gamble impacts their behaviour on the stock market, even though they underperform institutional investors.
Given the existing literature, it seems very likely that optimism somehow impacts financial decision-‐making. We would like to confirm or reject this hypothesis using data from Germany. If we find evidence that optimism has an influence on portfolio
composition, we would like to assess that impact.
3. DATA AND METHODOLOGY
The data we use to conduct this study is issued from the German Socio Economic Panel (SOEP). Wagner, Frick and Schupp (2007) describe this panel in detail. The SOEP exists since 1984 and the DIW Berlin, the German Institute for Economic Research, hosts it. Every year, more than 10000 private households are sampled.
In this study, we focus especially on data concerning the year 2002, as it is this year over which we get all the relevant data on expected life satisfaction in five years and portfolio composition that are crucial for this study. Some questions, such as the life satisfaction in five years, are not asked in every questionnaire but only in some waves. So, most of our variables are included in the SOEP core study of 2002 and the questionnaire of 2003, asking about the behaviour of individuals and households in the previous year. We assume that the decision-‐making is done jointly at the household level. We therefore consider all individuals within a household, but we will cluster the standard errors to take intra-‐household correlation into account. Furthermore, we use a self-‐assessed risk attitude measure of the individuals. This self-‐assessment was done in 2004. In order to construct the optimism measure, we also have to take into account the self-‐assessed life satisfaction in 2007.
a) How to measure optimism?
The common measure of optimism that is often used in psychology is the Life Orientation Test LOT (Scheier and Carver (1985)) and its successor LOT-‐R, the Life Orientation Test Revised (Scheier, Carver and Bridges (1994)). As we do not dispose of data from LOT or LOT-‐R tests, we need to create a proxy for optimism.
Puri and Robinson (2007) developed an indicator of dispositional optimism based on the difference between self-‐reported life expectancy and the life expectancy implied by actuarial life-‐tables. This miscalibration was used then to investigate the relationship between optimism and a wide range of economic outcomes.
Following this idea, we also compare two variables to measure optimism. Whereas Puri and Robinson used expected and actuarial life expectation, we focus on life satisfaction.
First, in 2002, people were on an individual level asked to assess their expected life satisfaction in five years on a scale from 0 to 10. Five years later, in 2007, people were asked about their current life satisfaction, again on a scale from 0 to 10. We compare the expected future life satisfaction as it is reported in 2002 and the actual life satisfaction in 2007. The difference can thus range between -‐10 and 10. This is a way to measure the gap between an expectation in 2002 and the real situation five years later. It can thus be interpreted as optimism if it is positive -‐ and as pessimism otherwise. In figure 1, we show the distribution of self-‐reported life satisfaction in 2007. In figure 2, the expected life-‐satisfaction in 2007, reported in 2002, is shown.
Figure 1: histogram of the self-‐reported life satisfaction in 2007
Figure 2: histogram of the expected life satisfaction in 2007, reported in 2002
In this setting, we need to have data from the same individuals in 2002 and 2007. We furthermore only can use those observations where there is also information available on the assets owned by their household and on the control variables described later in this section. For these reasons, we can use only 14566 observations, even though almost 24000 reported their life satisfaction in 2002. If we consider only the individuals who live in a household that reported to own at least some assets, we are left with 13211 observations. The descriptive statistics of these samples are reported in detail in
even higher than they expected it. The opposite holds for people who reported 0
initially. About 1% of our observations reported twice a life satisfaction of 10, less than 0.1% reported twice a life satisfaction of 0. We will consider these observations as „neutral optimists“. In table 1, we present the frequencies of the self-‐reported levels of expected life satisfaction in five years and the current life-‐satisfaction in 2007.
Table 1: Frequencies of self-‐reported expected life satisfaction in five years from now in 2002 and actual life satisfaction in 2007:
Exp. Act. 0 1 2 3 4 5 6 7 8 9 10 Total 0 4 4 2 6 2 9 4 4 4 1 0 40 1 2 5 7 6 5 10 5 12 7 2 2 63 2 6 4 19 18 24 25 22 19 33 11 5 186 3 6 3 23 26 34 80 48 57 67 17 6 367 4 1 2 26 35 51 101 74 93 81 18 7 489 5 7 7 31 96 117 350 241 285 272 84 30 1520 6 1 6 19 41 95 227 270 370 419 126 45 1619 7 5 5 22 45 96 301 371 719 1006 360 98 3028 8 4 1 10 32 60 227 283 723 1643 818 280 4081 9 1 3 4 6 7 40 51 178 449 503 199 1441 10 1 0 0 1 7 15 19 38 77 83 136 377 Total 38 40 163 312 498 1385 1388 2498 4058 2023 808 13211
If we consider these 13211 observations, which are our most restrictive sample, we identify an average level of optimism of 0.257. So, the average observation of this dataset is what we will qualify later as a neutral optimist.
In order to distinguish between mild and extreme optimism, Puri and Robinson
some observations with an optimism level if 3 as strong optimists and some others as mild optimists, which would in our opinion not make that much sense.
As our measure of optimism is a discrete one, we prefer to replace this definition of extreme optimism by an absolute one.
If the level of optimism computed in this way is 3 or higher, we qualify individuals as „strongly optimistic.“ In the same way, we qualify people with a negative level of optimism of -‐3 or lower as „strongly pessimistic.“
According to this measure, 9.36% of these 13211 observations are strong optimists and 6.21% are strong pessimists. The remaining observations are subdivided into weak optimists, who have a level of optimism of 1 or 2, neutral optimists with a level of optimism of 0, and weak pessimists for which we found a level of optimism of -‐1 or -‐2. 32.76% of the population are weakly optimistic, 28.20% neutral and 23.46% weakly pessimistic.
As alternative optimism measure, we focus on the gap between the self-‐reported estimated life-‐satisfaction in five years and the current life satisfaction. As the current life satisfaction and the expected life-‐satisfaction in five years have both been rated on a scale from 0 to 10, the difference can range here once again between -‐10 and 10. The idea behind this measure is that optimistic people are likely to think that their life satisfaction will go up in the future, using their current life satisfaction as a benchmark. The interpretation of the optimism levels can be done in the same way than with the optimism measure described earlier. Both indicators have been constructed in the same way by Abolhassani and Alessie (2013).
The mean optimism level in this sample excluding the observations without any assets is however smaller, 0.008.
Note that using this alternative measure of optimism, and the same definitions of strong-‐ weak-‐ and neutral optimism as earlier, we have a much larger fraction of neutral
optimists: 50.27% of the individuals can be qualified as neutral, 23.28% as weakly optimistic, 20.32% as weakly pessimistic. Only 2.80% are strongly optimistic and 3.33% strongly pessimistic.
Indeed, the correlation coefficient between the self-‐reported life satisfaction in 2002 and the estimation for the life satisfaction in five years, done in 2002 is 0.7249, whereas the correlation coefficient between the estimation for the life satisfaction in five years and the actual life satisfaction five years later is only 0.4473.
Thus in total, we have four different samples that we will use to run regressions: two samples include observations of life satisfaction in 2007 and two do not. In both cases, one of the samples includes the observations without any assets and the other does not. Detailed descriptive statistics tables of the four samples we use are reported in appendix A.
To check for validity, Puri and Robinson compared their optimism measure with the outcome of the LOT-‐R questionnaire. As we do not dispose from data about this test, we verify whether three kinds of worries of the individuals in 2002 are correlated and have a significant impact on our optimism measure.
The three kinds of worries we take into account are the following: 1) Economic worries, 2) Job worries and 3) Peace worries. Worries about economic development may be a proxy for pessimism in financial matters. Being worried about the own job security is strongly linked to the personal situation of the individuals and will therefore probably, just as the general view on economic development, influence the financial decision-‐ making. The peace worries are rather driven by the character or the attitude of a person, as the risk exposure to military conflicts in Germany is actually the same for each
from 1 to 3, where 3 means very concerned, 2 somewhat concerned and 1 not concerned at all.
We find that there is a negative statistical significant pairwise correlation between our measure of optimism and each of the three types of worries, which means that people who are less concerned by these worries are rather optimistic. Furthermore, running a linear regression of optimism on the three types of worries and a bundle of control variables, we find that each of the three worries is significant at the 1% level. People who worry more are significantly less optimistic according to our optimism measure. The same holds for our alternative measure of optimism. These correlations and linear regressions will be reported in the appendix B.
b) How to measure portfolio composition?
In order to assess the portfolio composition of the households, we subdivided the assets of households into three different risk classes, in a similar way as it was done by
Barasinska, Schäfer and Stephan (2008): saving accounts and home ownership saving contracts are considered as safe assets, life insurance policies and fixed interested securities are considered as relatively risky assets and other securities and operating assets are considered as risky assets.
The data on portfolio composition is collected at the household level. One person of each household reported the types of assets. We assume through the whole analysis that the decisions are taken jointly within a household, so that we do not consider the
individuals who reported the portfolio composition only. We include all the individuals of the households over which we have data on the portfolio composition into the sample.
who hold assets that we earlier qualified as relatively risky but no risky assets, no matter whether they hold safe assets or not, and 3) the risky assets holders, who hold among others also other securities and operating assets. Note that the four categories, no assets, safe assets only, relatively risky asset holders and risky asset holders are mutually exclusive.
Including the individuals of households that report not to possess any assets in the sample of our initial measure of optimism, 9.33% are non-‐asset holders, 17.28% have safe assets only, 35.51% are qualified as relatively risky asset holders and the remaining 37.91% do have risky assets in their portfolio. Excluding the observations without any assets, we have 19.06% safe asset holders, 39.15% relatively risky asset holders, and 41.80% risky asset holders.
As already mentioned, in most of our settings we drop all the no asset holders from the sample and we only take into account the individuals of households that possess at least some assets as we think that households that report not to hold any assets are not useful to draw any conclusions about the link between optimism and portfolio composition. In this case, we only have three categories, safe, relatively risky and risky, which are still mutually exclusive. Nonetheless, we will include the individuals of households without assets in some settings in order to check for the robustness of our findings.
In some set-‐ups, we will merge the categories of relatively risky and risky asset holders.
c) What control variables do we include?
In a first step, we will consider the control variables used by Puri and Robinson (2007) in order to replicate their results as closely as possible. However, several adaptations have to be made, given that we use a different optimism measure and have a different set of data. The independent variables used by Puri and Robinson are: optimism, age, college, excellent health, male, net worth, risk tolerance, self-‐employed and white.
variable when it comes to decision-‐making about portfolio composition. We furthermore replace the variable „white“ by dummy variables of the German regions. We think that in Germany in 2002, the regions where people live have a more important impact than the ethnical background of the people on their portfolio composition.
The age of the individuals in our sample of the initial measure of optimism, excluding the individuals without assets, ranges from 17 to 92 years. The average age is approximately 46 and a half year. We do not take into account the different types of higher education that exist in Germany, such as for example universities or Fachhochschulen, but we only consider whether or not the individuals got a degree of higher education in our college dummy. Also higher education degrees obtained outside Germany are included in this dummy variable. In the most restrictive sample, 21.66% of the asked individuals do have a higher education degree. Furthermore, we control for male individuals as well as for self-‐employed ones. Self-‐employed people may have another perception of risk in financial matters and therefore behave in a different way while making financial decisions than people who are employed or retired. For these reasons, it may be interesting to control for self-‐employment.
48.12% of the participants of the survey are male and 5.96% are self-‐employed in the sample with 13211 observations.
Puri and Robinson mainly control for health because their measure of optimism is directly linked to life expectancy, which is among others driven by the current state of health of a person.
Even though our measure of optimism is not directly linked to health, we will also control for health in our regressions. It is possible that people who have a good state of health are more likely to have a longer planning horizon. Therefore they may make other financial decision than people who suffer from illnesses. We include a dummy variable into the regressions that takes the value 1 if people reported that they have a very good state of health and 0 otherwise.
Income and wealth are likely to play a crucial role in portfolio composition. The income we take into account is the monthly net household income. The monthly average
again the household level. As only few participants of the SOEP core study provide complete and detailed information about their wealth missing values were imputed. Both of these measures are expressed per thousands of euros in all our regressions.
As our dataset is representative for Germany around the year 2000, it seems more interesting to control for the region where people are from rather than for the skin colour, as it was done by Puri and Robinson, who used data from the United States. We therefore subdivide Germany into five regions: North, South, East, West and Berlin. As North, we define the Bundesländer Schleswig-‐Holstein, Niedersachsen, Hamburg and Bremen, as West Nordrhein-‐Westfalen, Hessen and Rheinland-‐Pfalz, as South Saarland, Baden-‐Württemberg and Bayern, as East Thüringen, Sachsen, Sachsen-‐Anhalt,
Mecklenburg-‐Vorpommern and Brandenburg. In the most restrictive sample, 13.36% of the individuals in are from the North, 27.48% from the South, 31.35% from the West, 24.21% from the East and 3.60% are from Berlin. The proportions are roughly
unchanged in the three other samples.
In the SOEP questionnaire of 2002, there is no question included concerning the self-‐ reported risk attitude of the individuals. We therefore assume that the risk attitude did not change between 2002 and 2004 and measure the risk-‐attitude using the scale from 0 to 10 on which the individuals were asked in 2004 to rate their own willingness to take risk in financial matters. A high reported numbers reflects a high willingness to take risks in financial matters.
d) What methodology do we use?
The first regressions we will run are logit regressions. We prefer logit to probit regressions in order to stay consistent with the multinomial logit regressions that we will run later. As dependent variable, we will use mainly the dummy variable risky assets. We will also run the same regressions using another dummy variable that we will call „relatively risky and risky“ and takes the value of 1 if an observation is qualified as relatively risky or risky asset holder and 0 otherwise.
measure of optimism. Then, in a second step, we replace that one by our alternative measure of optimism.
In order to get a clearer view on the impact of optimism, we will then in a third step replace the measures of optimism going from -‐10 to 10 by five dummy variables: strong optimism, weak optimism, neutral optimism, weak pessimism and strong pessimism. As already stated, we qualify people as strongly optimistic or pessimistic if their level of optimism is in absolute terms larger than 3. Weak optimism means that their level of optimism is 1 or 2, weak pessimism -‐1 or -‐2 and neutral optimism 0. Finally, we will run multinomial logit regressions. There, we use the three asset classes safe, relatively risky and risky as dependent variables and regress them on the measure of optimism and the usual bundle of control variables. As already stated, we will cluster the standard errors to take intra-‐household correlation into account. The results of these regressions will be provided in the following section.
4. RESULTS
First, we regressed risky as well as relatively risky and risky assets onto the discrete optimism variable and the bundle of control variables we described earlier, including the five different regions we defined in section (3). Table 2 shows the outcome of these regressions excluding the observations that hold no assets at all. The same regressions, including the observations without any assets can be found in appendix D1. In order to stay consistent with the multinomial logit regressions we will run later in this section, we will use logit regressions to estimate the impact of the different variables onto the
Table 2: Logit regression of the dummy variables risky assets and “relatively risky and risky assets“ onto our standard measure of optimism and the usual bundle of control variables excluding the observations without assets. Standard errors are clustered at the household level. Marginal effects are measured at the mean. Coefficients: Marginal Effects: VARIABLES RISKY REL. RISKY+
RISKY RISKY REL. RISKY + RISKY Optimism -‐0.019 -‐0.051*** -‐0.005 -‐0.005*** (0.012) (0.015) (0.003) (0.002) Age -‐0.016*** -‐0.027*** -‐0.004*** -‐0.003*** (0.002) (0.002) (0.000) (0.000) Male -‐0.196*** -‐0.022 -‐0.048*** -‐0.002 (0.026) (0.032) (0.006) (0.003) College 0.578*** 0.275*** 0.142*** 0.028*** (0.056) (0.081) (0.014) (0.008) Self-‐Employed 0.440*** 0.623*** 0.108*** 0.063*** (0.097) (0.169) (0.024) (0.017) Risk Tolerance 0.206*** 0.125*** 0.051*** 0.013*** (0.011) (0.014) (0.002) (0.002) Household Income 2 0.221*** 0.501*** 0.054*** 0.051*** (0.026) (0.051) (0.007) (0.005) Net Wealth 3 0.218*** 0.237*** 0.054*** 0.024*** (0.021) (0.035) (0.005) (0.003) Very Good Health -‐0.221*** -‐0.288*** -‐0.054*** -‐0.030*** (0.077) (0.102) (0.019) (0.010) North -‐0.245 -‐0.102 -‐0.060 -‐0.010 (0.158) (0.192) (0.039) (0.020) West -‐0.396*** -‐0.029 -‐0.097*** -‐0.003 (0.147) (0.179) (0.036) (0.018) South -‐0.462*** -‐0.284 -‐0.114*** -‐0.029 (0.149) (0.182) (0.037) (0.018) East -‐0.197 0.146 -‐0.048* 0.015 (0.148) (0.179) (0.036) (0.018) Constant -‐0.881*** 0.985*** (0.173) (0.224) Observations 13,211 13,211 Log likelihood Pseudo R2 -‐7572.895 0.157 -‐5440.802 0.154
Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
The unreported regression of safe assets only onto the same explanatory variables provides the same significance levels and the opposite sign for each and every variable than the relatively risky and risky regression, as the three categories of asset holders are mutually exclusive and the non-‐asset holders are dropped from the sample.
We find that optimism is negatively correlated to holding relatively risky or risky assets, but the impact on risky assets only is not significant. The marginal effect of optimism at the mean is in the case of relatively risky assets -‐0.005. The marginal effect is thus negative, but really close to 0.
We furthermore find that age reduces the likelihood of holding risky assets and that female participants of the study are more likely to hold risky assets than males. The gender has however no significant effect any more when we consider risky and relatively risky assets at the same time. As we assumed that members of a household decide together on the portfolio composition, we will not put too much weight on this finding.
College education, self-‐employment, a self-‐reported risk loving attitude in financial matters, high income and high net wealth increase the probability of holding risky assets.
The fact that people with higher income and higher net wealth are more likely to hold risky assets is probably linked to the fact that they need to spend a lower share of their income or wealth on consumption goods. As they can save more, they are likely to have a rather diversified portfolio – including risky assets.
are more likely to have risky assets. Being able to take the right financial decisions is a required quality in both cases: if people want to become self-‐employed and if people want to become active on the stock market, so it is not surprising that in many cases, self-‐employed people also own risky assets.
A very good health makes it less likely to hold risky assets, which is at the first view maybe counter-‐intuitive. It could be that this is just due to the reporting style of the people, that some are just more „enthusiastic“ about reporting their health. However, it is also possible that reporting a very good health includes another dimension of risk aversion. People who report that their health is „very good“ instead of „good“ are more likely to have done regular medical checks which give them the possibility to have a clearer picture of their state of health. This attitude may capture a non-‐financial type of risk aversion. To check whether this explanation holds, we replace in the regression the dummy variable very good health by a dummy variable „healthy“, which includes all people that reported to have a good or very good health. In this case, there is no significant effect of health on portfolio composition any more. These regressions are reported in appendix C.
Concerning the regions of Germany, the people from the South and from the West tend to have a lower probability of holding risky assets than people in Berlin. Controlling for all the other variables we have reported in this table, it is possible that regional cultural differences in financial matters are the reason for this finding. It may be that people in the South and in the West have a more conservative way of dealing with their savings than people in other parts of Germany. In appendix G, we decompose the regions into the different Bundesländer. We see that the coefficient is negative and significant at the 1-‐percent-‐level in Nordrhein-‐Westfalen (West), Rheinland-‐Pfalz (West) and Baden-‐ Württemberg (South), but also in Schleswig-‐Holstein (North). In Bayern (South), the coefficient is negative, but significant at the 10-‐percent-‐level only. In the other Länder that we consider as South or West, i.e. Saarland (South) and Hessen (West), there is no significant effect. When we consider relatively risky and risky assets together, the effects of the regions in Germany are insignificant.
In a second step, we keep the same dependent variables and the same control variables, however we replace our initial measure of optimism by the alternative optimism
measure, which we defined as the difference between the expected life satisfaction in five years and the current life satisfaction. These results are reported in Table 3. The findings in this setting are similar to the ones of the initial optimism measure: optimism has again no significant impact on risky assets only, but there is still a negative impact of optimism onto holding relatively risky or risky assets. In this setting, the marginal effect of the alternative optimism measure at the mean is -‐0.008. The marginal effects at the mean of all the variables are reported in appendix H1. The coefficient of age and the excellent health dummy variable are once again negative and significant, whereas income, net wealth, college education, self-‐employment and a less risk averse attitude have a positive impact on the probability of holding risky assets. The pattern does again not change much when considering risky assets only or relatively risky and risky assets simultaneously. In this setting, we excluded just as it was done earlier the individuals who reported not to own any assets from the sample. The results including these
Table 3: The initial measure of optimism is replaced by the alternative measure of optimism; the methodology we use is still the same than for the regressions of table 2.
VARIABLES RISKY REL. RISKY + RISKY Alternative Optimism -‐0.007 -‐0.077*** (0.017) (0.019) Age -‐0.017*** -‐0.029*** (0.002) (0.002) Male -‐0.190*** -‐0.047* (0.023) (0.028) College 0.610*** 0.328*** (0.052) (0.073) Self-‐Employed 0.423*** 0.697*** (0.088) (0.154) Risk Tolerance 0.196*** 0.118*** (0.010) (0.013) Household Income 0.198*** 0.488*** (0.023) (0.046) Net Wealth 0.228*** 0.248*** (0.019) (0.032) Very Good Health -‐0.251*** -‐0.360*** (0.069) (0.090) North -‐0.327** -‐0.198 (0.146) (0.180) West -‐0.475*** -‐0.126 (0.135) (0.170) South -‐0.489*** -‐0.341** (0.137) (0.172) East -‐0.274** 0.005 (0.137) (0.170) Constant -‐0.721*** 1.171*** (0.158) (0.213) Observations Log likelihood Pseudo R2 16,299 -‐9302.450 0.158 16,299 -‐6286.784 0.165
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
In order to get a more precise picture of how optimism is related to the different asset classes, we subdivide in a next step the individuals into five different subcategories. As already mentioned in section 3, we qualify people as very optimistic if their level of optimism is equal to or larger than 3. In the same fashion, they are qualified as very pessimistic if their level of optimism is -‐3 or smaller. The remaining individuals are qualified as weakly optimistic if their level of optimism is 1 or 2, neutral if it is 0, and weakly pessimistic if the level of optimism is -‐1 or -‐2. We use neutral optimism as benchmark and regress risky assets as well as relatively risky and risky assets on these classes of optimism and the control variables that we also used in the previous models. The outcome of these models can be found in table 4. The same regressions are done
Table 4: Replacing the discrete initial optimism variable by the subcategories strong optimism, weak optimism, neutral optimism, weak pessimism and strong pessimism.
VARIABLES RISKY REL. RISKY + RISKY Strong Optimism -‐0.280*** -‐0.353*** (0.081) (0.098) Weak Optimism -‐0.136*** -‐0.236*** (0.053) (0.067) Weak Pessimism -‐0.080 -‐0.013 (0.058) (0.071) Strong Pessimism -‐0.237** -‐0.099 (0.099) (0.110) Age -‐0.016*** -‐0.027*** (0.002) (0.002) Male -‐0.198*** -‐0.024 (0.026) (0.032) College 0.571*** 0.271*** (0.056) (0.081) Self-‐Employed 0.443*** 0.626*** (0.097) (0.169) Risk Tolerance 0.206*** 0.125*** (0.011) (0.014) Household Income 0.218*** 0.497*** (0.026) (0.051) Net Wealth 0.219*** 0.238*** (0.021) (0.035) Very Good Health -‐0.226*** -‐0.290*** (0.077) (0.102) North -‐0.252 -‐0.105 (0.158) (0.192) West -‐0.400*** -‐0.032 (0.147) (0.180) South -‐0.468*** -‐0.285 (0.149) (0.182) East -‐0.198 0.141 (0.148) (0.179) Constant -‐0.770*** 1.115*** (0.177) (0.229) Observations Log likelihood Pseudo R2 13,211 -‐7654.884 0.158 13,211 -‐5435.028 0.155 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
