barasinska et al. - individual risk attitudes and the composition of financial portfolios evidence...

The Quarterly Review of Economics and Finance 52 (2012) 1 14

Contents lists available at SciVerse ScienceDirect

The Quarterly Review of Economics and Finance

jo u rn al hom epage: www.elsev ier .co

Individual risk attitudes and the composition of Evidence from German household portfolios

Nataliya Barasinskaa,, Dorothea Schferb, Andreas Stephanc

a DIW Berlin anb DIW Berlin anc Jnkping Int

a r t i c l

Article history:Received 19 NReceived in reAccepted 18 OAvailable onlin

JEL classicatioD14G11

Keywords:Private householdsPortfolio diversicationRisk aversion

tweeios oanel ggestk-fre

liqunive

1. Introdu

Accordinpricing mowealth acrosied portfportfolio cothat a largeportfolios c(Brsch-SupHochguerteYunker & M

The litedence of un

We thankEuropean EconAnnual Meetinthe Jnkping Research (DIWand Dorothea Communitys 7number 21726

CorresponE-mail add

1062-9769/$ doi:10.1016/j.ction

g to modern portfolio theory and the capital assetdel (CAPM), investors should allocate their nancialss all assets available in the market, leading to diver-olios. However, numerous empirical studies nd thatmposition varies signicantly across investors and

portion of private investors hold under-diversiedomprised of only a small subset of available assetsan & Eymann, 2000; Burton, 2001; Campbell, 2006;l, Alessie, & Van Soest, 1997; King & Leape, 1998;elkumian, 2010).rature offers a number of explanations for the inci-der-diversied portfolios. Specically, it is conjectured

Agostino Manduchi, Martin Wallmeier and the participants of theomic Association Annual Congress in Milan, the German Associationg, the German Finance Association Annual Meeting and of seminars atInternational Business School, and the German Institute for Economic

Berlin) for valuable comments and insights. Nataliya BarasinskaSchfer gratefully acknowledge nancial support from the Europeanth Framework Programme (FP7/20072013) under grant agreement6.ding author.ress: [email protected] (N. Barasinska).

that high transaction and search costs (King & Leape, 1987;Merton, 1987), preferential tax treatment of certain assets (King& Leape, 1998), lack of information about investment oppor-tunities (King & Leape, 1987), and investors lack of nancialsophistication (Goetzmann & Kumar, 2008) all result in portfo-lio under-diversication. Empirical tests prove that these factorsdo indeed play an important role in portfolio composition deci-sions, but they do not fully explain the differences in portfoliocomposition and the high incidence of under-diversied portfo-lios. For instance, it is hard to imagine that it is transaction coststhat lead rich people to have under-diversied portfolios or thatlack of information affects the portfolio choices of experienced andsophisticated investors.

In this paper we consider that, in addition to the factors men-tioned above, investors propensity to hold incomplete portfoliosis strongly affected by another factor, namely, individual risk atti-tude. Risk aversion inuences investors preference for a specicportfolio composition because the level of portfolio risk dependson its composition. The direction of the relationship between riskaversion and willingness to hold an incomplete portfolio dependson the composition of the portfolio. For instance, if an incompleteportfolio is comprised of a single or very few risky assets, the pref-erence for such a portfolio should be negatively related to theinvestors risk aversion. The negative relationship should emergebecause investors can reduce the risk by allocating wealth among a

see front matter 2011 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.qref.2011.10.001d Freie Universitt Berlin, Germanyd Jnkping International Business School, Germany

ernational Business School, DIW Berlin, CESIS Stockholm, Sweden

e i n f o

ovember 2010vised form 8 July 2011ctober 2011e 30 October 2011

n:

a b s t r a c t

This paper explores the relationship bepropensity to hold incomplete portfoldata from the German Socioeconomic Ptoward nancial risk. Our ndings suportfolios consisting mainly of a few risassets is highly dependent on whether

2011 The Board of Trustees of the Um/locate /qre f

nancial portfolios:

n the self-declared risk aversion of private investors and theirf nancial assets. The analysis is based on household survey(SOEP) that provides a reliable measure of individual attitudes

that more risk averse households tend to hold incompletee assets. We also nd that the propensity to acquire additionalidity and safety needs are met.rsity of Illinois. Published by Elsevier B.V. All rights reserved.

2 N. Barasinska et al. / The Quarterly Review of Economics and Finance 52 (2012) 1 14

larger number of assets (Markowitz, 1952). A positive relationshipbetween risk aversion and willingness to hold an incomplete port-folio should emerge if the portfolio is comprised of only risk-freeassets. In this case, adding more assets implies investing in riskyassets, which results in higher portfolio risk. Hence, the effect ofrisk aversion depends on what assets are combined in a portfolio.Several empirical studies have included risk attitude as an explana-tory variable in their models of explaining portfolio composition(King & Leape, 1987, 1998; Kelly, 1995). However, these studies donot discuss the effect risk attitude has on the probability of holdinga particular

Our stuinvestors rbased on daby the Germterm portfsions: (1) thand (2) the cate to eachabout the aanalyze onldifferent asnation of ditheir portfo

Specicarisks to theclasses of xed-interenon-listed rst measufolio. Despiindividualsputting all tby nonprofeall availablethe entire pof portfoliocated invesassets accoassigns thesication, thhold.

The infothe SOEP sunancial rismeasure ofit is predicthave severaship of diffedecisions anpossible in indicators oprovides dewell as for yearly and Finally, a sigEven after w

1 Other studtors of actual i2002).

2 Vlaev, Stemore risk-avechoices in labo

cases where a decision-maker could not be identied, we have asample of 2628 individuals observed across 4 years 20042007 which amounts to a total of 10,512 observations.

The resunicant effeMoreover, effect. To prassuming tprobability sion. Under

entn inio coegatiortfoeringbabiersioed, r

a subthermompely ray bs nolysis., theple erce

the of w

also of sas behness, thety ait cot peride n be

Thus be p, this

an inleteegredingsets o

likeneeds, dutfolip betn is p

this ortanongios.

rem, weversi

on thicatoypot combination of assets.dy takes a closer look at the relationship betweenisk attitude and portfolio composition. The analysis ista on the asset holdings of German households collectedan Socioeconomic Panel (SOEP). In the literature, the

olio composition usually refers to two types of deci-e ownership decision, i.e., what kinds of assets to own,allocation decision, i.e., what portion of wealth to allo-

of these assets. Our data do not provide informationmounts invested in the individual assets and thus wey one aspect of portfolio composition: the ownership ofsets. We ask: Given a certain risk attitude, what combi-fferent asset types do household heads tend to hold inlio?lly, we relate individuals attitudes toward nancial

ir propensity to hold various combinations of six broadnancial assets: saving deposits, mortgage savings plans,st securities, shares of listed companies, and equity ofrms. Portfolio composition is measured two ways. The

re is the number of distinct asset types held in a port-te its simplicity, this measure reects the decisions of

who follow a naive diversication strategy of notheir eggs in one basket. Such a strategy is often engagedssional investors who split their wealth evenly among

assets types, hoping that this will reduce the risk ofortfolio (Benartzi & Thaler, 2001). The second measure

composition is designed to capture more sophisti-tment strategies. A sophisticated investor differentiatesrding to their return and risk properties and therebym to different return-risk classes. Based on this clas-e investor then decides what combination of assets to

rmation about risk attitude that we use is collected inrvey by asking respondents how willing they are to takeks. Dohmen et al. (2005) show that the SOEP survey

risk attitude is behaviorally relevant, in the sense thative of actual risk-taking behavior.1 The SOEP data alsol other advantages. First, information about the owner-rent asset types allows us to investigate actual portfoliod hence provide more reliable evidence than would be

an experimental setting.2 Second, the data set includesf who is the main decision-maker in a household andtailed socioeconomic information on this individual asthe whole household. Third, the survey is conductedallows tracing individuals and households over time.nicant advantage of the data is the size of the sample.e drop all observations with missing data and exclude

ies also demonstrate that self-declared risk attitudes are good predic-nvestment behavior (Fellner & Maciejovsky, 2007; Kapteyn & Teppa,

wart, and Chater (2008) present evidence that people behave in arse manner when investing in real life than when making investmentratory experiments.

investmity of aportfol

A nplete pconsidthe prorisk avexpectited to

Furfolio cnegativfolio mcontainthe anaple (i.esubsam75th paffectsdently

Weconsistreasonwilling(1936)by safeis credagainsto provfolio cabuffer.shouldindeedaverseincompwhen rthe holsafe asis moresafety

Thuthe portionshiaversiolio. Foran imption amportfol

Thesectionfolio didetailsthe indmain hlts of our analysis show that risk aversion has a sig-ct on the propensity to hold an incomplete portfolio.we nd both a positive and a negative relationshipeview our results, we nd, as hypothesized, that whenhat investors follow a naive investment strategy, theof holding only one asset type increases with risk aver-

the assumption that investors follow a sophisticated strategy, a positive relationship between the probabil-complete portfolio and risk aversion is found for thensisting of only risk-free assets.ve relationship between risk attitude and an incom-lio is found in only one instance. Specically, when

the sophisticated investment strategy, we nd thatlity of owning an incomplete portfolio decreases withn if the portfolio consists of only risky assets. Hence, asisk-averse investors dislike incomplete portfolios lim-set of risky assets.ore, the probability of holding a fully diversied port-

rised of all asset types available in the market iselated to risk aversion. In this case, an incomplete port-e preferred to a diversied portfolio when the latter

risky assets at all. This relationship is also found when is performed on a subsample of relatively wealthy peo-ir wealth exceeds the sample median wealth) or on aof the richest people (those whose wealth exceeds thentile of the sample distribution). Hence, risk attitudepropensity to hold an incomplete portfolio indepen-ealth.

nd that the majority of under-diversied portfoliosfe assets only. This observation sheds some light on theind the negative relationship between risk aversion and

to hold a fully diversied portfolio. As argued by Keynes economic activity of private households is dominatednd liquidity concerns. For an average household thatnstrained, nancial wealth works as a safety bufferiods of low income. Since asset holdings are intendedsafety in the rst place, adding risky assets to a port-

viewed as adding more risk and reducing the safety, the tendency to view asset holdings as a safety bufferositively related to a decision-makers risk aversion and,

is exactly what we discover. We nd that the more riskvestor, the more he or she will be inclined to hold an

portfolio consisting of a few safe assets. Furthermore,ssing the number of risky assets held in a portfolio ons of safe assets, we nd a positive effect of the number ofn the number of risky assets. Hence, a decision-makerly to add some risky assets to his or her portfolio whens have been met.e to the important role precautionary motives play ino decisions of private households and the positive rela-ween risk aversion and accumulation of safe assets, riskositively correlated with holding an incomplete portfo-reason, individual risk attitude should be considered ast factor in explaining differences in portfolio composi-

households and the high incidence of under-diversied

ainder of the paper is organized as follows. In the next review the literature on the role of risk aversion in port-cation. Section 3 describes our data and provides moree portfolio diversication measures. Section 4 presentsr of individual risk aversion. In Section 5, we test thehesis and discuss the results. In Section 6, we analyze

N. Barasinska et al. / The Quarterly Review of Economics and Finance 52 (2012) 1 14 3

the role precautionary motives play in diversication, followed bya concluding last section.

2. Literatu

Academtion dates Markowitz assets if thfolio returnwith high rate expectereturns becwith variantal assets prmeanvariarisk aversiothat investodegree of rideterminat

Despite show that ihold incom(Brsch-SupHochguerte2010). A grwhy so ma(Benartzi &Viceira, 2002005; KellyHowever, oto portfolio

A theoreprobability function of rlevels of rising risky inrisk-loving researchersused to hedhedging feaate levels ohump. Oninvestors, werant invesTherefore, vdiversied risk-loving but their powith modersied portfo

Gomes aporal portftypes: a pora portfolio analysis imis an increafor this relamoney manare their risnancial resassets. Risklittle wealthcover the xrisk-free as

There is very little empirical evidence on how risk attitudesaffect ownership of particular asset combinations. Kelly (1995)uses data from the 1983 Survey of Consumer Finances to assess

el of diversication in the nancial portfolios of U.S. house-iversication is measured in terms of the number of distinct

held in a portfolio. While controlling for a large number ofr chn on. Kingnshipets. s in the abfoliop becal eliterlevel

enc

e dat

analed inn Socpaner asonoual lhe himpoent

ns in we uriablholdnditiation

vari muln is:ith t?: (1ney after tney

of tselvere reoldsableortfoualsnagin

ner

SOEwns

depo

Germare review

ic research into determinants of portfolio diversica-back to Markowitzs (1952) meanvariance analysis.develops a model that explains how investors selectey care only about the mean and variance of port-s. One of the models implications is that investorsisk aversion prefer diversied portfolios with moder-d returns to undiversied portfolios with high expectedause diversication reduces the portfolio risk associatedce of returns on individual assets. However, the capi-icing model (CAPM), which is derived from Markovitzsnce analysis, does not predict any relationship betweenn and level of diversication. This model conjecturesrs should hold diversied portfolios regardless of theirsk aversion, and the investors willingness to take risk isive only for the fraction of risky assets in the portfolio.the predictions of CAPM, numerous empirical studiesnvestors and especially private households oftenplete portfolios consisting of a few risk-free assetsan & Eymann, 2000; Burton, 2001; Campbell, 2006;l et al., 1997; King & Leape, 1998; Yunker & Melkumian,eat deal of empirical work is aimed at understandingny private investors hold under-diversied portfolios

Thaler, 2001; Blume & Friend, 1975; Campbell, Chan, &3; Goetzmann & Kumar, 2008; Gomes & Michaelides,, 1995; King & Leape, 1998; Polkovnichenko, 2005).nly a few scholars analyze how risk aversion is related

holdings.tical study by Campbell et al. (2003) shows that theof holding multiple assets might be a hump-shapedisk aversion. Specically, individuals with intermediatek aversion are predicted to hold multiple assets, includ-vestments. In contrast, both extremely risk-averse andinvestors should hold less diversied portfolios. The

explain this idea by noting that some risky assets can bege against uctuations in their own future returns. Thisture should be attractive for investors with intermedi-f risk aversion, thus forming the middle of the demand

the slopes of the hump are the very conservativeho tend to avoid any risk, and the extremely risk tol-

tors, who have little interest in intertemporal hedging.ery risk averse investors should choose to hold under-portfolios consisting mainly of safe assets; extremelyinvestors should hold under-diversied portfolios too,rtfolios will contain only risky assets. Finally, investorsate risk aversion are expected to hold the most diver-lios consisting of all available assets.nd Michaelides (2005) formulate a model of intertem-olio choice explaining preferences for two portfoliotfolio comprised of a risk free asset and a risky one, andcomprised of a risk-free asset only. The results of theply that the probability of owning both types of assetssing function of risk aversion. The explanation giventionship is that risk-averse investors are more prudentagers and thus more likely to accumulate wealth thank-loving counterparts. The availability of considerableources in turn motivates investors to acquire additional-prone investors, in contrast, tend to accumulate very

and thus most of them do not have enough means toed costs of market participation and hence hold only a

set.

the levholds. Dstocks investoaversiopeoplerelatioent assmatter

As tof porttionshiempirito the micro-

3. Evid

3.1. Th

OurticipatGermaanced alone osocioecindividabout t

An investmdecisiomakerrst vahousethe coinformsecondwithinquestioto do wreceivethe molooks athe moa sharefor our1 or 2 ahousehidentiwith pindividfor ma

3.2. Ow

Thehold osaving

3 The aracteristics, the author nds a negative effect of risk the number of stocks held in the portfolios of wealthy

and Leape (1987, 1998) also nd evidence of a negative between risk aversion and holding a mix of differ-Yet, none of these studies discusses why risk attitudehis instance or why the relationship is negative.ove literature survey demonstrates, theoretical models

decision making are in disagreement as to the rela-tween risk aversion and diversication. Furthermore,vidence on the issue is scarce. This paper contributesature by examining the effects of risk aversion using

data from the SOEP survey.

e on household portfolios from the SOEP

a set

ysis is based on a sample of 2628 individuals who par- four subsequent waves, 2004 through 2007, of theioeconomic Panel (SOEP) survey. The data set is a bal-l. The unit of observation is the individual, either living

a member of a multi-person household. Most of themic data, including the risk attitudes, are collected at theevel, but the survey also collected detailed informationousehold to which the surveyed individuals belongs.rtant question that needs to be addressed in a study of

behavior has to do with who is making the investment a multi-person household. To identify the decision-se two indicator variables provided by the SOEP. The

e indicates who is the household head. The SOEP denes head as the person with the best knowledge aboutons under which the household functions. Using this, we retain only household heads in our sample. Theable provides information about money managementti-person households. The exact wording of the survey

How do you and your partner (or spouse) decide whathe income that either you or your partner or both of you) Everyone looks after his own money, (2) I look afternd provide my partner with a share of it, (3) My partnerhe money and provides me with a share of it, (4) We puttogether and both of us take what we need, (5) We puthe money in together, and both of us keep a share of its. Only those individuals who chose either alternativetained in our sample. We conne our analysis to these

because in the other cases the decision-maker is not and, hence, we cannot connect individual risk attitudelio composition decisions. Thus, the sample consists of

who are household heads and are primarily responsibleg the households money.

ship of nancial assets

P survey contains information on whether a house-any of the following six types of nancial assets: banksits, mortgage savings plans,3 life insurance policies,

n term is Bausparvertrag.


ypes i

xed-interebonds issupapers of lwarrants helisted rmsclass is not

Fig. 1 doed asset tyBank deposmost frequeour sampleyears, altholife insuran

3.3. Portfol

Even thocommon apportfolios. Eon the datanumber of and Rouweinstrumentments in ormeasure is Markowitzmation abosecurity, inf

Most homost, whataddition tofrom privatinformationportfolios. Fhousehold sied. Liquirole for thelife insuran

4 Goetzmaninvestment stportfolio is no

oldssitiozi anuel ew sotegy

n aving icy oftive ect

sop

Naivehe fostmes of dggs i

naivle. ThFig. 2mberes wFig. 1. Ownership rates of different asset t

st securities (including federal savings bonds, savinged by banks, and mortgage-backed bonds), securityisted companies (including stocks, bonds, and equityld directly or through mutual funds), and equity of non-. Information about the amount invested in each assetprovided.cuments the fraction of households owning the speci-pes at the beginning and end of the observation period.its, life insurance, and mortgage savings plans are thently held types of assets for the private households in. The gures do not change very much over the fourugh a slight decline in ownership of bank deposits andce is observable.

io composition

ugh portfolio analysis has a long history, there is noproach to measuring the composition of householdmpirical studies suggest various methods, depending

at hand. Blume and Friend (1975) consider the totaldistinct securities in a portfolio. Goetzmann, Lingfeng,nhorst (2005) correct the total number of nancials for the correlation among returns on these instru-

househcompoBenartDeMigto follo1/n straamong

Taktendenalternato rereect

3.3.1. In t

to inveproleyour eingly, apossibtypes. the nuallocatder to account for passive diversication.4 The latterwell suited for portfolio analysis in the framework ofs meanvariance approach. However, it requires infor-ut the share of wealth allocated to each individualormation rarely provided in household surveys.usehold surveys report which assets are held or, at

amounts are invested in broad groups of assets. In the difculty of obtaining exact nancial informatione persons, another reason for the usually unspecic

collected is because most households hold very simpleor example, Campbell (2006) shows that the majority ofnancial portfolios in the United States are poorly diver-d assets (e.g., cash, demand funds) play the dominant

poor, while less liquid savings (e.g., savings accounts,ce contracts) dominate the portfolios of middle-class

n et al. (2005) use the term passive diversication with regard torategies when correlation between individual assets included in at taken into account and only the number of assets matters.

six-asset-ty

3.3.2. SophOur seco

capture moonly for thecombinatio

The six according toTable A.1). Basset, denis not possigorization dand Eyman

5 This appro(2000), BanksJappelli (2000n the sample.

. Carroll (1995) documents a similar pattern of portfolion among European households. Moreover, as shown byd Thaler (2001); DeMiguel, Garlappi, and Uppal (2009);t al. (2009), it is not rare for nonprofessional investorsme naive or heuristic diversication strategy, e.g., the, according to which investors split their wealth evenlyailable assets.nto account the specic attributes of our data and the

households to hold simple portfolios, we construct twomeasures of portfolio composition. The rst is intendednaive investment strategies and the second one tohisticated investment strategies.

investment strategyllowing, the term naive investment strategy refersnt behavior that ignores differences in the risk-returnifferent asset types, instead relying on a dont put alln one basket plan of action to diversify risk. Accord-e household tends to invest in as many asset types ase SOEP data allow identication of six distinct asset

shows the distribution of individuals in our sample by of asset types held. The largest fraction of individualsealth among two or three asset types; while owners of

pes portfolios make up less than 1% of the sample.

isticated investment strategynd measure of portfolio composition is constructed tore sophisticated investment patterns. It accounts not

number of assets, but also for their degree of risk andn in a portfolio. The measure is constructed as follows.available asset types are grouped into three classes

their riskiness: low risk, moderate risk, and high risk (seeecause we do not know the returns on each individual

ing riskiness according to the meanvariance approachble. Instead, we use a more simple, but feasible, cate-rawing on Blume and Friend (1975) and Brsch-Supann (2000).5

ach has also been applied by Alessie, Hochguertel, and van Soest and Smith (2000), Bertaut and Starr-McCluer (2002), Guiso and).


Fig. 2. Number of asset types held in portfolios.

Under this categorization, bank deposits are deemed to beclearly safeare guarantinterest assthe duratiopolicies do nreal return lower than listed rmswell as rmthe no freeto assets in moderate eto assets in are not perf

Based o(Table A.2).i.e., either sdiversicat

what asset type is held, an undiversied portfolio can have low risk), mcludeite d

nedType

safeky asclude

samFig.

safesets o divmerminod tooldsent

Fig. 3. Distribuportfolio of risType 6: quite d because their returns exhibit no variation and theyeed by the nancial institution. The returns on xed-ets are also stable; however, the real payoff depends onn and on the issuers rating. Holders of life insuranceot bear the risk of losing the entire investment, but the

upon termination is uncertain and can be signicantlythe expected return. Listed securities and equity of non-

are the most risky, since stock prices and dividends, as value, are volatile and uncertain. In accordance with

lunch principle, the lowest expected return is assignedthe safe class; relatively risky assets are assumed to havexpected returns; the highest expected return is assignedthe risky class. We assume that the dened asset classesectly positively correlated.n this classication, we dene seven portfolio types

A portfolio that consists of assets from only one class,afe, relatively risky, or risky, has the least degree of

ion and is referred to as undiversied. Depending on

(Type 1that into as quare detypes: sists ofand risthat in

Thetypes (towardsafe asals whalso nutute a assignehousehinvestmtion of individuals by portfolio types. Type 1: undiversied portfolio of safe assets; Type 2:ky assets; Type 4: quite diversied portfolio comprised of safe and relatively risky assets;iversied portfolio comprised of relatively risky and risky assets; Type 7: fully diversieoderate risk (Type 2), or high risk (Type 3). A portfolios assets from at least two different classes is referrediversied. Different types of quite diversied portfolios

according to the degree of risk of the included asset 4 includes safe and relatively risky assets, Type 5 con-

and risky assets, and Type 6 contains relatively riskysets. Finally, the fully diversied portfolio (Type 7) is ones assets from all three classes.ple distribution with respect to the seven portfolio3) indicates that households have a strong tendencyty: most of them hold either incomplete portfolios ofor a mix of safe and relatively risky assets. Individu-ersify their investments over all three asset classes areous. Owners of portfolios with few risky assets consti-rity in our sample. Hence, if the risk-return proles

the six asset types are correct, we can argue that most choose to forgo higher returns in favor of safety of theirs. undiversied portfolio of relatively risky assets; Type 3: undiversied Type 5: quite diversied portfolio comprised of safe and risky assets;d portfolio containing assets from all three risk groups.


Fig

4. Risk ave

As a meaattitudes tothe SOEP intheir willinwording of ingness to willing to tavalidity of texperimenttion did in risks.6

Two adjtudes to maconvert theof risk avershigher numfully prepadenotes noThe new disthe sample aversion in risk averse.

Becauseyear, a furtpanel-data able assum4-year periperiods of n

To paintattitudes anysis. For insubsamplesthe portfoliegy. Fig. 5 sholding a sfour, ve, a

6 For details7 We acknow

not unanimoutime-invariantdence that riskHelmenstein, Krisk attitudes d

number of those holding more than four asset types is too small toallow a reliable statistical inference. For each of the distributions,we report in Table 1 the mean value of risk aversion, the quartiles ofthe distribuof househoof differencare compartogether wi

Judging in a portfol

egrend thps eypes eren

of rihat . In pshiftn of verse ond, ticateepor6. Taof inithinn thordiniffert riskpriserepoatelyestorsets.(portvest

signssetsner

ple wriskyeopleatelyssetsio tyratelisk a. 4. Distribution of individuals by degree of risk aversion.

rsion

sure of risk aversion, we use individuals self-reportedward nancial risks. This information was collected by

2004 by asking respondents to assess the strength ofgness to take risks when investing money. The exactthe SOEP question is: How would you rate your will-take risks in nancial matters on a scale from 0 (notke any risks) to 10 (fully prepared to take risks)? Thehe individuals responses to the question was veriedally and it was shown that the self-reported informa-fact reect the individuals attitudes toward nancial

ustments are made to the original indicator of risk atti-ke it suitable for the purposes of our analysis. First, we

indicator from a measure of risk tolerance to a measureion. This is accomplished by reversing the scale so thatbers correspond to higher risk aversion: 0 denotesred to take risks, i.e., the lowest risk aversion, and 10t willing to take any risks, i.e., the highest risk aversion.crete variable that emerges is called FRA. Fig. 4 presentsdistribution of individuals according to their level of risk2004. Most respondents perceive themselves as highly

information about risk attitudes is available for only 1her adjustment is necessary to make it applicable in acontext. We treat the measure as a time-invariant vari-ing that attitudes toward risk remain stable over theod, which appears to be a reasonable assumption forormal economic conditions.7

mean dtype, aof grouasset tthe diffdegreeshow tgroupstion is fractiorisk av

Secsophistudes rin Fig. group sion wbetwee

Acctypes dhighes1, comsion is moderby invrisk asassets than inare notrisky athan ow3). Peohighly than pmoderrisky aportfol(modelower r a preliminary picture of the relationship between riskd portfolio composition, we conduct a descriptive anal-stance, we plot the distribution of risk attitudes for

of investors with distinct portfolio types. First, considero types dened according to the naive investment strat-hows the distribution of risk aversion among investorspecied number of distinct asset types. Investors withnd six distinct assets are grouped together because the

and discussion of the validity tests, see Dohmen et al. (2005).ledge that this assumption is quite restrictive as the literature does

sly support the presumption that risk attitude as an individual trait is. For instance, Barsky, Kimball, Juster, and Shapiro (1997) provide evi-

preferences are relatively stable over time, while El-Sehity, Haumer,irchler, and Maciejovsky (2002) challenge this nding and argue thato vary over time.

ever, they atype 3 (onlyrisky assetscomprised less risk avlios of typeassets); arefolio type 3 with regardof safe and highly riskydistributiononly with rerespect to thdistributionless risky potion of risk aversion within the group, and the numberlds in the group. We also report the results of a t-testes between the mean values of the groups. The groupsed pairwise, and the respective t-statistics are reportedth p-values.from the reported statistics, the more asset types heldio, the lower the degree of risk aversion reported. Thee of risk aversion decreases with each additional assete differences are statistically signicant for each pair

xcept for two: those who hold none of the consideredand those who hold only one asset type. Furthermore,ces between distinct groups are not limited to the meansk aversion. Figures obtained for the sample quartilesthe whole distribution of risk attitudes varies acrossarticular, as the number of assets increases, the distribu-ed toward the origin of the scale, implying a decreasingery risk averse people and an increasing fraction of lesspeople.consider the portfolio types dened according to thed investment strategy. The distributions of risk atti-ted by investors with distinct portfolio types are shownble 2 reports the mean degree of risk aversion for eachvestors, the quartiles of the distribution of risk aver-

the groups, and the statistical signicance of differencese group-specic mean values of risk aversion.g to the statistics, investors holding distinct portfolio

also with regard to their risk attitudes. On average, the aversion is reported by subjects holding portfolio typed of only safe assets. A signicantly lower risk aver-rted by investors holding portfolio type 2, consisting of

risky assets. An even lower risk aversion is reporteds with portfolio type 3, which is comprised of high-

Subjects holding a mix of safe and moderately riskyfolio type 4) are on average somewhat less risk averseors holding only safe assets (portfolio type 2), but theyicantly different from people holding only moderately

(portfolio type 2) and are signicantly more risk averses of portfolio comprised of only highly risky assets (typeith a fairly diversied portfolio containing safe and

assets (portfolio type 5) are on average less risk averse with portfolio type 1 (only safe assets), type 2 (only

risky assets), or type 4 (a mix of safe and moderately), but are signicantly more risk averse than owners ofpe 3 (only high-risk assets). Owners of portfolio type 6y and highly risky assets) report on average signicantlyversion than owners of portfolio types 1, 2, and 4; how-re not signicantly different from owners of portfolio

highly risky assets) or type 5 (a mix of safe and highly). Finally, investors with the most diversied portfoliosof all types of assets (portfolio type 7) are signicantlyerse compared to owners of under-diversied portfo-

1 (only safe assets) and type 2 (only moderately risky signicantly more risk averse than the owners of port-(only highly risky assets); and do not differ signicantly

to risk attitudes from owners of portfolio type 5 (a mixhighly risky assets) or type 6 (a mix of moderately and

assets). The reported quartiles of the group-specics of risk aversion indicate that the groups differ notspect to the mean value of risk aversion, but also withe form and location of the distribution: specically, the

shifts further toward the origin when we move fromrtfolios toward less risky portfolios.


0.2

.4

.2

.4

0 5 10

1

None N of asset types = 1 N of asset types = 2

N of asset types = 3 N of asset types = 4

Frac

tion

st, 1 of ass

Table 1Summary of ri

Asset holdin

None1 asset type 2 asset types3 asset types4 asset typ

*** Level of sig

In shortholding difftudes. Moreamong inveand what alarger numaverse thanrelationshipis more risk

Table 2Summary of ri

Portfolio typ

Type 1 Type 2 Type 3 Type 4 Type 5 Type 6 Type 7

* Level of sig** Level of sig

*** Level of sig0

0 5 10 0 5

Degree of risk aversion (0 loweFig. 5. Distribution of risk aversion by the numbersk aversion in household groups with different number of asset types.

gs Mean p25 p50 p75 N t-Test

None

8.16 7 9 10 4648.10 7 9 10 591 0.44

7.60 6 8 10 614 4.17*

7.03 5 7 9 579 8.30*

es 6.52 5 7 8 380 10.61*

nicance: p-value


0.2

.4

.4

Undiversified/ low risk Undiversified/ medium risk Undiversified/ high risk

Fairly diversified/ low risk Fairly diversified/ medium

t, 10f asset

5. Regress

5.1. The mo

Our maiand economplete portfohypothesis,combinatioconomic vahousehold-be importaof the variareported in

The twovariables wSpecicallyaccording tovalues, froma householdtions of asstakes on eigin Section 3types are he

To test tnaive inveslogistic reg

8 There is wconomic and decisions. In pin income, ageferent nanciasupports this Campbell (200

ent of thsumd annd Cion the Hnrm0.2

0.2

.4

0 5

0 5 10

Diversified

Frac

tion

Degree of risk aversion (0 lowesFig. 6. Distribution of risk aversion by number o

ion analysis

del

n hypothesis is that risk aversion has a statisticallyically signicant effect on the ownership of incom-lios by private households, ceteris paribus. To test this

we model the probability of observing a certain asset

dependresultssion asviolateUhler aregressmore, t(IIA) con as a function of risk aversion and a set of socioe-riables. The latter comprise various factors from the

and individual-specic level that are considered tont determinants of investment behavior.8 Descriptionbles is provided in Table A.4. Summary statistics are

Table 3. measures of portfolio composition are categoricalith J mutually exclusive and exhaustive alternatives., the measure reecting different combinations of assets

the naive investment strategy takes on ve successive 0 to 4, according to the number of asset types owned by. The second measure, reecting all possible combina-

ets according to the sophisticated investment strategy,ht values corresponding to the portfolio types dened.3.2, including the case when none of the specied assetld.he effects of risk aversion under the assumption of atment strategy, we should t the data to an orderedression model because of the ordinal nature of the

ide agreement in the empirical literature that investors socioe-demographic characteristics have signicant inuence on portfolioarticular, Uhler and Cragg (1971) and Tin (1998) nd that differences, and education explain a large portion of variation in number of dif-l assets held by U.S. households; evidence from more recent studiesnding. See, e.g., Brsch-Supan and Eymann (2000), Burton (2001),6), Hochguertel et al. (1997), King and Leape (1998).

in our case.The mod

where J = 5,nation, P(Yj

P(Yj) =ex

J

n=1

X is the vecof nancialdummies arpute robusestimator oindividuals

The effecated invesmial logistithe numberthe same as

5.2. Impact

The estimpredicted pdocumente10 0 5 10

risk Fairly diversified/ high risk

highest) types in a portfolio.

variable. However, after we estimated the model, thee Brant (1990) test indicated that the parallel regres-

ption (also called the proportional odds assumption) isd the data should be tted to another model. Similar toragg (1971), we employ a pooled multinomial logistichat relaxes the proportional odds assumption. Further-ausman test for independence of irrelevant alternativesed that a multinomial logit model is more appropriateel is specied as follows. For the case of J outcomes, the probability of observing a particular asset combi-), is:

p(Xj)

exp(Xn)n = 0, 1, 2, . . . , J; j = 0, 1, 2, . . . , J; j /= n.

(1)

tor of explanatory variables that includes the measure risk aversion and a range of control variables. Yeare included to control for time-specic effects. We com-t standard errors using the Huber-White sandwichf variance that allows for clustering of observations by.cts of risk aversion under the assumption of a sophisti-tment strategy are estimated using the same multino-c regression model with the sole difference being that

of outcomes, J, is now equal to 8. Control variables are in the case of employing a naive investment strategy.

of risk aversion under the naive investment strategy

ated marginal effects of explanatory variables on therobabilities of holding a given number of assets ared in Table 3. The marginal effects and probabilities are


Table 3The effects of nancial risk aversion on naive diversication.

Nassets = 0 Nassets = 1 Nassets = 2 Nassets = 3 Nassets 4FRA

ln(Househol

Household w

Household w

Household w

ln(Personal w

Property (d)

Male (d)

Age

Age2

University (d

Employed (d

Self-employ

Retired (d)

Married (d)

Separated (d

Adults

Children

Concerned

Year dummiProbability o

Probability(2

variable is a caof holding a giaversion). Thein parentheses


*** Level of sig

calculated asample mea

Overall, sample distasset types of 5 are moportfolios coprobabilitie

The estimimportant dThe probabwhen the leof FRA on thstatisticallymoderate rlio. The prorelated to rito invest inin four and by one unit0.005** 0.010***

(0.002) (0.002) d income) 0.116*** 0.140***

(0.008) (0.011) ealth50 (d) 0.129*** 0.004

(0.007) (0.016) ealth75 (d) 0.123*** 0.033

(0.010) (0.017) ealth100 (d) 0.077*** 0.052**

(0.013) (0.020) ealth) 0.010*** 0.006***

(0.001) (0.002) 0.022* 0.041**

(0.011) (0.013) 0.021** 0.009 (0.008) (0.010) 0.000 0.001 (0.002) (0.002) 0.000 0.000***

(0.000) (0.000)) 0.046*** 0.018

(0.009) (0.013) ) 0.071*** 0.026

(0.010) (0.015)

ed (d) 0.044* 0.010

(0.020) (0.024) 0.050*** 0.028 (0.013) (0.019) 0.022 0.010 (0.012) (0.017)

) 0.047*** 0.015 (0.011) (0.015) 0.026*** 0.004 (0.007) (0.010)0.030*** 0.023**

(0.005) (0.008) 0.033*** 0.025***(0.006) (0.007)

es Yes Yesf outcome 0.13 0.23

) = 0.00, Log Likelihood = 14, 085, Pseudo-R2 = 0.16, Nobs = 10, 512. This table reports tegorical variable that takes ve successive values, according to the number of asset typesven number of asset types. The variable FRA indicates the degree of nancial risk aversi

marginal effects and predicted probabilities are calculated at FRA = 5, while other variable.nicance: p-value


Table 4The effects of nancial risk aversion on sophisticated diversication.

No assets Undiversied Quite diversied Fully diversied

FRA

ln(Househol

Household w

Household w

Household w

ln(Personal

Property (d)

Male (d)

Age

Age2

University (d

Employed (d

Self-employ

Retired (d)

Married (d)

Separated (d

Adults

Children

Concerned


Probability(2

variable is a caof outcome is taversion) to 10standard error


*** Level of sig

sophisticatethe numbedetail in the

5.3. Impactstrategy

In this ssion on portmore sophimodel in whcorrespondThe estimatof given po

Househoto hold porprised of sais 32%. Thetive but staType 1 Type 2 Type 3

0.005* 0.016*** 0.000 0.002***(0.002) (0.002) (0.001) (0.000)

d income) 0.120*** 0.147*** 0.009 0.000 (0.008) (0.011) (0.006) (0.002)

ealth50 (d) 0.132*** 0.068*** 0.028*** 0.002 (0.008) (0.018) (0.005) (0.003)

ealth75 (d) 0.125*** 0.014 0.024*** 0.001 (0.010) (0.019) (0.006) (0.003)

ealth100 (d) 0.081*** 0.041* 0.003 0.002 (0.013) (0.021) (0.007) (0.003)

wealth) 0.010*** 0.004** 0.002*** 0.000 (0.001) (0.002) (0.001) (0.000) 0.021 0.006 0.013* 0.004 (0.011) (0.014) (0.005) (0.002)0.022** 0.003 0.005 0.007**

(0.008) (0.010) (0.004) (0.002) 0.000 0.006** 0.006*** 0.000 (0.002) (0.002) (0.001) (0.000) 0.000 0.000*** 0.000*** 0.000 (0.000) (0.000) (0.000) (0.000)

*** *) 0.047 0.019 0.000 0.007(0.010) (0.014) (0.006) (0.003)

) 0.072*** 0.017 0.003 0.007*(0.011) (0.015) (0.007) (0.003)

ed (d) 0.051* 0.067** 0.015 0.011*(0.021) (0.022) (0.010) (0.005) 0.053*** 0.000 0.000 0.008*(0.013) (0.020) (0.009) (0.003) 0.024 0.022 0.018* 0.001 (0.012) (0.017) (0.009) (0.003)

) 0.048*** 0.022 0.012 0.005 (0.012) (0.014) (0.006) (0.003) 0.028*** 0.018 0.004 0.001 (0.007) (0.010) (0.004) (0.002) 0.033*** 0.023** 0.006* 0.001 (0.005) (0.008) (0.003) (0.001) 0.033*** 0.017* 0.006 0.002 (0.006) (0.008) (0.003) (0.001)

es Yes Yes Yes Yes f outcome 0.15 0.22 0.04 0.02

) = 0.00, Log Likelihood = 15, 232, Pseudo-R2 = 0.16, Nobs = 10, 512. This table reports tegorical variable that takes eight different values corresponding to the seven portfolio tyhe predicted probability of holding a given portfolio type. The variable FRA indicates the

(highest risk aversion). The marginal effects and predicted probabilities are calculated ats are reported in parentheses.nicance: p-value


0

.1

.2

.3

.4

Prob

abilit

y

0

Degree

N=1N=2

Number ofasset types

Fig. 7. Effect number of ass

0

.1

.2

.3

.4

Prob

abilit

y

0 Degree

Fig. 8. Effect portfolio type versied portfsafe and relatiand risky asserisk groups.

Finally, ing fully divaversion.9 Aperfectly pocosts, we wattractive torisk averse almost linewith the p

9 We also esin a model whvalue of housethese variablesectional dataagain conrmof holding a di

Table 5The effects of the number of safe assets on the number of risky assets held.

Nsafe assets

FRA

ln(Househol

Household w

hold w

hold w

sonal w

ty (d)

d) (0.009) (0.009) (0.001)0.003 0.003 0.000(0.002) (0.002) (0.000)2 4 6 8 10

of risk aversion (0 lowest, 10 highest)

N=3N>=4 House

House

ln(Per

Proper

Male (

Age of nancial risk aversion on the probability of holding a particularet types in portfolio.

2 4 6 8 10of risk aversion (0 lowest, 10 highest)

Type 1Type 4Type 5Type 7

Portfolio type

of nancial risk aversion on the probability of holding a particularaccording to the sophisticated investment strategy. Type 1: undi-olio of safe assets; Type 4: quite diversied portfolio comprised ofvely risky assets; Type 5: quite diversied portfolio comprised of safets; Type 7: fully diversied portfolio containing assets from all three

the effect of risk aversion on the probability of hold-ersied portfolio Type 7 is negatively related to riskssuming that returns on different asset types are notsitively correlated and there are no transaction or entryould expect that a fully diversied portfolio is more

individuals with moderate risk aversion than to veryor risk tolerant investors. Instead we nd a strong andar negative relationship. Thus, our ndings disagreeredictions of Campbell et al. (2003) and Gomes and

timate the effects of risk aversion on the sophisticated diversicationere we additionally include ownership of commercial real estate andhold total assets and liabilities as control variables. As the data ons are available for 2007 only, the model is estimated with a cross-

set. Nevertheless, the results obtained for this specication once the negative relationship between risk aversion and the probabilityversied portfolio.

Age2

University (d

Employed (d

Self-employ

Retired (d)

Married (d)

Separated (d

Adults

Children

Concerned


Probability(2

This table repdent variable 2, according tothe predicted count variableindicates the daversion) to 10ities are calculrobust standar


*** Level of sig

Michaelides(1998).

6. Extensio

Our analifested indidiversied behind saviary needs hpersonal saprivate hou

A numb(1989), CaNo risky assets One risky asset Two risky assets

0.073*** 0.071*** 0.003***(0.006) (0.006) (0.001)0.024*** 0.023*** 0.001***(0.002) (0.002) (0.000)

d income) 0.155*** 0.149*** 0.006***(0.011) (0.011) (0.001)

ealth50 (d) 0.055** 0.056** 0.002(0.019) (0.018) (0.002)

ealth75 (d) 0.127*** 0.125*** 0.001(0.020) (0.020) (0.002)

ealth100 (d) 0.141*** 0.140*** 0.001(0.023) (0.023) (0.002)

ealth) 0.011*** 0.011*** 0.001***(0.001) (0.001) (0.000)

0.018 0.017 0.001(0.012) (0.012) (0.001)0.025** 0.023** 0.0010.000 0.000 0.000(0.000) (0.000) (0.000)

) 0.079*** 0.074*** 0.005***(0.012) (0.011) (0.001)

) 0.017 0.018 0.001(0.015) (0.014) (0.000)

ed (d) 0.045* 0.007 0.038***(0.020) (0.018) (0.008)0.008 0.010 0.003(0.020) (0.020) (0.002)0.001 0.000 0.001(0.014) (0.014) (0.001)

) 0.051*** 0.051*** 0.000(0.012) (0.012) (0.001)0.036*** 0.035*** 0.001(0.008) (0.008) (0.001)0.049*** 0.048*** 0.001*(0.007) (0.007) (0.000)0.043*** 0.042*** 0.000(0.007) (0.006) (0.001)

es Yes Yes Yesf outcome 0.80 0.19 0.01

) = 0.00, Log Likelihood = 5274, Pseudo-R2 = 0.25, Nobs = 10, 512.orts marginal effects after multinomial logit regression. The depen-is a categorical variable that takes three successive values from 0 to

the number of risky assets in a portfolio. Probability of outcome isprobability of holding a given number of asset types. Nsafe assets is a

indicating the number of safe assets in a portfolio. The variable FRAegree of nancial risk aversion and takes values from 0 (lowest risk

(highest risk aversion). The marginal effects and predicted probabil-ated at FRA = 5, while other variables are held at their means. Clusterd errors are reported in parentheses.nicance: p-value


Table 6The effects of nancial risk aversion on the portfolio composition of wealthy investors.

Outcome is the number of assets held

Nassets = 2 Nassets = 3 Nassets 4Wealth > 50

FRA 0.001 0.003 0.010***Probabilit 0.29 0.30 0.18

Wealth > 75FRA 0.006 0.005 0.010***Probabilit 0.25 0.33 0.22

e 3

Wealth > 50FRA 02***

Probabilit 1 Wealth > 75

FRA 00 Probabilit 1

d

One r

Wealth > 50FRA 0.03Nsafe assets 0.08Probabilit 0.37

Wealth > 75FRA 0.03Nsafe assetsProbabilit

Sample size foaversion FRA opeople with wincluded in thegender, higherand predicted

** Level of sig*** Level of sig

Eisenhauermotive for s

The indidetermine wture holds, be rst anddeposits. Oa householdstocks. Thuonly one asmatches whexpect thatwhen their

To test tlogit modelnumber of rinclude riskaddition, wNSafe assets. E

As expecthe numbeassets. Ceteber of safe refrains froasset increarisky assetsNassets = 0 Nassets = 1

th percentile0.005** 0.010***

y of outcome 0.06 0.17 th percentile

0.000 0.003 y of outcome 0.06 0.14

Outcome is the portfolio type

No assets Type 1 Type 2 Typ

th percentile0.005*** 0.017*** 0.000 0.0

y of outcome 0.06 0.18 0.02 0.0th percentile

0.000 0.010*** 0.000 0.0y of outcome 0.06 0.15 0.03 < 0.0

Outcome is the number of risky assets hel

No risky assets

th percentile0.037***

0.091***y of outcome 0.62 th percentile

0.037******0.107 0.10

y of outcome 0.58 0.41

r people with wealth >50th percentile = 5177. Sample size for people with wealth >75th pen the probability of specied outcomes. The effects are estimated by means of multinomiealth exceeding the sample median of 8000D , and on a sub-sample of people with wealt

regressions (but not reported) are: the logarithm of the household total wealth and the in education, employment status, ownership of residential property, marital status and th

probabilities are calculated at FRA = 5. Cluster robust standard errors are reported in parenicance: p-value


effects. Thus, it is possible that our main result regarding the neg-ative relationship between diversication and risk aversion is dueto collinearity between risk attitude and wealth.

To discodiversicatysis conducpeople. Speals: (1) thesample methe 75th pethe same rseparately variables isone exceptfor the percontinuoushousehold wtude on the are reporteresults obtanaive invesresults for investmentthe analysis

The resuricher porleast one risamong the is predictedof holding arisky assetsof risk attitconsiderablattitude affof an invest

8. Conclus

This papsion and thhouseholdsand demogprobability to the levelmeanvariaasset pricinirrespectivendings areLeape (1998the number

Our expare credit cassets as a higher expetion of nanneeds; thusreducing ththe safety bmore risky more likelyand liquid a

Variationto explain ttors, includi

also play an important role. Therefore, the role of risk attitudesshould be considered complementary to other factors important inexplaining portfolio composition.

dix A.

les A.1A.4.

1zation of asset types according to their riskiness.

sk Moderate risk High risk

eposits Life insurance policies Listed securitiesage savings plans Fixed-interest securities Equity of non-listed rms

2n of portfolio types according to strategies of sophisticatedcation.

lio type Level ofdiversication

Asset classes included in portfolio

Safe Relativelyrisky

Risky

Undiversied + Undiversied + Undiversied + Quite diversied + + Quite diversied + +

Quite diversied + +Fully diversied + + +

ates that at least one asset of particular type is owned, indicates thats of particular type are owned.

3ion of explanatory variables.

le Description

Degree of nancial risk aversion, on ascale from 0 (very low) to 10 (veryhigh)

hold income Net annual income of all householdmembers, in D

hold wealth Total value of nancial assets and realproperty owned by the household, inD a. In the regression analysis, fourdummy variables are used to indicatethe level of wealth: HouseholdWealth25 = 1 if household wealth is inthe lower quartile of sampledistribution and =0 otherwise;Household Wealth50 = 1 if householdwealth > 25th and 50th percentile ofthe sample distribution and =0otherwise; Household Wealth75 = 1 ifhousehold wealth is > 50th and 75thpercentile of the sample distributionand =0 otherwise; HouseholdWealth100 = 1 if household wealth is >75th percentile of the sampledistribution and =0 otherwise.

al Wealth The value of personal share of the

ropert

e (d)ver whether risk attitude is indeed a relevant factor inion decisions regardless of wealth, we perform the anal-ted in the preceding sections on a subsample of wealthycically, we construct two groups of wealthy individu-

relatively wealthy people with wealth exceeding thedian and (2) the rich people with wealth exceedingrcentile of the sample distribution. We then estimateegression models as reported in Tables 3 through 5for each subsample. The specication of explanatory

similar to regressions reported in Tables 3 through 5,ion being that instead of including dummy variablescentiles of the wealth distribution, we now include a

variable ln(Wealth), which is a natural logarithm of theealth. The results with respect to the effect of risk atti-

probability of holding a particular combination of assetsd in Table 6. The upper part of the table summarizesined for asset combinations dened according to thetment strategy; the middle section of the table reportsportfolio types dened according to the sophisticated

strategy; and the bottom part of the table is devoted to of precautionary motives.lts reveal that wealthy people are more likely to holdtfolios consisting of several distinct asset types with atky asset among them. An important nding is that, evenrelatively wealthy, the richest individuals risk attitude

to have a signicant negative effect on the probability diversied portfolio and on the probability of including

in the portfolio. Hence, our results regarding the effectsude also hold for the subsample of households withe nancial resources. Therefore, we conclude that riskects the portfolio composition decision independentlyors wealth.

ions

er explores the link between self-declared risk aver-e composition of nancial portfolios held by private. Taking into account a wide range of socioeconomicraphic characteristics of households, we nd that theof holding incomplete portfolios is positively related

of risk aversion. This result is at odds with both thence principle of Markowitz (1952), and the capitalg model, which predicts that diversication is optimal

of the investors level of risk aversion. However, our largely in agreement with Kelly (1995) and King and), who also nd a negative inuence of risk aversion on

of assets held in a portfolio.lanation of the nding is that most private householdsonstrained and hence prefer to hold safe and liquidsafety buffer against periods of lower income and/ornditures. Hence, for most individuals, the primary func-cial wealth is to meet their precautionary and liquidity, adding any risky asset to the portfolio is perceived ase safety buffer. The higher the risk aversion, the largeruffer a household desires and the less likely it will owntypes of assets. In effect, more risk averse people are

to hold incomplete portfolios consisting of only safessets.

in risk attitudes in the population itself does not sufcehe high incidence of incomplete portfolios. Other fac-ng poor nancial sophistication and participation costs,

Appen

Tab

Table A.Categori

Low ri

Bank dMortg

Table A.Denitiodiversi

Portfo

Type 1Type 2Type 3

Type 4Type 5Type 6

Type 7

+ indicno asset

Table A.Descript

Variab

FRA

House

House

Person

Real P

Femal

Age Age2 households total assets owned by thehousehold head, in D a

y (d) =1 if household owns real property, =0otherwise=1 if household head is female, =0 ifmaleAge of the household head in yearsSquare of Age


Table A.3 (Continued)

Variable Description

University (d) =1 if respondent has university degree,0 otherwise

Employed (d) =1 if household head is employed, 0otherwise

Self-employed (d) =1 if household head is self-employed,0 otherwise

Retired (d) =1 if household head is retired, 0otherwise

Adults Number of adult household members(older than 18 years)

Children Number of children up to 18Concerned A categorical variable indicating

whether the individual is concerned

Note: (d) denoa Data abou

2007 only. Foassumption th

Table A.4Summary stat

Variable

FRA Household iHousehold wPersonal weReal propertFemale AgeUniversity EmployedSelf-employRetired Married Separated AdultsChildren Concerned

Very concSomewhaNot conce

Number of ind

References

Alessie, R., HoNetherland

Banks, J., & SmWorking P

Barsky, R. B., Kters and beretiremen

Benartzi, S. & Tbution sav

Bertaut, C. & SHousehold

Blume, M. E. & Friend, I. (1975). The asset structure of individual portfolios and someimplications for utility functions. Journal of Finance, 30, 585603.

Brsch-Supan, A., & Eymann, A. (2000). Household portfolios in Germany. Institut frVolkswirtschaftslehre und Statistik, Universitt Mannheim, Discussion Paper603-01.

Brant, R. (1990). Assessing proportionality in the proportional odds model for ordi-nal logistic regression. Biometrics, 46(4), 11711178.

Burton, D. (2001). Savings and investment behaviour in Britain: More questions thananswers. Service Industries Journal, 21(3), 130146.

Caballero, R. J. (1991). Earnings uncertainty and aggregate wealth accumulation.American Economic Review, 81(4), 859871.

Campbell, J. Y. (2006). Household nance. Journal of Finance, 61(4), 15531604.Campbell, J. Y., Chan, Y. L. & Viceira, L. M. (2003). A multivariate model of strategic

asset allocation. Journal of Financial Economics, 67(1), 4180.Carroll, C. D. (1995). Why do the rich save so much. NBER Working Paper, no. 6549.DeMiguel, V., Garlappi, L. & Uppal, R. (2009). Optimal versus naive diversication:

How inefcient is the 1/n portfolio strategy? Review of Financial Studies, 22,19151953.

, T., Fidual ated s, T., H

t bias kets. JoG. & e from350.nn, W

nce, 12nn, Worrelaabout his or her nancial standing(1 = very concerned, 2 = somewhatconcerned, 3 = not concerned at all)

tes dummy-variables.t nancial and real assets were collected by the SOEP in 2002 andr years 2004 through 2006, we calculate total wealth based on theat its value changes linearly over time.

istics of explanatory variables.

2004 2007

DohmenIndivvalid

El-Sehitysighmar

Fellner, denc338

GoetzmaFina

Goetzmaket cMean Std. Dev. Mean Std. Dev.

7.53 2.28 7.53 2.28ncome 25,657 16,014 27,343 20,841ealth 12,883 44,021 13,917 50,304

alth 9,194 39,248 10,325 45,313y 0.34 0.47 0.36 0.48

0.44 0.50 0.44 0.5049 16.76 52 16.760.19 0.39 0.21 0.400.58 0.49 0.57 0.50

ed 0.06 0.24 0.07 0.250.30 0.46 0.33 0.470.28 0.45 0.29 0.450.42 0.49 0.45 0.501.66 0.72 1.64 0.740.38 0.76 0.34 0.70

erned 30.40 0.46 28.01 0.45t concerned 50.42 0.50 48.82 0.49rned at all 19.18 0.39 23.17 0.42

ividuals in the panel, N = 2628.

chguertel, S., & van Soest, A. (2000). Household portfolios in thes. Tilburg University, Discussion Paper, p. 55.ith, S. (2000). UK household portfolios. Institute for Fiscal Studies,

aper, p. 14.imball, M. S., Juster, F. T. & Shapiro, M. D. (1997). Preference parame-havioral heterogeneity: An experimental approach in the health and

t study. Quarterly Journal of Economics, 112(2), 537579.haler, R. H. (2001). Naive diversication strategies in dened contri-ing plans. American Economic Review, 91(1), 7998.tarr-McCluer, M. (2002). In L. Guiso, M. Haliassos, & T. Jappelli (Eds.),

portfolios in the United States, Household portfolios. MIT Press.

Gomes, F. & Ming the em

Guiso, L., & JapHochguertel, S

and bonds99(1), 81

Kapteyn, A., &choice. Tilb

Kelly, M. (1995holds. Jour

Keynes, J. M. University

King, M. and J.Working P

King, M. & LeaJournal of P

Markowitz, H.Merton, R. C. (1

informatioPolkovnichenk

dependenSkinner, J. S. (1

ings. JournTin, J. (1998). H

Review of EUhler, R. S. & C

Review of EVentura, L. & E

Economic RVlaev, I., Stew

relativity gambles a

Wilson, B. (20and Financ

Yunker, J. A. & sication:Economics

Zeldes, S. P. (1certainty ealk, A., Huffman, D., Sunde, U., Schupp, J., & Wagner, G. G. (2005).risk attitudes: New evidence from a large, representative, experimentally-urvey. IZA Discussion Paper 1730.aumer, H., Helmenstein, C., Kirchler, E. & Maciejovsky, B. (2002). Hind-and individual risk attitude within the context of experimental asseturnal of Psychology and Financial Markets, 3(4), 227235.Maciejovsky, B. (2007). Risk attitude and market behavior: Evi-

experimental asset markets. Journal of Economic Psychology, 28(3),

. N. & Kumar, A. (2008). Equity portfolio diversication. Review of(3), 433463.. N., Lingfeng, L. & Rouwenhorst, K. G. (2005). Long-term global mar-tions. Journal of Business, 78, 138.ichaelides, A. (2005). Optimal life-cycle asset allocation: Understand-pirical evidence. Journal of Finance, 60(2), 869904.pelli, T. (2000). Household portfolios in Italy. CSEF Working Paper 43.., Alessie, R. & Van Soest, A. (1997). Saving accounts versus stocks

in household portfolio allocation. Scandinavian Journal of Economics,97.

Teppa, F. (2002). Subjective measures of risk aversion and portfoliourg University, Center for Economic Research, Discussion Paper 11.). All their eggs in one basket: Portfolio diversication of U.S. house-nal of Economic Behavior & Organization, 27(1), 8796.(1936). The General Theory of Employment, Interest and Money. The

of Adelaide Library Electronic Texts Collection. Leape (1987). Asset accumulation, information and the life cycle. NBER,aper (2392).pe, J. (1998). Wealth and portfolio composition: Theory and evidence.ublic Economics, 69, 155193.

(1952). Portfolio selection. Journal of Finance, 7(1), 7791.987). A simple model of capital market equilibrium with incompleten. Journal of Finance, 42(3), 483510.o, V. (2005). Household portfolio diversication: A case for rank-

t preferences. Review of Financial Studies, 18(4), 14671502.988). Risky income, life cycle consumption, and precautionary sav-al of Monetary Economics, 22(2), 237255.ousehold demand for nancial assets: A life-cycle analysis. Quarterlyconomics and Finance, 38, 875897.ragg, J. G. (1971). The structure of the asset portfolios of households.conomic Studies, 38(115), 341357.isenhauer, J. G. (2005). The relevance of precautionary saving. Germaneview, 6(1), 2335.art, N. & Chater, N. (2008). Risk preference discrepancy: A prospectaccount of the discrepancy between risk preferences in laboratorynd real world investments. Journal of Behavioral Finance, 9, 132148.03). Diversication of risk and saving. Quarterly Review of Economicse, 43, 697712.Melkumian, A. A. (2010). The effect of capital wealth on optimal diver-

Evidence from the survey of consumer nances. Quarterly Review of and Finance, 50, 9098.989). Optimal consumption with stochastic income: Deviations fromquivalence. Quarterly Journal of Economics, 104(2), 275298.

Individual risk attitudes and the composition of financial portfolios: Evidence from German household portfolios1 Introduction2 Literature review3 Evidence on household portfolios from the SOEP3.1 The data set3.2 Ownership of financial assets3.3 Portfolio composition3.3.1 Naive investment strategy3.3.2 Sophisticated investment strategy

4 Risk aversion5 Regression analysis5.1 The model5.2 Impact of risk aversion under the naive investment strategy5.3 Impact of risk aversion under the sophisticated investment strategy

6 Extension 1: the role of precautionary motives7 Extension 2: wealthy investors8 ConclusionsReferences

barasinska et al. - individual risk attitudes and the composition of financial portfolios evidence...

Documents

investors propensity

portion of private investors

incomplete portfolio

composition of evidence

specicportfolio composition

underdiversied portfolios

modern portfolio theory

german institute