Accepted Manuscript

Responsibility and Limited Liability in Decision Making for Others - An ex-

perimental consideration

Sascha Füllbrunn, Wolfgang J. Luhan

PII: S0167-4870(18)30647-0
DOI: https://doi.org/10.1016/j.joep.2019.06.009
Reference: JOEP 2186

To appear in: Journal of Economic Psychology

Received Date: 1 November 2018

Revised Date: 23 May 2019
Accepted Date: 25 June 2019

Please cite this article as: Füllbrunn, S., Luhan, W.J., Responsibility and Limited Liability in Decision Making for
Others - An experimental consideration, Journal of Economic Psychology (2019), doi: https://doi.org/10.1016/
j.joep.2019.06.009

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Responsibility and Limited Liability in Decision Making

for Others - An experimental consideration

∗ †‡
Sascha Füllbrunn, and Wolfgang J. Luhan

June 28, 2019

Abstract
Agency in nancial markets has been claimed to foster excessive risk taking, ultimately
leading to bubble formation. The main driving factor appears to be the skewed bonus
system for agents who invest other people's money. The resulting excessive risk taking
on behalf of others would imply that such bonus systems crowds out responsible decision
making for others in order to serve egoistic self-interest. To test this implication, we
conduct laboratory experiments comparing decision making for others with and without
such a bonus system. First, we show that, in the absence of bonus systems, decision makers
invested signicantly less for others than for themselves. Second, we show that limited
liable decision makersparticipating only in gains but not in lossesinvested substantially
more for others than for themselves. Hence, our results suggest that indeed limited liability
outweighs responsibility.
JEL: C91, D03, D81, G11
Keywords: nancial decision making, responsibility, limited liability, decision making for others, risk preferences,
experiment

∗ Radboud University, Institute for Management Research, Department of Economics, Heyendaalseweg 141,

6525 AJ NIJMEGEN, The Netherlands (s.fullbrunn@fm.ru.nl)

† University of Portsmouth, Faculty of Business and Law, Richmond Building, Portland Street, Portsmouth

PO1 3DE, United Kingdom, (wolfgang.luhan@port.ac.uk)

‡ We thank seminar participants at the University of Dallas, Texas A& M, University of Göttingen, University

of Portsmouth, Radboud University, as well as conference participants of ESA meetings in Zurich, Dallas,
Berlin, and Florida and of the SEF meeting at Radboud Univerity for valuable comments; in particular, we
thank participants of the Decision Making for Others Meeting at Radboud University in 2018. We gratefully
acknowledge funding by the Ruhr-University Bochum's Young Researchers Funds.
 Responsibility and Limited Liability in Decision Making

for Others - An experimental consideration

∗
double-blind no author

June 28, 2019

Abstract
Agency in nancial markets has been claimed to foster excessive risk taking, ultimately
leading to bubble formation. The main driving factor appears to be the skewed bonus
system for agents who invest other people's money. The resulting excessive risk taking
on behalf of others would imply that such bonus systems crowds out responsible decision
making for others in order to serve egoistic self-interest. To test this implication, we
conduct laboratory experiments comparing decision making for others with and without
such a bonus system. First, we show that, in the absence of bonus systems, decision makers
invested signicantly less for others than for themselves. Second, we show that limited
liable decision makersparticipating only in gains but not in lossesinvested substantially
more for others than for themselves. Hence, our results suggest that indeed limited liability
outweighs responsibility.
JEL: C91, D03, D81, G11

Keywords: nancial decision making, responsibility, limited liability, decision making for others, risk preferences,

experiment

∗ double-blind, thanks to be added later

1
1 Introduction
Economic research on risk attitudes has traditionally focused on individual decision making

without much consideration for potential social inuences (see e.g., Dohmen et al., 2011; Eckel

and Grossman, 2008b; Harbaugh et al., 2010; Holt and Laury, 2002). As real world decisions are

embedded in a social context however, a decision maker is hardly ever the only person aected

by the consequences of his actions. In fact, many risky decisions are specically taken on behalf

of a third party, e.g. the decision maker's family or business partners. On a larger scale,

a CEO's decision might aect the company or even industry, and political decisions aect a

country's future. On nancial markets, investors usually put an investment adviser or a money

manager in charge of their risky investments. It has been a long standing claimespecially

since the last nancial crisisthat this practice of delegated portfolio investment, with its

skewed bonus systems, leads to excessive risk-taking and ultimately to bubble formation (e.g.,

Allen and Gorton, 1993). Taking excessive risks with the clients' money is largely attributed

to the limited liability payment schemes prevalent in these markets (Allen and Gorton, 1993;

Allen and Gale, 2000; Cheung and Coleman, 2014; Kleinlercher et al., 2014). However, the

question remains whether such payment schemes alter the money managers' behaviour towards

excessive risk taking or whether taking investment decisions on behalf of othersindependent

of the payment scheme already increases risk taking.

To answer this question, we study investment decisions for others in a laboratory experiment

in conditions with and without limited liability. First, we try to establish whether agency

without or with perfectly aligned monetary incentive schemes increases risk taking for others.

Second, we introduce limited liability with the propensity to privatize prots and socialize

losses to see whether this aects risk taking for others. We are the rst to consider such a

decision environment in the laboratory.

The terms risky shift and cautious shift introduced by Stoner (1961) describe situations

in which the initial individual level of risk preference is altered due to exogenous impacts.

Without any prior assumptions, taking responsibility for others could result in a shift in either

direction.

In the psychological literature, a prominent explanation for a risky shift is psychological

self-other distance (e.g., Beisswanger et al., 2003; Cvetkovich, 1972; Stone and Allgaier, 2008;

Trope and Liberman, 2010; Wray and Stone, 2005), in which the assessment of a potential loss

in a risky situation is decreasing in the distance of the aected party to the decision maker.

This nding translates directly to the results from economic experiments which report a risky

2
shift in hypothetical risk taking without monetary consequences (e.g., Harrison, 2006; Holt and

Laury, 2002, 2005). Albrecht et al. (2011) nd that making inter-temporal decisions for others

results in lesser activation of areas of the brain that are thought to be engaged in emotion and

reward-related processes, than when taking decisions for oneself. The resulting argument would

be that decisions made on behalf of a third party are equivalent to situations without any real

outcome.

In contrast, Charness and Jackson (2009) propose responsibility alleviation as an expla-

nation for a cautious shift. Taking responsibility for a third party's welfare induces pro-social

behavior which results in conservative risk taking (Charness, 2000; Charness and Jackson, 2009).

Several studies report a cautious shift. Physicians, for example, have been found to prefer treat-

ments with higher mortality rates for themselves than what they recommend to their patients

(Garcia-Retamero and Galesic, 2012; Ubel et al., 2011). Managers also try to avoid responsibil-

ity for decisions with even a minimal probability of hazardous outcomes (Swalm, 1966; Viscusi

et al., 1987). Using controlled laboratory experiments, we aim to ultimately determine whether

agency in risky nancial decisions leads to increased risk taking and what role limited liability

plays in this observation.

In our experiment, a decision maker (henceforth money manager) took investment decisions

for himself and for six other subjects (henceforth clients) using the investment environment

from Gneezy and Potters (1997). First, we tested whether the money manager invests more

or lesstook more or less riskfor himself than for his clients. In the individual decision

making treatment `IND', the money manager invested for himself only. Further, we needed a

treatment in which the money manager invests for his clients onlywithout any own payo

consequencesdenoted as `OTH'. Finally, we considered a treatment with perfectly aligned

payosdenoted as treatment `ALL'in which the money manager invests the same amount

for himself AND for his clients; here, IND and OTH were combined in one decision. With

this treatment, we can see whether the money manager invests in line with his egoistic pref-

erencesinvests the same amount in IND and in ALLor whether he steps back to invest in

line with his social preferencesinvests the same in OTH and ALL. Therefore, we adminis-

tered a within-subjects design to enable the analysis of individual heterogeneity controlling for

order eects. Our study is the rst to systematically compare these theoretically very dierent

situations. Second, we implemented limited liability by replacing OTH with a new treatment

`LIM'. Here, the money manager earns a xed percentage from each client's prot, while an

investment failure has no monetary consequence for the money manager. In general, such an

option-like payment scheme increases risk taking already in individual decision making. The

3
question is, however, whether the money manager takes excessive risks with the clients' money

to increase his own payo, suggesting rather anti-social behaviour. Here, we compare LIM to

OTH in a between-subject design.

Our aggregate results indicate investment behavior to be in line with responsibility alle-

viation, as we observed a clear cautious shift. The money managers invested signicantly

less when clients bore the consequences (OTH), even when the money manager's payo was

perfectly aligned (ALL). Limited liability, however, changes the picture drastically; money man-

agers show a signicant risky shift and invest signicantly more in LIM than in OTH. Limited

liability clearly crowded out the eects of responsibility.

2 Related Literature
In recent years a growing body of literature has studied risky decisions for others using economic

experimentsnding mixed results. Some studies nd evidence for a risky shift using rst price
sealed bid auctions and multiple price lists (Chakravarty et al., 2011) or investment decisions

(Polman, 2012; Pollmann et al., 2014; Sutter, 2009). In contrast, plenty of studies nd evidence

for a cautious shift using lottery choices (Reynolds et al., 2009; Bolton and Ockenfels, 2010),

investment decisions (Eriksen and Kvaløy, 2010), or strategic risk taking in stag-hunt games

(Charness and Jackson, 2009). Pahlke et al. (2015) nd both, a risky shift in the loss domain and

a cautious shift in the gain domain using a battery of lotteries. Finally, using a multiple price list

Andersson et al. (2016) nd little evidence for a signicant shift in either direction. All studies

listed above consider ` risk taking for others '. They focus on the dierence between making a

decision for oneself and making the same decision for an anonymous stranger. However, there

are important dierences in the decision maker's payo when making decisions for others. Either

the decision maker decides for the others only and earns a lump sum payment independent of

the client's earnings (OTH), or the decision maker makes the same decision for himself and

the others with equal payo consequences (ALL). The previous studies on decision making

for others have each exclusively considered either OTH or ALL.

1 While standard models of

rational behavior predict similar investments in IND and ALL, based on the decision maker's

utility function without other-regarding preferences, there is no standard-theoretical prediction

in OTH (Eriksen and Kvaløy, 2010). We conduct a systematic study of the impact of these two

types of decision making for others, by implementing both treatments using a within-subjects

design.

1 The only exception is the study by Andersson et al. (2016). However, they do not discuss theoretical
dierences nor do they compare OTH and ALL in their analysis.

4
 Convex incentive structures in nancial institutions seem to foster excessive risk taking in

nancial markets, which is supposed to be one of the causes for market bubbles (e.g., Bebchuk

and Spamann, 2009; Dewatripont et al., 2010; French et al., 2010; Gennaioli et al., 2012). Ra-

jan (2006) even argues that one of the main origins of instability in highly developed nancial

markets are widely used convex incentives structures. In many cases, the managers' incentives

are barely aligned with the investors' interests. The agency problem of misaligned incentives

of money managers and their clients and its eects on markets is theoretically considered by

Allen and Gorton (1993); money managers do not share the losses but earn a proportion of the

prots. Such a convex incentive structure induces the money managers to trade at prices above

fundamental value, causing a price bubble. Holmen et al. (2014) tested this conjecture exper-

imentally and found convex incentives to inate prices. However, the experimental literature

only s convex incentive structures and the eects on markets. What appears to be missing are

the eects on investment decisions when clients are immediately aected, i.e. when a money

manager is investing the clients' money facing convex incentive structures. We aim to ll this

gap.

3 Design and Procedures

In this experiment, subjects made investment decisions in line with Gneezy and Potters (1997)

which is an established tool to elicit investment decisions under dierent conditions. The amount

invested provides a good metric for capturing treatment eects and dierences in attitudes to-

ward risk taking between individuals; see Charness et al. (2013) for a detailed discussion. We

consider four dierent decision environments: individual decision making ('IND'), decision mak-

ing for others ('OTH'), decision making for others and for oneself ('ALL'), and decision making

under limited liability ('LIM') using a mixture of within- and between-subject comparisons.

3.1 Experimental Design

In our individual decision making treatment (' IND ', I ), each subject was endowed with nine

euro and was asked to decide on the amount to invest in a risky asset.
2 With a probability of

2/3 the amount invested was lost and with a probability of 1/3 the investment was returned in

addition to a gain of 2.5 times the amount invested. For X ∈ {0, 9} being the amount invested,

the subject earned either πI = 9 − X , in case of a loss, or π I = 9 + 2.5 X , in case of a win.

In our others treatment (' OTH ', O ), subjects were randomly organized in groups of seven,
2 Instructions can be found in the appendix section B.4

5
consisting of six passive members, the clients (c), and one active member, the money manager
(m). Each client was endowed with nine euro and the money manager decided on the amount

to invest in the risky asset for each of the six clients. The amount invested was identical for all

clients. Hence, each client earned either πcO = 9 − X , in case of a loss, or πcO = 9 + 2.5 X , in

O
case of a win. The money manager was unaected by the outcome of the investment, πm = 0.
O
Additionally, we ran some extra sessions ('OTH315') with πm = 31.5 to test whether a high

xed paymentthe highest possible payment for the clientaected the money manager's

investment decision. In contrast to the literature, we signicantly increase responsibility for

others as each money managers invested on behalf of six clients instead of only one.

In our aligned treatment ('ALL', A), we implemented the same group protocol as in OTH,
but also endowed the money manager with nine euro. Now, the money manager decided on the

amount to invest which was binding for each of the six clients and for himself. Each subject,

the clients and the money manager, earned either πcA = πm

A
= 9 + 2.5 X in case of a win or

πcA = πm
A
=9−X in case of a loss.

In our limited liability treatment (' LIM ', L), we also implemented the same group protocol
as in OTH. In contrast, the money manager earned a fee which equaled ve percent from each

clients' prot in case of a win. In case of a loss, however, the money manager faced no monetary

L
consequences. Hence, in case of a win the payo was Xm = 0.05 × 6 × 2.5X for the money

manager and XcL = 9 + (1 − 0.05) × 2.5X for each client. In case of a loss, the payo was

L
Xm =0 for the money manager and XcL = 9 − X for each client. Hence, the money manager

participated only in gains but not in losses. However, th fee is quite small; note that risking

the entire endowment for all clients (in total 54 euro) earns an expected payo of just about

2.25 euro for the money manager. To control whether the magnitude of the fee inuenced the

decision, we ran some extra sessions (treatment `LIM50' ) in which the fee equaled 50 percent
L
instead of ve percent. Now, the payo in case of a win was changed to Xm = 0.5 × 6 × 2.5X for

the money manager and XcL = 9 + (1 − 0.5) × 2.5X for each client. Hence, the clients' expected

earnings for X euro invested was negative: (1 − 0.5) × 2.5X × 1/3 − X × 2/3 = −1/4X . Any

amount invested yields an expected loss for the clients but high expected gains for the money

manager.

3.2 Implementation
The experiment was programmed and conducted using z-Tree (Fischbacher, 2007). Upon arrival

subjects were randomly placed at computer terminals separated by blinds. For each treatment,

6
instructions were read aloud separately and questions were answered privately. The experiment

only started once the instructor was sure that all subjects had comprehended the instructions.

Once the experiment started, each money manager was endowed with an on-screen calculator

which calculates potential payos for the money manager himself and/or the clients for arbi-

trarily entered investment levels (screenshots can be found in appendix B.5). Eventually, the

money manager selected one investment from the generated list and conrmed his choice.

In each session, the subjects made investment decisions in three separate treatments. To

exclude hedging eects from repetitions or cross-game eects, only one of the three treatments

was payo relevant (Charness et al., 2016). The subjects were informed that the experiment

would consist of three independent parts, without specifying the exact nature of each part

upfront. The instructions for each part were distributed only after the previous part was

concluded.

In OTH, ALL, and LIM, each subject made the investment decision in the role of the money
manager. At the end of the session, one of the seven subjects was randomly determined to be

the active money manager while the remaining six subjects became the passive clients. To avoid

accountability eects, we guaranteed anonymity (Pollmann et al., 2014). Neither the money

managers knew the identity of the clients nor did the clients know the identity of the money

manager.

To avoid psychological anchoring eects, such as gamblers fallacy or hot hand fallacy, sub-

jects did not receive any information on the outcome of their investments. To determine the

payos, one subject rst threw a dice to determine the payo relevant treatment and then

threw a dice for each group to determine the outcome of the investment. The computer nally

assigned the roles as money manager and clients in a group. Subjects were payed privately in

cash.

We additionally elicited sex, age, eld of study, self-reported risk preferences (Dohmen

et al., 2011), and a social responsibility indicator (Berkowitz and Lutterman, 1968). The latter

two provide information on the general willingness to take risks (likert scale, 1-10) and social

responsibility, the tendency to help others without expecting any immediate personal reward

(score from eight likert scales 1-7).

All sessions were conducted at the [ double-blind, location hidden ]. Our participants were

mainly bachelor students from all departments of [ double-blind, location hidden ]. Subjects

participated only once in this experiment. Figure 1 provides an overview of the implementation

and the comparisons together with the investment means for all treatments.

To answer our rst research question, whether money managers take more or less risk for

7
 Figure 1: Overview of experiments with strategy for comparison

Notes. The gure shows the treatments as conduced in the sessions with ALL as the last treatment; however, we also
tested for order eects. Additionally, we provide the investment means. The links with diamonds indicate within-
subject considerations and links with bullets between-subject considerations.
others than for themselves, we ran 15 sessions with a total of 140 participants. We implemented

two dierent setups. In setup one (70 observations), the treatment order was OTH-IND-ALL.

In setup two (70 observations), the treatment order was ALL-IND-OTH. Average payments

were 15.82 euro (max. 34.5, min. 3, SD 10.10) including a show-up fee of 3 euro. As indicated

in gure 1, we make use of a within-subject consideration (links with diamonds). Additionally,

we ran three high payo sessions in the order OTH315-IND-ALL (35 observations). Average

payments were 10.95 euro (max. 34.5, min. 3, SD 8.64) including a show-up fee of 3 euro. Here,

we aimed to compare OTH315 to OTH. All sessions lasted roughly half an hour.

To answer the question whether limited liability has an impact on decision making for

others, we ran ve sessions with a total of 105 participants. We implemented two dierent

setups. In setup one (49 observations), the treatment order was LIM-IND-ALL. In setup two

(56 observations), the treatment order was ALL-IND-LIM. Average payments were 16.43 euro

(max. 34.5, min. 3, SD 10.74) including a show-up fee of 3 euro. To see whether the size of

the fee plays an important role, we ran three additional sessions with high fees, 50 percent, in

the order LIM50-IND-ALL (21) and ALL-IND-LIM50 (28). Average payments were 16.27 euro

(max. 70.5, min. 8, SD 12.75) including a show-up fee of 3 euro. The experiments lasted about

half an hour. As indicated in gure 1, we make use of between-subject considerations (links

8
with bullets).

3.3 Predictions
Our experimental setup focuses on two considerations. The rst consideration is about the

dierences in decision making for others and for oneself using the sessions with the treatments

IND, ALL, and OTH (140 observations). Our second focus lies on the implementation of limited

liability, comparing mainly the 140 observations in OTH to the 105 observations in LIM.

For the rst consideration, we are less interested in individual investment levels; our focus lies

on the shift of investments comparing IND and OTH, and IND and ALL respectively. As stated

above, the decision in IND will be determined by the subjects' risk preferences (Charness et al.,

2013). For the decisions in ALL, economic standard models of perfectly rational and egoistic

agents make clear predictions. Without other-regarding preferences the money manager is just

investing the same amount as in IND, i.e. XI = XA . However, the standard models provide no

predictions for OTH as the payment for the money manager is not aligned to the investment

decision. Eriksen and Kvaløy (2010) as well as Andersson et al. (2016) pick up the social

distance hypothesis arguing that loss aversion is less pronounced when deciding for others than

when deciding for oneself, as these decisions are perceived as less important. Indeed Füllbrunn

and Luhan (2017) report experimental evidence for lower levels of loss aversion when decisions

are taken for others compared to decisions for oneself. However, our setup by design does not

emphasize losses as the payo on screen is framed only in the gain domain (the total payo is

displayed). Coming back to responsibility alleviation (Charness, 2000; Charness and Jackson,

2009), we should nd that money managers' behavior will be more conservative when investing

for others, i.e., XI > XO .3 When the money manager believes his clients to have similar

risk preferences as he has, he wouldin line with the false consensus eect (Ross et al., 1977)

invest the same amount for himself as for the clients, i.e., XI = XO .

Another prominent model to explain dierences between risky decisions for oneself and for

other, the self-others discrepancy eect, describes a bias when predicting others' risk aversion,

as this is perceived to be dierent from the individual's own risk preference. In the context of

our design this would state that money managers evaluate their own risk preferences dierently

than the risk preferences of their clients (Hsee and Weber, 1997; Eckel and Grossman, 2008a)

leading to an observed shift. The direction and magnitude of the predicted shift depends on the

3 As the investment in our experiment has a positive expected payo, it is not immediately clear that low
investments are a "responsible" choice. As one reviewer pointed out, it could be seen as responsible to maximize
the clients expected payos with a maximal investment. Our denition of a shift driven by responsibility,
however, follows Charness and Jackson (2009) who clearly predict a decrease in the risks taken on behalf of
others.

9
risk attitudes of the money managers relative to their clients. If money managers believe their

clients to be relatively risk averse, they would invest less for their clients than for themselves

(XI > XO ), while money managers who believe their clients to be relatively risk seeking, would

invest more for their clients than for themselves (XI < XO ).

The investment decision in ALL might induce a conict of preferences when the egoistic

preferences of the money manager and the social preferences for clients clash. On one side of

the spectrum, social preferences play no role (XA = XI ) inline with the economic prediction

of an egoistic decision maker who strictly follows his own preferences. On the other end of the

spectrum, social preferences for the clients would completely crowd out egoistic preferences,

i.e., XA = XO . When we assume that both preferences play a role in the consideration of the

money manager, the investment in ALL is expected to be in-between investments in IND and

OTH, i.e., either XI > XA ≥ XO (for a cautious shift) or XI < XA ≤ XO (for a risky shift).

Note that the interaction of social and/or egoistic preferences in risky decision making can only

be studied when all three treatments are considered in a within-subjects design.

In LIM, standard economic theory would predict the money manager to invest the full

amount. Such behaviour would also be optimal for risk averse clients. Expected returns from

investment are positive and higher investments lead to higher expected payos, at least when

the fee is merely ve percent. Hence, neglecting responsibility as in standard economic theory,

we predict subjects to invest the entire amount in the LIM treatments. However, there might

be a conict between the monetary incentives that trigger the egoistic preferences and other-

regarding preferencesthe former predicting high investment while the latter a rather cautious

investment. Nonetheless, high investment in LIM can still be judged as acting in line with risk

neutral clients. This argument does not hold any longer for a fee of 50 percent though. Here,

the clients clearly suer from excessive risk taking due to their negative expected payo (e.g.,

Allen and Gorton, 1993). In contrast to OTH, the decision maker now has an incentive to take

risks in order to earn a positive payo for himself. If responsibility plays the main role, however,

we expect similar investment levels in OTH and LIM. When egoistic preferences play the main

role, the money managers will not hesitate to risk the entire endowment of the clients.

4 Results
We rst consider the eect of risk taking for others by testing within-subjects whether dierences

exist between XI , XA , and XO . Then, we consider the eect of limited liability by testing

between-subjects whether dierences exist between XO and XL . We also consider a xed

10
payment eect in OTH (vs. OTH315) and an eect of a 50% fee in LIM (vs LIM50) between-

subjects. The p-values for the all tests conducted were derived from two sided permutation

tests.

4.1 Risk Taking for Others

To test whether money managers take more or less risk for others, we use each subject's shift
in investments as the relevant unit of observation, i.e., SA = XA − XI and SO = XO − XI .

While negative values indicate a cautious shift, positive values indicate a risky shift. Table 1

provides averages (and standard deviations) for investment levels and shifts for 140 independent

observations. .
4
Table 1: Average Investments in Euro

General Risk
below median risk above median risk
n = 140 (n = 53) (n = 64)
4.53 (2.42) 3.19 (1.75) 5.42 (2.49)
3.92 (1.91) 3.04 (1.50) 4.41 (1.96)
XI

3.87 (2.21) 3.24 (1.96) 4.27 (2.40)

XA
XO

-0.60 (1.57)
∗∗∗
-0.15 (0.85) -1.01 (1.86)∗∗∗

-0.65 (2.39) 0.05 (1.65) -1.15 (2.75)

SA
∗∗∗ ∗∗∗
SO

Notes. The second column contains averages for all observations. The next two columns categorize subjects in two
groups according to elicitation of risk attitudes in line with Dohmen et al. (2011). Subjects in the group below median
(above median ) stated below (above) median risk attitudes on a scale from 1 - 10. Rows report average investments
and average shifts together with the standard deviation in parentheses. The asterisks refer to the p-value from a
permutation test testing the Null that S equals zero (** = p <0.05, *** = p <0.01).

The table shows that investment levels in ALL and in OTH are on average about 13 and

14 percent lower than in IND, respectively. Using a two-sided permutation test, we conrm a

signicant cautious shift for ALL and for OTH which is on average at SA = −0.60 (p < 0.001)

and at SO = −0.65 (p < 0.001), respectively.

5 Thus, we state observation 1.

Observation 1. Money managers invest signicantly less for their clients than for themselves
in ALL and in OTH.

This result is a clear indication of acting in line with the responsibility alleviation hypothesis.

We nd a signicant cautious shift, not only in OTH but also in ALL, which is in contrast to

the prediction of the standard rationality models.

4 As we nd no signicant order eect in non-parametric tests and regressions, we pool all 140 observations
(see appendix B.1 and A)
5 We nd no signicant relationship between demographics and decision making for others nor was the social
responsibility score signicant (see appendix table A.5 and A.6).

11
 The self-other discrepancy might be seen as a renement of trying to act responsibly, as

the money manager tries to act according to the investors risk preferences while deviating from

his personal preferences. According to this view, the direction of the observed shift depends

on the perceived risk preferences of the clients in comparison to the money managers own

risk preferences. Therefore, we conjecture that relatively risk averse subjects invest more for

others while relatively risk seeking subjects invest less for others. To test this conjecture, we

consider the self-reported willingness to take general risks based on Dohmen et al. (2011) which

provides a number from 1 (not willing to take risks) to 10 (willing to take high risk). We

make use of a median split to compare two groups; the relatively risk averse subjects (reported

general risk below 5) and the relatively risk seeking subjects (reported general risk above 5;

we neglect 23 observations at the median). Table 1 indicates that in the below-median group

the shift not signicantly dierent from zero. In the above-median group, however, the general

pattern observed above is quite strong, i.e., we nd a signicant cautious shift in ALL and in

OTH (for both p < 0.001).6 As the averages in table 1 already suggest, we nd a signicant

eect comparing above-median group and below-median group for each of the ve variables

(p < 0.025). Thus, we state observation 2.

Observation 2. Investment levels in OTH and ALL do barely dier for rather risk averse
money managers, while rather risk seeking money managers show a signicant cautious shift.

Observation 2 appears to be in line with the self-other-distance theory. Money managers in

the above median risk group tend to assume that they are relatively risk seeking in comparison

to the population while at least the averages suggest the opposite for the risk averse money

managers. The decisions for their clients reect a propensity towards the perceived average

preference of their clients; nally, the above median risk level money managers are clearly

driving the aggregate results. The regression in the appendix (Table A.5) shows a clear picture:

the higher the willingness to take general risks, the higher the cautious shift.

These conclusions are only derived from observed behavior under the assumption that money

managers did indeed presume the average risk aversion to be higher or lower than their personal

risk preferences. To test whether this assumption was correct or a mere artifact, we let the

subjects estimate the investment level XI of the other subjects' (unincentivized) which we

denote XIb .7 Table 2 reports respective measures.

6 Implementing the same analysis by categorizing subjects according to their observed risk preference in the
IND treatment (XI ), we get similar results (see appendix B.3).
7 To elicit beliefs, we included the question  What would you say, how much do others in your group on
average invest for themselves?  in the questionnaire. We abstained from using incentivized believe elicitation
methods as this would increase the complexity and duration of the experiment with vague additional benets
(see Trautmann and Kuilen, 2015, for a discussion). Unfortunately, we have elicited the beliefs for 91 subjects

12
 Table 2: Average Beliefs about Investments of Clients

General Risk
below-median risk above-median risk
n = 91 (n = 36) (n = 40)
4.41 (1.62) 3.85 (1.14) 4.65 (1.95)
0.00 (2.22) 0.77 (1.32) -0.90 (2.36)
XIb
∗∗∗ ∗∗∗
XIb − XI
XIb − XO 0.58 (2.56)
∗∗∗
0.92 (1.88)
∗∗∗
0.05 (3.11)
Notes. The second column contains observations from 91 subjects from which we elicited beliefs. X , X , and X
denote the beliefs about the investments of others in IND, own investment in IND, and investment in OTH, respectively.
Ib I O

The next two columns categorize subjects in two groups according to elicitation of risk attitudes in line with Dohmen
et al. (2011). Subjects in the group below median (above median ) stated below (above) median risk attitudes on a
scale from 1 - 10. The asterisks refer to the p-value from a permutation test testing the Null that S equals zero (** =
p <0.05, *** = p <0.01).

Taking all observations into account, the average dierence between beliefs and own invest-

ment is not signicantly dierent from zero (p = 0.979). This indicates that money managers

on average do not believe others to take less or more risk then they take for themselves. Such

thinking might lead money managers to take similar risks for others as they would take for

themselves in line with behaviour of nancial advisers in Canada (Foerster et al., 2017).

On the individual level, however, the investment level in OTH can still deviate signicantly

from the expectation of the average risk preference. Thus, we again separate between subjects

with general risk below and above the median, and nd that the above median money managers

believe others to invest signicantly less than they do (XIb − XI = −0.90, p < 0.001), while the

below median money managers believe others to invest signicantly more (XIb − XI = 0.77,

p < 0.001). However, above-median risk takers generally believe others to take more risk than

below-median risk takers (p = 0.036).

The question remains, whether money managers in OTH invest in line with their beliefs,

i.e., does XIb − XO = 0 hold? Overall, we nd money managers to invest signicantly less for

their clients than what they believe their clients would invest for themselves (XIb − XO = 0.58,

p = 0.033). In particular, the below-median money managers invest signicantly less for their

clients than what they believe their clients would invest for themselves (XIb − XO = 0.92,

p < 0.001). In contrast, the above median money managers tend to invest in line with their

beliefs about others (XIb − XO = 0.05, p = 0.701).8 Overall, the results suggest that money

managers are relatively conservative in that they invest at most what they believe the clients

would invest for themselves. This supports Bolton et al. (2015), who nd that decision makers

act according to the preferences of their clients.

Finally, we compare investment levels between OTH and ALL, i.e., a situation in which the

only.
8 However, comparing XIb − XO between the two categories we nd no signicant dierence (p = 0.150).

13
payment of the money manager is perfectly aligned (ALL) and when the payment is not aligned

(OTH). Results in table 1 indicate that dierences between XA and XO are negligible when

considering all subjects (XA − XO = 0.05, p = 0.766), but also when considering subjects below

median risk attitudes (XA − XO = −0.14, p = 0.501) and above (XA − XO = 0.32, p = 0.246).

Hence, overall it seems that money managers in ALL put their egoistic preferences on hold to

match the preferences of the clients indicated by OTH.

9
One caveat in our design might be the unequal payments of the active money manager

and the passive client in OTH. While the clients' expected earnings are positive, the money

managers' earnings are zero. When fairness preferences come into play, the money manager

might change his behavior to match his preferences for a fair allocation of payments (Fehr and

Schmidt, 1999; Bolton and Ockenfels, 2000; Charness and Rabin, 2002, 2005; Fudenberg and

Levine, 2012). However, models that incorporate risk preferences and fairness preferences are

missing. Nonetheless, we wanted to control whether the money manager's xed payment in

OTH has an inuence on decision making for others. Therefore, we compared the investment

levels in treatment OTH315 and OTH. We nd no signicant dierence in investments compar-

ing the two treatments (see appendix section B.2 and the regression in appendix table A.5 and

A.6) suggesting that the money manager's own unrelated payo when deciding for his clients

in OTH plays a minor role. However, the power is not sucient to claim that investment levels

in OTH315 and OTH come from the same population.

10

4.2 Limited Liability

Section 4.1 shows that money managers in OTH invested less for others than for themselves.

In this section, we compare investment levels and shifts in OTH and in LIM between-subjects.

Figure 2 shows average investments (A) for IND (as the benchmark), OTH, and LIM and the

histograms (B) of OTH and LIM.

11 Both gures show a clear dierence between OTH and

LIM, i.e. subjects invest more in LIM than in OTH.

Table 4 reports the descriptives of the investments and shifts together with test statistics

for dierent categories of subjects. Considering the full data set, we nd a clear and signicant

dierence between investing in OTH and in LIM. On average, the money managers invested

2.79 euro more in LIM than in OTH, a 77 percent increase. Even though order played a role

9 However, even with n = 140 we would not be able to detect a small eect size of 0.2 as the power would
be only 0.63. Merging the data with observations from OTH315, i.e. having 175 observations, we still nd no
signicant dierence but for the same parameters we get a decent power of 0.73.
10 For a small eects size of 0.2, the power for a given number of observations of 70 and 35 respectively (same
order sessions) would be 0.28.
11 We neglect ALL as the relevant comparison is between OTH and LIM.

14
 Figure 2: Comparing Investments: OTH vs LIM

(a) Average investments (b) Histogram of investments

Notes. Panel (a) displays average investment levels separated by treatments. The column labelled IND displays the
average investment for Experiments 1 and 2 combined. Error bars show the standard error. Panel (b) displays the
histogram of investments in OTH and LIM. The respective curves show the kernel densitiesthe solid line for OTH
and the dashed line for LIM.
in the sessions with LIM, i.e., money managers invested about one euro more when LIM was

played last than when LIM was played rst (p = 0.014), investment levels and shifts remain

signicantly higher in LIM than in OTH in both orders.

Additionally, we conducted a full analysis of session and treatment eects in a panel regres-

sion for which the results are reported in table 3. The dependent variable is the investment

level in euro. To compare the treatments in the sessions with OTH, `OTH-sessions', to the ses-

sions with LIM, `LIM-sessions' (row one vs. row two in the grey area in gure 1), we introduce

LIM session as a dummy for the latter sessions. Further, the dummy ALL equals 1 if ALL is
played (and zero otherwise) and the dummy OTH* equals 1 if the decision is made for others

(and zero otherwise), which is OTH in the OTH-sessions and LIM in the LIM-sessions. The

references was therefore the investment in IND in the OTH-sessions. Model 2 adds subjects

characteristics.

15
 Table 3: Session and treatment eects on investment

independent: investment (1) (2)

∗∗∗ ∗∗∗
OTH* -0.65 -0.65
(0.20) (0.20)
∗∗∗ ∗∗∗
ALL -0.60 -0.60
(0.13) (0.13)
LIM sessions 0.020 0.026
(0.29) (0.25)
∗∗∗ ∗∗∗
ALL × LIM session 0.56 0.56
(0.17) (0.17)
∗∗∗ ∗∗∗
OTH* × LIM session 2.95 2.95
(0.33) (0.33)
∗∗∗ ∗∗∗
Constant 4.53 3.04
(0.20) (0.93)
Observations 735 735
Controls NO YES
Wald test ALL × LIM session = OTH* × LIM session:
p-value < 0.001 < 0.001
χ2 64.67 64.32

Notes. Random-eects panel regression with robust standard errors, variables explained above. Column 1 contains
the estimation results without controlling for individual characteristics, column 2 contains the results from the same
estimation controlling for gender, age, self-reported risk preferences (Dohmen et al., 2011), and a social responsibility
indicator (Berkowitz and Lutterman, 1968). Standard errors in parentheses.(* = p <0.1, ** = p <0.05, *** = p
<0.01).

Firstly, the LIM session dummy is is not signicant and close to zero. Hence, investments

in IND are not dierent comparing LIM- and OTH-sessions. Secondly, the OTH* coecient

is signicantly negative, which conrms the cautious shift in OTH, while the interaction co-

ecient OTH* × LIM-sessions is signicantly positive which conrms the risky shift in LIM.

Third, the ALL coecient is signicantly negative, which conrms the cautious shift in ALL.

However, this eect is diminished in the LIM-sessions, suggested by the signicantly positive

interaction coecient ALL × LIM-sessions. Maybe money managers more generally invest

more for others in the LIM-sessions than in the OTH-sessions and the risky shift in LIM is just

a consequence. However, the interaction with ALL is signicantly lower than the interaction

with OTH* conrmed by a Wald test (p < 0.001). This result conrms that even if money

managers invest generally more for others in the LIM-sessions than in the OTH sessions, the

eect is much stronger in OTH* than in ALL. These results remain unchanged if we control

for individual characteristics, as reported in the second column of table 3.

The signicant increase of investments between OTH and LIM still holds if we compare

LIM to OTH separately for below- and above-median general risk group (see table 4), i.e.

independent of risk attitudes, investments in LIM exceed investments in OTH. To keep this

section concise, we provide respective regressions with shifts and investment levels as dependent

16
 Table 4: Investment in OTH and LIM

No Limited Liability Limited Liability

XO SO XL SL H0 : XO = XL

∗∗∗ ∗∗∗
All 3.87 -0.65 6.85 2.30 <0.001
(n=140/105) (2.22) (2.3) (2.15) (2.68)

∗∗∗
Below Median risk 3.24 0.05 6.80 3.13 <0.001
(n=53/40) (1.96) (3.13) (2.25) (2.84)

∗∗∗ ∗∗∗
Above Median risk 4.27 -1.15 6.85 1.46 <0.001
(n=64/49) (2.4) (2.75) (2.11) (2.43)

∗∗∗ ∗∗∗
OTH/LIM rst 3.95 -0.77 6.30 1.52 <0.001
(n=70/49) (1.85) (2.11) (2.13) (2.49)

∗ ∗∗∗
OTH/LIM last 3.84 -0.53 7.33 2.06 <0.001
(n=70/56) (2.54) (2.67) (2.98) (2.67)

Notes. The table reports averages (SD) of investment levels and shifts in the treatments OTH and LIM for all
observations, subjects in the category below and above median risk, and in the orders OTH-IND-ALL/LIM-IND-ALL
(`rst') and in the orders ALL-IND-OTH/ALL-IND-LIM (`last'). The number of observations are in brackets for
OTH/LIM. The last column shows the p-values from a permutation test testing the Null X = X in the dierent
categories. The asterisks refer to the p-value from a permutation test testing the Null that the shift equals zero (* =
O L

p <0.1, ** = p <0.05, *** = p <0.01).

17
variables and a LIM dummy plus controls as independent variables in the appendix (tables A.5

and A.6); these conrm our results as the LIM dummy is signicantly positive. Hence, we state

observations 3 and 4.

Observation 3. Money managers invest signicantly more for their clients than for themselves
in LIM.

Observation 4. Money Managers show a signicant risky shift independent of personal risk
attitudes.

This appears to be an indication of egoistic-, monetary incentives crowding out the previ-

ously observed eects of responsibility. However, these higher investment levels for others in

LIM could in principle stem from the fact that the money managers believe their clients to prefer

higher investments. We have shown in section 4.1 that money managers in OTH acted rather

cautiously, investing less than, or at most what they believe their clients would invest them-

selves. While there is no obvious reason for a change in believes, such a shift would in principle

explain increased investments even if money managers still try to act responsibly. We therefore

again compare the investments with the elicited beliefs. The average dierence between XL and

XIb equals 1.57 euro (SD 2.52) which is signicantly positive (p < 0.001).12 Recall, in OTH

the average dierence between XO and XIb was -0.58 (SD 2.36) which is signicantly negative

(p = 0.035). Hence, without limited liability the money manager invests rather cautiously for

others, as he invests less than he believes others would invest for themselves. With limited lia-

bility, however, such cautious investment behavior has disappeared and investment levels were

higher than what the money managers believed their clients would prefer.

We implemented additional sessions to test whether a higher fee with negative expected

earnings changes the decisions for others for the money manager. The increase to a fee of 50

percent yields an average investment of 6.03 euro (SD 2.90) which is slightly lower than in

LIM with a ve percent fee (6.85, p = 0.053).13 However, the shift is still signicantly positive

(SLIM 50 =1.64 euro, SD 2.72, p < 001) and not signicantly dierent to LIM (p = 0.158). Again,

investments are signicantly higher than in OTH (p < 0.01). Hence, the money managers in

our setting seem to care less about the expected negative consequences for their clients.

As a side eect, the new LIM50 treatment also allows us to consider ex-ante and ex-post

fairness. If the money manager strives for equal expected earnings (ex-ante fairness), he should

have invested 3.27 euro to have an expected earning of 8.17 euro for each client and himself.

If instead the money manager strives for equal earnings (ex-post earnings), he should have

12 Unfortunately, we have only 49 observations for the beliefs in this treatment.

13 See also appendix table A.5.

18
invested 1.44 euro to earn 10.80 euro for clients and himself in case of success; there was no way

for equal earnings in case of a loss. Nevertheless, as shown above, the investment in LIM50 is

signicantly higher than 3.27 euro and of course than 1.44 euro (p < 0.001).

Looking at the clients' expected earnings given the investments, we see that they were

highest in LIM (average 9.86 euro, SD 0.27), slightly but signicantly lower in OTH (9.65 euro,

SD 0.37, p < 0.001), and signicantly lower in LIM50 (7.49 euro, SD 0.73, p < 0.001). In

this particular setting, limited liability with a ve percent fee increased the clients' expected

earnings. On the one hand, the fee is suciently low to keep expected earnings positive, and

on the other hand, the convex incentives increased the investment levels. However, limited

liability with a 50 percent fee makes the investment opportunity unprotable for the clients due

to negative expected returns.

5 Discussion
5.1 Risk Taking for Others
Our rst question was, whether making decisions for others, in the absence of limited liability,

would lead to a risky or cautious shift as compared to decisions for oneself. In particular,

in contrast to the literature, our decision makers make decisions for a substantial number of

clients. This is a necessary rst step in establishing whether it is limited liability that leads to

increased risk taking on nancial markets. Overall, we nd clear evidence of a cautious shift in

both OTH and ALL, which in our experiment is mainly driven by relatively risk seeking money

managers. They believe others to be more risk averse than themselves and act accordingly. Our

aggregate results are line with the results from Eriksen and Kvaløy (2010) who use a similar

design in a between-subject setting with only one client. However, our results dier from those

experiments in the literature who nd rather a risky shift. How can we explain those mixed

results in the literature? In the following, we discuss some suggestions.

The conclusions from the literature are based on aggregate results only and the heterogeneity

of subjects with respect to risk attitudes has barely been considered. Due to our within-subject

design we are able to take the relative risk attitudes of the money manager into account. We

nd that our results are driven by the relatively risk seeking subjects. Therefore, any study

with a rather risk averse subject pool would nd an aggregate risky shift, of course. Dierences

to Andersson et al. (2016), for instance, might be due to the fact that their subject pool is

taken from the general Danish population which has been found to be more risk averse than

19
the common student population (von Gaudecker et al., 2012).

Among others, Eriksen and Kvaløy (2010) report that hypothetical decision making for

others the most extreme social distanceleads to higher risk taking in comparison to a

situation with monetary consequences. Thus, experiments with higher social distance, which is

the case for internet experiments as opposed to laboratory experiments, might lead to higher risk

taking for others. Furthermore, our experimental design allows the potential money managers

to put themselves into the shoes of their clients, as the money manager becomes a client with

a probability of 6/7. This might lead to a higher empathy for the others leading to a cautious

shift (as in the equal opportunity mode treatment in Bolton and Ockenfels, 2006).

Experiments in which the subjects pick lotteries rather than making investments, the results

seem to support a risky shift in the loss domain or in the mixed domain, while lotteries in the

gain domain support a cautious shift (Pahlke et al., 2015). One reason might be that loss

aversion is weaker when making decisions for others (see e.g. Füllbrunn and Luhan (2017)).

In our Gneezy and Potters (1997) investment game, however, we cannot control the subjects'

reference point, as we have no record of the editing phase (Kahneman and Tversky, 1979).

When the endowment is integrated, the decision takes place in the gain domain only (9 + 2.5X

vs. 9 − X ). When the endowment is segregated, the decision takes place in the mixed domain

(2.5X vs. −X ). As we provide integrated outcomes on the decision screen, the subjects might

have perceived the task in the gain domain (see screenshot appendix B.5).

While money managers are assumed to know their own preferences, they are uncertain

about their clients' preferences; in particular when estimating the preferences of six clients.

This creates an ambiguous situation when deciding for others in contrast to when deciding for

oneself. From this point of view, our results are in line with ambiguity aversion, as subjects take

less risk in a situation with higher ambiguity (e.g., Trautmann and Van De Kuilen, 2015). This

eect might be amplied due to comparative ignorance (Fox and Tversky, 1995) as in a within-

subject design subjects are able to compare decisions for others and for themselves. When

money managers observe the decision of their clients in OWN beforehand, we would reduce

ambiguity and thus can see whether they behave indeed in line with the client's preferences as

our results on following their believes suggest, Bolton et al. (2015) also nd that the decision

maker follows their client's own choice (for one single client). However, how money managers

select the level of investment when they know the preferences of several clients in a group

remains a question for future research. So far, subjects in a group had to discuss and agree on

one common amount to invest; the results support a risky shift (Sutter, 2009).

Further design elements might be interesting to reconsider. For example, we only consider a

20
situation under anonymity; the money manager's identity is not revealed nor is s/he accountable

for his/her decision. When accountability comes into play, subjects might behave dierently.

Pahlke et al. (2012) nd that accountability leads to reduced risk aversion in mixed lotteries

with one 'client'. We consider a situation in which one subject is randomly chosen to be

the money manager, i.e. the decision is implemented with probability 1/7. When the money

manager knows for sure that his/her decision is implemented, s/he might be even more cautious.

Finally, we use a student subject pool for our experiments. It might be interesting to analyze the

same situation using the 'relevant decision maker'; Kirchler et al. (2019), for example, consider

nancial professionals who make investment decisions for others (in a dierent setting though).

5.2 Limited liability

Our second research question asked whether limited liability triggers self-interest. Money man-

agers who participate in gains only do not risk their own money but the money of each of the

six clients. A risky shift in comparison to OTH would provide evidence that social preferences

of money managers are crowded out by egoistic monetary incentives. To be able to answer

this question, we invented a new experimental setting which has not been considered in the

literature before; in particular, only a substantial number of clients allows for a 'privatizing

gains and socializing losses' environment.

We do indeed nd a clear indication for a risky shift once we introduce limited liability.

This result is not driven by the beliefs about the clients preferences; as money managers invest

signicantly more for their clients than what they believed their clients would prefer. The

investment levels taken for clients are signicantly higher than for oneself. This result is quite

robust testing for the size of the bonus (5 percent or 50 percent), risk preferences (below- or

above-median risk), and order (LIM/OTH played rst or last).

14
The size of the bonus does not appear to play any role for this consideration. We compare

limited liability treatments with a ve percent and a 50 percent bonus and nd no dierence in

investments. Given that the expected payos for their clients were negative with a 50 percent

bonus, the money managers' decisions are clearly motivated by their own payo expectations

with social preferences playing only an inferior role. In this treatment, investments could be

chosen such that expected earnings for the money manager and the clients were equal, both,

in line with ex-ante and ex-post fairness. However, we found that the investment levels under

limited liability were signicantly higher than any fairness benchmark. We only consider two

14 The result is also robust when separating the subjects by sex. For the interested reader, we have added
appendix D.11 considering gender eects.

21
dierent bonus conditions. Future research might consider dierent bonus conditions to carve

out whether indeed the size of the bonus has an inuence on the money manager, in particular

if higher bonuses have negative consequences for the clients.

One argument against our interpretation of the eects of limited liability on investment levels

might be that the observations are driven by a few participants who, for whatever reason, take

the maximum risk, i.e. invest the entire endowment. Without limited liability only six percent

invest the entire endowment; with limited liability 37 percent invest the entire endowment (41

percent in the 50 percent condition). The remaining money managers might behave in a similar

with and without limited liability. However, we still nd a signicant risky shift (p < 0.001).

6 Conclusion
Coming back to our real-world motivation, we asked the question whether limited liability

payment schemes on nancial markets lead to money managers taking excessive risks when

investing their clients' money (e.g., Allen and Gorton, 1993; Allen and Gale, 2000; Cheung and

Coleman, 2014; Kleinlercher et al., 2014). Such a situation involves two possible motivational

factors that could increase the willingness to take risks, the mere fact that the decision is taken

for somebody else, and limited liability. We decided to study both separately in an incremental

design. In the rst condition, we aimed to establish whether money managers take higher risks

for their six clients in absence of limited liability. In a second condition, we introduced limited

liability to test whether behaviour changes in comparison to the rst condition. We found that

agency in itself leads to a cautious shift, i.e. decision makers took less risk with other people's

money. However, if gains are shared, while losses remain with the clients, we observe a risky

shift, i.e. decision makers took more risk with other people's money. Our results clearly suggest

that decision makers on one hand have social preferences and invest more cautiously for others

than for themselves. On the other hand these social preferences are crowded out by egoistic

preferences as soon as we introduce limited liability.

Our ndings of initially decreased risk taking are in line with responsibility alleviation, and

we found that investors seek to act according to the perceived preferences of their clients. Our

participants were aware of their relative risk preferences in that money managers with low

risk aversion sought lower risks for others than for themselves and money managers with high

risk aversion increased the risks taken for their clients. When the decision maker bears the

same consequences as their clients (ALL), the clients' preferences seem to outweigh the money

manager's own preferences. This situation changed drastically once we introduced limited

22
liability in the form of convex incentives. Promising a ve percent share of the possible returns

on investment, with losses being incurred only by the clients, was sucient to increase the

investment levels by more than 77 percent. While the perception of what the clients would

prefer did not change, money managers nonetheless follow their own egoistic preferences and

invest much more. The emerging picture is that while decision makers initially try to accomplish

what their clients want and overall take lower risks for others, the convex incentives reverse the

decision. Apparently, social preferences are largely crowded out by egoistic monetary incentives.

The fact that in one treatment the expected prot was even negative for the clients did not

signicantly change the money managers' behaviour. A clear sign that egoistic preferences are

the main driver.

If one would want to draw conclusions for the nancial markets, the recommendation is

straightforward. A fee based on the success of the investment, without consequence in case of

losses increases risk taking and potentially fosters bubble formation. Both a at fee, as well as

a bonus including negative consequences in case of losses for the client leads to investments that

aim to follow the client's preferences. However, the results stem from a laboratory experiment

and are not directly transferrable to the real world.

Our limited liability design is the rst that applies convex incentives with consequences for

other people's earnings in an experiment. This experimental design can be applied to answer

more questions regarding `privatizing prots and socializing losses' scenarios. Such a design

could for example consider 'trust' in the fund industry when money managers compete which

each other under convex incentives (Gennaioli et al., 2015; Agranov et al., 2014) but can also be

applied to asset market experiments to test whether such an environment indeed fosters asset

market bubbles (Holmen et al., 2014).

References
Agranov, M., Bisin, A., Schotter, A., 2014. An experimental study of the impact of competition

for other people's money: the portfolio manager market. Experimental Economics, 564585.

Albrecht, K., Volz, K. G., Sutter, M., Laibson, D. I., Von Cramon, D. Y., 2011. What is for me

is not for you: brain correlates of intertemporal choice for self and other. Social Cognitive

and Aective Neuroscience 2, 218225.

Allen, F., Gale, D., 2000. Bubbles and crises. The Economic Journal 110 (460), 236255.

23
Allen, F., Gorton, G., 1993. Churning bubbles. The Review of Economic Studies 60 (4), 813

836.

Andersson, O., Holm, H. J., Tyran, J.-R., Wengström, E., 2016. Deciding for others reduces

loss aversion. Management Science 62 (1), 2936.

Bebchuk, L. A., Spamann, H., 2009. Regulating bankers' pay. Georgetown Law Journal 98,

247287.

Beisswanger, A. H., Stone, E. R., Hupp, J. M., Allgaier, L., 2003. Risk taking in relationships:

Dierences in deciding for oneself versus for a friend. Basic and Applied Social Psychology

25 (2), 121135.

Berkowitz, L., Lutterman, K. G., 1968. The traditional socially responsible personality. Public

Opinion Quarterly 32 (2), 169185.

Bolton, G. E., Ockenfels, A., 2000. Erc: A theory of equity, reciprocity, and competition.

American Economic Review 90 (1), 166193.

Bolton, G. E., Ockenfels, A., 2006. Inequality aversion, eciency, and maximin preferences in

simple distribution experiments: comment. The American economic review 96 (5), 19061911.

Bolton, G. E., Ockenfels, A., 2010. Betrayal aversion: Evidence from brazil, china, oman,

switzerland, turkey, and the united states: Comment. The American Economic Review

100 (1), 628633.

Bolton, G. E., Ockenfels, A., Stauf, J., 2015. Social responsibility promotes conservative risk

behavior. European Economic Review 74 (0), 109  127.

Chakravarty, S., Harrison, G. W., Haruvy, E. E., Rutström, E. E., 2011. Are you risk averse

over other people's money? Southern Economic Journal 77 (4), 901913.

Charness, G., 2000. Responsibility and eort in an experimental labor market. Journal of Eco-

nomic Behavior & Organization 42 (3), 375384.

Charness, G., Gneezy, U., 2012. Strong evidence for gender dierences in risk taking. Journal

of Economic Behavior & Organization 83 (1), 5058.

Charness, G., Gneezy, U., Halladay, B., 2016. Experimental methods: Pay one or pay all.

Journal of Economic Behavior & Organization 131, 141150.

24
Charness, G., Gneezy, U., Imas, A., 2013. Experimental methods: Eliciting risk preferences.

Journal of Economic Behavior & Organization 87, 4351.

Charness, G., Jackson, M. O., 2009. The role of responsibility in strategic risk-taking. Journal

of Economic Behavior & Organization 69 (3), 241247.

Charness, G., Rabin, M., 2002. Understanding social preferences with simple tests. The Quar-

terly Journal of Economics 117 (3), 817869.

Charness, G., Rabin, M., 2005. Expressed preferences and behavior in experimental games.

Games and Economic Behavior 53 (2), 151169.

Cheung, S. L., Coleman, A., 2014. Relative performance incentives and price bubbles in exper-

imental asset markets. Southern Economic Journal 81 (2), 345363.

Cvetkovich, G., 1972. Eects of sex on decision policies used for self and decision policies used

for other persons. Psychonomic Science 26 (6), 319320.

Dewatripont, M., Rochet, J.-C., Tirole, J., 2010. Balancing the banks: Global lessons from the

nancial crisis. Princeton University Press.

Dohmen, T., Falk, A., Human, D., Sunde, U., Schupp, J., Wagner, G. G., 2011. Individual

risk attitudes: Measurement, determinants, and behavioral consequences. Journal of the

European Economic Association 9 (3), 522550.

Eckel, C. C., Grossman, P. J., 2008a. Forecasting risk attitudes: An experimental study using

actual and forecast gamble choices. Journal of Economic Behavior & Organization 68 (1),

117.

Eckel, C. C., Grossman, P. J., 2008b. Men, women and risk aversion: Experimental evidence.

Handbook of experimental economics results 1 (7), 106173.

Eriksen, K. W., Kvaløy, O., 2010. Myopic investment management. Review of Finance 14 (3),

521542.

Fehr, E., Schmidt, K. M., 1999. A theory of fairness, competition, and cooperation. The Quar-

terly Journal of Economics 114 (3), 817868.

Fischbacher, U., 2007. z-tree: Zurich toolbox for ready-made economic experiments. Experi-

mental Economics 10 (2), 171178.

25
Foerster, S., Linnainmaa, J. T., Melzer, B. T., Previtero, A., 2017. Retail nancial advice: does

one size t all? The Journal of Finance 72 (4), 14411482.

Fox, C. R., Tversky, A., 1995. Ambiguity aversion and comparative ignorance. The Quarterly

Journal of Economics 110 (3), 585603.

French, K., Baily, M., Campbell, J., Cochrane, J., Diamond, D., Due, D., Kashyap, A.,

Mishkin, F., Rajan, R., Scharfstein, D., et al., 2010. The squam lake report: xing the

nancial system. Journal of Applied Corporate Finance 22 (3), 821.

Fudenberg, D., Levine, D. K., 2012. Fairness, risk preferences and independence: Impossibility

theorems. Journal of Economic Behavior & Organization 81 (2), 606612.

Füllbrunn, S. C., Luhan, W. J., 2017. Decision making for others: The case of loss aversion.

Economics Letters 161, 154156.

Garcia-Retamero, R., Galesic, M., 2012. Doc, what would you do if you were me? on self

other discrepancies in medical decision making. Journal of Experimental Psychology: Applied

18 (1), 38.

Gennaioli, N., Shleifer, A., Vishny, R., 2012. Neglected risks, nancial innovation, and nancial

fragility. Journal of Financial Economics 104 (3), 452468.

Gennaioli, N., Shleifer, A., Vishny, R., 2015. Money doctors. The Journal of Finance 70 (1),

91114.

Gneezy, U., Potters, J., 1997. An experiment on risk taking and evaluation periods. The Quar-

terly Journal of Economics 112 (2), 631645.

Harbaugh, W. T., Krause, K., Vesterlund, L., 2010. The fourfold pattern of risk attitudes in

choice and pricing tasks. The Economic Journal 120 (545), 595611.

Harrison, G. W., 2006. Hypothetical bias over uncertain outcomes. In: List, J. A. (Ed.), Using

epxperimental methods in environmental and resource economics. Northampton. MA: Elgar,

pp. 4169.

Holmen, M., Kirchler, M., Kleinlercher, D., 2014. Do option-like incentives induce overvalua-

tion? evidence from experimental asset markets. Journal of Economic Dynamics and Control

40, 179194.

Holt, C. A., Laury, S. K., 2002. Risk aversion and incentive eects. American Economic Review

92 (5), 16441655.

26
Holt, C. A., Laury, S. K., 2005. Risk aversion and incentive eects: New data without order

eects. The American Economic Review 95 (3), 902904.

Hsee, C. K., Weber, E. U., 1997. A fundamental prediction error: Selfothers discrepancies in

risk preference. Journal of Experimental Psychology: General 126 (1), 45.

Kahneman, D., Tversky, A., 1979. Prospect theory: An analysis of decision under risk. Econo-

metrica: Journal of the Econometric Society, 263291.

Kirchler, M., Lindner, F., Weitzel, U., 2019. Delegated decision making and social competition

in the nance industry. Available at SSRN 3177459.

Kleinlercher, D., Huber, J., Kirchler, M., 2014. The impact of dierent incentive schemes on

asset prices. European Economic Review 68, 137150.

Pahlke, J., Strasser, S., Vieider, F. M., 2012. Risk-taking for others under accountability. Eco-

nomics Letters 114 (1), 102105.

Pahlke, J., Strasser, S., Vieider, F. M., 2015. Responsibility eects in decision making under

risk. Journal of Risk and Uncertainty 51 (2), 125146.

Pollmann, M. M., Potters, J., Trautmann, S. T., 2014. Risk taking by agents: The role of

ex-ante and ex-post accountability. Economics Letters 123 (3), 387390.

Polman, E., 2012. Selfother decision making and loss aversion. Organizational Behavior and

Human Decision Processes 119 (2), 141150.

Rajan, R. G., 2006. Has nance made the world riskier? European Financial Management

12 (4), 499533.

Reynolds, D. B., Joseph, J., Sherwood, R., 2009. Risky shift versus cautious shift: determining

dierences in risk taking between private and public management decision-making. Journal

of Business & Economics Research 7 (1).

Ross, L., Greene, D., House, P., 1977. The "false consensus eect": An egocentric bias in social

perception and attribution processes. Journal of Experimental Social Psychology 13 (3), 279

301.

Stone, E. R., Allgaier, L., 2008. A social values analysis of selfother dierences in decision

making involving risk. Basic and Applied Social Psychology 30 (2), 114129.

27
Stoner, J. A. F., 1961. A comparison of individual and group decisions involving risk. Ph.D.

thesis, Massachusetts Institute of Technology.

Sutter, M., 2009. Individual behavior and group membership: Comment. The American Eco-

nomic Review 99 (5), 22472257.

Swalm, R. O., 1966. Utility theory-insights into risk taking. Harvard Business Review 44 (6),

123136.

Trautmann, S. T., Kuilen, G., 2015. Belief elicitation: A horse race among truth serums. The

Economic Journal 125 (589), 21162135.

Trautmann, S. T., Van De Kuilen, G., 2015. Ambiguity attitudes. The Wiley Blackwell hand-

book of judgment and decision making 1, 89116.

Trope, Y., Liberman, N., 2010. Construal-level theory of psychological distance. Psychological

Review 117 (2), 440.

Ubel, P. A., Angott, A. M., Zikmund-Fisher, B. J., 2011. Physicians recommend dierent

treatments for patients than they would choose for themselves. Archives of Internal Medicine

171 (7), 630634.

Viscusi, W. K., Magat, W. A., Huber, J., 1987. An investigation of the rationality of consumer

valuations of multiple health risks. The RAND Journal of Economics, 465479.

von Gaudecker, H.-M., van Soest, A., Wengström, E., 2012. Experts in experiments. Journal of

Risk and Uncertainty 45 (2), 159190.

Wray, L. D., Stone, E. R., 2005. The role of self-esteem and anxiety in decision making for self

versus others in relationships. Journal of Behavioral Decision Making 18 (2), 125144.

28
Appendix
A Regressions
In the tables A.5 and A.6, we regress the investment levels for the clients (XO and XL ) and

the shifts (SO and SL ) in OTH and LIM respectively on dierent treatments (dummy for LIM

( LIM ), dummy for high payment in OTH (High Payment ), dummy for high fee in LIM (High
Fee )), on order (whether decision making for others (OTH or LIM) was rst or last (Order = 1

if last), and on subjects' background (general risk ( Risk ), female dummy (Female ), age (Age ),
dummy for the eld of study being economics ( Econ ), and social responsibility (SRS )).
General risk equals the self-reported risk preference in form of a likert scale question from

one to ten with a higher number indicating a higher willingness to take risks(Dohmen et al.,

2011). The social responsibility score was taken from Berkowitz and Lutterman (1968)).

Table A.5 considers three OLS regressions considering the 140 observations without limited

liability (model 2) and the 105 observations with limited liability (model 3), and a model

including both (model 1) clustered at session level. The rst model neglects the additional

treatments with higher payments in OTH (xed payment was 31.50 euro instead of zero) or

higher fees in LIM (fee was 50 percent instead of 5 percent). The results reect our observations.

Limited liability increases the shift signicantly (observation 3. General risk decreases the shift

signicantly 1 in model 2. However, general risk plays no signicant role in model 3. The

additional treatments High Payment and High Fee show no or a weak correlation with the

shift, i.e. high payments for the money managers in OTH do not change risk taking for others

but a higher fee in LIM tends to reduces risk taking for other. The social responsibility score

has no inuence on the shift.

29
 Table A.5: Investment shifts in OTH (SO ) and LIM (SL )

All No Limited Limited Liability

Liability

∗∗∗ ∗∗∗
Order 0.80 0.41 1.18
(0.24) (0.24) (0.32)
∗∗∗
LIM 2.92
(0.25)
∗∗∗ ∗∗∗
General Risk -0.36 -0.34 -0.11
(0.100) (0.11) (0.22)
Social Responsibility Score 0.16 0.025 -0.27
(0.26) (0.33) (0.39)
Female 0.099 -0.10 0.044
(0.29) (0.36) (0.32)
Econ -0.31 -0.35 0.15
(0.27) (0.34) (0.46)
∗
Age 0.023 0.091 -0.073
(0.043) (0.046) (0.053)
High Payment 0.044
(0.26)
∗
High Fee -0.66
(0.35)
∗∗
Constant -0.073 -1.08 4.70
(1.61) (1.93) (1.83)
Observations 245 175 154
Adjusted R2 0.320 0.076 0.045

Notes. Variables explained above. Standard errors in parentheses. The asterisks refer to the p-value from a permutation
test testing the H that the shift equals zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

Table A.6 considers similar regressions as table A.6. However, we now consider investment

levels instead of investment shifts and make use of truncated regressions censored at X = 0 and

X = 9. Money managers invest signicantly more for their clients when limited liable. Further

on, order plays a role in the experiments with limited liability, i.e. subjects invest more when

LIM is played last. The money makers willingness to take risk is also reected in the decision

for others, i.e. the higher the willingness to take risks, the higher the risk money managers take

for others. Finally, females seem to take less risk for their clients when limited liable.

30
 Table A.6: Investment levels in OTH (XO ) and LIM (XL )

All No Limited Limited Liability

Liability

∗∗ ∗∗∗
Order 0.74 0.094 1.46
(0.30) (0.25) (0.39)
∗∗∗
LIM 3.52
(0.31)
∗∗
General Risk 0.16 0.28 0.25
(0.12) (0.12) (0.29)
Social Responsibility Score 0.24 0.32 0.17
(0.36) (0.43) (0.26)
∗∗
Female -0.72 -0.28 -1.74
(0.45) (0.45) (0.67)
∗∗ ∗
Econ -0.90 -0.81 0.095
(0.39) (0.43) (0.92)
∗ ∗
Age 0.0094 0.057 -0.14
(0.046) (0.033) (0.082)
High Payment 0.22
(0.41)
High Fee -0.68
(0.74)
∗∗∗
Constant 2.47 0.49 9.01
(1.94) (1.80) (3.05)

Observations 245 175 154

Notes. Variables explained above. Standard errors in parentheses. The asterisks refer to the p-value from a permutation
test testing the H that the shift equals zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

B Sessions with IND, ALL, and OTH

B.1 Order Eects
To test whether an order eect has an impact on investment levels, we compare 70 observations

in which subjects made investment decisions in the order OTH-IND-ALL and 70 observations

in the order ALL-IND-OTH.

We make use of a permutation tests to evaluate the Null that investment levels and shifts do

not dier comparing the two dierent orders. As table B.7 indicates, the order of the treatments

has no eect on the main variables of interest. Further on, observation 1 also holds each of the

two subsets.

31
 Table B.7: Order Dierences

OTH-IND-ALL (70) ALL-IND-OTH (70) Dierence (p-value)

XI 4.68 (2.59) 4.37 (2.25) 0.30 (0.461)
XA 3.85 (1.95) 4.00 (1.88) -0.15 (0.637)
XO 3.91 (1.85) 3.84 (2.54) 0.06 (0.864)

∗∗∗ ∗∗
SA -0.83 (1.61) -0.38 (1.51) -0.45 (0.088)
∗∗∗ ∗
SO -0.77 (2.11) -0.53 (2.67) -0.24 (0.558)

Notes. The table reports averages in investments and shifts in the three treatments separated by order. The last
column shows the p-values from a permutation test comparing the two previous columns. The asterisks refer to the
p-value from a permutation test testing the H that dierences equal zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

B.2 Fixed Payment Eects

To test whether the payment condition for the money manager in OTH has an impact on

investment levels we compare 70 observations with a zero payment for the decision maker and

35 observations with a payment of 31.50 euro for the decision maker (both in the order OTH-

IND-ALL). We nd no signicant dierence comparing the two dierent payment conditions;

the investment level in OTH and the shift are almost equal in both conditions. Table B.8 shows

and overview of the results.

Table B.8: Fixed Payment Dierences

πm = 0 (70) πm = 31.5 (35) Dierence (p-value)

XO 3.91 (1.85) 4.03 (2.26) -0.12 (0.701)
∗∗∗
SO -0.77 (2.11) -0.63 (2.37) -0.14 (0.769)

Notes. The table reports averages in investments and shifts in the three treatments separated by order. The last
column shows the p-values from a permutation test comparing the two previous columns. The asterisks refer to the
p-value from a permutation test testing the H that dierences equal zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

B.3 Risk Categorization via IND

In the main text, we consider general risk as out measure for subjects' risk attitudes. However,

we can also use XI as a measure for risk attitudes (Charness et al., 2013). If we replicate table

1 by using the categories below median XI  and above median XI  we nd similar results

which support observation 2.

32
 Table B.9: Average Investments in euro

Investment in IND

below median investment above median investment

n = 140 (n = 53) (n = 64)
XI 4.53 (2.42) 2.27 (0.87) 6.67 (1.73)
XA 3.92 (1.91) 2.46 (1.18) 5.24 (1.73)
XO 3.87 (2.21) 2.60 (1.62) 4.92 (2.16)

∗∗∗
SA -0.60 (1.57) 0.20 (0.83) -1.43 (1.83)
∗∗∗
SO -0.65 (2.39) 0.34 (1.62) -1.75 (2.61)
Notes. The second column contains all observations. The next two columns categorize subjects in two groups according
to XI which can be seen as a measure of risk attitudes. Subjects in the group below median (above median ) invested
below (above) median investment levels in IND. Rows report average investments and average shifts together with the
standard deviation in parentheses. The asterisks refer to the p-value from a permutation test testing the Null that S
equals zero (** = p <0.05, *** = p <0.01).

B.4 Instructions
Find below the translated instructions for the order OTH-IND-ALL. They start with general in-
structions, followed by the three separate instructions for each treatment which were distributed
only when the preceding treatment was concluded. The German instructions are available upon
request.

INSTRUCTIONS

Welcome to the experiment. Please do not talk to any other participant from now on. We
kindly ask you to use only those functions of the PC that are necessary for the conduct of
the experiment. The purpose of this experiment is to study decision behavior. You can earn
real money in this experiment. Your payment will be determined solely by your own decisions
according to the rules on the following pages. The data from the experiment will be anonymized
and cannot be related to the identities of the participants. Neither the other participants nor
the experimenter will nd out which choices you have made and how much you have earned
during the experiment.

SUB EXPERIMENTS

You will participate in three independent sub experiments followed by a short questionnaire.
For each sub experiment you receive a new set of instructions. Of the three sub experiments
only one will be paid out at the end of the experiment. The payo relevant experiment will be
randomly determined by the roll of a die.

EXPERIMENT 1 [ Treatment OTH ]

Groups - At the begin of the experiments you will be randomly organized in groups of seven
participants. Your group aliation has no impact on your tasks or our payment.

Role - In this part participants are either active or passive members. In each group there
is only one active member. This member decides for the other six members and, thereby,
determines their payo. The active group member will randomly be determined at the end of
the experiment. First, all participants decide as the active member for all other group members.
At the end of the experiment the real active member will be determined and his decision will
be implemented.

Task - In the following your decision as an active member will be explained. The passive
members receive 9 euro each. You now decide for each of the other members how much of
The investment is the same for each passive group
their 9 euro to invest in a risky project.
member, i.e., when you invest a certain amount then you invest this amount for each passive
group member. The remaining amount (9 euro - Investment) will be paid out to each passive
member independent of the project's success.

33
The project is either a success or a failure. In case of a success each passive member gets her
invested amount back and in addition receives 2.5 times of the investment as a gain:
Payment in case of success = 9 + 2.5×Investment.
In case of a failure the investment is lost:
Payment in case of failure = 9−Investment.
Whether the project is successful will be determined by the throw of a six-sided die at the end
of the experiment. In case of a ve or six, the project is a success, in case of a one, two, three
or four the project is a failure. The probability of success is therefore 33.33%.

The active member receives no payo in this sub experiment.

Procedure - Details on how to enter the investments - calculation of potential payments, elds
of entry, etc - will be displayed on the upper part of your screen once the experiment has started.

EXPERIMENT 2 [Treatment IND ]

In this experiment you decide only for yourself, independent of the other participants. You
receive 9 euro and decide how much of their 9 euro to invest in a risky project. The remaining
amount (9 euro - Investment) will be paid out independent of the project's success.

The project is either a success or a failure. In case of a success You will get your invested
amount back and in addition receive 2.5 times of the investment as a gain:
Payment in case of success = 9 + 2.5×Investment.
In case of a failure the investment is lost:
Payment in case of failure = 9−Investment.
Whether the project is successful will be determined by the throw of a six-sided die at the end
of the experiment. In case of a ve or six, the project is a success, in case of a one, two, three
or four the project is a failure. The probability of success is therefore 33.33%.

Procedure - Details on how to enter the investments - calculation of potential payments, elds
of entry, etc - will be displayed on the upper part of your screen once the experiment has started.

EXPERIMENT 3 [Treatment ALL]

Groups - At the begin of the experiments you will be randomly organized in groups of
seven participants. You will be regrouped, this means that the group members are not the
same as in the rst sub experiment. Your group aliation has no impact on your tasks or your
payment.

Role - In this part participants are either active or passive members. In each group there is
only one active member. This member decides for himself and the other six members and, there
by, determines the payos for the whole group. The active group member will randomly be
determined at the end of the experiment. First, all participants decide as the active member for
all group members. At the end of the experiment the real active member will be determined
and his decision will be implemented.

Task - In the following your decision as an active member will be explained. Each group member
(active and passive) members receives 9 euro each. You now decide for each member of the
The investment
group, including yourself, how much of the 9 euro to invest in a risky project.
is the same for each passive group member, i.e., when you invest a certain amount then you
invest this amount for yourself and for each passive group member. The remaining amount (9
euro - Investment) will be paid out to each group member independent of the project's success.

The project is either a success or a failure. In case of a success each group member gets her
invested amount back and in addition receives 2.5 times of the investment as a gain:
Payment in case of success = 9 + 2.5×Investment.
In case of a failure the investment is lost:
Payment in case of failure = 9−Investment.
Whether the project is successful will be determined by the throw of a six-sided die at the end
of the experiment. In case of a ve or six, the project is a success, in case of a one, two, three
or four the project is a failure. The probability of success is therefore 33.33%.

34
Procedure - Details on how to enter the investments  calculation of potential payments,
elds of entry, etc will be displayed on the upper part of your screen once the experiment has
started.

END OF EXPERIMENT

At rst we will determine, by the roll of a die, which experiment will determine your payo.
Thereafter, a separate dice roll for each group will determine whether the project was successful
or not. After you answered a short questionnaire your payment will be shown at your screen.
Please enter the amount on your receipt. You will be called individually to the payo desk.
Please bring the small number plate and the signed receipt with you. The payment will be in
cash, private and anonymous.

B.5 Decision Screens

Figures B.9 to B.9 provide screenshots of the investment decisions in each treatment. Subjects

were able to enter arbitrary investment levels in the eld EINSATZ in euro (Investment in

euro). A click on the gray button (Generate Payos) added a new line to a table. The table

listed the chosen investment and the potential payos for the passive and the active members

together depending on the treatments together with the respective probabilities. The ultimate

investment was chosen by marking one line in the list and by clicking the red button (CONFIRM

YOUR DEFINITIVE INVESTMENT). Then a pop-up asked whether the decision is ultimate or

whether the subject wants to revise it. In this example, the subject rst entered an investment

amount of 2.86 euro, then 1.75 euro, and then 8.62 euro. The amount 5.67 euro was not

generated yet. However, the investment 1.75 euro was chosen and the red button opened the

dialog for a nal conrmation of the choice. The z-Tree code is available upon request.

35
Figure B.9: Screenshot of Investment Decision: IND

36
Figure B.9: Screenshot of Investment Decision: ALL

37
 Figure B.9: Screenshot of Investment Decision: OTH

C Sessions with IND, ALL, and LIM

C.1 Order Eects
To test whether an order eect has an impact on investment levels in LIM, we compare 49

observations in which subjects made investment decisions in the order LIM-IND-ALL and 56

observations in the order ALL-IND-LIM.

We make use of a permutation tests to evaluate the Null that investment levels and shifts do

not dier comparing the two dierent orders. As table B.7 indicates, the order of the treatments

has an eect on the main variables of interest. However, observation 1 also holds for the two

subsets.

38
 Table C.10: Order Dierences

LIM-IND-ALL (49) ALL-IND-LIM (56) Dierence (p-value)

XI 4.77 (1.85) 4.34 (2.19) 0.34 (0.284)
XL 6.30 (2.12) 7.33 (2.06) -1.03 (0.013)
∗∗∗ ∗∗∗
SL 1.52 (2.49) 2.98 (2.67) -1.46 (0.005)

Notes. The table reports averages in investments and shifts in LIM separated by order. The last column shows the
p-values from a permutation test comparing the two previous columns. The asterisks refer to the p-value from a
permutation test testing the H that dierences equal zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

C.2 Instructions
The instructions for LIM simply replace the instructions for OTH (Section B.4). Instructions

for LIM50 just replaced the 5 percent by 50 percent.

EXPERIMENT 1 [ Treatment LIM ]

Groups - At the begin of the experiments you will be randomly organized in groups of seven
participants. Your group aliation has no impact on your tasks or our payment.

Role - In this part participants are either active or passive members. In each group there
is only one active member. This member decides for the other six members and, thereby,
determines their payo. The active group member will randomly be determined at the end of
the experiment. First, all participants decide as the active member for all other group members.
At the end of the experiment the real active member will be determined and his decision will
be implemented.

Task - In the following your decision as an active member will be explained. The passive
members receive 9 euro each. You now decide for each of the other members how much of
The investment is the same for each passive group
their 9 euro to invest in a risky project.
member, i.e., when you invest a certain amount then you invest this amount for each passive
group member. The remaining amount (9 euro - Investment) will be paid out to each passive
member independent of the project's success.

The project is either a success or a failure. In case of a success each passive member gets her
invested amount back and in addition receives 2.5 times of the investment as a gain. The active
member receives 5 % from the gain of each passive member:
Payment in case of success
Prot = 2.5 × Investment
passive member = 9 + 95% × Prot
active member = 6 × 5% × Prot

In case of a failure the investment is lost and the active member receives no payment
Payment in case of failure
passive member = 9 − Investment
active member =0

Whether the project is successful will be determined by the throw of a six-sided die at the end
of the experiment. In case of a ve or six, the project is a success, in case of a one, two, three
or four the project is a failure. The probability of success is therefore 33.33%.

Procedure - Details on how to enter the investments - calculation of potential payments, elds
of entry, etc - will be displayed on the upper part of your screen once the experiment has started.

39
C.3 Decision Screen
Figures C.10 provides a screenshots of the investment decisions for treatment LIM similar to

section B.5.

Figure C.10: Screenshot of Investment Decision: LIM

40
D Gender Eects
We added this section for those researchers who are interested in gender eects in risk taking.

Charness and Gneezy (2012) provide evidence for a gender eect in the Gneezy and Potters

(1997) environment with varying payments and probabilities. To test whether a gender eect,

or rather a sex eect, has an impact on investment levels, we compare the observations in

which females made investment decisions to the observations in which males made investment

decisions. As table D.11 indicates, males invest signicantly more than females in IND, ALL,

and LIM, i.e. in all treatments in which the money managers faces true outcome consequences.

Investments in OTH, however, are not dierent. The general patterns from observation 1 and

3 and holds for both sex though.

Table D.11: Gender dierences

Males Females Dierence (p-value)

XI (162/132) 4.99 (2.41) 3.91 (1.82) 1.08 (<0.001)
XA (162/132) 4.59 (2.21) 3.72 (1.54) 0.87 (<0.001)
XO (75/65) 4.14 (2.37) 3.56 (1.99) 0.58 (0.119)
XL (59/46) 7.20 (2.12) 6.39 (2.11) 0.81 (0.055)

∗∗∗
SA (162/132) -0.40 (1.39) -0.19 (1.38) -0.21 (0.193)
∗∗ ∗∗
SO (59/46) -0.69 (2.62) -0.61 (2.13) -0.08 (0.842)
∗∗∗ ∗∗∗
SL (75/65) 1.84 (2.56) 2.89 (2.69) -1.05 (0.45)

Notes. The table reports averages in investments and shifts separated by sex. The numbers in column one report the
number of observations for males and females (males/females). The last column shows the p-values from a permutation
test comparing the two previous columns. The asterisks refer to the p-value from a permutation test testing the H
that dierences equal zero (* = p <0.1, ** = p <0.05, *** = p <0.01).
0

In a debrieng questionnaire, we ask several questions on risk aversion in line with Dohmen

et al. (2011). Using a Mann-Whitney U test, we nd signicant gender dierences in questions

about risk taking in general (p = 0.004), driving a car (p = 0.019), and making nancial

decisions (p = 0.002). We nd no eect in questions on risk taking in sports and leisure

(p = 0.074). in career (p = 0.319), health (p = 0.937), trust in strangers (p = 0.567), or in a

hypothetical investment decision (p = 0.132).

41
Highlights
• We consider financial decision making for others with and without limited liability
• Without, decision makers take less risks for others than for themselves
• With limited liability decision makers take excessive risks for others
• Conclusion: Convex incentives crowd out social preferences
• New experimental design allows for private gains and social losses

Page | 1