Journal of Financial Economics 128 (2018) 403–421

Contents lists available at ScienceDirect

Journal of Financial Economics

journal homepage: www.elsevier.com/locate/jfec

Time varying risk aversionR

Luigi Guiso a,b, Paola Sapienza b,c,d, Luigi Zingales b,d,e,∗
a
EIEF, Italy
b
CEPR, UK
c
Northwestern University, USA
d
NBER, USA
e
University of Chicago, USA

a r t i c l e i n f o a b s t r a c t

Article history: Exploiting portfolio data and repeated surveys of an Italian bank’s clients, we test whether
Received 16 July 2015 investors’ risk aversion increases following the 2008 crisis. We find that, after the cri-
Revised 17 October 2016
sis, both qualitative and quantitative measures of risk aversion increase substantially and
Accepted 18 October 2016
that affected individuals divest more stock. We investigate four explanations: changes in
Available online 19 March 2018
wealth, expected income, perceived probabilities, and emotion-based changes of the utility
JEL codes: function. Our data are inconsistent with the first two channels, while they suggest that
G11 fear is a potential mechanism underlying financial decisions, whether by increasing the
G41 curvature of the utility function or the salience of negative outcomes.
D91 © 2018 Published by Elsevier B.V.

1. Introduction (i.e., not from stock prices) that aggregate risk aversion in-
deed fluctuates over time?
As Campbell and Cochrane (1999) show, to fit histori- Aggregate risk aversion can fluctuate either because the
cal data, asset pricing models require large fluctuations in risk aversion of the typical investor changes or because the
the aggregate risk aversion. Yet, what is the direct evidence distribution of wealth among investors with different risk
aversion changes. In this paper we test the first channel
and analyze whether individual risk aversion increases fol-
R
We thank Nick Barberis, John Campbell, John Cochrane, James Dow,
lowing the major financial crisis of the last 80 years—the
Stefan Nagel, Andrei Shleifer, Ivo Welch, Jeffrey Wurgler, and an anony-
mous referee for very helpful comments. We also benefited from com- 2008 one. We do so by exploiting portfolio choices and
ments from participants at seminars at the University of Chicago Booth, some survey-based measures of risk aversion elicited in a
Boston College, University of Minnesota, University of Michigan, Hong sample of clients (labeled investors from now on) of a large
Kong University, London Business School, Statistics Norway, European Italian bank (hereafter, the bank) in 2007 and repeated on
Central Bank, University of Maastricht, Warwick University, University of
Montreal, the 2011 European Financial Association Meetings, the 2012
the same set of people in 2009.
European Economic Association Meetings, the April 2013 NBER Behav- We find that both qualitative and quantitative measures
ioral Finance Meeting, UCLA Behavioral Finance Association, and Stanford of risk aversion exhibit large increases following the crisis.
University. Luigi Guiso gratefully acknowledges financial support from The risk premium required to accept a risky gamble with a
PEGGED, Paola Sapienza from the Zell Center for Risk and Research at
50% chance of winning 10,0 0 0 euros increases from 10 0 0
Kellogg School of Management, and Luigi Zingales from the Stigler Center
and the Initiative on Global Markets at the University of Chicago Booth euros to 2500 euros. Similarly, the fraction of investors
School of Business. We thank Filippo Mezzanotti for excellent research who say they do not want to take any financial risk goes
assistantship, and Peggy Eppink for editorial help. from 16% to 43%. Individuals who experience an increase in
∗
Corresponding author. Present address: The University of Chicago risk aversion are four times more likely to sell their stock
Booth School of Business, 5807 S. Woodlawn Avenue, Chicago IL 60637,
holdings during the worst moment of the crisis than peo-
USA.
E-mail addresses: [email protected] (L. Guiso),
ple who do not.
[email protected] (L. Zingales).

https://doi.org/10.1016/j.jfineco.2018.02.007
0304-405X/© 2018 Published by Elsevier B.V.
404 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

There are many theories that can explain large changes nitude similar to what we observe in naturally occurring
in individual risk aversion. The best way to classify them is data.
according to the channel that leads from the shock (large To simulate in the lab this change in state, we rely
drop in stock prices) to the outcome (increase in indi- on a fear conditioning model. As in the classical Pavlov
vidual risk aversion). The most prominent mechanism in (1927) experiment, the fear response can be triggered by
the literature is changes in wealth, as predicted by the conditioning factors, which have little or nothing to do
classical von Neumann-Morgenstern utility function and with the experience itself. Kinreich et al. (2011) show that
by the habit persistence model (Campbell and Cochrane, watching a horror movie stimulates the amygdala in a way
1999). Prospect theory (Barberis et al., 2001) can also work consistent with the arousal of fear. Thus, to generate the
through this channel. fear produced by a stock market crash, we treat a sam-
Changes in background risk are the second most com- ple of students with a five-minute excerpt from the movie
mon explanation. Changes in the outside environment can Hostel (2005, directed by Eli Roth), characterized by stark
affect an individual’s expected income (Heaton and Lucas, and graphic images. It shows a young man inhumanly tor-
20 0 0; Guiso and Paiella, 20 08) and in so doing modify the tured in a dark basement.
risk aversion of the value function. We find that students treated with the horror movie ex-
A third possibility is that a major shock has an ef- hibit a higher risk aversion (both according to the quantita-
fect on the expected distribution of returns as in Bordalo tive and the qualitative measure) that is very similar to the
et al. (2012). In their model individuals’ attention is di- one experienced by the Italian bank’s investors in 2009.
rected to some particular realization that receives dis- The treated subjects’ risk premium is $672 (27%) higher
proportionate weighting (salience). Therefore, individuals than the untreated ones’. Interestingly, the effect is entirely
may become more risk averse because the financial melt- concentrated among students who dislike horror movies.
down has made the worst stock market realization more The ones who like them seem unaffected.
salient. Such an experiment shows that fear causes an increase
Finally, a major shock can affect the emotions of in- in our measures of risk aversion, even in the absence of
vestors and alter their decisions about their willingness to any change in the outside environment (which is the same
take risks because it changes their perceived utility loss of for the treated and non-treated sample) and in their en-
bad outcomes (Loewenstein, 20 0 0). In economic language, dowment (which is unaffected by the treatment). Obvi-
a major shock leads to a state-contingent increase in the ously, the experiment cannot prove in any way that such
curvature of the utility function. a causal link exists among bank investors in our sample.
Consistent with the wealth channel, we find that indi- Nevertheless, it does provide evidence that such a large
viduals who experience extraordinarily big losses exhibit a increase in measured risk aversion can indeed occur even
greater increase in the quantitative measure of risk aver- when not mediated by wealth changes and in absence of
sion. Yet, we also find that risk aversion increases substan- background risk. The psychology model based on fear is
tially even among those individuals who did not experi- consistent with both the survey and the experimental data.
ence any loss, suggesting that not all the changes in risk Our result is consistent with Cohn et al. (2015). In a lab
aversion occur via changes in wealth. experiment with a sample of financial professionals, they
We do not find much support for the changes-in- show that those “treated” with a stock market crash sce-
expected-income channel. For example, the increase in risk nario become more risk averse and report an increase in
aversion of retirees (who in Italy enjoy a public pension) fear, even though they do not experience any direct finan-
and of public employees (who at the time faced little or cial loss. This nice result is complementary to ours. Like us,
no risk of layoffs) is no smaller than that of the rest of the they show that risk aversion can fluctuate with the stock
population. market performance. Yet, we can show that an actual stock
We test the salience theory by looking at the individ- market crash, caused by the financial crisis, increases risk
ual responses on the expected distribution of returns. For aversion and induces a change in portfolio allocation. Since
those subjects willing to answer the question in both pe- Cohn et al. are limited to lab data, they are only able to
riods, we do find evidence of changes in the expected dis- show changes in the lab. However, they can successfully
tribution of returns. Furthermore, we do find a significant establish a causal link between the fear induced by the
increase in the fraction of people unwilling to answer that crash and a more conservative portfolio allocation, while
question. we can only establish a correlation.
Our evidence is also consistent with the Loewenstein Our paper is also related to Weber et al. (2013). They
(20 0 0) hypothesis that, faced with a negative shock, indi- survey online customers of a brokerage account in Eng-
viduals are affected by an emotion (fear) that alters their land between September 2008 and June 2009, asking them
willingness to take risk in both financial and nonfinan- how they would allocate 10 0,0 0 0 pounds between a risk
cial domains. However, with naturally occurring data it free asset and the UK stock market index and a few mea-
is difficult to design a direct test with any power to re- sures of risk attitudes. Similarly to us, they find that risk
ject this hypothesis. For this reason, we run a labora- taking decreases between September and March, but, un-
tory experiment. While previous experiments have already like us, their measures of risk attitudes do not change. One
shown that emotions can increase risk aversion (Kuhnen likely explanation for this difference is that their baseline
and Knutson, 2005, 2011; Knutson et al., 2008), our goal is measures are taken in September 2008 when the situa-
to test whether the fear associated with a negative shock tion is already problematic, while our baseline measures
can indeed change our measures of risk aversion by a mag- are taken long before the inception of the crisis.
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 405

Finally, our paper is also related to the literature on minute interview.1 Specifically, investors were asked the
market sentiments (see Baker and Wurgler, 2007 for a two risk aversion questions (see below), a question about
summary), on fear and risk aversion (e.g., Lerner and Kelt- trust in their bank advisor or financial broker, and a ques-
ner, 20 0 0, 20 01), and that on the effect of emotions and tion about stock market expectations using exactly the
anxiety on risk attitudes, portfolio choice, and stock re- same wording that was used to ask these questions in the
turns (Kamstra et al., 2003; Kramer and Weber, 2012; Kan- 2007 survey. Before asking the questions, the interviewer
dasamy et al., 2014; and Bassi et al., 2013). Several of these made sure that the investor was the same person who an-
papers establish that risk preferences vary over time and swered the 2007 survey by collecting a number of demo-
with emotions. graphic characteristics and matching them with those from
The rest of the paper continues as follows. Section the 2007 survey.
2 presents the data and our measures of risk aversion.
Section 3 reports the results about the changes in risk 2.2. Risk aversion questions
aversion. Section 4 tests for possible explanations of these
changes. Section 5 discusses how fear can be induced in a If we want to test whether changes in risk aversion can
lab experiment and reports its results. Section 6 concludes. explain movements in asset prices, we need a way to infer
risk aversion that is independent of asset prices. For this
reason, we resort to survey-based measures.2 We have two
2. Data description and measures such measures. The first, patterned after a question in the
US Survey of Consumer Finances, is a qualitative indicator
2.1. Sample of risk tolerance. Each participant is asked: "Which of
the following statements comes closest to the amount
Our main data source is the second wave of the in- of financial risk that you are willing to take when you
vestors’ survey run between June and September 2007 make your financial investment: (1) a very high return,
done by a large Italian bank. The survey is comprised of with a very high risk of losing money; (2) high return and
interviews with a sample of 1686 Italian customers. The high risk; (3) moderate return and moderate risk; (4) low
sample was stratified according to three criteria: geograph- return and no risk."
ical area, city size, and financial wealth. To be included in In a world in which people face the same risk-return
the survey, customers must have had at least 10,0 0 0 euros tradeoffs and make portfolio decisions according to Mer-
worth of assets with the bank at the end of 2006. The sur- ton’s formula, their risk/return choice reflects their degree
vey is described in greater detail in Section A1 of the In- of relative risk aversion. In such a world, the answers to
ternet Appendix, in which we also compare it to the Bank the above questions can fully characterize people’s risk
of Italy survey, which is representative of the Italian popu- preferences. However, if people differ in beliefs about stock
lation. market returns and/or volatility, these differences will con-
Besides collecting detailed demographic information, taminate their answers to the above question. This bias
data on investors’ financial investments, information on would affect not only cross-sectional comparisons, but also
beliefs, expectations, and risk perception, the survey col- intertemporal ones, possibly revealing a change in risk
lected data on individual risk attitudes by asking both preferences when none is present. While we elicit expecta-
qualitative questions on people’s preferences regarding tions about stock market returns and volatility and control
risk/return combinations in financial decisions as well for them, the controls are not without errors.
as their willingness to pay for a (hypothetical) risky As Fig. 1A shows, in 2007, only 16% of the sample
prospect. chooses the “low return and no risk” answer to the quali-
For the sample of investors who participated in the tative risk aversion question. So most investors are willing
2007 survey, the bank gave us access to the administrative to accept some risk if compensated by a higher return, but
records of the assets that these investors have with them. very few (1.8%) are ready to choose very high risk and very
Specifically, we can merge the survey data with adminis- high return. From the answers to this question we con-
trative information on the stocks and on the net flows of struct a categorical variable ranging from 1 to 4 with larger
26 asset categories that investors have at the bank. We de- values corresponding to greater risk aversion.
scribe in detail the data set and its content in the Internet The second measure of risk aversion is designed to
Appendix, Section A2. These data are available at monthly be independent of subjective beliefs. Each investor was
frequency for 35 months beginning in December 2006, and presented with several choices between a risky prospect,
we use them to obtain measures of variation in wealth which paid 10,0 0 0 euros or zero with equal probability
and portfolio investments over time. Since some house-
holds left the bank after the interview, the administrative 1
Since the second survey was conducted during the same season as
data are available for 1541 households instead of the 1686
the first, the differences in risk aversion cannot be due to season varia-
in the 2007 survey. tions in the length of day (see Kamstra et al., 2003).
To study time variations in risk attitudes, in the spring 2
A potential alternative, followed by Friend and Blume (1975), is to in-
of 2009, we asked the same company that ran the 2007 fer an individual’s relative risk aversion from his share of investments in
survey to run a telephone survey on the sample of 1541 risky assets. This method is not appropriate to study time series changes
in risk aversion because the necessary maintained assumption is that
investors interviewed in 2007 who were still clients of the portfolio shares are instantaneously adjusted. If not, any adjustment costs
bank. The telephone survey was fielded in June 2009 and will be reflected in the estimated changes in risk aversion (Bonaparte and
asked a much more limited set of questions in a short 12- Cooper, 2009).
406 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

3000 4000 3500

2215
Risk Premium:

Risk Premium:
Premium:

3000
2000
Median
mean

973 2000
1000 1000
1000
0 0
2007 2009 2007 2009

Fig. 1. Frequency distribution of the level of risk aversion indicators in 2007 and 2009. Panel A reports the frequency distribution of the qualitative measure
of risk aversion in 2007 and 2009. The qualitative indicator elicits the investment objective of the respondent, offering them the choice among “Very high
returns, even at the risk of a high probability of losing part of my principal”; “A good return, but with an ok degree of safety of my principal”; “An ok
return, with good degree of safety of my principal”; “Low returns, but no chance of losing my principal.” Responses are coded with integers from 1 and
4, with a higher score indicating a higher aversion to risk. Panel B shows the frequency distribution of the risk aversion indicator based on the answers
to the lottery that delivers 10,0 0 0 euros or zero with equal probability in 2007 and 2009. The risk premium for this gamble is computed as the difference
between the expected value of the lottery (50 0 0 euros) and each respondent’s certainty equivalent. Panel C reports the average and median risk premium
for this gamble in the two years.
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 407

Table 1
Comparing the sample of non-participants and participants to the second interview.
This table shows summary statistics for a set of variables for the two samples of respondents to the 2007
bank survey: those who did not participate in the 2009 survey and those who did. The third column reports the
p-value for the null that the means of the two samples are the same. The variables are defined in the Internet
Appendix, Section A4.

Nonparticipants Participants p-value of test of

(N. 1020) (N. 666) equality

Age 55.02 54.5 0.39

Male 0.7 0.7 0.77
Married 0.69 0.67 0.40
North 0.53 0.49 0.12
Center 0.24 0.25 0.61
Education 12.44 13.18 0.00
Trust advisors 2007 3.74 3.83 0.05
Qualitative measure of risk aversion 2.88 2.85 0.31
Quantitative measure of risk aversion: risk premium 865 792 0.66
Financial assets 2007 150,980 158,950 0.22
Financial assets 2009 139,720 142,290 0.73
Stock ownership Jan 2007 0.44 0.44 0.93
Stock ownership Jun 2009 0.41 0.42 0.80
Share in stocks Jan 2007 0.1 0.11 0.54
Share in stocks Jun 2009 0.08 0.08 0.51
Risky assets ownership Jan 2007 0.79 0.81 0.41
Risky assets ownership Jun 2009 0.74 0.73 0.63
Risky assets share Jan 2007 0.56 0.58 0.29
Risky assets share Jun 2009 0.50 0.50 0.90

and a sequence of certain sums of money. These sums We will refer to the measure based on preferences for
were progressively increasing between 100 euros and risk-return combinations as the qualitative indicator and to
90 0 0 euros. More risk-averse people will give up the risky the one based on the lottery as the quantitative indicator.
prospect for lower certain sums. Thus, the first certain The first is a measure of relative risk aversion, while the
sum at which an investor switches from the risky to the second is a measure of absolute risk aversion. These risk-
certain prospect identifies (an upper bound for) his/her aversion measures should be thought of as measures of the
certainty equivalent. risk aversion for the investor’s value function and as such
Specifically, investors were asked: “Imagine being in a are potentially affected by any variable that impacts peo-
room. To get out you have two doors. Behind one of the ple’s willingness to take risk, such as their wealth level or
two doors there is a 10,0 0 0 euro prize, behind the other any background risk they face.
nothing. Alternatively, you can get out from the service In the 2009 survey we also ask “After the stock mar-
door and win a known amount. If you were offered 100 ket crash did you become more cautious and prudent
euros, would you choose the service door?” in your investment decisions?” The possible answers are:
If the investor accepted 100 euros, the interviewer “More or less like before,” “A bit more cautious,” or “Much
moved on to the next question. Otherwise the interviewer more cautious.” Thirty-five percent of the investors declare
asked whether the investor would accept 500 euros to exit themselves to have become “much more” and 18% “a bit
the service door and if not, 1500 and if not, progressively more” cautious. Therefore, we create a “change in cautious-
30 0 0, 40 0 0, 50 0 0, 550 0, 70 0 0, 90 0 0, and more than 90 0 0 ness” variable equal to zero if the response is no change,
euros. one if the response is a bit more, and two if it is much
The question was framed so as to resemble a pop- more. The summary statistics for these measures and all
ular TV game (Affari Tuoi, the Italian version of the TV the other variables are contained in Table 2.
game Deal or No Deal), analyzed by Bombardini and Trebbi
(2012). Incidentally, it is similar to the Holt and Laury 2.3. Sample attrition
(2002) strategy which has proved particularly successful in
overcoming the under/over-report bias implied when ask- Of the 1541 who were contacted, roughly one-third
ing willingness to pay/accept. agreed to be re-interviewed so that we end up with a two-
We code answers to this question as the certainty year panel of 666 investors. Table 1 compares the charac-
equivalent value required by the investor to give up the teristics of responders and non-respondents to the 2009
risky prospect. We then compute a risk premium as the survey along several dimensions. In the first part of the ta-
difference between the expected value of the gamble and ble, we compare the two samples according to the demo-
an investor’s certainty equivalent. Fig. 1B presents these graphic characteristics collected in the 2007 survey such
risk premia. Interestingly, roughly one-third of the in- as age, gender, marital status, geographical location, and
vestors appear risk loving in 2007. The extreme risk-averse education. The differences are small and not statistically
people (with a risk premium equal to 4900) are 17% in significant, with the exception of education, for which we
2007. cannot statistically reject the hypothesis that the two sam-
408 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Table 2
Summary statistics of risk aversion measures, other variables, and controls.
Panel A reports the summary statistics for the risk aversion measures. The qualitative measure of risk aversion elicits
the investment objective of the respondent, offering them the choice among “Very high returns, even at the risk of a
high probability of losing part of the principal,” “A good return, but with an ok degree of safety of the principal,” “An ok
return, with good degree of safety of the principal,” “Low returns, but no chance of losing the principal.” The responses
are coded with integers from 1 to 4, with a higher score meaning a higher risk aversion. The quantitative measure of
risk aversion is calculated by eliciting the certainty equivalent for a gamble that delivers either 10,0 0 0 euros or zero
with equal probability; the risk premium is then obtained as the difference between the expected value of the gamble
(50 0 0 euros) and each respondent’s certainty equivalent. Panels B, C, and D report the summary statistics for all the
other variables used later in estimates and defined in the Internet Appendix, Section A4.

Panel A: Risk aversion measures in 2007 and 2009

Quantitative measure (risk

premium in euros) Qualitative measure

Mean Median Sd Mean Median Sd

Level in 2007 792 10 0 0 3248 2.87 3 0.72

Level in 2009 2215 3500 2815 3.28 3 0.73
Change 20 09/20 07 1423 2500 3994 0.42 0 0.81
Fraction of people with 0.55 0.46
increase in risk aversion
Fraction of people with 0.18 0.44
unchanged risk aversion
1- Fraction of people with a 0.73 0.90
decrease in risk aversion

Panel B: Other variables: levels

Mean Median Sd

Male 0.70 1 0.46

Age 54.81 57 12.3
Education (years) 12.73 13 4.25
Retired 0.33 0 0.47
Government employee 0.33 0 0.47
Size of the financial loss −0.05 −0.01 0.07
Log net wealth: 2007 13.11 13.10 0.59
Log net wealth: 2009 13.05 13.03 0.64
Risky asset ownership 2007 0.79 1 0.40
Risky asset share 2007 0.57 0.70 0.37
Knightian uncertainty 0.29 0 0.46
Change in cautiousness 2.13 2 0.90
Trust advisors 3.78 4 0.91
Change trust advisors −0.24 0 1.11

Panel C: Other variables: first differences

Mean Median Sd

 Log net wealth 20 09–20 07 −0.06 −0.051 0.27

 Log net wealth 20 09–20 08 −0.04 −0.005 0.20
 Ownership of risky assets −0.06 0 0.35
 Share risky assets −0.04 0 0.24
 in Expected gross stock return in euros −432 −168 3442
 in Range gross stock market return in euros −249 −50 7200

Panel D: Variables in the rebalancing equation

Mean Median Sd
γ 
Risk aversion ratio: γ − 1 ωiM −0.05 0.0 0.21
i
Mean risky asset share 2007: ωiM 0.52 0.59 0.36
 pωiM 
Post-shock share: pωiM +1−ωiM
0.65 0.72 0.28

ples differ. Still, the economic magnitude of the difference 865 euros among non-respondents to the 2009 telephone
is small (less than a year of education). survey and 792 euros among respondents (p-value = 0.66).
In the middle part of the table, we compare the two While the two samples do not differ in observable char-
samples according to their risk attitudes, as measured in acteristics in 2007, they might differ in time-varying char-
2007. Along this dimension, which is the most important acteristics. For example, the crisis might have affected the
one for our analysis, participants in the 2009 survey do two groups differentially, in a way that is correlated with
not differ from non-participants. For instance, the average their willingness to be re-interviewed. Fortunately, we
2007 risk premium for the hypothetical risky prospect is have the 2007 and 2009 administrative data (and hence
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 409

Table 3
Correlation between the various measures of risk aversion.
The table reports the correlation between the two measures of risk aversion for the two survey waves 2007 and 2009, the cor-
relation between their changes, and the correlations between their changes and a measure of change in cautiousness in investing
defined in the Internet Appendix, Section A4.

Correlations between measure of risk aversion

Qualitative and Qualitative and Change in Change in Change in

quantitative quantitative qualitative and qualitative quantitative
indicator: 2007 indicator: 2009 change in indicator and indicator and
quantitative change in change in
indicator: 2007– cautiousness cautiousness
2009

0.116 0.160 0.118 0.119 0.074

p-value 0.00 0.00 0.002 0.002 0.056

portfolio choices) of both the respondents and the non- 2009 cross sections. Also, as documented in the literature,
respondents. Hence, the last part of Table 1 compares these men are less risk averse than women (Byrnes et al., 1999).
choices. The stock of financial assets, before and after the Third, we document that our measures have predic-
crisis, does not differ between the two groups, nor does tive power on investors’ financial choices. Panel B of
the fraction of financial wealth invested in stock or risky Table 4 shows that the qualitative indicator of risk aversion
financial assets. Similarly, there are no differences in the is strongly negatively correlated with ownership of risky
percentage of people who own stock or risky financial as- financial assets (a dummy variable equal to one if an in-
sets. We, thus, conclude that there does not seem to be dividual owns stocks and corporate bonds in her portfo-
any systematic selection in the investors’ decisions to be lio). The correlation with the lottery-based measure is neg-
re-interviewed in June 2009. ative but weaker. This is partly due to some investors pro-
viding noisy answers in the quantitative measure, which
is more difficult to understand. When we drop inconsis-
2.4. Validating the risk aversion measure tent answers—those who are highly risk averse according
to the first indicator (a value greater than 2), but highly
A large and increasing literature shows that questions risk lovers on the basis of the lottery question (a risk pre-
like the ones above predict risk taking behavior in various mium less or equal to −40 0 0 euros)—we also find that the
domains (see, for instance, Dohmen et al., 2011; Donkers quantitative measure significantly predicts risky asset own-
et al., 2001; Barsky et al., 1997; Guiso and Paiella, 2006, ership (column 3). Furthermore, the change in risk aver-
2008). The risk aversion measures elicited in this way are sion predicts the change in assets ownership: those whose
also robust to the specific domain of risk: using a panel risk aversion increased more between 2007 and 2009 are
of 20,0 0 0 German consumers, Dohmen et al. (2011) show more likely to become non-stockholders over the same pe-
that indicators of risk attitudes over different domains tend riod (Table 4, panel C). In the Internet Appendix (Table A.2,
all to be correlated, with correlation coefficients of around panels A and B), we also document a similar pattern for
0.5—a feature that is consistent with the idea that risk the level and in the change in the share of wealth held in
aversion is a personal trait. risky assets.
To validate our measures, we run various tests. First,
in Table 3 we document that our qualitative and quanti- 2.5. Measure of subjective probabilities
tative measures are positively correlated either when us-
ing the 2007 cross section (correlation coefficient 0.12) or Depositors were also asked to state what (in their view)
the 2009 cross section (correlation 0.16) or when looking the minimum and maximum value of a 10,0 0 0 euro invest-
at the correlation between the changes in the two mea- ment in a fully diversified stock mutual fund would be 12
sures between 2007 and 2009 (correlation coefficient 0.12). months later. Next, they were asked to report the probabil-
We also find that the change in cautiousness variable has ity that the value of the stock by the end of the 12 months
a 12% correlation (p-value 0.002) with the changes in the was above the mid-point of the reported support. Under
qualitative measure of risk aversion and a 7.4% correlation very simple assumptions about the shape of the distribu-
(p-value 0.056) with changes in the quantitative measure tion, this parsimonious information allows computing the
of risk aversion. subjective mean and variance of stock market returns. We
Second, we document that our measures tend to be cor- have computed these moments assuming the distribution
related in expected ways with classical covariates of risk is uniform, but results are the same assuming it is triangu-
attitudes.3 As panel A of Table 4 shows, risk aversion de- lar.
creases with total wealth levels in both the 2007 and the Some investors claim that the maximum one-year-ahead
value of a 10,0 0 0-euro investment in the stock market is
3
zero. This is a sign that they might have misunderstood the
These patterns of correlations have been documented in several
studies, either using surveys or experiments [e.g., Croson and Gneezy,
question, raising some doubt on the quality of this mea-
(2009) for gender; Barsky et al. (1997), Guiso and Paiella (20 06, 20 08), sure. To address these concerns, we drop all the responses
Hartog et al. (2002)]. where the maximum is below 30 0 0 euros (i.e., maximum
410 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Table 4
Validation.
Panel A, columns 1 and 2 report estimates of an ordered probit model where the dependent variable is the qualitative measure of risk
aversion for the two different survey waves, 20 07 and 20 09. Columns 3 and 4 report interval regressions; the dependent variable is the
interval of the risk premium obtained from the lottery question as described previously, divided by the expected value of the lottery.
Panel B reports marginal effects of probit models, where the dependent variable is a dummy variable equal to one if the individual
holds risky assets in her portfolio. The quantitative indicator of risk aversion is the risk premium defined previously divided by 50 0 0 (the
expected value of the lottery) in 2007. The last column reports the results eliminating individuals who reported inconsistent answers to
the risk aversion question (those who are highly risk averse according to the first measure—a value greater than 2—but risk lover on the
basis of the quantitative question—a certainty equivalent greater or equal to 90 0 0 euros).
Panel C reports the marginal effects for ordered probit regressions; the dependent variable is the change in a dummy variable equal to
one if an individual owns risky assets between June 2009 and June 2008 (just before the financial collapse). The change in risk aversion
is calculated as the difference between the reported answers in the 2009 and 2007 surveys. The change in the quantitative indicator of
risk aversion is the change in risk premium divided by 50 0 0 (the expected value of the lottery). All the other variables are defined in the
Internet Appendix, Section A4. Robust standard errors are in parentheses. ∗ /∗ ∗ /∗ ∗ ∗ indicates statistical significance at the 10%, 5%, and 1%
level, respectively. In Panel C, changes in wealth have been trimmed out at the 1st and 99th percentiles.
Panel A: Cross-sectional correlates of risk aversion
Risk aversion qualitative Risk aversion quantitative
Whole sample Eliminate inconsistent answers
2007 2009 2007 2009 2007 2009
Male −0.338∗∗∗ −0.497∗∗∗ 0.006 0.150∗∗ −0.031 0.011
(0.063) (0.109) (0.046) (0.075) (0.041) (0.059)
Age −0.047∗∗ −0.011 −0.001 0.026 0.010 0.031∗
(0.020) (0.032) (0.014) (0.021) (0.013) (0.017)
Age2/100 0.049∗∗∗ 0.020 0.008 −0.015 −0.004 −0.022
(0.019) (0.031) (0.014) (0.021) (0.012) (0.017)
Education −0.035∗ ∗ ∗ −0.044∗∗∗ −0.013∗ ∗ −0.010 −0.021∗∗∗ −0.016∗ ∗ ∗
(0.007) (0.012) (0.005) (0.008) (0.005) (0.006)
Log net wealth: 2007 −0.139∗∗∗ −0.057 −0.042
(0.047) (0.036) (0.033)
Log net wealth: 2009 −0.147∗∗ 0.013 0.001
(0.074) (0.050) (0.040)
Observations 1494 584 1494 584 1311 548

Panel B: Risk aversion and risky assets ownership

Whole sample Eliminate inconsistent answers
(1) (2) (3)
Risk aversion qualitative: 2007 −0.122∗∗∗
(0.032)
Risk aversion quantitative: 2007 −0.001 −0.055∗∗∗
(0.005) (0.004)
Male 0.129∗∗∗ 0.154∗∗∗ 0.162∗∗∗
(0.016) (0.028) (0.027)
Age 0.022∗ ∗ 0.025∗∗∗ 0.026∗∗∗
(0.010) (0.008) (0.009)
Age2/100 −0.020∗ ∗ −0.023∗∗∗ −0.023∗∗∗
(0.009) (0.007) (0.008)
Education 0.018∗ ∗ ∗ 0.020∗ ∗ ∗ 0.019∗∗∗
(0.005) (0.002) (0.002)
Trust advisor 2007 0.039∗∗∗ 0.047∗∗∗ 0.049∗∗∗
(0.011) (0.007) (0.008)
Log net wealth: 2007 0.145∗∗∗ 0.152∗∗∗ 0.137∗∗∗
Observations 1494 1494 1311

Panel C: Effect of changes in risk aversion on changes in ownership of risky assets

Whole sample Eliminate inconsistent answers

(1) (2) (3)
 Risk aversion: Qualitative measure −0.172
(0.105)
 Risk aversion: Quantitative measure −0.154∗ −0.243∗∗
(0.088) (0.100)
Male 0.367∗∗ 0.379∗∗ 0.308
(0.172) (0.171) (0.189)
Age 0.074 0.071 0.070
(0.062) (0.062) (0.070)
Age2/100 −0.069 −0.066 −0.071
(0.058) (0.058) (0.065)
Education 0.006 0.008 0.011
(0.019) (0.020) (0.020)
 in Advisor trust −0.065 −0.082 −0.088
(0.072) (0.072) (0.082)
 Log net wealth 20 09–20 07 1.467∗∗∗ 1.351∗∗∗ 1.214∗∗∗
(0.371) (0.366) (0.462)
Observations 569 569 500
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 411

Fig. 2. Distribution of expected gross stock returns. The figure shows the cross-sectional distribution of one-year ahead subjective expected stock returns
in 2007 and 2009. Expected returns are obtained from the answers to a question asking the minimum and maximum value of an investment of 10,0 0 0
euros in a fund representative of the Italian stock market one year later and the probability that the value is below the mid-point of this range. We drop
the observations where the respondents claimed that the maximum one-year-ahead value of a 10,0 0 0 euros investment is 30 0 0 euros or less. The reported
distributions are for the respondents to both the 2007 and 2009 surveys.

net return of −70% or below). In so doing, the number of from zero at the 1% level. In 2007, only 16% of the respon-
observations drops from 470 to 337. The resulting distribu- dents chose the most conservative option “low return and
tion of individual expected returns is presented in Fig. 2. no risk,” in 2009, 43% did. In the Internet Appendix (Ta-
ble A.3) we show the transition matrix of the responses.
2.6. Changes in wealth and financial losses There is a homogeneous shift toward more conservative
combinations of risk and return. Of the people who chose
For all the participants in the survey, we have access to the most aggressive option (“Very high returns, even at the
the administrative data, which include the amount of de- risk of a high probability of losing part of the principal”),
posits at the bank, the amount and composition (by broad 83% change toward a more conservative one. Of those who
categories) of their brokerage account at the bank, the pro- had chosen the second more risky combination (“Moderate
portion of financial wealth represented by their holdings Risk/Medium Return”), 77% move to more conservative op-
at the bank, and the value of their house. Thanks to these tions, while only 2% move to the most aggressive one. Of
data we can infer the changes in investors’ total wealth those who chose “Small Risk/Some Return,” 44% move to
and the losses incurred on their financial portfolio. “Low Return and No Risk,” while only 9.5% move to more
The change in total wealth is computed as the sum of aggressive options. Note that these very stark results are
the actual changes in their financial wealth held at the present in spite of a censoring in the data. The 16% of
bank (divided by the proportion of financial wealth held the respondents who chose the most conservative option
at the bank to obtain an estimate of total household as- in 2007 cannot become more risk averse.4
sets) and the imputed changes in home equity. To impute Fig. 1B compares the distribution of the risk premium
these changes we look at the variation in local indexes of before and after the crisis and Fig. 1C the mean and me-
real estate prices. dian risk premium before and after the crisis (the transi-
The losses on the financial portfolio are computed by tion matrix is in table A.4 of the Internet Appendix). As
multiplying the holdings of risky securities (stocks, stock Fig. 1C shows, before the crisis the average risk that pre-
mutual funds, corporate bonds, and corporate bond funds) mium investors were willing to pay to avoid a gamble of-
before August 2008 by the proportional change in their fering 10,0 0 0 euros and zero with equal probability was
price between September 2008 (before the Lehman col- 973 euros. In 2009, the risk premium for the same group
lapse) and February 2009 (when the stock market started of people increased to 2215 euros. The median increased
to rebound) and then scaling by the stock of financial as- from 10 0 0 to 350 0. All these changes are statistically dif-
sets before August 2008. ferent from zero. Interestingly, the large surge in the risk
premium is driven by a much higher number of people
who choose the lowest certainty equivalent (and thus the
3. Changes in risk aversion
highest risk premium).
Since the risk premium is proportional to the investor
3.1. Changes in individual risk aversion
risk aversion, these estimates imply that the (absolute) risk
aversion of the average investor has increased by a factor
Fig. 1A compares the distribution of the qualitative
of two, and that of the median investor by a factor of 3.5!
measure of risk aversion before and after the crisis. Before
the crisis the average response was 2.87, after the crisis
it has jumped to 3.28 (recall, a higher number indicates 4
The effect of this censoring is considered in the Data Appendix, Fig.
higher risk aversion). This change is statistically different A1.
412 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Table 5
Net sales of risky assets.
The table reports the average monthly net-purchases (purchases minus sales) of risky assets (as proportion of the assets held at the
beginning of the period) before the Lehman collapse (June 2008–August 2008, column 1) and after (October 2008–April 2009, column
2). The third column reports the difference in net sales in the post-and pre-Lehman collapse (p-values in parentheses). Row A includes
all the households who responded to both surveys and own some risky assets at the beginning of the period. Rows B and C refer to
the subsample of Row A, that respectively did not exhibit an increase in the qualitative measure of risk aversion and those who did.
Similarly, Rows D and E report the net purchase of individuals who respectively did not and did exhibit an increase in the quantitative
measure of risk aversion.

Difference
Pre-Lehman Post-Lehman Oct Post – Pre Lehman
Jun-Aug 08 08-Apr 09 (p-values)
(1) (2) (3)

A. Subsample who own risky assets at the beginning of period −0.022 −0.038 −0.06
(0.224)
B. Subsample of A without an increase in qualitative risk aversion −0.023 −0.021 0.002
(0.922)
C. Subsample of A with an increase in qualitative risk aversion −0.020 −0.052 0.032
(0.057)
p-values of the difference between B and C (0.596) (0.089) –
D. Subsample of A without an increase in quantitative risk aversion −0.012 −0.020 −0.008
(0.616)
E. Subsample of A with an increase in quantitative risk aversion −0.029 −0.046 −0.017
(0.250)
p-values of the difference between D and E (0.594) (0.424) –

One benign reason why risk aversion might have in- assets held at the beginning of the period) before the
creased is that from the first to the second survey our in- Lehman collapse (June to August 2008) and after it (Octo-
vestors became older. While true, this effect is likely to be ber 2008 to April 2009). The sample in row A includes all
small, since only two years went by. Nevertheless, we com- the households who responded to both surveys and own
puted the average risk aversion by age and then took the some risky assets at the beginning of the period (so that
difference of risk aversion between the first and the second they can sell them). In the 2008 months leading to the
surveys, keeping the age constant (i.e., between the aver- Lehman collapse, households were net sellers of risky as-
age of people who were 30 in 2009 and the people who sets (2.2% average net sales). It is important to notice that
were 30 in 2007). The results are unchanged. the level of net purchases is relatively small, because most
Such an increase cannot be attributed solely to a wors- households at any given time are inactive. After Lehman,
ening of expectations about the distribution of future in- the net sales almost doubled (3.8%), but this difference is
vestments since it manifests itself also in the quantitative not statistically significant (column 3 of the table).
measure, which is unrelated to the stock market. In fact, In rows B and C we report the net purchase before and
the probability distribution underlying the gamble in the after Lehman of the individuals who did not exhibit an in-
quantitative measure is objective, not subjective. crease in the qualitative measure of risk aversion and of
those who did. While the net purchases are almost iden-
3.2. Changes in portfolios tical before (see the first column), they are very different
after Lehman (column 2). People with no increase in risk
As Table 2.C shows, the share of risky assets in individ- aversion sell 2.1% of their assets, while individuals with an
ual portfolios dropped between 2007 and 2009. This drop increase sell 5.2%. This level of net purchases is statistically
could be the mechanical effect of a decrease in the value different from the one exhibited by the same people be-
of risky assets held in the portfolio or the consequence of fore Lehman (p-value of 0.057) and that of the investors
an active sale of risky assets by individuals (or both). In who did not exhibit an increase in qualitative risk aversion
the current section we focus on the second component: (p-value of 0.089).
the active sale of risky assets. In a standard (Mertonian) In rows D and E we report the net purchase before and
model of portfolio choice with constant risk aversion, ex- after Lehman of the individuals who did not exhibit an in-
pected value, and volatility, individuals should buy more crease in the quantitative measure of risk aversion and of
risky assets after a drop in their value. We showed that in- those who did, respectively. While the pattern is similar to
dividuals do not exhibit a change in the expected return the one observed in rows B and C, the differences here are
or volatility after the crisis. Thus, at least the individuals not statistically significant at the conventional levels.
who expressed opinions on the return distribution should
not be selling risky assets after the shock more intensively 3.3. A reality check on the magnitude of the changes
than the whole sample (indeed the data show that if any-
thing they seem to sell less). Our sample is representative of Italian individual in-
In Table 5 we report the average monthly net-purchases vestors, but not of all investors: institutional investors and
(purchases minus sales) of risky assets (as proportion of professional traders are not represented. Yet, if we treat it
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 413

Table 6
Aggregate risk aversion.
This table computes measures of the aggregate absolute risk aversion for the two
sample years 2007 and 2009 using as weights the total net wealth of the individuals
in the sample. The absolute risk aversion measure is obtained from the elicited risk
premium assuming an exponential utility function.

Total sample Stockholders

wealth weights: wealth weights:

2007 2009 2007 2009

ARA 2007 1.30 1.30 1.22 1.22
ARA 2009 2.28 2.25 2.42 2.40
Change in ARA 0.98 0.95 1.20 0.98

as a representative sample, we can compute the aggregate sharp increase in risk aversion is quantitatively sufficient
risk aversion and check whether the change in the aggre- to explain the severe drop in stock prices during the crisis.
gate risk aversion is large enough to explain the large drop Yet, it begs the question of what caused such an increase
in stock prices. in risk aversion.
To compute the aggregate risk aversion we start by
mapping the risk premium computed from the quantitative 4. What causes the changes in risk aversion?
question into a coefficient of absolute risk aversion by us-
ing a constant relative risk aversion (CRRA) utility function. 4.1. Changes in wealth
Then, we compute the aggregate risk aversion by weight-
ing these coefficients by the net total wealth of each indi- A characteristic that standard expected utility models
vidual. As Table 6 shows, the aggregate absolute risk aver- have in common with the non-standard ones (habit forma-
sion (ARA) in 2007 was 1.3. If we maintain the individual tion and prospect theory) is that any change in risk aver-
risk aversions estimated in 2007 and multiply them by the sion is mediated by changes in wealth. For this reason, we
2009 wealth weights, the aggregate risk aversion does not start by analyzing whether there is any relation between
change at all. By contrast, if we use the 2009 estimated changes in risk aversion and changes in wealth.
individual risk aversion, the aggregate risk aversion almost Fig. 3 plots a nonparametric estimation of the rela-
doubles. If we repeat the analysis restricting the sample to tionship between changes in risk aversion and the size
people who were stockholders in 2007, the results are the of financial losses incurred between September 2008 and
same. February 2009 (if we use total wealth the results are the
Now that we have computed the variation in aggregate same). As Fig. 3A shows, there is no consistent relation-
risk aversion, we can estimate whether this change is suf- ship between the increase in the qualitative measure of risk
ficiently large to justify the severe drop in stock prices aversion and the size of the losses in the financial portfo-
that took place. What is relevant for asset prices is the lio during the financial crisis. For losses between zero and
relative risk aversion. Since the change in total wealth is 20%, the increase in risk aversion is stable at 0.4, around
small (a relatively small fraction was invested in equity), 14% of the sample mean in 2007. For losses above 20%, the
all the increase in absolute risk aversion translates into an increase in risk aversion seems to first decrease and then
increase in the relative risk aversion. To compute how this increase.
increase could affect stock prices we make the (strong) as- As Fig. 3B shows, for the quantitative measure there
sumption that the only source of variation was a (tempo- seems to be a negative relation between the size of the
rary) increase in risk aversion. This implies that the future financial loss and the relative risk premium (the risk pre-
expected cashflow remains unchanged and that after one mium divided by the expected value of the lottery), consis-
year even the risk aversion returns to normal. Then, next tent with a wealth channel. Yet, even people with no losses
year stock price P1 should remain unchanged and all the exhibit a significant increase in the relative risk premium
adjustment should take place in today’s stock priceP0 . By (by 20 percentage points), which seems to contradict the
using the Merton (1969) model, we can write wealth channel.
P1 In Table 7 we revisit this issue in regression format,
re = − rf = γ σ 2 (1) which allows us to control for individual characteristics.
P0
The dependent variable in panel A is the change in the
where the left-hand side is the equity premium and σ 2 the qualitative risk aversion between 2007 and 2009. As con-
variance of stock returns. If the expected variance of re- trol variables in column 1 we use the initial level of risk
turns does not change and the risk aversion γ doubles to aversion, gender, two dummy variables for the age groups,
γ  = 2γ , as it does in our sample, then the initial stock and education. Our explanatory variable of interest is the
price P0 should be size of the financial loss, calculated as the loss in value of
P0 1 the risky investments between September 2008 and Febru-
= γ   (2) ary 2009, scaled by the value of financial assets held in
P0 1 − P0
r + P0
r
γ P1 f P1 f September 2008. As in the figure, we find no evidence of
which is roughly half of what it was before. Hence, stock correlation between this variable and the increase in risk
price roughly halves if risk aversion doubles. Thus, the aversion. In column 2 we re-estimate the same specifica-
414 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Fig. 3. Financial loss and change in risk aversion. The figure plots the relation between potential losses in the financial portfolio between September
2008 and February 2009 and the change in the qualitative indicator of risk aversion (Panel A) and in the risky premium of the quantitative lottery (Panel
B). The change in risk premium is scaled by the expected value of the lottery (50 0 0 euros). The figures show the 95% confidence interval around the
estimated polynomial. The relation is estimated using a kernel-weighted local polynomial regression. The financial loss is computed as loss in value of
risky investments held at the end of September 2008 between September 2008 and February 2009, scaled by the initial value of financial assets.
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 415

Table 7
Change in risk aversion, financial loss, and Knightian uncertainty.
Panel A reports ordered probit estimates for the first difference of the qualitative measure
of risk aversion. Panel B reports interval regressions estimates for the changes in the risk
premium scaled by the expected value of the lottery. All the other variables are defined
in the Internet Appendix, Section A4. Robust standard errors are in parentheses. ∗ /∗ ∗ /∗ ∗ ∗
indicates statistical significance at the 10%, 5%, and 1% levels, respectively. Size of financial
loss is computed as loss in value of risky investments held at the end of September 2008
between September 20 08 and February 20 09, scaled by the initial value of financial assets.
Change in wealth is computed as the log difference in the value of net wealth in the second
quarter of 2009 and the value in the second quarter of 2008. Changes in net wealth have
been trimmed out at the 1st and 99th percentiles.

Panel A

(1) (2) (3)

Risk aversion qualitative: 2007 −1.165∗ ∗ ∗ −1.172∗ ∗ ∗ −1.178∗ ∗ ∗

(0.083) (0.083) (0.083)
Male −0.422∗ ∗ ∗ −0.423∗ ∗ ∗ −0.350∗ ∗ ∗
(0.103) (0.102) (0.104)
Age ≤ 45 −0.695∗ −0.685∗ −0.604
(0.366) (0.372) (0.375)
Age ≥ 65 −0.553 −0.545 −0.501
(0.372) (0.379) (0.381)
Education −0.041∗ ∗ ∗ −0.042∗ ∗ ∗ −0.037∗ ∗ ∗
(0.012) (0.012) (0.012)
Size of financial loss 0.293
(0.583)
Change log net wealth 2009–Q2 0.513 0.503
2008
(0.399) (0.413)
Knightian uncertainty 0.441∗ ∗ ∗
(0.105)

Observations 572 572 572

Panel B

(1) (2) (3)

∗∗∗ ∗∗∗
Risk aversion −0.860 −0.860 −0.860∗ ∗ ∗
quantitative: 2007
(0.043) (0.043) (0.043)
Male 0.126∗ ∗ 0.126∗ ∗ 0.131∗ ∗
(0.057) (0.057) (0.058)
Age ≤ 45 −0.060 −0.069 −0.063
(0.170) (0.171) (0.174)
Age ≥ 65 0.071 0.071 0.075
(0.175) (0.176) (0.178)
Education −0.003 −0.003 −0.003
(0.005) (0.005) (0.005)
Size of financial loss −0.094
(0.326)
Change log net wealth 0.014 0.013
20 09–Q2 20 08
(0.232) (0.233)
Knightian uncertainty 0.029
(0.052)

Observations 503 503 503

tion with changes in wealth in place of size of the finan- (either because they gained or because they did not have
cial loss. Also in this case, there is no effect. In Table A.5 any risky assets and thus did not experience any loss).5
we show the robustness of these results to use a quadratic
form to model the effect of age. 5
The people with inconsistent answers are 59. We focus on financial
In panel B we repeat the same exercise for the quan-
losses; households could have suffered losses on housing wealth. This is
titative measure of risk aversion. As in the figure, we find not the case: between the second quarter of 2008 and the first quarter
no relation between the increase in risk aversion and ei- of 2009, house prices increased in all local markets but one where they
ther the size of the financial loss or the change in wealth. dropped by 3.7%. A nonparametric estimate of the relation between the
change in the qualitative and quantitative measures of risk aversion and
To better understand the robustness of these results we
the proportional change in total wealth leads to a similar conclusion. The
focus on the 295 people who did not experience any fi- certainty equivalent increases by a similar amount even for households
nancial loss between September 2008 and February 2009 whose total wealth holdings do not change or even increase.
416 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Before the crisis the qualitative measure of risk aversion younger people (who have most of their wealth in human
of this subsample was 2.82 (statistically not different from capital) than for older people. Table 8, panel B compares
the rest of the sample). After the crisis their qualitative the changes in the qualitative and quantitative measures
measure jumped to 3.25, again not statistically different for young (age below 45) and old (age above 65) peo-
from the jump in the rest of the sample. ple who did not suffer any financial loss. The change in
The same is true for the quantitative measure. Before risk aversion is not statistically different between the two
the crisis, the mean risk premium for such subsample was groups.
649 euros, not statistically different from the rest of the Thus, we find no evidence consistent with a change in
sample. After the crisis, the mean risk premium rises to labor income or other changes in background risk being
2260 euros, not statistically different from the rest of the the proximate factor that leads to a surge in risk aversion.
sample. Therefore, investors who did not experience any Yet, the only way to completely rule out this hypothesis is
loss exhibit an increase in risk aversion equal to those who to conduct a lab experiment, where the background risk is
did. perfectly controlled for.
Our measure of financial losses is based on the wealth
investors have deposited at the bank. We cannot exclude, 4.3. Changes in probability distribution
thus, that they might have faced a loss in other invest-
ments. To check this possibility we restrict attention to in- Bordalo et al. (2012) develop a theory according to
dividuals who declare that they only have financial wealth which individuals overweight the probability of salient
at the bank (184 observations). The results are very similar. payoffs. The collapse of Lehman and the fall in stock prices
The qualitative measure of the risk aversion increases from that ensued might have increased the salience of the neg-
2.89 before the crisis to 3.28 after the crisis—a change not ative payoffs, increasing their subjective probability. Given
different from that in the whole sample. The risk premium these probabilities, investors optimally behave in a more
on the quantitative measure increases from 757 euros to risk-averse manner, i.e., as we observe in our survey and
2382 euros, again a change similar as in the whole sam- our portfolio data.
ple. As described in Section 2.2, our survey data allow us
to calculate the subjective probability investors have about
4.2. Changes in expected future income future returns. Despite the limitations of this measure, it
is interesting to analyze the changes in the cross-sectional
What we observe is the value function risk-aversion. distribution of expected returns between 2007 and 2009
Thus, it can increase not only for a drop in wealth but (see Fig. 2). The 2007 mean (median) gross return is
also for an increase in the variability or a drop in the level 1.073 (1.057), and the 2009 one is 1.045 (1.042). In the
of future income. Since the income from financial assets is Kolmogorov-Smirnov test of the equality of the 2007 and
generally small relative to labor income, the main suspect 2009 distributions of returns, the distance parameter is
is the expected labor income. 0.1365, with a p-value of 0.003. Thus, we cannot reject the
With field data it is hard to test this channel, since the hypothesis that the two distributions are different. In par-
expected income depends upon many unobservable vari- ticular, Fig. 2 shows an increase in the weight of the dis-
ables. To gain some insights on the plausibility of this hy- tribution in the negative net returns domain. This shift is
pothesis we focus on people who face very little (possi- consistent with the salience hypothesis. However, if this ef-
bly no) labor income risk, such as government employ- fect was the primary reason for the observed increase in
ees. Note that our second survey (June 2009) predates the risk aversion in our sample, we would expect a correla-
Greek (October 2009) and euro crises (2010–12), and the tion between the change in risk aversion and the change
Italian government solvency was not seriously in doubt (at in expected returns. When we do so, the correlation is not
that time the spread between the 5-year Italian bond and statistically significant (the result is also confirmed in an
the German one was around 60 basis points). unreported regression).
As Table 8 shows, among people who did not experi- Fig. 2 hides an important fact. That is, 27% of the house-
ence any financial loss, government employees exhibit a holds who were willing to give an answer to the distri-
surge in the qualitative measure of risk aversion higher (al- butional questions in 2007 refused to do so in 2009. This
beit not statistically significant) than non-government em- change in behavior might reflect an increase in Knight-
ployees, while they exhibit an increase in the quantitative ian uncertainty, which might lead investors to behave in a
measure similar to that of non-government employees. more risk-averse way (Caballero and Krishnamurthy, 2008).
We repeat the same analyses dividing the sample of in- To test this hypothesis, in the third columns of
vestors who did not face financial losses between retirees Table 7 (panels A and B) we insert an indicator variable
and non-retirees. In Italy retirees enjoy a defined benefit equal to one if a household did not answer this question in
plan backed by government guarantee. Thus, the same con- 2009. When as a left-hand side variable we use our qual-
siderations above apply. We find that retired people have itative measure of risk aversion, this indicator variable has
statistically the same increase in the quantitative and qual- a positive coefficient, which is statistically different from
itative measures of risk aversion as the non-retired; if any- zero at the 1% level. Depositors who did not answer the
thing, the magnitude is bigger for the retired, contrary to distributional questions in 2009 have a risk aversion twice
the background risk hypothesis. as big as the mean. By contrast, when as a left-hand side
If risk aversion increases because of a change in the fu- variable we use the quantitative measure of risk aversion,
ture expected income, it should increase much more for we do not find any effect.
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 417

Table 8
Background risk and expectations about future income.
Panel A computes changes in the qualitative and quantitative measures of risk aversion for groups non-exposed (government-
employed and retired) and exposed (non-government employees and non-retired) to background labor income risk among
investors facing no financial losses. The third and sixth columns compute the difference between the two groups and the p-
value of the null hypothesis that the difference is zero (in parentheses). Panel B does a similar calculation for the groups whose
permanent income should respond more (the younger) or less (the older).

Panel A: Background risk

No financial losses

Government Non- Difference Retired Non-retired Difference

employees government (p-value) (Nobs = 94) (Nobs = 201) (p-value)
(Nobs = 96) employees
(Nobs = 180)
Change in 0.479 0.399 0.085 (0.420) 0.468 0.423 0.045
qualitative (0.661)
measure
Change in 0.316 0.318 −0.002 0.423 0.275 0.149
quantitative (0.989) (1.132)
measure (risk
premium/50 0 0)

Panel B: Expected future income: Sample no losses

No financial losses

Age < = 45 Age > = 65 Difference

(Nobs = 70) (Nobs = 48) (p-value)
Change in qualitative 0.358 0.438 −0.081
measure (0.585)
Change in 0.189 0.388 −0.199
quantitative measure (0.134)
(risk premium/50 0 0)

4.4. Changes in utility individual i would be given by

pωiM
The last possible channel is a change in utility. Most Ri = ωiM − (4)
economists are reluctant to accept as explanations changes pωiM + 1 − ωiM
in the utility function because—without any specific the-
If the risk aversion moves to γ i , then the portfolio re-
ory for the changes—we lack testable restrictions. We need
balancing of individual i is
a theory of why and how utility might change after a neg-
ative shock. Loewenstein et al. (2001) recognize that emo- γ M pωiM
Ri = ω − (5)
tions could affect decisions. This is tantamount to a state- γi i
pωiM + 1 − ωiM
contingent increase in the curvature of the utility func-
tion. We have already seen in Section 3.2 that investors We can nest these two specifications as
increased their sales of risky assets after Lehman. In what    
γ pωiM
follows we test whether the observed changes in risk aver- Ri = α − 1 ω M
+ βω M
+ δ (6)
sion can explain the financial decisions to rebalance the
γi i i
pωiM + 1 − ωiM
portfolio of risky assets. where if α = 0, β = 1, δ = −1 we obtain (4), i.e., the opti-
Let’s assume that before the crisis individual portfo- mal rebalancing under the standard Mertonian model with
e
lios were at the optimal Mertonian share ωiM = γ rσ 2 . This no changes in risk aversion, while ifα = 1,β = 0,δ = −1 we
i
assumption is realistic given that before the crisis stock obtain (5), i.e., the optimal rebalancing when the risk aver-
prices were fairly flat for a while and thus investors did sion parameter changes.
have the time to adjust. We denote with p the value of To test which expression fits the data best, we build
stocks after the shock relative to their value before, p < 1. empirical counterparts of the terms on the right-hand side
Then, after the severe market downturn the actual share of of (6); the details are reported in the Internet Appendix,
risky assets became Section A.4. We define the shock as the drop in stock
prices that occurs after August 2008, i.e., the pre-Lehman
pωiM month. Since prices continue to fall until February 2009,
ωi = (3) we define various measures of the drop in risky asset
pω + 1 − ωiM
M
i prices since August 2008, computed at different months
from September 2008 until February 2009. Importantly,
We have seen that after the crisis the distribution of ex- we construct an investor-specific measure of p by taking
pected returns did not change much. If the individual risk portfolio-weighted means of the drop in different com-
aversion did not change either, the portfolio rebalancing of ponents of the risky portfolio, using as weights the risky
418 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

Table 9
Portfolio rebalance.
Panel A reports regressions where the dependent variable is the flow of risky assets bought (positive) or sold (negative) in the three
months following the period specified in each column scaled by the value of total financial assets at the end of August 2008, prior
to the Lehman Brothers collapse. The risk aversion ratio is the ratio between the risk aversion before and after the Lehman collapse
multiplied by the average share of risky assets in 2007; the mean risky asset share is computed as the average share in risky assets
over the months from January to December 2007. The post-shock share is the risky share implied by the level of the risky asset prices
at the various dates specified in the columns after the collapse of Lehman. All the other variables are defined in the Internet Appendix,
Section A4. Robust standard errors are in parentheses. ∗ /∗ ∗ /∗ ∗ ∗ indicate statistical significance at the 10%, 5%, and 1% levels, respectively.
Change in wealth is computed as the log difference in the value of net wealth in the second quarter of 2009 and the value in the
second quarter of 2008. Changes in wealth have been trimmed out at the 1st and 99th percentile.

(1) (2) (3) (4) (5) (6)

September 08 October 08 November 08 December 08 January 09 February 09
γ 
Risk aversion ratio: γi − 1 ωiM 0.048∗ ∗ 0.043∗ 0.036 0.077∗ ∗ 0.088∗ ∗ 0.083∗
(0.023) (0.026) (0.027) (0.036) (0.038) (0.044)
Mean risky asset share 2007: ωiM 0.005 0.015 −0.001 0.001 0.006 −0.0 0 0
(0.012) (0.015) (0.017) (0.020) (0.022) (0.024)
 pωiM 
Post-shock share: pωiM +1−ωiM
−0.029∗ ∗ −0.028 −0.016 −0.030 −0.044∗ −0.033
(0.013) (0.017) (0.019) (0.021) (0.024) (0.025)
Male −0.005 −0.020 −0.026∗ −0.028∗ −0.034∗ ∗ −0.031∗
(0.010) (0.013) (0.014) (0.015) (0.016) (0.017)
Age 0.007∗ ∗ −0.001 0.002 −0.0 0 0 −0.001 −0.003
(0.003) (0.006) (0.006) (0.006) (0.006) (0.007)
Age2 −0.007∗ ∗ 0.001 −0.002 0.0 0 0 0.001 0.003
(0.003) (0.005) (0.006) (0.006) (0.006) (0.007)
Education 0.001 0.001 0.002 0.002 0.003 0.002
(0.001) (0.001) (0.002) (0.002) (0.002) (0.002)
 Log net wealth 20 09–20 08 0.221∗ ∗ ∗ 0.313∗ ∗ ∗ 0.309∗ ∗ ∗ 0.342∗ ∗ ∗ 0.383∗ ∗ ∗ 0.450∗ ∗ ∗
(0.072) (0.087) (0.089) (0.088) (0.091) (0.099)
Observations 502 502 502 502 502 502
R-squared 0.087 0.098 0.089 0.101 0.123 0.124

portfolio compositions of each individual as of August 5. The experiment

2008. Ri (t)is computed as the net flow of risky assets (pos-
itive for net purchases and negative for net sales), scaled While suggestive, this hypothesis cannot be tested with
by the value of total financial assets in August 2008. our data because we do not have any direct measure of
The results are reported in Table 9. In all regressions fear. Does the TV reporting of Lehman’s fired employees
we add some demographic controls and the change in to- trigger an emotional fear response, or does it increase the
tal assets; results are invariant to these controls. The left- subjective probability of a very bad outcome? And if it
hand side variable represents the active reallocation in the triggers a fear response, is this response sufficiently strong
period that goes from August 2008 to the date written to explain the increase in risk aversion that we have docu-
at the top of each column. Thus, in column 1 the real- mented in Section 4?
location considered is the one during the period August To separate the emotional response from a Bayesian re-
to September 2008. In all the specifications except one, sponse and establish whether an emotional response can
the α coefficient is significantly bigger than zero, albeit also generate large increases in risk aversion, we rely on a lab-
significantly less than one. In all the specifications, the oratory experiment in which the outside environment is
coefficient β is not different from zero, while the coef- controlled for. As long as the treatment provides no in-
ficient δ is negative and sometimes significantly different formation about the real world, the probability of an ex-
from zero, albeit always significantly different from −1. treme event should remain constant between treated and
Thus, neither of the two models perfectly fits the data. Yet, untreated samples. To discriminate between the two hy-
considering that the noise in the data tends to downward potheses, the key feature of such an experiment is to in-
bias the coefficient, the data seem more consistent with duce fear in the lab without altering a subject’s perception
model (5)—in which the changes in risk aversion impact of her financial and economic prospects. To achieve such a
the portfolio rebalancing—than with model (4). goal we rely on the fear conditioning model used in psy-
Loewenstein’s model can explain why, in the 2008 fi- chology. Notice that our intent is not to prove whether fear
nancial crisis context, investors who did not lose any causes an increase in risk aversion. This link has already
money became more risk averse even with respect to a been established (see, e.g., Cohn et al., 2015). Our purpose
gamble with given known odds such as our quantitative is to test whether the fear channel is powerful enough to
measure. The terrifying news appearing on television, the generate an increase in risk aversion of a magnitude that
interaction with friends who lost money in the market, resembles the one induced by the financial crisis.
or the pictures of fired people leaving their failed banks As for the classical Pavlov (1927) experiment, the fear
might have triggered an emotional response. response can be triggered by conditioning factors, which
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 419

Table 10
Experimental evidence: comparison between the group of treated and untreated.
This table shows the summary statistics for treated and untreated subjects in the experiment run at Northwestern Uni-
versity and t-tests for the differences (last column). The risk aversion measures are elicited as described in Table II. The
other variables are defined in the Internet Appendix, Section A4. The indicator for low risk investment is constructed from
the qualitative question, setting it equal to one if the person chose “An OK return, with good degree of safety of my prin-
cipal” or “Low returns, but no chance of losing my principal,” and zero otherwise. ∗ /∗ ∗ /∗ ∗ ∗ indicate statistical significance
at the 10%, 5%, and 1% levels, respectively.

Variable Obs. tot. Mean treated Mean non-treated Difference

Quantitative measure of risk aversion (risk premium) 207 3198 2526 672∗ ∗
Qualitative measure of risk aversion 210 2.54 2.41 0.13
Low risk investment 210 0.53 0.39 0.14∗
Male 206 0.39 0.34 0.05
Age 203 19.77 19.83 −0.06
Caucasian 206 0.41 0.40 0.01
Income (thousands of dollars) 210 111.68 120.96 −9.28

have little or nothing to do with the experience itself. As Given the nature of the video, which potentially dis-
Pavlov’s dog salivates when a bell rings, the fear response turbs some of the subjects, we had to give them the option
arises in the presence of stimuli associated to past trau- to skip the video at any moment. We dropped the observa-
matic events. This evidence suggests that a fear-based re- tions of the subjects (27) who decided to skip the video in
sponse can be triggered by fear stimuli in an unrelated do- the first minute of the five-minute presentation, since they
main. For example, Kinreich et al. (2011) show that watch- did not really experience much horror. This choice might
ing a horror movie stimulates the amygdala in a way con- underestimate the effect of the treatment, since those most
sistent with the arousal of fear. Yet, they do not provide ev- sensitive to the treatment dropped out.
idence that this experience can alter a risk aversion mea- Another possible concern is that, if a subject has al-
sure like ours, nor that it can alter it to the extent we ob- ready watched the video, its perceived effect would be dif-
serve after the financial crisis. This is what we try to test. ferent from the true effect. We therefore decided to drop
We chose a brief horrifying scene from a movie that those 13 subjects who declared to have already watched
was sufficiently recent to be really scary for undergradu- it.
ates accustomed to the scariest videogames (Psycho would To guarantee the reliability of the results, the experi-
not cut it), but sufficiently old to minimize the chance they ment was designed in such a way that the participants
had already seen it. We chose a five-minute excerpt from were not aware that the treatment was not identical for
the 2005 movie, Hostel, directed by Eli Roth, which is char- everyone. As measures of risk aversion, we use answers to
acterized by stark and graphic images and that shows a the very same questions that were used in the bank sur-
young man inhumanly tortured in a dark basement. The vey, in which we translated euros into dollars at a 1:1 ra-
movie won "Best Horror" at the Empire Awards in 2007. tio.
Our experiment was run at Northwestern University in As Table 10 shows, the random assignment assump-
March 2011 in three different sessions. A total number of tion cannot be rejected: none of the main personal char-
249 students took part. The participants were recruited acteristics and demographic information has been found
through an internal mailing list service that is normally to be statistically different between treatment and control
employed for experiments at Northwestern.6 A compensa- groups. Furthermore, around 60% of the participants were
tion of $5 was paid in cash to each subject taking part in female and the average age is 20, which is not surprising
the experiment, which in general takes around 10–15 min. given that the sample is composed of undergraduate stu-
All the participants were asked to complete a question- dents.
naire of approximately 40 questions. Its main scope is to When we look at the risk aversion measure, we find
construct some measures of risk aversion, as well as to that there is a large and statistically significant difference
provide other controls. To identify the effect of fear on in the quantitative measure of risk aversion. Among the
the subjects, we relied on a simple treatment and con- treated students, the risk premium they are willing to
trol framework. In particular, around half of the partici- pay for avoiding the risky lottery is 672 dollars (i.e., 27%)
pants were asked to watch a short video before completing higher. This holds true without controls and controlling for
the questionnaire. Since the subjects were randomly as- observables (Internet Appendix, Table A.6 columns 1 and
signed to watch the video, the idea is that the difference 2).
in risk aversion between the two groups should be com- In the qualitative measure we observe an increase, but
pletely driven by this difference in the treatment. this increase is not statistically significant at the conven-
tional level (p-value = 0.11). In part, this phenomenon is
due to the fact that students bunched their choices in the
two central values: 96% of the responses are either two or
6
The students can freely enroll on the mailing list and, after they three. Hence, the scale 1–4 is probably better reduced to a
have completed an introductory demographic survey, they receive peri-
odic communications on the experiments that are going on at the Uni-
dichotomous choice: low risk aversion (1 and 2) and high
versity. risk (3 and 4). When we look at the proportion of people
420 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421

3500 3256

3000
2435 2437
2500
Risk premium 2124
2000
Non treated
1500 Treated

1000

500

0
Dislike horror movies Like horror movies

0.7 0.65
Fracon choosing low risk investmnet

0.6 0.58

0.5
0.41 0.4
0.4
Non treated
0.3 Treated

0.2

0.1

0
Dislike horror movies Like horror movies

Fig. 4. Effect of fear on risk aversion. The figure shows the average risk aversion of subjects “Treated” with the horror movies or “Not treated,” for groups
of subjects that differ in how much they like horror movies. “Like horror movies” is calculated from a survey measure asking subjects whether they like
horror movies in a scale ranging from zero to 100 increasing in liking. Subjects who replied with a value of 20 (the median) or more are classified as liking
horror movies. “Dislike horror movies” is the group that reports less than 20 in liking of horror movies. Panel A shows the effect on the risk premium of
the lottery; Panel B presents the effect on the risky investment choice.

choosing the low risk option, this proportion increases by treatment (Panel A). This difference is statistically signifi-
13.5% (30% of the sample mean) among the treated group cant at the 1% level.
(Columns 5 and 6). This difference is large and statistically The second group is formed by those subjects who
significant at the 5% level. moderately like horror movies (liking indicator above 20).
In the second half of the sample, we asked people how Here the treatment has no effect (the risk premium goes
much they liked horror movies on a scale from zero to from 2435 to 2437) and this difference is not statistically
100. Roughly a third of the sample declared they do not significant.
like them at all (i.e., like = 0) and 50% report a value of We get a similar result when we look at the qualita-
liking below 20. In Fig. 4 we split the sample on this ba- tive measure of risk aversion, where we bunched the re-
sis. In the first group, there are students who do not like sponses into two groups. Among people who dislike hor-
horror movies (liking indicator below median). Their risk ror movies, the treatment effect increases the probability
premium rises from 2124 to 3256 dollars as a result of the of buying risky assets by almost 25%. Among those who
 L. Guiso et al. / Journal of Financial Economics 128 (2018) 403–421 421

moderately like horror movies, the increase is significantly Bonaparte, Y., Cooper, R., 2009. Costly Portfolio Adjustment. NBER Work-
smaller by 7%. ing Paper No. 15227.
Bordalo, P., Gennaioli, N., Shleifer, A., 2012. Salience theory of choice un-
These results seem to be inconsistent with the back- der risk. Q. J. Econ. 127 (3), 1243–1285.
ground risk hypothesis and suggest that fear is a poten- Byrnes, J., Miller, D., Schafer, W., 1999. Gender differences in risk taking:
tial (and understudied) mechanism that influences finan- a meta-analysis. Psychol. Bull. 125 (3), 367–383.
Caballero, R., Krishnamurthy, A., 2008. Collective risk management in a
cial decisions, whether it does so by increasing the curva- flight to quality episode. J. Finance 63 (5), 2195–2230.
ture of the utility function or the salience of negative out- Campbell, J., Cochrane, J., 1999. By force of habit: a consumption-based
comes (Bordalo et al., 2012). explanation of aggregate stock market behavior. J. Polit. Econ. 107 (2),
205–251.
Cohn, A., Engelmann, J., Fehr, E., Maréchal, M., 2015. Evidence for counter-
6. Conclusions cyclical risk aversion: an experiment with financial professionals. Am.
Econ. Rev. 105 (2), 860–885.
Croson, R., Gneezy, U., 2009. Gender differences in preferences. J. Econ.
In our view, the paper has two main contributions. The
Lit. 47 (2), 448–474.
first is methodological. Most papers use either naturally Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G., 2011.
occurring data or lab/field data, but not both. We think Individual risk attitudes: measurement, determinants, and behavioral
that well designed lab tests can be a useful complement consequences. J. Eur. Econ. Assoc. 9 (3), 522–550.
Donkers, B., Melenberg, B., Van Soest, A., 2001. Estimating risk attitudes
to the analysis of naturally occurring data, when we are using lotteries: a large sample approach. J. Risk Uncertain. 22 (2),
faced with very important questions that are impossible 165–195.
to answer with naturally occurring data. In particular, it is Friend, I., Blume, M., 1975. The demand for risky assets. Am. Econ. Rev. 65
(5), 900–922.
impossible to disentangle whether the surge in risk aver- Guiso, L., Paiella, M., 2006. The role of risk aversion in predicting indi-
sion we observe in the data is due to fear or background vidual behavior.. In: Chiappori, P., Gollier, C. (Eds.), Competitive Fail-
risk. If we were unable to reproduce that surge in the lab, ures in Insurance Markets: Theory and Policy Implications. MIT Press,
Boston, MA, pp. 213–250.
we could have ruled out the fear explanation. The fact we Guiso, L., Paiella, M., 2008. Risk aversion, wealth and background risk. J.
were able to does not prove that the cause of the surge in Eur. Econ. Assoc. 6 (6), 1109–1150.
the data is fear, but it makes it more plausible. Thus, tar- Hartog, J., Ferrer-i-Carbonell, A., Jonker, N., 2002. Linking measured risk
aversion to individual characteristics. Kyklos 55 (1), 3–26.
geted lab experiments can help in identifying effects diffi-
Heaton, J., Lucas, D., 20 0 0. Portfolio choice in the presence of background
cult to sort out in naturally occurring data. risk. Econ. J. 110 (460), 1–26.
The second contribution is to provide some evidence Holt, C., Laury, S., 2002. Risk aversion and incentive effects. Am. Econ. Rev.
92 (5), 1644–1655.
consistent with a fear-based explanation of the increase of
Kamstra, M., Kramer, L., Levi, M., 2003. Winter blues: a SAD stock market
risk aversion during the financial crisis. cycle. Am. Econ. Rev. 93 (1), 324–343.
A question we are unable to answer in this paper is Kandasamy, N., Hardy, B., Page, L., Schaffner, M., Graggaber, J., Powlson, A.,
how persistent such fear-induced change in risk aversion Fletcher, P., Gurnell, M., Coates, J., 2014. Cortisol shifts financial risk
preferences. Proce. Nat. Acad. Sci. 111 (9), 3608–3613.
is. The evidence of Malmendier and Nagel (2011), who find Kinreich, S., Intrator, N., Hendler, T., 2011. Functional cliques in the amyg-
a cohort effect of “Depression era babies” in the risk aver- dala and related brain networks driven by fear assessment acquired
sion measure of the Survey of Consumer Finances, suggests during movie viewing. Brain Connect. 1 (6), 484–495.
Knutson, B., Wimmer, E., Kuhnen, C., Winkielman, P., 2008. Nucleus ac-
it might be long-lasting. With our sample we are unable cumbens activation mediates the influence of reward cues on finan-
to answer whether fear provokes long-term consequences cial risk taking. NeuroReport 19 (5), 509–513.
because of the subsequent events in the eurozone, which Kramer, L., Weber, M., 2012. This is your portfolio on winter: seasonal
affective disorder and risk aversion in financial decision making. Soc.
made the 2008 shock not an isolated crisis. Psychol. Personal. Sci. 3 (2), 193–199.
Finally, our results raise an interesting question. If the Kuhnen, C., Knutson, B., 2005. The neural basis of financial risk taking.
behavior we document is typical and during severe down- Neuron 47 (5), 763–770.
Kuhnen, C., Knutson, B., 2011. The impact of affect on beliefs, preferences
turns investors are caught by fear and sell their risky as-
and financial decisions. J. Financial Quant. Anal. 46 (3), 605–626.
sets at the worst times, the effective return on equity in- Lerner, J., Keltner, D., 20 0 0. Beyond valence: toward a model of emo-
vestment is much lower. Can this feature explain—at least tion-specific influences on judgment and choice. Cognit. Emot. 14 (4),
473–494.
in part—the famous equity premium? Only future research
Lerner, J., Keltner, D., 2001. Fear, anger, and risk. J. Personal. Soc. Psychol.
would be able to tell. 81 (1), 146–159.
Loewenstein, G., 20 0 0. Emotions in economic theory and economic be-
References havior. Am. Econ. Rev. 90 (2), 426–432.
Loewenstein, G., Weber, E., Hsee, C., Welch, N., 2001. Risk as feelings. Psy-
Baker, M., Wurgler, J., 2007. Investor sentiment in the stock market. J. chol. Bull. 127 (2), 267–286.
Econ. Perspect. 21 (2), 129–151. Malmendier, U., Nagel, S., 2011. Depression babies: do macroeconomic ex-
Barberis, N., Huang, M., Santos, T., 2001. Prospect theory and asset prices. periences affect risk-taking? Q. J. Econ. 126 (1), 373–416.
Q. J. Econ. 116 (1), 1–53. Merton, R., 1969. Lifetime portfolio selection under uncertainty: the con-
Barsky, R., Juster, T., Kimball, M., Shapiro, M., 1997. Preference parameters tinuous-time case. Rev. Econ. Stat. 51 (3), 247–257.
and behavioral heterogeneity: an experimental approach in the health Pavlov, I., 1927. Conditioned Reflex: An Investigation of the Psychological
and retirement Study. Q. J. Econ. 112 (2), 537–579. Activity of the Cerebral Cortex. Oxford University Press, New York.
Bassi, A., Colacito, R., Fulghieri, P., 2013. O sole mio: an experimental anal- Weber, M., Weber, E., Nosić, A., 2013. Who takes risks when and why:
ysis of weather and risk attitudes in financial decisions. Rev. Financial determinants of changes in investor risk taking. Rev. Finance 17 (3),
Stud. 26 (7), 1824–1852. 847–883.
Bombardini, M., Trebbi, F., 2012. Risk aversion and expected utility theory:
an experiment with large and small stakes. J. Eur. Econ. Assoc. 10 (6),
1348–1399.