Very preliminary draft do not cite without author s permission.

Risk Preferences and (Behavioral) Consequences –

Effects of Changes in Risk Attitude on Individuals’ Health-related and Financial Decision Making

Verena Jaeger ^*

Previous research reveals that an individual’s risk preferences have a significant impact on his or her decision making under risk. However, previous research reveals that these preferences are not constant, but vary over time. As risk preferences are so important for decision making, it is crucial to understand if and to what extent individuals incorporate shifts in risk aversion into their behavior and decision making.

Our analysis uses the German Socio Economic Panel (GSOEP), a representative longitudinal sample dataset of the German population that includes rich information on individuals’ behavior and decision making, such as insurance purchases, savings behavior, stock holdings, as well as frequency of sport exercises and alcohol and smoking consumption. In addition, the survey includes information on individuals’ self-reported risk attitude. We combine changes in risk attitude with information on risky behaviors and decision making. Our results indicate that individuals who showed an increase in risk aversion from one period of observation to another are more likely to purchase term life policies and are also more likely to put more money into their emergency funds.

Keywords: Variable risk preferences, self-reported risk attitude, demand for insurance, savings behavior, health-related decision making, self-protection and self-insurance

JEL classification: D14, D81, G220, I130

Very preliminary draft – do not cite without author’s permission.

*

Munich Risk and Insurance Center, Munich School of Management, Ludwig-Maximilians-Universitaet, Schackstr. 4,

80539 Munich, Email: [email protected], Phone: +49 89 21803883

1. Motivation

Previous research has shown that risk preferences have a significant impact on individuals’

decision making under risk. Individuals’ attitudes towards risk have been shown to be important, particularly when it comes to financial decisions, such as insurance purchases, allocation of funds between savings and consumption, or investing in risky assets (see, for instance, Markowitz (1952), Mossin (1968), or Cohen and Einav (2007)). Furthermore, other studies (e.g., Viscusi (1990, 1991)) lend evidence that health and lifestyle related objectives are additional domains that are associated with individuals’ risk perception. However, risk preferences are not constant but change over the course individuals’ lifetime. As risk preferences have such significant impact on decisions and behavior, it is crucial to understand in which domains and to what the extent individuals alter their decision making once they alter their risk attitude. In particular, financial and health-related decisions have a long-lasting impact on an individual’s resources over his or her lifetime; thus consequences of changing risk preferences have important implications for public policy and social security programs as well as for the well-being of businesses and individuals.

In the face of risk, individuals often consider risk management to mitigate the probability and

severity of undesired events. In this respect, a classical distinction has been made between self-

insurance (loss reduction), which reduces the severity of a loss, and self-protection (loss

prevention), which reduces the probability of a loss (Ehrlich and Becker (1972)). Whereas the

first activity is a substitute for insurance, the latter can be either a substitute or a complement for

insurance coverage. Furthermore, more risk-averse agents invest more in self-insurance but not

necessarily in self-protection (Dionne and Eeckhoudt, 1985; Briys and Schlesinger, 1990; Jullien

et al., 1999). Individuals’ financial and health-related decisions can be classified as being actions

of self-insurance, self-protection, or a combination of both, and prior literature lends evidence of

how risk preferences are associated with these domains. In this respect, Raviv (1979), Doherty

and Schlesinger (1983), and Chesney and Loubergé (1986) find that changes in the willingness

to insure depend on relative risk aversion and the wealth elasticity of insurable risky wealth. In

addition, Cleeton and Zellner (1993) investigate the relationship between insurance demand and

income when there is a change in the degree of risk aversion. Insurance demand in the

aftermath of natural disasters has been studied by others--e.g. Browne and Hoyt (2000), who

find that recent flood events increases the demand for insurance. Weinstein (1989) suggests

that feelings of worry increase following the personal experience of a traumatic event, which

increases individuals’ attempt to protect themselves from future harm.

There is a rich body of literature which examines individuals’ risk preferences and decision making under risk. Allen et al. (2005) provide evidence that more risk-averse individuals show higher intentions-turnover links when it comes to health-related changes in behavior.

Furthermore, Viscusi (1991) and Liu and Hsieh (1995) find that risk perception influences smoking behavior. Determinants of healthy dietary habits were studied by Johansson et al.

(1999). They find that not only gender, socio-economic status and education but also the extent to which an individual pays attention to other domains of risky decisions impact dietary habits.

Moreover, Weber et al. (2002) observe that individuals’ degree of risk aversion differs among financial and health/safety related decisions.

To our knowledge, our paper is the first longitudinal study investigating changes in individual risk attitudes and their domain-specific consequences. We track changes in risk attitude over a multi- year observation period and are particularly interested in analyzing effects of increases in risk aversion on individuals’ financial and health-related decisions. In this respect, we investigate impacts on insurance purchases, savings behavior, and retirement planning as well as health- related measures, such as body mass index (BMI), smoking and alcohol consumption, frequency of sport exercise and the number of doctor consultations during a specific period of time. In addition, we empirically analyze whether an increase in risk aversion leads to the previously found increase in self-insurance but not necessarily self-protection.

Prior literature primarily utilizes three methods to identify risk preferences in empirical studies.

First, many experiments and some datasets elicit risk preferences by asking for hypothetical

choices in lotteries (see, for instance, Donkers et al. (2001) and Hartog et al. (2002)). Bartsy et

al. (1997) and Kimball (2008) provide a detailed explanation of how to elicit preferences from

these choices. Second, many studies rely on self-reported risk preferences (see, for instance,

Dorn and Hubermann (2005), Dorn and Hubermann (2010), and Dohmen et al. (2011)). To elicit

an individual’s risk attitude, they use survey questions where respondents have to self-assess

their risk attitude; for example, the survey respondents are asked to rate their willingness to take

risks on a scale from 0 to 10. Third, risk preferences are estimated with a variety of different

instrumental variables. In the field of agricultural economics, risk attitudes of farmers are

estimated by technology choice and time allocation in studies by Bar-Shira et al. (1987) and

Antle (1988). Chetty (2006) uses labor supply and for insurance, Cicchetti and Dubin (1994) use

the decision to insure certain risks, while Cohen and Einav (2005) estimate risk preferences from

the deductible choice.

Our study utilizes the second identification method: individuals self-report their risk attitude over multiple years which we track in our dataset. Dohmen at al. (2011) confirm the behavioral validity of this method by matching the self-reported risk preferences in the German Socio Economic Panel to elicited preferences from an experimental set up with paid lottery choices. Kapteyn (2011) finds that subjective risk measures can even outperform more sophisticated choices between income streams as proposed by Barsky et al. (1997). Accordingly, we are confident that our results will be meaningful for the study of risk preferences in general, regardless of the way in which preferences are measured.

2. Data and Methodology

For our analysis we use the German Socio Economic Panel (GSOEP). The GSOEP is a representative panel survey dataset of the adult population. ² The survey has been conducted on a yearly basis since 1984 and asks individuals for a wide range of personal and household information, including income, lifestyle, and health status, and for their attitudes on assorted topics, including political and social issues. In addition, individuals are asked to give a global assessment of their willingness to take risks. Individuals can indicate on a scale from 0 to 10 how they see themselves with respect to risk taking, with ‘0’ indicating ‘not at all willing to take risks’ and ‘10’ meaning ‘very willing to take risks’. ³ In our dataset the average willingness to take risks during our observation period is 4.25.

The GSOEP initially surveyed individuals’ risk attitude in 2004 and 2006. Starting in 2008, risk attitude has been available on a yearly basis, and we are able to include all survey waves up to and including 2012. ⁴ As the years 2005 and 2006 do not include information on risk attitude, we start our analysis in 2008.

2

For a detailed description, see e.g.Wagner et al. (1993) or Schupp and Wagner (2002).

3

The exact wording (translated from the German questionnaire) is as follows: ‚How do you see yourself:

Are you generally a person who is fully prepared to take risks or do you try to avoid taking risks? Please tick a box on the scale where the value 0 means: ‘not at all willing to take risks‘ and the value 10 means:

‘very willing to take risks’.

4

The 2013 wave of the GSOEP will become available during the first quarter of 2015. Once the data is

available, we will include the year of 2013 in our analysis.

Changes in risk attitude, which is our predominant control variable, are calculated as follows:

Δ 𝑟𝑖𝑠𝑘 𝑎𝑡𝑡𝑖𝑡𝑢𝑑𝑒 _𝑖,𝑡 = 𝑟𝑖𝑠𝑘 𝑎𝑡𝑡𝑖𝑡𝑢𝑑𝑒 _𝑖,𝑡 − 𝑟𝑖𝑠𝑘 𝑎𝑡𝑡𝑖𝑡𝑢𝑑𝑒 _{𝑖,𝑡−1}

with t=2009, …, 2012 indicating the years of observation and i=0,…,N indicating the number of individuals. Figure 1 displays the distribution of changes.

Figure 1: Distribution of changes in risk attitude during observation period of 2008-2012

Since we are interested in consequences of changes in risk attitude on financial and health- related decision making, we include the following dependent variables in our analysis: purchase of (term) life insurance policy, amount of monthly savings towards emergency funds (in Euro), purchase of supplemental health insurance policy, ⁵ daily smoking consumption, alcohol consumption, number of doctor consultations during the last three months, and frequency of sport exercise. Summary statistics for all dependent variables are displayed in Table 1.

5

In Germany, health insurance is predominantly provided through a statutory system and only a minority

of people qualifies for and chooses health insurance coverage provided through the private insurance

market. In general, the statutory system provides basic treatments and additional coverage can be

obtained by purchasing supplemental health insurance policies in the private health insurance market. In

this respect, holding a supplemental health insurance policy reveals information on an individual’s risk

management or the level of precaution that has been taken.

Variable Description Years Mean Std. Dev. Min Max life insurance

purchase (1) if purchase 2004-2012 0.06 0.24 0 1

monthly savings amount of monthly

savings in Euro 2004-2012 337.51 700.74 0 3,000

supplemental health

insurance (1) if purchase 2004-2008, 2010-

2012 0.15 0.36 0 1

smoking

daily (aggregated) consumption of cigarettes, cigars and

pipes

2004, 2006, 2008,

2010, 2012 3.72 7.85 0 71

alcohol

(0) never, (1) rarely, (2) sometimes,

(3) frequently

2006, 2008, 2010 0.83 0.52 0 3

doctor consultations # consultations last 3

months 2004-2012 2.58 3.83 0 27

frequency sport exercise

(0) never, (1) rarely, (2) monthly, (3) weekly, (4) daily

2005, 2007, 2008,

2009, 2011 1.49 1.36 0 4

Table 1: Summary statistics for dependent variables included in our analysis

Because not all of the above listed variables are available for the entire observation period of 2008-2012, we restrict our analysis to a reduced number of observation years in the case of smoking and alcohol consumption and sport participation. After cleaning for missing values, our balanced dataset consists of 8,809 individuals which provides us with a large number of observations from which we can draw statistical inferences.

We also include control variables for those factors that have been found to be important by prior literature. These first include age, gender, body height, marital status and the region where individuals live, as well as financial and income related indicators, information about family, health and employment status and individuals’ school degrees. We capture (real) income by including individuals’ monthly household income (after taxes), and we also include (real) income that they receive from interest and dividend payments. ⁶ In order to account for individual savings behavior, we include a dummy variable indicating whether or not individuals have a savings account. We also include another dummy variable that indicates homeownership. ⁷

Marital status is included by differentiating between single, married, divorced, and widowed individuals; we use being single as the omitted category. We further capture the number of children by counting the number of children the individual receives child allowances for. These include all children below age 18, all job-seeking children below age 21 and all children who are

6

We use 2012 numbers in our analysis and account for inflation by referring to http://de.inflation.eu/inflationsraten/hvpi-inflation.aspx

7

The homeownership variable is one either if the individual owns the house or flat he or she lives in or if

the individual is a landlord.

in education below age 25. We believe that the number of children that qualify for child allowances is a better measure of the financial impact of having children than is the overall number of children or the number of children living in the household. In addition, we include a dummy variable to indicate whether the individual provides care to an elderly or sick family member.

To include individuals’ level of educational attainment, we split our dataset into three educational levels. Low level of school includes individuals who did not graduate from school or graduated with the lowest certificate of secondary education (Hauptschulabschluss). ⁸ Medium level of school consists of individuals with Realschulabschluss, while high level of school refers to individuals with Abitur. Abitur is the highest school leaving certificate that allows enrollment into a university in Germany. It is comparable to A-levels in the U.K. and the baccalauréat in France.

Realschulabschluss and Hauptschulabschluss do not qualify for university enrollment. The main difference between the lowest and the medium school degree in Germany is related to the fact that most white-collar positions require a medium school degree, whereas certain blue-collar workers only need to have the lowest school degree. The omitted category in our analysis is the lowest level of educational attainment (Hauptschulabschluss).

Differentiating between blue-collar employees, white-collar employees, civil servants and self- employed individuals incorporates individuals’ occupational status. We also control for trainees and retirees, as well as for individuals that have no job, either because they are currently seeking work, which we refer to as unemployed individuals, or because they intentionally have no job in the wage economy (e.g. housewives). The latter we refer to as individuals with “no job.”

The omitted category in our analysis is blue-collar workers.

In addition, we are able to include several health-related indicators, such as individual’s overall health status, which is measured by an integer variable taking values between 1 (very good health status) and 5 (poor health status). We further capture how often the individual has been sick for more than 6 weeks during a year of observation with our disability variable. To include the type of health insurance, we use a dummy variable to capture all individuals that have full private health insurance coverage and a dummy variable for those who own a supplemental health insurance policy. Table 2 displays summary statistics of all control variables used in our analyses.

8

Note that in Germany, there are three types of degrees of secondary education. Hauptschulabschluss is

obtained after 9-10 years of schooling (depending on the federal state) and enables individual to take up

jobs such as handymen or sales personnel. Realschulabschluss is obtained after 10 years of schooling

and enables individuals to take up qualified jobs in a local bank such as office clerks or tellers. Abitur is

obtained after 12 years of schooling and allows the individual to enroll in a university.

Variable Description Mean Std. Dev. Min Max

risk attitude

scale from 0 to 10

(0): no risk tolerance and (10): high willingness to take risk

4.25 2.22 0 10

inc_risktaking* (1): if individual increased willingness

to take risk 0.36 0.48 0 1

decr_risktaking* (1): if individual decreased willingness

to take risk 0.37 0.48 0 1

abs_decr_risktaking absolute downward shift on 0-10-scale 0.82 1.38 0 11

male* (1): if imale 0.47 0.49 0 1

age age in years 53.93 15.57 21 102

health

scale from 1 to 5

(1): very good health status and (5):

poor health status

2.72 0.92 1 5

bodyheight height in cm 170.81 9.92 74.5 205

bodyweight weight in kg 86.92 24.30 21.5 210

private health insurance* (1): if full private health insurance

coverage 0.15 0.35 0 1

nojob* (1): if intentionally no job 0.06 0.23 0 1

trainee* (1): if trainee 0.01 0.13 0 1

unemployed* (1): if seeking job 0.05 0.20 0 1

retired* (1): if retired 0.31 0.46 0 1

bluecollar* (1): if blue-collar worker 0.14 0.35 0 1

whitecollar* (1): if white-collar worker 0.30 0.46 0 1

selfemployed* (1): if self-employed 0.57 0.23 0 1

civilservant* (1): if civil servant 0.04 0.21 0 1

east_germany* (1): if individual lives in the Eastern

Germany 0.29 0.45 0 1

married* (1): if martial status married 0.68 0.46 0 1

single* (1): if martial status single 0.15 0.36 0 1

divorced* (1): if martial status divorced 0.08 0.27 0 1

widowed* (1): if martial status widowed 0.07 0.26 0 1

aftertaxincome (hh) household income after taxes (Euro) 2,934 2,131.10 0 200,000 supportpersoncare* (1): if individual provides care to family

member 0.06 0.24 0 1

savingsaccount* (1): if individual has savings account 0.73 0.44 0 1

disability being sick more than 6 weeks 0.03 0.20 0 2

lowlevelschool* (1): individual has low level school

leaving certificate 0.36 0.48 0 1

mediumlevelschool* (1): individual has medium level school

leaving certificate 0.35 0.47 0 1

highlevelschool* (1): individual has high level school

leaving certificate 0.27 0.44 0 1

propertyownership* (1): individual owns house or flat 0.58 0.49 0 1 number_children number of children qualifying for child

allowance 0.56 0.92 0 10

interestdividendincome income from interest and dividend

payments in EURO 1,484 10,971.97 0 1,038,222

Table 2: Summary statistics of all control variables during the observation period of 2008-2012

(* denotes dummy variables)

As mentioned above, we use the 2008-2012 waves of the GSOEP for our analysis as they started to survey risk attitude on an annual basis in 2008. Our dataset is a balanced dataset and consists of 9,543 individuals older than 18 years (or who turned 18 in the year they were first surveyed and included in the dataset). After dropping individuals with missing data, we are left with a panel dataset of 8,809 individuals yielding 35,236 observations. ⁹

In our preliminary analysis we test consequences of changes in risk attitudes with respect to (term) life insurance purchasing decisions and savings behavior. In this respect, we are first interested if increases in risk aversion (i.e., downward shifts on the 0-10 scale) are associated with a greater propensity to purchase life insurance coverage. The GSOEP surveys whether or not an individual owns a life insurance policy on an annual basis and provides a respective dummy variable that indicates the ownership. Based on that, we create a dummy variable to display whether or not an individual has bought a life insurance policy from one year to another.

To control for potential endogeneity problems, we consider insurance purchases (and later savings patterns) in the consequent year of observation. ¹⁰ As a result, we use lagged changes in risk attitude as control variables, which are denoted by a preceding ‘L.’

Our first model investigates general changes in risk attitude by including dummy variables for individuals that increased and decreased their willingness to take risk, respectively. We fit the following probit model with clustered standard errors. ¹¹

𝑃(𝑙𝑖𝑓𝑒𝑖𝑛𝑠𝑢𝑟𝑎𝑛𝑐𝑒 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒| 𝑋) = 𝐺(𝛽 ₀ + 𝛽 ₁ 𝐿. 𝑑𝑒𝑐𝑟 _{𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔} + 𝛽 ₂ 𝐿. 𝑖𝑛𝑐𝑟 _{𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔} + 𝜙 ∗ 𝑋 _{𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠} ) (1)

9

Since we calculate changes in individuals’ willingness to take risk from one year of observation to the following year of observation, we end up with a panel consisting of four years (2009-2012). This is attributed to the fact that the ‘first change’ in willingness to take risk (Δ 𝑟𝑖𝑠𝑘 𝑎𝑡𝑡𝑖𝑡𝑢𝑑𝑒) is computable for t=2009.

10

To ensure that the increase in risk aversion precedes the event of interest, e.g. purchasing a life insurance policy, we use changes in risk attitude from, e.g. year 2008 to year 2009 as control variable for insurance purchasing decisions made between year 2009 and year 2010. We do so in order to exclude cases of (potential) reversed causalities in the sense that individuals become more risk averse when holding a life insurance policy. We are aware that we might lose observations where individuals both altered their risk attitude and bought a life insurance policy before they were surveyed. The average survey month in our data set, however, is 3.23 (i.e. beginning of March) and none of the individuals was surveyed later than May. In this respect, we argue that our methodology is the most valid approach for dealing with causality issues.

11

Clustered standard errors account for possible correlations within a cluster and asymptotically equal

unclustered standard errors. Since we cannot exclude that clustered standard errors are not necessary,

we include them to err on the side of caution.

Since we are particularly interested in investigating increases in risk aversion, i.e. decreases in individuals’ willingness to take risks, we fit a second probit model with clustered standard errors that analyzes the impact and extent of downward shifts on the 0-to-10 scale on the likelihood of life insurance purchases.

𝑃(𝑙𝑖𝑓𝑒𝑖𝑛𝑠𝑢𝑟𝑎𝑛𝑐𝑒 𝑝𝑢𝑟𝑐ℎ𝑎𝑠𝑒| 𝑋) = 𝐺(𝛽 ₀ + 𝛽 ₁ 𝐿. 𝑎𝑏𝑠 𝑑𝑒𝑐𝑟_𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔 + 𝜙 ∗ 𝑋 _{𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠} ) (2)

Secondly, we investigate whether changes in individuals’ willingness to take risks are associated with modified savings patterns. The GSOEP surveys savings behavior by asking individuals how much money they put into their emergency funds monthly. We first analyze increases and decreases in willingness to take risk and run the following OLS model, which includes individual and year fixed effects as well as clustered standard errors.

𝑚𝑜𝑛𝑡ℎ𝑙𝑦𝑠𝑎𝑣𝑖𝑛𝑔𝑠 𝑖,𝑡 = 𝛽 0 + 𝛽 1 𝐿. 𝑑𝑒𝑐𝑟 𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔

+ 𝛽 2 𝐿. 𝑖𝑛𝑐𝑟 𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔

+ 𝜙 ∗ 𝑋 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝜀 𝑖,𝑡 (3)

with t=2009, …, 2012 indicating the years of observation and i=0,…,N indicating the number of individuals.

Correspondingly, we investigate the impact and extent of increases in risk aversion and fit the following OLS model, which also includes individual and year fixed effects and clustered standard errors.

𝑚𝑜𝑛𝑡ℎ𝑙𝑦𝑠𝑎𝑣𝑖𝑛𝑔𝑠 𝑖,𝑡 = 𝛽 0 + 𝛽 1 𝐿. 𝑎𝑏𝑠 𝑑𝑒𝑐𝑟_𝑟𝑖𝑠𝑘𝑡𝑎𝑘𝑖𝑛𝑔

+ 𝜙 ∗ 𝑋 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 + 𝜀 𝑖,𝑡 (4)

with t=2009, …, 2012 indicating the years of observation and i=0,…,N indicating the number of individuals.

3. Preliminary Results

Table 3 and Table 4 report the results of our estimations. Model (1) and model (2) summarize coefficient estimates of the regressions on life insurance purchase, with model (1) being the model that analyzes the impact of general increases and decreases in willingness to take risk.

Model (2) controls for the degree that risk aversion has been increased.

VARIABLES

Model (1) Probit (marginal effects)

dependent variable:

Lifeinsurance Purchase

Model (2) Probit (marginal effects)

dependent variable:

Lifeinsurance Purchase

L.decr_risktaking 0.0039 [0.0032]

L.incr_risktaking -0.0003 [0.0033]

L.abs_decr_risktaking 0.0026*** [0.0009]

age -0.0011* [0.0006] -0.0011* [0.0006]

age2 0.0001 [0.0001] 0.0001 [0.0001]

male 0.0002 [0.0035] 0.0003 [0.0035]

nojob -0.0047 [0.0063] -0.0047 [0.0063]

trainee -0.0292*** [0.0105] -0.0292*** [0.0105]

whitecollar -0.0081* [0.0042] -0.0080* [0.0042]

civilservant -0.0038 [0.0076] -0.0038 [0.0076]

unemployed -0.0077 [0.0069] -0.0080 [0.0069]

retired -0.0033 [0.0055] -0.0033 [0.0055]

selfemployed -0.0008 [0.0064] -0.0009 [0.0064]

mediumlevelschool -0.0092*** [0.0031] -0.0091*** [0.0031]

highlevelschool -0.0099*** [0.0038] -0.0097** [0.0038]

married -0.0007 [0.0045] -0.0007 [0.0045]

widowed 0.0161** [0.0067] 0.0160** [0.0067]

divorced 0.0091 [0.0058] 0.0089 [0.0058]

number_children 0.0001 [0.0017] 0.0001 [0.0017]

supportpersoncare 0.0046 [0.0052] 0.0047 [0.0052]

health 0.0028* [0.0015] 0.0028* [0.0015]

bodyheight 0.0000 [0.0002] 0.0000 [0.0002]

disability 0.0048 [0.0058] 0.0048 [0.0058]

east_germany 0.0049* [0.0029] 0.0050* [0.0029]

ln_real_aftertaxincome 0.0013 [0.0029] 0.0012 [0.0029]

ln_real_interestdividendincome 0.0086*** [0.0007] 0.0086*** [0.0007]

savingsaccount -0.0193*** [0.0035] -0.0191*** [0.0035]

privatehealthinsurance -0.0095** [0.0046] -0.0093** [0.0046]

supphealth -0.0046 [0.0038] -0.0045 [0.0038]

propertyownership -0.0125*** [0.0029] -0.0125*** [0.0029]

year2010 0.0016 [0.0037] 0.0014 [0.0037]

year2011 0.0099*** [0.0035] 0.0105*** [0.0035]

year2012 -0.0030 [0.0036] -0.0024 [0.0036]

Observations 35,236 35,236

Pseudo R-squared 0.0168 0.0172

Standard errors in brackets, *** p<0.01, ** p<0.05, * p<0.1

Table 3: Regression results for life insurance purchases. Reference categories: low level school

degree, blue-collar worker, single, year 2009.

With respect to life insurance purchases, the findings provided by model (1) indicate that individuals who decreased their willingness to take risk from one period of observation to another (i.e. became more risk averse) are more likely to have purchased a life insurance policy in the subsequent period. The coefficient estimate is 0.0039. Correspondingly, we find that individuals who increased their willingness to take risk (i.e. became less risk averse) are less likely to have purchased a policy in the subsequent period, and the respective coefficient estimate is -0.0003. The signs of our coefficients are as expected but are not statistically significant at the tested levels. Our second model is used to investigate the impact and the extent of increases in risk aversion on life insurance purchases. The coefficient of (lagged) absolute downward shifts on the 0-10 scale is 0.0026, which indicates that individuals that showed larger decreases in their willingness to take risk are more likely to purchase a life insurance policy. The coefficient estimate is statistically significant at the 1% level. Similar to findings provided by prior literature, we find that control variables for individuals’ education, age and wealth are associated with the demand for life insurance.

We next investigate how changes in risk attitude are associated with individuals’ savings behavior. Similarly to the case of life insurance purchases, we first investigate general increases and decreases in risk attitude (model (3)) and focus on decreases in willingness to take risk afterwards (model (4)). Respective regression results of model (3) and model (4) are summarized in table (4).

Results obtained with model (3) report that individuals who decreased their willingness to take

risk, i.e. became more risk averse, put a higher amount of money towards their emergency

funds. The reverse holds for individuals who became more risk loving. Our coefficient estimates

are 0.0039 and -0.0003 but are both not statistically significant at the tested levels. With respect

to model (4), we find that the more individuals decreased their willingness to take risks, the

higher are the monthly savings towards emergency funds. The reported coefficient is 0.0026 and

is significant at the 1% level. We further find that wealth, occupational status and education as

well as age and health status are statistically significant associated with individuals’ savings

behavior. This is consistent with prior literature (see, e.g., Modigliani (1986)).

VARIABLES

Model (3)

Individual Fixed Effects with Clustered Standard Errors -

dependent variable:

monthly savings

Model (4)

Individual Fixed Effects with Clustered Standard Errors -

dependent variable:

monthly savings

L.decr_risktaking 3.1225 [5.7326]

L.incr_risktaking -5.6538 [6.0203]

L.abs_decr_risktaking 3.9737* [2.0707]

age2 -0.1437** [0.0599] -0.1434** [0.0600]

nojob 1.4144 [18.1971] 1.5850 [18.1912]

trainee 24.6680 [39.3029] 24.9201 [39.2759]

whitecollar 17.1935 [10.9489] 17.3004 [10.9552]

civilservant -25.5314 [38.9899] -25.1252 [38.9770]

unemployed 6.3510 [12.4891] 6.2068 [12.4864]

retired -21.2356 [18.8533] -21.1120 [18.8391]

selfemployed 58.2349 [46.0219] 58.0658 [46.0185]

mediumlevelschool 0.4466 [76.8252] 3.5365 [77.4713]

highlevelschool -23.5909 [14.7814] -21.4723 [15.0011]

married -25.0262 [28.8515] -24.4242 [28.7681]

widowed -30.8659 [99.0149] -30.7929 [99.2830]

divorced -17.8754 [30.6649] -17.3197 [30.6721]

number_children -19.4488* [11.6494] -19.5133* [11.6525]

supportpersoncare -0.3496 [9.4322] -0.3417 [9.4230]

health -4.5324 [3.7688] -4.5244 [3.7697]

disability -7.2747 [7.6533] -7.2346 [7.6572]

east_germany -24.0467 [67.8512] -23.3081 [67.8220]

ln_real_aftertaxincome 427.1734*** [42.1292] 427.2124*** [42.1424]

ln_real_interestdividendincome 7.2162*** [2.0499] 7.2096*** [2.0512]

savingsaccount 10.7589 [10.7840] 10.7460 [10.8113]

privatehealthinsurance 175.3919** [88.1725] 175.0588** [88.1313]

supphealth -14.0238 [12.1746] -14.2492 [12.1932]

propertyownership -4.8613 [20.0135] -4.8157 [20.0193]

year2010 24.6731*** [8.6982] 24.6237*** [8.6814]

year2011 64.2461*** [14.7273] 64.1208*** [14.7178]

year2012 93.7771*** [21.5304] 93.9137*** [21.5037]

Constant -2,592.8348*** [407.8265] -2,599.7992*** [408.7162]

Observations 35,236 35,236

Number of individuals 8,809 8,809

adjusted R-squared 0.0582 0.0582

Standard errors in brackets, *** p<0.01, ** p<0.05, * p<0.1

Table 4: Regression results for savings behavior. Reference categories: low level school

degree, blue-collar worker, single, year 2009.

4. Conclusion

Our paper investigates changes in individual risk attitudes and their domain-specific consequences. We find empirical evidence that individuals, who became more risk averse, have a greater propensity to purchase life insurance coverage and put a higher monthly amount of money to their emergency funds. In addition, our results indicate that the extent to which individuals decreased their willingness to take risks have an economically significant impact.

Consequently, risk attitudes and particularly changes in risk attitudes have an material impact on

individuals decision making under risk. As financial and health-related decisions have a long

lasting impact on individuals’ resources over his or her lifetime, consequences of changing risk

preferences for public policy and social security programs.

5. References

Allen, D. G., Weeks, K. P., & Moffitt, K. R. (2005). Turnover intentions and voluntary turnover:

the moderating roles of self-monitoring, locus of control, proactive personality, and risk aversion. Journal of Applied Psychology, 90(5), 980-990.

Antle, J. M. (1989). Nonstructural risk attitude estimation. American Journal of Agricultural Economics, 71(3), 774-784.

Bar-Shira, Z., Just, R. E., & Zilberman, D. (1997). Estimation of farmers' risk attitude: an econometric approach. Agricultural Economics, 17(2), 211-222.

Barsky, R. B., Juster, F. T., Kimball, M. S., & Shapiro, M. D. (1997). Preference parameters and behavioral heterogeneity: An experimental approach in the health and retirement study.

The Quarterly Journal of Economics, 112(2), 537-579.

Briys, E., & Schlesinger, H. (1990). Risk aversion and the propensities for self-insurance and self-protection. Southern Economic Journal, 458-467.

Browne, M. J., & Hoyt, R. E. (2000). The demand for flood insurance: empirical evidence.

Journal of risk and uncertainty, 20(3), 291-306.

Chesney, M., & Loubergé, H. (1986). Risk aversion and the composition of wealth in the demand for full insurance coverage. Swiss Journal of Economics and Statistics (SJES), 122(III), 359-370.

Chetty, R., & Szeidl, A. (2007). Consumption commitments and risk preferences. The Quarterly Journal of Economics, 122(2), 831-877.

Cleeton, D. L., & Zellner, B. B. (1993). Income, risk aversion, and the demand for insurance.

Southern Economic Journal, 146-156.

Cicchetti, C. J., & Dubin, J. A. (1994). A microeconometric analysis of risk aversion and the decision to self-insure. Journal of Political Economy, 169-186.

Cohen, A., & Einav, L. (2007). Estimating risk preferences from deductible choice. The American Economic Re-view, 97(3), 745-788.

Dionne, G., & Eeckhoudt, L. (1985). Self-insurance, self-protection and increased risk aversion.

Economics Letters, 17(1), 39-42.

Doherty, N. A., & Schlesinger, H. (1983). Optimal insurance in incomplete markets. The Journal of Political Economy, 1045-1054.

Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., & Wagner, G. G. (2011). Individual risk attitudes: Measurement, determinants, and behavioral consequences. Journal of the European Economic Associa-tion, 9(3), 522-550.

Donkers, B., Melenberg, B., & Van Soest, A. (2001). Estimating risk attitudes using lotteries: A

large sample approach. Journal of Risk and Uncertainty, 22(2), 165-195.

Dorn, D., & Huberman, G. (2005). Talk and action: What individual investors say and what they do. Review of Finance, 9, 437-481.

Dorn, D., & Huberman, G. (2010). Preferred risk habitat of individual investors. Journal of Financial Economics, 97, 155-173.

Ehrlich, I., & Becker, G. S. (1972). Market insurance, self-insurance, and self-protection. The Journal of Political Economy, 623-648.

Hartog, J., Ferrer‐i‐Carbonell, A., & Jonker, N. (2002). Linking measured risk aversion to individual characteristics. Kyklos, 55(1), 3-26.

Johansson, L., Thelle, D. S., Solvoll, K., Bjørneboe, G. E. A., & Drevon, C. A. (1999). Healthy dietary habits in relation to social determinants and lifestyle factors. British Journal of Nutrition, 81(03), 211-220.

Jullien, B., Salanie, B., & Salanie, F. (1999). Should more risk-averse agents exert more effort?.

The Geneva Papers on Risk and Insurance Theory, 24(1), 19-28.

Kapteyn, A., & Teppa, F. (2011). Subjective measures of risk aversion, fixed costs, and portfolio choice. Journal of Economic Psychology, 32(4), 564-580.

Liu, J. T., & Hsieh, C. R. (1995). Risk perception and smoking behavior: Empirical evidence from Taiwan. Journal of Risk and Uncertainty, 11(2), 139-157.

Markowitz, H. (1952). Portfolio Selection. The Journal of Finance, 7(1), 77-91.

Modigliani, F. (1986). Life cycle, individual thrift, and the wealth of nations. The American Economic Review, 297-313.

Mossin, J. (1968). Aspects of rational insurance purchasing. Journal of Political Economy, 76(4), 553-568.

Raviv, A. (1979). The design of an optimal insurance policy. The American Economic Review, 84-96.

Schupp, J., & Wagner, G. G. (2002). Maintenance of and innovation in long-term panel studies:

The case of the German Socio-Economic Panel (GSOEP). Allgemeines Statistisches Archiv 86 (2), 163-175.

Viscusi, W. K. (1990). Do smokers underestimate risks?. Journal of political Economy, 1253- 1269.

Viscusi, W. K. (1991). Age variations in risk perceptions and smoking decisions. The Review of Economics and Statistics, 577-588.

Wagner, G. G., Göbel, J., Krause, P., Pischner, R., & Sieber, I. (2008). Das Sozio-

oekonomische Panel (SOEP): Multidisziplinäres Haushaltspanel und Kohortenstudie für

Deutschland–Eine Einführung (für neue Datennutzer) mit einem Ausblick (für erfahrene

Anwender). AStA Wirtschafts-und Sozialstatistisches Archiv, 2(4), 301-328.

Weber, E. U., Blais, A. R., & Betz, N. E. (2002). A domain‐specific risk‐attitude scale: Measuring risk perceptions and risk behaviors. Journal of Behavioral Decision Making, 15(4), 263- 290.

Weinstein, N. D. (1989). Optimistic biases about personal risks. Science, 246(4935), 1232-1233.