ON THE ROBUSTNESS OF HIGHER ORDER RISK PREFERENCES

• Author: Cary Deck,Harris Schlesinger
• DOI: http://doi.org/10.1111/jori.12217
• Date: 01 June 2018
• Published date: 01 June 2018

ON THE ROBUSTNESS OF HIGHER ORDER

RISK PREFERENCES

Cary Deck

Harris Schlesinger

ABSTRACT

Economists have begun to recognize the role that higher order risk

preferences play in a variety of settings. As such, several experiments

have documented the degree of prudence, temperance, and, to a lesser

extent, edginess and bentness that laboratory subjects exhibit. More recently,

researchers have argued that higher order risk preferences generally

conform to mixed risk-averse and mixed risk-loving patterns that arise

from a preference for disaggregating or aggregating harms, respectively.

This article examines the robustness of this pattern in three ways. First, it

attempts to directly replicate previous results with compound lotteries over

monetary outcomes. Second, it compares behavior in compound lotteries

with behavior in reduced-form lotteries. And third, it evaluates choices over

monetary and nonmonetary risks. While previous results are replicated for

compound lotteries over monetary outcomes and aggregate behavior with

reduced-form lotteries has a similar pattern, individuals clearly treat

compound and reduced-form lotteries differently. Further, behavior differs

between monetary and nonmonetary outcomes.

INTRODUCTION

Notions of risk have longbeen central to discussions of economic behavior, but to date

relatively little attention has been paid to the role of higher (than second) order risk

attitudes. One exception is the relevance of prudence, a third-order risk property, in

precautionary savings (see Leland, 1968; Sandmo, 1970; Dr

eze and Modigliani, 1972;

Kimball, 1990; Lee, 2017). Recently, there has been a flurry of behavioral interest in

measuring higher order risk attitudes. This effort largely stems from the theoretical

work of Eeckhoudt and Schlesinger (2006) and Eeckhoudt, Schlesinger, and Tsetlin

†

Due to his untimely passing, Harris Schlesinger was unable to see this project through to

completion. The article is dedicated to his memory, but he should not be held accountable for

its errors. We thank Sebastian Ebert, Timo Heinrich, and Gijs van de Kuilen for helpful

feedback.

Cary Deck is at the University of Alabama and Chapman University, but was at the University

of Arkansas when this research was conducted. He can be contacted at cdeck@cba.ua.edu.

Harris Schlesinger was at the University of Alabama.

© 2017 The Journal of Risk and Insurance. Vol. 85, No. 2, 313–333 (2018).

DOI: 10.1111/jori.12217

313

(2009), which developed a method for classifying higher order risk attitudes based on

binary comparisonsthat involved apportioning lower order risks.In the laboratory, the

most common finding is that people tend to be risk averse, prudent, and to a lesser

degree temperate (see Deck and Schlesinger, 2010; Maier and R€

uger, 2012; Ebert and

Wiesen, 2011, 2014; Noussair,Trautman, and van de Kuilen, 2014; Baillon, Schlesinger,

and van de Kuilen, 2016).

1

Deck and Schlesinger (2014) go further,reporting evidence

of choices consistent with either mixed risk aversion or mixed risk loving. The term

mixed risk aversion, introduced by Caball

e and Pomansky (1995), applies to an

individual who dislikes an increase in risk of any degree, while the term mixed risk

lover, introducedby Crainich, Eeckhoudt, and Trannoy (2013),describes an individual

who dislikes increased risk for odd ordersbut likes increased risk in even orders. That

the two types have similar preferences for odd orders but not even may help explain

why the experimental evidence for prudence is stronger than that for temperance.

As researchers are exploring the theoretical and behavioral properties of higher order

risk preferences, it is important to understand the robustness of these preferences.

This article attempts to do so in three ways. First, it offers a direct replication of some

of the tasks used in Deck and Schlesinger (2014). Second, it examines the degree to

which the presentation format of the lottery impacts observed higher order risk

attitudes. Specifically, we compare higher order risk apportionment tasks where

choices are presented as compound lotteries with tasks where the choice is presented

in reduced form. Harrison, Martinez-Correa, and Swarthout (2015) demonstrate that

(second-order) risk-taking behavior varies based on whether or not compound or

reduced-form lotteries are used. This design choice is potentially even more

important for higher order risk attitudes where the study instrument is designed as a

choice between aggregating and disaggregating harms. In fact, Deck and Schlesinger

(2014, pp. 1921–1922) state “(Compound lottery) presentation admittedly also

facilitates viewing the problem as ‘combining good with bad’ or ‘combining good

with good,’ rather than presenting the lotteries in a reduced form, which might

obfuscate this interpretation.” While most studies of higher order risk uses

compound lotteries, Maier and R€

uger (2012) present subjects with reduced-form

lotteries, but do not directly compare the two formats.

2

To our knowledge, the only

other article that directly compares presentation formats for higher order risk

problems is Haering, Heinrich, and Mayrhofer (2015).

3

The results of their

experiments suggest that subjects choose the prudent and temperate options less

frequently when lotteries are presented in reduced form. Further, the reduction in

temperate choices is sufficiently strong for the authors to conclude that with the

reduced-form presentation there is no evidence that subjects are generally temperate.

1

Deck and Schlesinger (2010) and Baillon, Schlesinger, and van de Kuilen (2016) find a majority

of respondents were intemperate.

2

Maier and R€

uger (2012) find strong relationships between risk aversion and temperance. Their

article also compares higher order risks in the domain of gains and losses.

3

Haering, Heinrich, and Mayrhofer (2015) also test for effects of stake size and geographic

locations. They report behavior that is largely consistent with Deck and Schlesinger (2014) for

subjects in the United States, Germany, and China as well as with the use of higher stakes.

314 THE JOURNAL OF RISK AND INSURANCE