use the objectively specified probabilities. To this end, we consider gamble (10, 4), which
is the reference gamble in the second part of the experiment, and confront it with gamble
(7, 6). We show that using the above restrictions alongside the estimated parameter of loss
aversion leads to a preference reversal across the two frames of choice.
We start by evaluating the utilities of the two gambles in part 1 of the experiment. In this
part, no endowment is given to the subjects and it is not immediately clear which gamble
should be used as the reference point. Two natural candidates are the degenerate gamble
paying the average expected value of the gambles in our experiment (which is $7) in all
states and the degenerate gamble that pays $0 in all states. We carry the exercise using
$7 as the reference point.
22
Under the neutral frame, the two gambles have utility given
by the following:
U ((10, 4)|7) = 0.5u(10|7) + 0.5u(4|7) = 0.5[10 + (10 − 7)] + 0.5[4 + 3.4(4 − 7)] = 3.4
U ((7, 6)|7) = 0.5u(7|7) + 0.5u(6|7) = 0.5[7 + (7 − 7)] + 0.5[6 + 3.4(6 − 7)] = 4.8
Thus, the agent described by the KR model will choose the gamble (7, 6) over (10, 4) in
part 1 of the experiment. We now turn to part 2, where the gamble (10, 4) serves as the
agent’s endowment and is therefore used as the reference point. The evaluations of the two
gambles are as follows:
U ((10, 4)|(10, 4)) = 0.25u(4|4) + 0.25u(10|10) + 0.25u(4|10) + 0.25u(10|4) = 3.4
U ((7, 6)|(10, 4)) = 0.25u(7|10) + 0.25u(6|10) + 0.25u(7|4) + 0.25u(6|4) = 1.8
Thus, in the second part of the experiment the agent would choose (10, 4) over (7, 6)
22
A similar exercise shows that if $0 is used as a reference the model predicts two choice reversals (in
this case, for the gambles (13, 2) and (14, 2)).
27
exhibiting a preference reversal. Under the same assumptions, gamble (8, 6) is preferred
to (10, 4) in part 1 and is indifferent to it in part 2. Assuming ties are broken evenly,
and according to the loss aversion model of KR the average subject in our experiment will
therefore exhibit a status quo bias index equal to 1.5, contradicting our finding of no bias
in the R-R treatment.
23
Appendix B
Predictions of Bewley’s Model
According to Bewley’s inertia hypothesis, the presence of ambiguity is a necessary condi-
tion for status quo biased behavior. Therefore it correctly predicts the absence of status
quo bias in the R-R treatment. However this model is not well suited to explain the ab-
sence of ambiguity in the A-A treatment together with the presence of ambiguity in the
asymmetric treatments A-R and R-A. Intuitively, if inertia stems from ambiguity, it stands
to reason that the more ambiguity there is, the more likely it is to observe the bias. In this
appendix we show this intuition formally under a fairly mild assumption on the set of pri-
ors and using the axiomatic development of Bewley’s approach taken by Ortoleva (2010).
24
The Model
We briefly summarize the model developed in Ortoleva (2010).
25
Using the standard
Anscombe-Aumann framework, there is a finite set S of possible states of the world and a
set X of consequences, which is assumed to be a convex and compact subset of a Banach
space. Let ∆(X) stand for the set of all Borel probability measures (lotteries) on X. By
23
Using a lower degree of loss aversion of λ = 3, which has often been used as a benchmark (Koszegi
and Rabin 2006, 2007), the model would still predict the same 1.5 points of reversal.
24
We adopt the model by Ortoleva (2010) rather than the original formulation by Bewley (1986) because
the former adopts a complete preference relation in the absence of an endowment. This allows to derive
precise predictions for choices under the neutral frame.
25
We
specifically
make
use
of
Theorem
1
of
an
earlier
version
of
the
paper
given
at
http://gtcenter.org/Archive/Conf07/Downloads/Conf/Ortoleva498.pdf.
28
F we denote the set of all acts, that is, the set of all functions f : S → ∆(X) where f (s)
denotes the consequence (lottery) of act f in state s ∈ S.
Within this set up and under the axioms imposed in Ortoleva (2010) the agent may
be thought of as if he has a utility function u : X → R and a set Π of possible priors
over the states of the world. In addition, he has a single prior ρ which he uses when
choosing absent an endowment. According to the model, we have that ρ ∈ Π. Facing acts
absent an endowment, the agent evaluates them on the basis of their subjective expected
utilities (by using the single prior ρ and utility function u), just like a standard Savagean
agent. However, when the agent is endowed with an act f he does not act as an expected
utility maximizer anymore. Rather, he becomes uncertainty averse and may be described
as a constrained utility maximizer. His maximization takes place over a constrained set of
alternatives. The constrained set comprises only the options that carry higher utility than
the endowment according to all priors in the set Π. We follow Ortoleva (2010) and denote
the constrained set by D
Π,u
(f ). Formally it takes the form:
D
Π,u
(f ) :=
g ∈ F |
s∈S
π(s)E
g(s)
(u) >
s∈S
π(s)E
f (s)
(u), for all π ∈ Π
where E is the expectation operator. The agent’s choices can be summarized formally as
follows. When facing a choice set A without an endowment he maximizes:
arg max
g∈A
s∈S
ρ(s)E
g(s)
(u)
When facing a choice set A with an endowment f ∈ A (denoted by (A, f )) the agent
29