Adv Behav Econ pdf

138
S T A R M E R
Conventional theory can claim a success here: a one-parameter extension to
EU can offer significantly improved predictive power for a large body of data
generated mainly from triangle experiments. If we want to predict behavior over
simple choices like this, we know a good deal about how to improve on EU.
Notwithstanding this success, it is important to note that there is a wide range of
evidence that conventional theories stand little chance of digesting. For example,
there is considerable evidence revealing systematic failures of monotonicity and
transitivity in risky choice experiments. Some of this evidence is reviewed in
Starmer (2000).
5.2. Evidence from the Field
I have heard some economists argue that they would take more notice of non-EU
models if they could be shown cases where they help to explain real-world phe-
nomena of practical interest to economics. It is a fair point, but proponents of
nonexpected utility theory can muster some strong responses. Let me illustrate
this by way of a couple of examples.
The standard theory of insurance based on EU has some implications that have
long been regarded as highly implausible. For example, a risk-averse expected-
utility maximizer will not buy full insurance in the presence of positive marginal
loading (see Mossin 1968). This implication, Karl Borch (1974) suggests, is
“against all observation.” More recently, Wakker, Thaler, and Tversky (1997)
have made a similar point in relation to “probabilistic insurance.” Think of prob-
abilistic insurance as a policy with some fixed probability 
q
that a claim will not
be paid in the event of an insured loss. Wakker, Thaler, and Tversky show that an
expected-utility maximizer willing to pay a premium 
c
for full insurance against
some risk should be willing to pay a premium approximately equal to the actuar-
ially adjusted premium (1
2
q
)
?
c
for probabilistic insurance. Survey evidence,
however, shows that people are extremely averse to probabilistic insurance and
their willingness to pay for it is much less than standard theory allows.
If expected utility can’t explain insurance behavior, can nonexpected-utility
theory do any better? Part of the answer is provided by Segal and Spivak (1990),
who show that a number of implications of EU for insurance and asset demand
that are widely recognized to be counterintuitive have a common origin. They
arise because, with any smooth (i.e., differentiable) utility function, EU implies
that agents will be approximately risk neutral for small risks (since the utility
function will be almost linear). This theoretical property is at odds with peoples’
actual risk attitudes as revealed through their reactions to probabilistic insurance
and so on: people demand a much greater reduction in premium than the actuari-
ally fair adjustment for accepting a small positive risk of claim nonpayment.
Segal and Spivak go on to show that the counterintuitive implications of EU carry
through to nonexpected-utility theories which have similar smoothness properties.
This captures a large number of alternatives to EU and, in fact, only a single type of

---