Ellsberg's paradox: why do the common preferences between bets on an urn of known and an urn of unknown...
decision/ellsberg
Problem
Ellsberg's paradox: why do the common preferences between bets on an urn of known and an urn of unknown composition violate the independence axiom of expected utility theory? Suppose that: Urn I contains 100 red and black balls, in a ratio entirely unknown to you. In Urn II, you confirm that there are exactly 50 red and 50 black balls. An observer who is ignorant of the state of your information about the urns wants to measure your subjective probabilities by asking you questions. 1. Which would you rather bet on, Red[I] or Black[I]? (That is, drawing a ball from Urn I, do you prefer a bet that wins if the ball is Red, or one that wins if the ball is Black?) 2. Which would you prefer to bet on, Red[II] or Black[II]? 3. Which would you prefer to bet on, Red[I] or Black[II]? 4. Which would you prefer to bet on, Black[I] or Black[II]? Suppose you say that in (1) and (2) you are indifferent. Most people say so. And suppose you also prefer Red[II] to Red[I], in (3), and Black[II] to Black[I], in (4). Again, most people have those preferences. The 'justification' (I'm not saying that there is any real justification!) is supposed to be that betting on the draw from the urn of unknown composition is 'riskier'. Now you are in trouble with the independence axiom. (ObPuzzle: demonstrate.) Daniel Ellsberg, _Quarterly Journal of Economics_ 75 (1961) 643-669. -- Jamie Dreier <pl436000@brownvm.brown.edu> [ Editor: the first paragraph is the editor's summary. See also decision/allais and decision/zeckhauser. ]
Show solution
The demonstration. Draw one ball from each urn, so that there are four
states of the world, written (colour from urn I, colour from urn II):
RR, RB, BR, BB. Each bet pays the same prize if it wins and nothing
otherwise. The four bets in question win in these states:
Red[I] RR RB
Black[I] BR BB
Red[II] RR BR
Black[II] RB BB
Compare f = Red[II] and g = Red[I]. They agree on the event
E = {RR, BB}: both win on RR and both lose on BB. They differ only
off E: f wins on BR, g wins on RB.
Now change both bets on E, in the same way: make them lose on RR and
win on BB. Off E nothing changes. f becomes the bet that wins on
{BR, BB}, which is Black[I]; g becomes the bet that wins on {RB, BB},
which is Black[II].
The independence axiom (Savage's "sure-thing principle", the axiom
that the Allais preferences also violate) says that a preference
between two acts cannot depend on what they give where they agree:
if f is preferred to g, then the modified f is preferred to the
modified g. So
Red[II] preferred to Red[I] implies Black[I] preferred to Black[II],
but the common preferences are Red[II] over Red[I] in (3) and
Black[II] over Black[I] in (4). That is the contradiction.
A shorter way to see that something is wrong: if you were maximizing
expected utility with subjective probabilities, indifference in (1)
would mean P(Red[I]) = P(Black[I]) = 1/2, and in (2)
P(Red[II]) = P(Black[II]) = 1/2. Preferring Red[II] to Red[I] then
means 1/2 > P(Red[I]), and preferring Black[II] to Black[I] means
1/2 > P(Black[I]), so P(Red[I]) + P(Black[I]) < 1, although one of the
two must happen. No probability measure fits these preferences, which
is why the observer in the story cannot "measure your subjective
probabilities" at all. The sure-thing argument above shows which
axiom carries the blame.
About question 3: as printed it compares Red[I] with Black[II], while
the preference stated is Red[II] over Red[I]. By (2) Red[II] and
Black[II] are equally good, so either reading gives the same argument.
The preferences are usually explained as aversion to ambiguity
(unknown probabilities), as distinct from risk (known probabilities).
Decision rules that allow it give up the sure-thing principle, e.g.
the maxmin expected utility of Gilboa and Schmeidler, which evaluates
a bet on urn I by its worst case over all possible compositions: each
of Red[I] and Black[I] is then worth a 0% chance, while Red[II] and
Black[II] are worth 50%.
References:
D. Ellsberg, "Risk, ambiguity, and the Savage axioms", Quarterly
Journal of Economics 75 (1961) 643-669.
L. J. Savage, The Foundations of Statistics (1954).
I. Gilboa and D. Schmeidler, "Maxmin expected utility with
non-unique prior", Journal of Mathematical Economics 18 (1989)
141-153.
-- Claude (rec.puzzles Oracle, 2026-11-02)
[ Editor: the material collected earlier follows. ]
[ Editor: the problem was posted, with Zeckhauser's problem, as one of
"two relatives of the Allais paradox" (see decision/allais). ]
[ Editor: further material on this puzzle, edited separately from other collected postings (batch 19); it may repeat some of the above. ]
[ Editor: problem as stated there: ]
Urn I contains 100 red and black balls in an unknown ratio; urn II
contains exactly 50 red and 50 black. Most people are indifferent
between betting on Red[I] or Black[I], and between Red[II] or
Black[II], yet prefer Red[II] to Red[I] and Black[II] to Black[I].
Show that these preferences violate the independence axiom of expected
utility theory (Ellsberg's paradox).
[ Editor: problem statement reconstructed from the posting. ]
ELLSBERG'S PARADOX
Suppose that:
Urn I contains 100 red and black balls, in a ratio entirely unknown to
you. In Urn II, you confirm that there are exactly 50 red and 50 black
balls.
An observer who is ignorant of the state of your information about the
urns wants to measure your subjective probabilities by asking you
questions.
1. Which would you rather bet on, Red[I] or Black[I]? (That is, drawing a
ball from Urn I, do you prefer a bet that wins if the ball is Red, or one
that wins if the ball is Black?)
2. Which would you prefer to bet on, Red[II] or Black[II]?
3. Which would you prefer to bet on, Red[I] or Black[II]?
4. Which would you prefer to bet on, Black[I] or Black[II]?
Suppose you say that in (1) and (2) you are indifferent. Most people say
so. And suppose you also prefer Red[II] to Red[I], in (3), and Black[II]
to Black[I], in (4). Again, most people have those preferences.
[ Editor: question 3 reads "Red[I] or Black[II]" in the original, but
the answer compares Red[II] with Red[I]; the intended question is
probably "Red[I] or Red[II]". ]
The 'justification' (I'm not saying that there is any real justification!)
is supposed to be that betting on the draw from the urn of unknown
composition is 'riskier'.
Now you are in trouble with the independence axiom. (ObPuzzle: demonstrate.)
Daniel Ellsberg, _Quarterly Journal of Economics_ 75 (1961) 643-669.
-- Jamie Dreier <pl436000@brownvm.brown.edu>