From Overcoming Bias:
Imagine that you have a large pool of cases, where in each case you
weakly suspect some sort of villainous stink. But you have a limited
investigative resource, which you can only apply to one case, to sniff
for stick there.
For example, you might have one reporter, who you could assign for
one month to investigate the finances of any one member of Congress. Or
you might have one undercover actor, whom you could assign to offer a
bribe to one member of the police force of a particular city. Or you
might assign a pretty actress to meet with a Hollywood producer, to
check for harassment.
Imagine further that you are willing to invite the world to weigh in,
to advise you on where to apply your investigative resource. You are
willing to say, “Hey world, which of these cases looks stinky to you?”
If this is you, then I offer you villain markets.
In a villain market, some investigative resource will be applied at
random to one case out of a set of cases. It will report back a verdict,
which in the simplest case will be “stinky” or “not stinky”. And before
that case is selected for investigation, we will invite everyone to bet
anonymously on the chances of stickiness in each case. That is, anyone
can bet on the probability that the verdict of case C will be found
stinky, given that case C is selected for investigation. So if
you have reason to suspect a particular member of Congress, a particular
police officer, or a particular Hollywood producer, you might expect to
gain by anonymously betting against them.
Imagine that we were sure to investigate case C87, and that the
market chance of C87 being found stinky was 2%, but that you believed
C87’s stinkiness chances were more like 5%. In this situation, you might
expect to profit from paying $3 for the asset “Pays $100 if C87 found
stinky”. After your bet, the new market chance might be 4%, reflecting
the information you had provided the market via your bet.
Now since we are not sure to investigate case C87, what you’d really
do is give up “Pays $3 if C87 investigated” for “Pays $100 if C87
investigated and found stinky.” And you could obtain the asset “Pays $3
if C87 investigated” by paying $3 cash and getting a version of this
“Pays $3 if C investigated” investigation asset for every possible case C.
So you could reuse the same $3 to weigh in on the chances of
stinkiness in every possible case from the set of possible cases. And
not only could you bet for and against particular cases, but you could
bet on whole categories of cases. For example, you might bet on the
average stickiness of men, or people older than 60, or people born in
Virginia.
To get people to bet on all possible cases C, there needs to be at
least some chance of picking every case C in the set of possible cases.
But these choice chances do not need to be equal, and they can even
depend on the market prices. The random process that picks a case to
investigate could set the choice chance to be a strongly increasing
function of the market stinkiness chance of each case. As a result, the
overall chance of the investigation finding stink could be far above the
average market chance across the cases C, and it might even be close to
the maximum stinkiness chance.
So far I’ve describe a simple version of villain markets, but many
variations are possible....MORE
In news completely unrelated, except for the stank (and recency):
China lifts ban on stinky cheese imports—CNN
And thanks but I'm good without the
Casu marzu.
