Quote:
Originally Posted by ChrisV
I hate you guys for making me think this hard, but here's a disproof by counterexample of the idea that mixed strategies can always be varied without penalty.
Imagine a game between players A and B, where A has either been dealt a 1 or a 0, with 50% probability each.
In round 1 of the game, if A has a 1, he may choose to turn it face up or keep it hidden. If he chooses to turn it face up, B has to pay him 1 coin.
In round 2 of the game, B may wager that A has a 0. If he is correct, A has to pay him 10 coins. If he is incorrect, he has to pay A 20 coins.
It's clear that A cannot turn up his 1 100% of the time. If he does, B can always wager in round 2 whenever A does not turn his card up, always winning 10 coins off A. What A has to do is ensure that B cannot exploit him in round 2. To break even in that round, he needs B to be wrong once for every two times he is correct. Therefore, he needs to have 1 in 3 odds of having a 1. He can achieve this by mixing his strategy such that he only turns over his 1 50% of the time. This means that heading into round 2, he will have a 0 50% of the time, a hidden 1 25% of the time and an exposed 1 25% of the time. This prevents B from having a profitable wager in round 2. A's expectation with this strategy is 0.25 coins per iteration, gaining 1 coin on the 25% of occasions he turns his 1 over.
So here we have a game where A is playing a mixed strategy. Yet it is not the case that it doesn't matter what his frequencies are in round 1. If he unilaterally changes his strategy such that he never turns over his 1, his expectation in the game is now zero.
EDIT: I noticed after I posted this that plexiq gave the same solution in the Poker Theory forum thread.
I think you made the assumption that B will always guess 0 on the second round. This can't be right, though, because then A would exploit this by hiding 1s more often. OTOH, if B never guesses 0 on the second round, then A would show 1s more often. So B will mix between sometimes guessing 0 and sometimes not.
Assuming B faces no penalty for not guessing, B will guess 0 on the second round exactly 5% of the time. If A has a 1, then A's EV will be 1 for showing in the first round, and 1 for hiding it (5% * 20 coins = 1). A is now indifferent to showing or hiding when he has a 1. And equilibrium is reached. (A made B indifferent to guessing 0, and B made A indifferent to showing.)
Now fix B's strategy. B guesses a 0 exactly 5% of the time on the second round... If B does this, then A will have the following EVs
EV(show 1) = 1
EV(hide 1) = 1
EV(hide 0) = -0.5 [5% * -10 coins]
This yields an overall EV of (0.5 + -0.25)=0.25, like you said.
However, since B's strategy is now fixed, A can choose to show or hide the 1 however often he likes. A does not need to stick to the strategy of hiding 50% of the time when B is not trying to exploit A.
On the flip side, if we fix A's strategy so that A hides 50% of the time, then B may choose to guess 0 at any frequency B wants.. Because EV(guess 0) = EV(no guess) = 0.
BUT... If A and B continue trying to maximally exploit each-other, they would remain in equilibrium only by mixing their strategies at the optimal frequencies.
Last edited by pocketzeroes; 08-04-2017 at 04:59 AM.