I learned about this paradox recently through an interview question.
The paradox comes from the following game. more here at wiki:
The casino flips a coin until it gets heads. The number of tails is squared and payed to the player. How should they price this game?
Having never seen this question, I proceeded to take the expectation. That is the weighted average of the payouts. Well the answer is infinity, this can be seen intuitively from the fact that the payout grows exponentially at the same rate as the diminishing probability of the event happening.
The paradox is that no one will place huge sums of money to play this game, even though theoretically, they should – since it has an unlimited expected payoff.
The variance of this game is also infinite which stems from the fact that the payouts are finite but the expectation is infinite. E(X) = (X-E(X))^2
There are several resolutions:
A utility function approach which says we need to discount the payoffs by the utility function. In other words, the first million dollars brings me more utility then each next million. However, if this is not bounded, you can simply keep increasing the speed of payouts given any unbounded utility function.
Another approach suggests that people discount the probability of negligible events. However, lottery ticket sales undermine this argument, since those seem to be low probability events that people pay too much for. However, this counter argument neglects to mention that certain events are discounted completely if they are below a certain probability. As an example, the chance that all the air will decide to stay in one side of my tire, nobody will pay any amount of money for that event. Same goes for the law of gravity to stop applying, there is some negligible probability for that event, but no matter how large the payoff no one will by that ticket.
An experimental approach, suggests that you should pay around 10 dollars. Having played the game 2048 times.
Another approach suggests that you would need the casino to be good for an infinite sum of cash and since they aren’t no one would place that money.
A combination of utility and the previous reason, gives an answer that is close to the experimental result. Suppose you would be willing to play and suppose it takes a minute to flip a coin. You have a magic genie that has guaranteed that the coin will flip tails until you say heads after which it will be heads. How long will you play?
Most likely, you will stop after you are the richest person in the world, but that only will take an hour. After that you apparently have more money than the value of earth by three orders of magnitude. If you discount to that best case scenario, you get no paradox at all, in fact the most you would then pay is 1/4*60 = 15 dollars. If you understand, that casino can’t possibly guarantee more than a billion in winnings, that brings the value down to 29.8973529 ~ 30, which says it’s closer 7.5 dollars. If you are playing against another person you can’t expect more than a million dollar payout so you shouldn’t bet more than 1/4*20, unless you are playing against Harry Kakavas.
One last way to think about this problem. What is the expected value of the game itself. The answer to this is 1. That is the total number of flips will be 2 but you will only expect one tail. In which case, you expect 1 dollar. So perhaps you should simply pay 1 dollar because that’s how the game will play out.This is called the median value of the game and is in fact what many people bet.
The fact is, the more money you have the more valuable your time becomes, and it makes less and less sense to keep playing the game. So you will tell your genie to stop, because now that you have earned all this money you want to go out and spend it.