View Single Post
  #3  
Old 11-28-2007, 07:41 PM
pzhon pzhon is offline
Senior Member
 
Join Date: Mar 2004
Posts: 4,515
Default Re: An original math problem

[ QUOTE ]

1) What is the optimal strategy? (not hard)


[/ QUOTE ]
<font color="white">Bet on whatever is favored to show up. When the deck is even, what you bet on doesn't matter.</font>

[ QUOTE ]
3) What is the EV/value of this game?

[/ QUOTE ]
<font color="white">The amount you win (ignoring the $7 fee) is $k if you guess correctly k times while the deck is even. When the deck is unbalanced by n cards, you will win precisely $n by the time the deck becomes even again.

So, the expected win is $0.50 times the expected number of times the deck is even, or $0.50 times the sum of the probabilities that the deck is even after 0,2,4,...,50 cards. The probability the deck is even after 2i cards is (2iCi)((52-2i)C(26-i))/(52C26). The sum is not a smooth number, so I don't think this sum simplifies. The sum of the probabilities is about 8.08, so the expected gain within the game is $4.04. You lose $2.96 by paying $7 to play.</font>

[ QUOTE ]
2) What are the chances you'll only guess 26 right, i.e.,
lose $7 (playing optimally)?


[/ QUOTE ]
<font color="white">I've seen this part before. Without loss of generality, guess one color whenever the deck is even. Then you break even after the fee only if the cards seen so far is never in favor of the color you are guessing. That's a Catalan-type problem, and the number of such shuffles is the 26th Catalan number, (52C26)/27, so the probability is 1/27.</font>

[ QUOTE ]
4) Let n be the number of cards for each color (n=26 in the
original game). What is a good approximation for the
number of correct guesses above n/2 for large n? For which
values of n is the $7 fee too high for the game to be +EV?


[/ QUOTE ]
<font color="white">Asymptotically, (2kCk) ~ 4^k/sqrt(pi k). So, we can approximate (2iCi)(2n-2iCn-i)/(2nCn) by sqrt(pi n)/(sqrt(pi i)sqrt(pi n-i)) ~ 1/sqrt(pi n) 1/sqrt((i/n)(1-i/n)). The point of that last manipulation is that the sum of f(i/n)/n as i ranges from 0 to n-1 is about the integral from 0 to 1 of f(x) dx, so this sum is about sqrt(n)/sqrt(pi) times the integral from x=0 to x=1 of 1/sqrt(x(1-x)) dx, or sqrt(n)/sqrt(pi) * pi ~ sqrt(n pi). The expected value of this game is $0.50 times that, or about sqrt(n pi)/2 dollars, which is high by about 48 cents for n=26. It's high by 47 cents for n=1000.

The value with 71 cards of each color is $6.98. With 71 or fewer, it is not worth paying $7. The value with 72 cards of each color is $7.03. With any more, the game is worth playing.
</font>
Reply With Quote