A resolution of the ‘exchange paradox’ (Q4359960)

The ``exchange paradox'' may be described as follows: You are presented with two sealed envelopes, which you know contain amounts of money \(X\) and \(2X\), where \(X\) is unknown. You and another player are each given one of the envelopes, at random. You each look inside your envelope. You find, let us say, ten dollars in your envelope. You know that the other envelope contains either 5 or 20. Since the choice was random, you assign them each probability \(1/2\). Then you compute that the expected amount in the other envelope is \((1/2)5+ (1/2)20= 12.5\). This is more than you have, so you ask the other player if she would like to switch. She has made the same calculation, come to the same conclusion, so she is also eager to make the exchange. This paper is one in a long line of attempts to explain the paradox away as a failure to regard explicitly the Bayesian prior distribution that the player places on the quantity \(X\). It begins by correcting an elementary computational error which completely undermined an earlier work (by other authors) and then goes on to discuss briefly the ``optimal strategies'' when the player places various prior distributions on \(X\) -- in particular, the uniform distribution on a finite interval and the Pareto distribution. In some of these cases, though, the fundamental paradox remains -- the fact that an optimal strategy seems to be to make the exchange, no matter what is actually in the envelope -- a point which is passed over in silence by the author.