Bayes' Theorem: Exchange Paradox

Let’s discuss a famous unresolved paradox that comes up in day-to-day probability theory. It is important that you do the probability calculations before reporting your reasonings!

Two envelopes are handed out to you and your friend. One envelope contains double the money as the other.

  1. You open the envelope and see $X. Is it beneficial for you to switch the envelope? If yes, what is the expected gain?
  2. Your opponent opens the envelope and sees $Y. Is it beneficial for them to switch the envelope? If yes, what is the expected gain?
  3. Let’s say you both swap envelopes, but do NOT see the amount yet. Is it beneficial to swap again?
  4. Let’s say after swapping, you both see the amount in your envelops. Now, would it make sense to conduct the swap again?
  5. If you answered yes to 1 and 2, how can a game exist where both parties have an advantage in swapping?
  6. If you answered yes to 3, when do you actually stop swapping? Does swapping keep increasing the value of the envelop indefinitely?
5 Likes

I think the paradox stems from the two different ways we calculate the probabilities, and while one is in line with our intuition, the other isn’t.

One way to think about this is by stating the random variable being sampled is the total value of both envelopes combined, say the random variable is T.
Now, one of the envelopes will contain \frac{T}{3} and the other will contain \frac{2T}{3}.
Now we can apply conditional expectation on the gain from the exchange, G.

E[G] = P(X = \frac{T}{3}) \cdot \text{Gain}(X=\frac{T}{3}) + P(X = \frac{2T}{3}) \cdot \text{Gain}(X=\frac{2T}{3})

Which simplifies to:

E[G] = \frac{1}{2} \cdot \left(\frac{T}{3}\right) + \frac{1}{2} \cdot \left(-\frac{T}{3}\right) = 0

This aligns with our intuition as well, as in a symmetric setup like this, exchanging shouldn’t bring any advantage.

The paradox stems from a different calculation, where we can calculate the gain from the exchange as:
E[G] = P(\text{X = Bigger Value})*(\text{Gain(X is bigger)}) + P(\text{X = Smaller Value})*(\text{Gain(X is smaller)})
E[G] = \frac{1}{2} \cdot \left(\frac{-X}{2}\right) + \frac{1}{2} \cdot \left(X\right) = \frac{X}{4}
SInce getting a positive gain regardless of the value of X, information of the amount we have should have no affect on our decision to switch since it will be positive anyways. Even if we don’t ‘peek’ this leads to the paradox of both players having advantage in a symmetric game. Which seems to go against the intuition that a symmetric setting like this would offer an advantage to switching.

From what I could gather (I could be wrong) the problem stems from:

a) Assigning equal probabilities to X being the bigger value and X being the smaller value, the implicit assumption here is that any X is equally likely to be the bigger value, which is problematic as the sample space is the real line and assigning equal probabilities to all numbers isn’t a valid pdf. Which means this problem is ill defined for this method since we don’t know the distribution of X which will change the chances of a particular X being the bigger value.

b) This is a guess, but there do exist other paradoxes where the way we define the sampling gives different probability values, for eg. the problem of finding the probability a randomly chosen chord is bigger than the side of the inscribed equilateral triangle, the answer changes depending on how you ‘select’ these random chords.
Similarly in this example, I think the first calculation with the T being the random variable reflects the section in which you take a certain amount of money and then put it into two envelopes such that one is double of other (which is what is described in the question).
While for the second calculation you first put X amount in an envelope and then flip a coin to decide whether to put double the amount of half the amount in the other envelope. If the player knows he has the initial envelope then switching makes sense, and this doesn’t lead to infinite switching as he knows the expected value of the unknown envelope is \frac{5X}{4} which is greater than X.

This was my attempt, please do correct me if I’m wrong.

I think the paradox arises when we consider our initial value to be X and the value when swapped to beequal to 2x or x/2 with equal probability(in which case the expected gain value of the swap would be x/4).But that wouldn’t be the correct mathematical representation of the puzzle as we would be considering 2different situations in that case: So if we took x=100 then we are taking into account the situation where we have[100, 50] and also the situation where we have [100, 200]. These are 2 separate cases and shouldn’t be merged into one. Like, if you win in the [100, 50] game, your gain is 50 and if you win in the [100, 200] game,your gain is 100.

Let’s assume that the player gets $K in the envelope. There’s a 0.5 probability that the other envelope contains $K/2 and 0.5 that it contains $2K. So, the expected gain would be 0.25K. However, this is just the expected gain and NOT an accurate measure of risk involved. There is an equal probability of both outcomes so the expected gain alone cannot be used to conclude whether it’s a beneficial trade-off. However, based solely on the expected gain, the opponent also has the same gain. So it’d make sense that both believe it is better to switch.

After switching, the expected gain remains the same (percentage-wise). So, it’s again better to switch. Once both see the amounts of the exchanged envelopes, one of them would have gained 100% over the initial value and the other would’ve lost 50%. Obviously, the person who profited wouldn’t want to switch and his opponent would. So, there is no possibility of a switch happening after the envelopes are opened post-swapping.

This actually shouldn’t be happening and doesn’t happen because the probabilities are not uniformly distributed. What I mean is that the probability of the other envelope containing a larger amount goes on decreasing as the value in your envelope increases. So, for example, if you get 10$ in the envelope, let the probability of the other envelope containing double the amount (20$) be p1. And let it be p2 for 100$ in your envelope. In practice, p1>p2. And vice versa for the probability of the other envelope containing half the amount in your envelope. So, the game actually doesn’t have equal expected gains for both contestants. However, I don’t know how we can calculate the actual expected gains as my approach is purely intuition based.

There wouldn’t be any stopping if we go this way as we’re essentially assuming a uniform distribution of probabilities. But this doesn’t mean that we’re inflating the values in the envelope indefinitely as we already know the value in the first envelope. So, we’re basically saying that the other envelope contains double the initial envelope and when we actually swap and get the second envelope, we say that the first one has double the amount and hence halvjng the value in the second envelope (the one we have as of now). But this doesn’t seem to be the correct inference so please correct me if I’m wrong.

So what i am thinking first is we have two scenarios suppose i have ₹X then the other person could have ₹2X or ₹X/2 so in that case

E[G]=₹X/4 so this implies benefit in switching

But if we see here the total amount in play is different in one case it is ₹3X and in other ₹3X/2

And now suppose we start from having total amount as ₹3X then if we have ₹X then other has ₹2X and if we have ₹2X then other has ₹X which leads to

E(G) =0 so there arises a paradox

Please correct me wherever I am Wrong

So according to me the answer should be No as the game by intuition is symmetric but .