Problem 22.8 Medium Free

Three Cards of Unknown Offset

You have three face-down cards bearing the values nn, n+1n+1, n+2n+2 in some unknown order, where nn itself is unknown to you. You turn over one card at a time, in a fixed order, revealing only its numerical value. After seeing a card’s value you may either stay (collecting that value as your payout) or discard it and reveal the next card; you must stay after the third card is revealed if you have not already stopped. Using the optimal stopping strategy, the expected payout equals n+cn+c for a constant cc. Find cc.

Work it out, then compare with the answer.

SolutionShow the worked solution

Key idea. Since nn is unknown, a card’s face value tells you nothing by itself. Only differences between revealed cards carry information. So we assume the order of the cards is uniformly random, and we restrict attention to strategies that depend only on these differences. Then every payout has the form n+Xn+X, where X∈{0,1,2}X\in\{0,1,2\} is the offset of the card you keep. We maximise E[X]\mathbb{E}[X] over the 3!=63!=6 equally likely orders of the offsets 0,1,20,1,2.

Stopping at once. At the first card you have no information. Stopping gives E[X]=1\mathbb{E}[X]=1.

Decision at the second card. Suppose the first card is aa and the second is bb, and write d=b−a∈{±1,±2}d=b-a\in\{\pm1,\pm2\}. Each value of dd is consistent with some of the orders below. In each case, compare the offset you get by staying with the offset of the third card.

  • d=2d=2: the order is (0,2,1)(0,2,1). Staying gives 22 and the third card gives 11, so stay: 22.
  • d=−2d=-2: the order is (2,0,1)(2,0,1). Staying gives 00 and the third card gives 11, so continue: 11.
  • d=1d=1: the order is (0,1,2)(0,1,2) or (1,2,0)(1,2,0). Staying gives 1,21,2 (total 33) and continuing gives 2,02,0 (total 22), so stay: total 33.
  • d=−1d=-1: the order is (1,0,2)(1,0,2) or (2,1,0)(2,1,0). Staying gives 0,10,1 (total 11) and continuing gives 2,02,0 (total 22), so continue: total 22.

Summing the best totals over all six orders gives 2+1+3+2=82+1+3+2=8. So after passing the first card, the optimal continuation gives E[X]=86=43\mathbb{E}[X]=\tfrac{8}{6}=\tfrac43. This exceeds the value 11 of stopping at once, so the optimal strategy is as follows:

  • always pass the first card;
  • keep the second card if it is higher than the first;
  • otherwise take the third card.

Its expected payout is n+43n+\tfrac43, so

c=43\boxed{c=\tfrac{4}{3}}