The Card-Pack Trading Game
Cards are drawn one at a time from a shrinking, known deck, and you must quote a market on what's left after each draw — the interview classic for testing whether you can update expected value as the sample space depletes without replacement.
Prerequisites: Expected-Value Games
A standard deck of 52 cards, values Ace (1) through King (13), suits ignored. Cards are drawn one at a time, face up, without replacement, and after each draw you must quote a market on the value of the next card. The game tests something dice games can't: cards are drawn without replacement, so every reveal permanently changes the deck you're pricing against, unlike a die which resets every roll.
Why "without replacement" changes everything
With a die, each roll is independent — the expected value of the next roll never depends on past rolls. With a card pack, every card removed shifts the composition of what's left. If you've seen a run of low cards come out, the remaining deck is, on average, richer in high cards than a fresh deck — not because of any mystical "due for a high card" effect, but because the high cards genuinely haven't been removed yet and there are fewer cards left to share them among.
Worked example
Simplify to a small pack: cards valued 1, 2, 3, 4, 5 (one each), no replacement. Unconditional expected value of a single draw is . Suppose the first card drawn is a 1. The remaining pack is {2, 3, 4, 5}, and the expected value of the next draw is now — it rose, because the lowest card is gone and can't be drawn again. If instead the first card drawn had been a 5, the remaining pack {1, 2, 3, 4} has expected value — it fell. Each draw mechanically shifts the average of what remains in the opposite direction of the card removed, and by an amount that depends on how few cards are left: removing one card from a five-card pack moves the average by half a unit; removing one from a fifty-card pack barely moves it at all. That shrinking sensitivity, not the direction of the last card, is the part worth stating explicitly when quoting.
What this means in practice
This is a scaled-down version of a real problem: pricing a claim on a diminishing, known pool — a lottery with a fixed prize structure and shrinking ticket count, or an auction where bidders drop out and reveal information about the remaining field. The reflex being trained is the same one: after every observation, recompute the expectation from the actual remaining set, not from an intuition about "streaks" or "what's due."
Without replacement, every observed draw permanently changes the composition — and therefore the expected value — of what remains, in a direction opposite the card removed and by an amount that shrinks as the remaining pack grows larger.
The classic error is gambler's-fallacy reasoning ("we've seen a lot of low cards, so a high one is due") applied to a game where it happens to be numerically correct for the wrong reason. Say the right reason out loud: it's not that high cards are "due," it's that they're the only ones left in a shrinking, enumerable set — the same logic gives the opposite answer in a with-replacement game like dice, where nothing is ever "due."
Related concepts
Practice in interviews
Further reading
- Xinfeng Zhou, A Practical Guide to Quantitative Finance Interviews, ch. 2
- Weber, Ernst, Fifty Challenging Problems in Probability