Mathematics
Optional Stopping Theorem and Fair Game Paradoxes
Quick fact
In a fair game, even if you use a clever stopping rule to 'quit while you're ahead,' your expected winnings are still exactly zero — you cannot outsmart a fair game. However, this only holds if your stopping rule has a bounded horizon or other technical conditions; otherwise, you can encounter paradoxes like the St. Petersburg lottery.
Why this is interesting
Imagine a perfectly fair coin flip where you win $1 on heads and lose $1 on tails. Surely you can't beat it... but what if you decide to stop after you're ahead by $1? Does that guarantee you walk away a winner?
Read the full explanation
Understanding Optional Stopping Theorem and Fair Game Paradoxes
Think of a fair game as a walk where each step is equally likely to go up or down. Your expected position after any fixed number of steps is where you started — that's what 'fair' means. The optional stopping theorem extends this: if you decide to stop based on what has happened so far (a stopping time), your expected position at the stop is still the starting value, provided you don't stop after an unbounded number of steps. So the strategy 'stop when I'm up $1' sounds like a sure win, but the catch is there's no guarantee you'll ever hit that point — you might wander downward forever first. The theorem says your expected result is zero, meaning the chance of being up $1 is balanced by the chance of being down a huge amount.
A deeper explanation
The optional stopping theorem is a formal result about martingales — sequences where the expected next value equals the current value. A stopping time is a rule for ending the game that depends only on past and present information. The theorem states that for a martingale and a bounded stopping time T, the expected value at T equals the initial expected value: E[XT] = E[X0]. Why does this hold? Because you can write XT as an average of future steps, and the martingale property says each future step has expected change zero. Summing these zero-expectation increments up to T (which is bounded) gives zero. If the stopping time is unbounded, the theorem can fail — for example, you could wait until you are ahead by $1, but that might take forever with probability 1. This is why the classic doubling strategy fails: it requires unbounded capital and an infinite time horizon, both impossible in practice. Understanding this theorem dismantles many 'get rich quick' schemes and underscores the fundamental nature of fair games.