48%与52%：大数定律如何将随机性转化为确定性

a 48% game can feel random forever, a 52% game can build a casino

one blackjack hand tells you almost nothing. the casino can lose 10 hands, 100 hands, even have a terrible night, because short-term variance is designed to hide a tiny statistical edge. but Lee's example gives the player roughly a 48% chance and the house 52%; repeat that bet across enough players and the randomness starts disappearing. the player experiences noise, while the casino experiences the law of large numbers.

this is the same mathematics behind systematic trading. a strategy does not need to predict every trade correctly; it needs a real positive expectation and enough independent repetitions for the average to reveal it. Lee explicitly makes the connection to hedge funds and high-frequency trading: even a tiny edge can become meaningful if you can repeat it enough times, while variance can make a winning strategy look completely broken over a small sample.

the numbers get stranger at scale. one random outcome can land almost anywhere, but take 100 million independent observations and their distribution becomes highly predictable. averaging n independent observations leaves the expected value unchanged while dividing the variance of the average by n; increase the number of trials 100× and the variance falls 100×. randomness does not disappear - its ability to move the average does.

then the Central Limit Theorem goes one step further. start with almost any distribution with the required conditions, combine enough independent observations, normalize the result, and the distribution moves toward the same bell curve. this is one reason the normal distribution appears everywhere in finance: completely different microscopic outcomes can produce remarkably similar statistical behavior once they are aggregated at scale.

but there is a trap hidden inside the theorem that traders almost never talk about: IID - independent and identically distributed. the math works when your repeated trials are actually generated by the same underlying process. if your “edge” changes with liquidity, volatility or market regime, then doing 10,000 trades does not automatically turn bad assumptions into certainty; you may simply be sampling several different games and pretending they are one.

the lecture is free. 48% can feel indistinguishable from 52% in the short run, 100 million observations can turn chaos into a visible distribution, and enough repetitions can expose an edge too small for any human intuition to feel. the real advantage is not predicting the next outcome - it is knowing whether you are actually playing the same positive-expectation game enough times for probability to take over.