LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
Monte Carlo replaces an integral you cannot finish with an average of random draws. The law of large numbers is why that average settles: independent draws from one distribution push the sample mean toward the population mean. The method fails when that distribution will not sit still.
Why it matters
A contract price and a risk number are both expectations. If there is no tidy formula, simulation is what remains. The lecture uses it for a dice game, a hitting probability, and a European option under geometric Brownian motion. In each case the rules do not change between draws, so a long average can stand in for the theoretical mean. A live return series can move while you are still averaging.
The idea in plain words
A random variable is a weighted set of outcomes, not one algebra symbol. A bell curve puts more weight near the center than in the tails. Draw from it and the histogram fills in. Ten thousand draws in the lecture sit on the curve they were drawn from. That match is convergence in distribution, and it assumes you keep using the same curve.
The law of large numbers is the matching statement for averages. Keep drawing independently and the sample mean approaches the mean of the curve. The lecture's pictures start with a handful of draws, where the mean wanders, and run out to many thousands, where it sits on the center. One animation starts at two draws and grows from there. If you cannot compute the center by hand, you can still draw and average. That is the whole method.
A fair die has a known mean and variance. About 10,000 rolls move the sample mean and variance onto those formulas. The useful case is a game you would rather not expand by hand. Roll a six-sided die. On an odd face, flip a weighted coin and take its payout. On an even face, flip a fair coin. The rules never change. About 10,000 plays reproduce the theoretical mean, which the lecture computes as 325.
How to use it
Price the game above 325 if the house wants a positive expected gain per play. At 325 the long-run average is a wash. Below 325 the average favors the player. How many plays you get decides how much room a losing streak has. That is the same family of questions as gambler's ruin.
An indicator turns the same average into a probability. Mark a win as 1 and a loss as 0. The mean of those marks is the win rate, because a Bernoulli random variable has mean equal to its success probability. After 1,000 plays the lecture's estimate is moving toward a theoretical win probability of 0.55. A second question uses the same trick: start at 10,000, pay the fair 325 each game, and ask for the chance of reaching 15,000 before ruin. Paths that hit 15,000 are marked one way and paths that go broke another. The simulated chance is about 42.8 percent. The lecture treats that figure as a poor reason to play at the fair price.
An option price is the same average. Under geometric Brownian motion the terminal law is fixed once the parameters are set. Simulate, discount the payoff, and average. The lecture's average approaches the Black–Scholes reference. An indicator answers smaller questions, such as expiring in the money or touching a barrier. Variance, skewness, and kurtosis are the same kind of sample statistic.
What usually goes wrong
The draws have to come from one distribution. A return series does not offer that. Nothing requires a young company's returns and the same name in a later boom to share a curve. In the lecture's animation the true curve keeps moving while the histogram chases it, and the chase works better only after the curve stops. You never see that true curve. You see the histogram, and returns sit with inflation, jobs, and policy.
A printed mean can then be tempting and still not be the mean of anything stable. A wrong model does not by itself decide gain or loss. It does mean the simulated number is not approaching a fixed theoretical value. The lecture flags value-at-risk style levels for the same reason: they target a threshold that may not exist as a stable population number.
Questions people ask
What has to be true for the average to settle?
Independent draws from one distribution that does not change. Break either condition and the sample mean is no longer approaching a single population mean.
Why is the house price above 325?
The lecture's fair value of that game is 325. A higher price is the positive expected gain per play. At 325 the long-run average does not favor the house.
What did the 15,000 target show?
From 10,000, paying 325 a game, about 42.8 percent of the simulated paths reached 15,000. The rest were the ruined paths.