LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
An option price that can be written as an expectation can be estimated by averaging simulated payoffs. The law of large numbers is why the average converges. Black-Scholes is the case where the expectation also has a formula. In the lecture's matched demo the simulated price was 10.4504 and the formula price was 10.45. The gap is sampling error, not a second theory. When no formula exists, the same average is the price you have, and it is slow.
Why it matters
Analytical pricing means substituting a mass function or a density into the definition of expectation and simplifying. A Bernoulli trial that pays 1 on success has expectation p. A fair coin has expectation 0.5. A coin with p = 0.7 has expectation 0.7. An exponential waiting time with rate lambda has expectation 1/lambda, so lambda = 2 gives 0.5. Those are formulas. Some sums and integrals do not simplify. Monte Carlo is the fallback: draw from the distribution, average the draws, and let the sample size grow. The lecture's small loops produced a Bernoulli average of 0.711 against 0.7, and an exponential average of 0.5023 against 0.5. The law of large numbers says the sample mean approaches the expectation as the number of draws goes to infinity. The central-limit story about fluctuations is a different theorem. Convergence of the average is the one this pricing argument uses.
The idea in plain words
A geometric count of trials until the first success has expectation 1/p. For a fair die, the chance of a six is 1/6, so the expected number of rolls until the first six is 6. Simulate by rolling until a six appears, record the count, and average many games. A small number of games sat near 6.4. More games moved the average onto 6. The approach is in the number of games, not in calendar time. A European call under Black-Scholes is the same shape of statement with a harder expectation. The stock follows geometric Brownian motion. Under the risk-neutral measure used for pricing, the call price is the discounted expected payoff, and the payoff is the larger of zero and the terminal stock minus the strike. Deriving the formula is a hedging argument, Itô's lemma, a partial differential equation, and a boundary condition. The lecture points at that stack and does not repeat it. Once the formula exists, parameters go in and a price comes out. The simulation locks the same parameters: spot, strike, maturity, rate, and volatility. It draws terminal prices, computes payoffs, discounts, and averages. The histogram of terminal prices and the histogram of payoffs are intermediate pictures. The number at the end of one lecture run was 10.4504, next to a formula value of 10.45.
How to use it
If a formula exists, use it and treat a simulation as a test. Match every input. A 10.4504 against 10.45 is a passed test. A large gap means the discount, the payoff, or the distribution of the terminal stock does not match the formula's assumptions. If a formula does not exist, the average is the estimator. Increase the path count until the average settles the way the die average settled from 6.4 onto 6. The lecture notes that a known error analysis can put a band on that average. It does not compute the band for the option demo.
The research thread the lecture adds is to simulate many parameter sets offline and train a model to map parameters to the simulated price, so a live calibration evaluates a function instead of a new set of paths. The lecture calls the results promising for families without formulas. It does not report an error table for that map here.
What usually goes wrong
Simulating geometric Brownian motion to price a plain European call and preferring the noisy average to the formula. The lecture calls that average redundant when the formula is available. Changing volatility between the formula and the paths, then treating the gap as a mystery. And reading 0.711 as a failure to hit 0.7. It is a finite-sample average. The theorem is about the limit, which the exponential's 0.5023 is also approaching. Using too few option paths and quoting 10.4504 as exact is the same mistake at a higher price.
Questions people ask
What did the coin and the exponential average?
The coin with success probability 0.7 averaged 0.711, against an expectation of 0.7. The exponential with rate 2 averaged 0.5023, against 0.5. Both expectations are formulas. The averages are checks.
Why is the wait for a first six equal to six?
The number of trials until the first success has mean 1/p, and a fair die has p = 1/6. A short simulation sat near 6.4. A long one moved to 6.
How close were the option prices?
The Monte Carlo price in the lecture was 10.4504. The Black-Scholes price on the same inputs was 10.45. The lecture treats that as agreement, not as a new model.