LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
You can believe a stock is going to the moon, and the person quoting you can believe it is not, and the option price in this lecture still does not split the difference. In the Black-Scholes hedge the drift drops out, and the underlying is modeled as earning the riskless rate in expectation. That is the move from the real-world measure P to the risk-neutral measure Q. Once the market is incomplete, there are many such measures, and the lecture picks one by calibrating a richer model to liquid option prices.
Why it matters
A European call or put has a known payoff at maturity T, a function of the spot then and the strike. One simulated path finishes below the strike and pays zero, after having spent time above it. Pricing is about that unresolved payoff. A geometric Brownian motion needs a drift mu and a volatility sigma. The optimist might put mu at 0.20 and the pessimist at 0.05. Those choices move the cloud of terminal prices, and a higher volatility spreads it further. Estimate both parameters on a bull run and more simulated paths finish in the money. Estimate them on a crisis sample, the lecture cites the financial crisis or COVID, and more finish at a zero payoff. Each person would then name a different premium. That is speculation, and it is also unstable, because the parameters are not constant across regimes.
Misspecification means the model class is wrong: prices jump or remember, and geometric Brownian motion does not. Misparameterization means the class is tolerable and the numbers are not. The lecture treats both as ordinary. A model that knew the true dynamics would not be needed.
The idea in plain words
A change of measure is a change of distribution, a reassignment of probabilities on the same outcomes. In the animation, P has a drift of 15 percent. Q earns 5 percent, the riskless rate used there, in expectation. Volatility remains. Only the center of the terminal cloud moves. The reason you are allowed to make that change, in this lecture, is replication. Hold the Black-Scholes hedge and the randomness cancels. A riskless portfolio must earn the riskless rate, or an arbitrage appears. The drift you argued about is gone. Simulate with that rate and the averaged, discounted payoff converges to the formula. Simulate with a zero drift and it does not. Feynman-Kac is the lecture's name for the bridge between the hedging equation and that expectation. The history of the idea was not a straight line. Replication came first. The measure language was tidied up afterward.
How to use it
Read the smile as the market's charge for what the one-shock hedge cannot remove. The lecture's surface is not flat. Out-of-the-money puts are priced at higher implied volatility than the corresponding calls. Downside risk is paid differently. Fear, in this telling, dominates the equilibrium price. Agents are not the rational agents of the textbook, so a clearing price and a fair value can diverge, and the lecture says that gap is where alpha can live. The 1987 crash is the historical moment attached to models that had to fit the skew.
What usually goes wrong
Pricing from a personal drift, bull-market or crisis, reopens the disagreement the hedge was built to close. Forgetting the riskless drift in the simulation misses the formula even when the hedging story is right. Treating calibration as a claim that the model is true misses the point of an incomplete market. The fitted surface is a choice of measure consistent with today's liquid prices, including a premium for unhedgeable risk. It will be the wrong choice when those prices move and you do not refit.
Questions people ask
Why can the drift be ignored?
In the complete-model hedge, the risky part is cancelled and the leftover portfolio must earn the riskless rate. Your forecast of mu never enters that price.
What were the two drifts in the picture?
An optimist at 0.20 and a pessimist at 0.05 for the real-world mean, and a separate animation with P at 15 percent against a risk-neutral mean of 5 percent.
How do you choose a measure when many exist?
By calibration to the implied-volatility surface, not by averaging the two forecasts. The smile is the lecture's evidence that one volatility cannot hedge every risk.