LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
A handful of public comments, answered in one lecture, cover more ground than a single formula. Backtests here are built in-house, not in TradingView or QuantConnect, with data from Interactive Brokers, Databento, and Norgate so delisted names stay in the universe. A 93 percent win rate is called possible and uninformative until you see the average win, the average loss, and the left tail. The law of large numbers is not a promise that more trades will reveal an edge once regimes move.
Why it matters
One comment asks about a politician's portfolio and, by name, treats it as an outlier. The lecture generalizes. Politics is one risk. Stocks are one instrument. A model cannot hold every risk, because complexity trades effectiveness against speed. Whatever you leave out is being treated as noise. If political risk is in the noise bucket and a political crisis then dominates returns, the omission shows up as a tail event. The same framing covers any factor you hoped would stay small.
The speaker's own book, in this reply, is cross-sectional equities with a large share of the return described as premium capture, especially volatility. Short-volatility variants and zero-days-to-expiry options are mentioned as ways people try to harvest a volatility risk premium. The usual insult is picking up pennies in front of a steamroller. The reply is to move. The lecture does not claim the premium is free. It claims the failure mode is refusing to get off the trade.
The idea in plain words
A résumé attack says the speaker only worked at Reuters and therefore cannot comment. The lecture says that job never happened. A separate remark blames someone named Nick for citing the central limit theorem when the law of large numbers was the right tool. The speaker does not know who Nick is and does not adjudicate the citation. The substantive correction is about volume. More trades are encouraging if you are estimating something stable. They do not force an empirical average onto a data-generating distribution that has changed regime. Non-stationarity is the reason a citation of the law of large numbers plus a large sample is not, in this reply, a proof the edge is real.
Another comment asks why the material is taught instead of being used inside a top firm. Three answers are given. The draw was the mathematics, not the pay. Teaching students who want the math, and hearing that the notes showed up in an interview or a job, is ranked above a bonus. And the speaker does not want to be managed. A demand for "proof of results," paired with the claim that stochastic calculus is useless in this field, gets a different reply. The lecture says trillions in notional rely on that calculus, that some quants use it daily and some never do, and that calling it useless is like calling a baseball bat useless because you play football. The teaching claim is narrower still: the notes are math and statistics, not a trading product, and the audience can stay or leave.
How to use it
A mathematics student asking how to move into finance is told to build the skill the seat actually uses. Quant trading, research, and developer roles want models. Banking or a sales-and-trading seat may want a discounted-cash-flow habit: enterprise value, the statements, and whether you could forecast a lemonade stand's cash flows and say what that implies for a share price. Catch up on probability if that is the hole. The advice is matched to the destination, not to a single major.
What usually goes wrong
Outsourcing the backtest to a platform, then dropping delisted names, changes the universe you think you tested. The lecture's stack is a point-in-time return panel and a walk-forward. Treating a famous portfolio as a puzzle about one person misses the modeling point, which is which risks you included. Treating a large sample as automatic convergence misses regime change. Treating a high win rate as skill misses the one-trade caricature and the tail.
Questions people ask
Where do the backtests in this reply get their data?
From Interactive Brokers, Databento, and Norgate, the last especially for delisted names. The tests themselves are in-house, with a point-in-time panel and a walk-forward.
Does more volume prove an edge?
Only if the distribution you are averaging stays put. The lecture says regimes change, so the law of large numbers does not turn volume into convergence by itself.
What is missing from a 93 percent win rate?
The average win, the average loss, and the loss tail. One win in 22 years would also be a perfect rate, which is why the percentage alone is not the strategy.