LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
A win rate, a monthly cash figure, and an average return do not say whether a strategy was compensated for risk, or whether that result will show up again. A Sharpe ratio, a Sortino ratio, and a maximum drawdown add the risk comparison. The lecture still calls them incomplete. The missing piece is stability: if you traded the book from today, would the live path still look like the backtest?
Why it matters
Every one of these numbers is computed on history, and history is not a schedule for the future. Nvidia is the lecture's simple case. A ten-year average mixes a different company. Inside one year, a tariff stretch produced a year-to-date average near minus 0.4 percent a day. After that stretch cooled, the average was about 0.52 percent a day. The year-to-date gain was roughly 22 percent, and it hides the drawdown at the start. Two strategies with the same 30 percent average return can have taken very different risks. The average alone cannot separate them.
The idea in plain words
Win rate is worse. Three books: a perfect win rate gains 2.97 percent, a 99 percent win rate gains 65 percent, and a 40 percent win rate gains more than 137 percent. The high rate is insurance. Quiet premiums, then one gap larger than all of them. The rate stays flattering. The wealth does not.
Cash per month has the same hole. Three thousand dollars a month from a million-dollar account can be a Treasury coupon. If the Treasury stopped paying, the proxy would be the least of the problems. A steady coupon chart is not a risky book. One contrasted book earns 36 percent with risk that at least existed. Without account size, the monthly dollar figure cannot be judged.
Risk is the expected squared deviation from an expected return. If the expected return is a moving target, the deviation is too. Sharpe and Sortino put the two together and subtract a financing benchmark. If a Treasury pays about 3 percent, a strategy that returns 2 percent has given that benchmark up. Sharpe penalizes upside and downside deviations. Sortino penalizes downside only. In the lecture's pair, strategy A and strategy B finish with similar wealth and comparable Sharpe ratios. From 100,000 dollars, B has a loss of more than 5,000 dollars along the way, almost 12 percent. A's Sortino is 2.09 and B's is 1.77, and A's maximum drawdown is lower. You would rather have sat in A. That comparison is still about the path you already have.
How to use it
Ask what happens if the book is traded from now. One attractive path shows Sharpe 3.1. Its Sortino is 5.6, and the deepest hole is 12 percent. It might be a real economic exposure, a run of discretion, or a fit to noise. The next stretch can degrade. A book the lecture treats as reasonably stable has a backtest Sharpe of 2.12 and a live Sharpe of 1.71. A book that fails the test has a backtest Sharpe of 2.31 and a live Sharpe of 0.51. Degradation can be a crowded mispricing, a risk factor that mean-reverts, an overfit, a backtest that ignored costs, or a bias in the data. A live resemblance is not permanent. Crowding and regime changes still arrive, and the lecture rejects the picture of one person finding an edge, levering it, and walking away.
The ratios remain inputs. They are not the decision. A Sharpe ratio above 10 with no evidence that live trades would resemble it is, in the lecture's close, not a tradable profile. The question it substitutes is whether you are trading a mispricing or bearing a risk, and whether that thing is still there.
What usually goes wrong
Leading with win rate, monthly dollars, or average return leaves out the payoff size, the account size, or the risk. Leading with Sharpe, Sortino, or drawdown and stopping there treats a historical risk adjustment as a forecast. Splitting one history into a train set and a test set is not the test the lecture means. The test is a new live window. And a stable live window is still a thing to watch, not a machine you can leave running.
Questions people ask
How can a 40 percent win rate beat a perfect one?
In the lecture's three charts it gains more than 137 percent, against 2.97 percent for the perfect win rate. The perfect book never says how large the wins and losses are.
What is the Sortino ratio doing differently?
It penalizes downside deviations and not upside ones. That is why strategy B, with the deeper hole, scores 1.77 against A's 2.09 while their Sharpe ratios stay comparable.
What counts as stable in the examples?
A move from a backtest Sharpe of 2.12 to a live 1.71 is the lecture's reasonable range. A move from 2.31 to 0.51 is the degradation it says not to trade.