LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
The Sharpe ratio is expected portfolio return divided by portfolio standard deviation. You can raise it by increasing the numerator or by decreasing the denominator. The mechanical route to a smaller denominator is to hold risks whose correlation is below one, because that shrinks portfolio variance. The catch is that a correlation estimated from returns can jump to one in a crisis. Physical independence, two processes that do not share a cause, is what keeps the variance benefit when you need it.
Why it matters
Everyone wants a better risk-adjusted score. Fewer people write down the two-asset variance that produces it. Take risky assets A and B, stocks, bonds, or anything else. The portfolio return is a weight w on A plus a weight q on B, where q is one minus w if the weights exhaust the book. Variance is the expected squared gap between the random portfolio return and its mean. Because expectation is linear, the mean of the portfolio is the same weighted mix of the two means.
Subtract the mean equation from the return equation and the weights factor out. What remains inside the variance is a weighted sum of each asset's surprise. Square that sum and take the expectation. You get three pieces: w squared times the variance of A, q squared times the variance of B, and twice w times q times the covariance of A and B. Covariance is correlation times the two standard deviations. Substitute that in, and the cross term contains the correlation explicitly.
The idea in plain words
Portfolio standard deviation is the square root of that variance. At a correlation of one half, the cross term is already cut relative to the perfectly locked case. At zero, the cross term drops out. At a negative correlation, it subtracts. You cannot edit the stand-alone variances of A and B. You can change weights, and you can choose different assets. That shift is arithmetic. The lecture's warning is to check the numerator after you celebrate the denominator.
Independence has a direction. If two things are physically independent, like successive rolls of a die, they are also uncorrelated. The correlation is zero because one roll does not act on the next. Stochastic independence in a textbook, a property of random variables, also drives correlation to zero in the limit. The reverse is the failure. A sample correlation near zero does not mean the zero will survive. When that correlation rises, portfolio variance rises, standard deviation rises, and the Sharpe ratio falls. The calm-sample benefit was real and temporary.
A sports market-making algorithm next to a Nvidia position is the lecture's example of the stricter case. The contracts settle on the sport. A technology selloff does not rewrite them. Physical independence is what lets you skip the backtest as a proof of the zero correlation.
How to use it
Write the variance with the correlation visible before you add a diversifier. If the only candidates are equities that share a factor, assume the cross term can inflate later. If you add a sleeve, recompute both pieces of the Sharpe ratio, not just the volatility. The lecture's standard for the sleeve is orthogonal return: a driver that is not the driver's cousin. Statistical decoration, two stock charts that happened to be flat against each other last year, is the version that breaks.
Do not read the algebra as a promise about future wealth. It says how the ratio is built from means, variances, and a correlation. The future correlation of two stocks is the object the algebra cannot police.
What usually goes wrong
Raising expected return and accidentally raising variance by as much or more. Diversifying two names that share a crash factor and calling the result independence. Assuming a classroom proof about stochastic independence applies to Apple versus Nvidia. And forgetting that a zero cross term with a near-zero expected return on the new sleeve can leave the ratio no better, because the numerator fell with the denominator.
Questions people ask
Does any correlation below one raise the Sharpe ratio?
It lowers the variance relative to a correlation of one, holding the other inputs fixed. The ratio improves only if the expected return does not fall enough to offset that. The lecture treats the variance result as mechanical and the return check as mandatory.
Why is a die a better analogy than two stocks?
One die roll does not cause the next. That is physical independence, which implies a zero correlation. Two stocks can print a near-zero correlation in a sample and still share a cause that appears in a crisis. The implication does not run backward.
Is this a backtest result?
No. The lecture derives the variance identity and then interprets it. The sports-versus-equity example is there to name a pair you do not need a historical correlation to classify. A backtest might illustrate the pair. It is not the proof.