LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
Beating a benchmark is not alpha. This lecture's nine-stock portfolio makes the distinction concrete. Firm-specific losses can be diluted. Sector bets can be diluted. What remains tracks the market, and in a stress window the correlations that made diversification work all rise together. A tech sleeve that returned 87 percent while the S&P 500 returned 49 percent is, in the regressions, extra market exposure, not a separate skill.
Why it matters
The notebook uses nine real names, three each from technology, health care, and consumer staples. Inside health care, an equal-weight sleeve finishes ahead while UnitedHealth finishes the holding period down 45 percent. That loss is idiosyncratic: it belongs to the firm. A scandal or an operating miss can do the same thing, and the lecture says the market tends to overreact, especially on the way down. An equal-weight sleeve of the three health-care names, or of tech, or of staples, already softens the single-name hit. The sleeve can still suffer if the whole industry is hit by a rule change the other sectors do not share. That is sector risk, and it too can be diluted by holding all nine names.
The residue is market risk. The lecture highlights the tariff stress at the start of 2025. Tech and staples, which had looked weakly related or even opposed, move together. The statistical trick behind diversification, low correlation, is unavailable while that lasts. Afterward the correlations ease. The lecture calls the episode mean-reverting. You cannot diversify it away by adding more sectors. A put is a hedge, and the lecture separates hedging from diversification, which it still calls the only free lunch, and one that closes in a crisis.
The idea in plain words
Inside each sector the correlation matrix is positive. Across a tech name, a health-care name, and a staples name, the pairwise correlations are described as statistically indistinguishable from zero. Walmart and Broadcom are the sort of pair that makes that plausible, until the stress window. Correlation estimates also refuse to settle. Thirty days, three months, or last year are different answers, because the expectations inside covariance do not get the law of large numbers when the source is non-stationary.
The equal-weight nine-stock book hugs SPY, and the rolling correlation of the two is described as very close to one. The single-name upside and downside have been averaged off. The lecture notes that, in the cross-section, idiosyncratic exposure is typically a poor deal, because many holders chase a lottery ticket in one name. One name can still be the right personal bet. The pricing statement is about the average, not about your case. The nine-stock book beat the market and returned 46 percent. That sentence is not yet an alpha claim.
How to use it
Regress each sleeve's excess return on the market's excess return. The intercepts are described as statistically indistinguishable from zero. The slopes differ. Tech is overexposed. Health care has about a quarter of that exposure, and staples about a third. Tech's R-squared against the market is higher than health care's, with staples in between. The market factor is the first principal component's 27 percent, not the whole panel. The rest is other priced risks and whatever is left over. Fama-French three- and five-factor models, and Carhart, are the lecture's next shelves, along with momentum, size, and value. A significant intercept would be the statistical version of selection skill. In these sleeves it is not there.
What usually goes wrong
Calling the 46 percent or the 87 percent alpha ignores intercepts that were indistinguishable from zero. Freezing a beta estimated in a calm window misses the jump when correlations spike. Adding names until the book matches SPY and then expecting it to survive a crisis by diversification repeats the tariff chart: the correlations rise, and the market factor is what you have left.
Questions people ask
What happened to UnitedHealth in the sample?
It finished the holding period down 45 percent while an equal-weight health-care sleeve of three names was still ahead. The lecture treats that gap as firm-specific risk.
How much variance did the first components explain?
The first principal component explained 27 percent, the first three explained 60 percent, and five or six components reached almost 90 percent.
Why is the tech outperformance not alpha?
The sleeve's regression intercept was indistinguishable from zero, and its beta was the high one. The 87 percent versus the index's 49 percent is described as extra market exposure.