Why are there 300 factors, and how many are real?
Once three characteristics were shown to earn more than beta explained, every characteristic was tried. By 2016 Harvey, Liu and Zhu had catalogued more than three hundred published factors, and the title of their paper — … and the Cross-Section of Expected Returns — is the joke: fill in the blank with anything. This lesson is the two additions most practitioners accept, and the rule for doubting the rest.
Quality: profitable beat unprofitable
Novy-Marx showed in 2013 that gross profitability — revenue minus cost of goods, over assets — sorted shares nearly as well as book-to-market, and in the opposite direction from a naive value screen: profitable companies are expensive for a reason, and the ones that are expensive and profitable did not underperform the way pure value said they should. Fama and French folded a profitability factor and an investment factor into their 2015 five-factor model. In funds it is sold as quality: high return on equity, stable earnings, low debt.
Measured on 2026-09-04, the quality fund QUAL.US holds 118 names and turned over 58% of them in a year — a rule that re-sorts on published accounts, which change every quarter.
Low volatility: boring beat exciting
The market model says a low-beta share should earn less. In the data the lowest-volatility groups earned about as much as the market with far less turbulence — Baker, Bradley and Wurgler in 2011; Frazzini and Pedersen's betting against beta in 2014. The candidate explanation is a constraint: investors who cannot borrow buy high-beta shares to get more market exposure, overpaying for them, and leave the low-beta ones cheap.
Measured on the same footing as the rest: over ten years to 3 September 2026 the low-volatility fund USMV.US returned about 164% against 316% for SPY.US. Less return, in a decade that rewarded taking every risk on offer. What it bought is visible in the fund's own reported three-year volatility, which is lower, and the risk-and-return course's was-it-worth-the-ride question is the only fair way to compare the two.
The zoo, and the rule for doubting it
Three hundred factors were not found because markets have three hundred risks. They were found because researchers ran thousands of sorts on the same fifty years of US data, and one sort in twenty clears the conventional significance bar by chance. Harvey, Liu and Zhu's response was to raise the bar: a factor should show a t-statistic of at least 3, not 2, before anyone believes it — and even then, most of the catalogue fails to survive out-of-sample.
The practical rule that falls out of this, and which the rest of the course uses, has three parts. A factor deserves attention when it has an economic reason to exist — a risk someone is paid to bear, or a behaviour that is hard to arbitrage; when it survives in other markets and other decades than the ones it was found in; and when it survives costs, because a premium that turns over the whole portfolio each year has to clear the trading bill before it clears anything else. Size, value, momentum, quality and low volatility pass the first two tests with arguments — and the second test has real evidence behind it: Asness, Moskowitz and Pedersen's Value and Momentum Everywhere (2013) found both premia in eight markets and asset classes at once, and Fama and French's 2012 international study found size, value and momentum in four regions outside the sample they were discovered in. Most of the other 295 have no such record.
In the data
A quality sort ranks on return on equity, margins and debt, all in a company's published accounts; a low-volatility sort needs nothing but the price history and a standard deviation. Two companies on the same exchange, at the two ends of a quality sort:
NVIDIA earns more than its equity in a year; Pfizer earns about a twentieth of it (29 September 2026). A quality fund holds the first kind, and pays up for it, which is why quality and value so often disagree about the same name.
Try it now
- Read the annual turnover of the quality fund and the minimum-volatility fund below, and write down which rule re-sorts more often and why that follows from what each one sorts on.
- The market both funds are measured against, at the range where a decade of "every risk rewarded" is visible:
Measure 2 September 2016 to 3 September 2026 and compare with the 164% the low-volatility fund delivered. Then say what a t-statistic of 3 would have to do with whether you trust that comparison.