Lesson 8 of 8
What is left standing
Take away the data mining, the publication decay, the costs and the drawdown. Write down what the evidence still supports, in terms someone can disagree with.
By Javier Sanz
Founder of ValueScreener. Built Ninety Nine, a securities services provider, and sold it to a tier-one global bank; ran operations and internationalization at Alpaca, the brokerage API.
Who writes thisEditorial processLinkedIn (opens in a new tab)X (opens in a new tab)
In one minute
- The long record shows a premium of about 3.6 percent a year since 1926 and about 3.1 percent since 1963, with two negative decades inside it.
- After publication decay, costs and the measurement choices, the part a person can expect to keep is a fraction of the printed figure.
- What survives best is not the factor. It is the practice of paying attention to price, which needs no premium to be worth doing.
This course has taken the evidence apart. The last job is to say what is left, in a form a reader can argue with rather than a slogan they can repeat.
The case, in the order this course made it
- 1926MeasuredThe book-to-market factor compounds at about 3.6 percent a year from July 1926, and about 3.1 percent from July 1963.
- 1994ContestedWhether the premium is payment for risk or a recurring error is unsettled, by the admission of the people who measured it.
- 2006InterruptedThe factor fell 57.8 percent between December 2006 and September 2020 and had not regained that level by July 2026.
- 2016DiscountedAcross 97 published predictors, returns were 26 percent lower out of sample and 58 percent lower after publication.
- 2016Crowded316 factors had been published, and the researchers counting them argue most claimed findings in the field are likely false.
The four statements the evidence supports
Say these, and you can defend every word
-
A long, repeatedly measured record shows companies priced cheaply against their accounts earning more on average than companies priced expensively, in the United States and in several other markets.
-
Nobody has settled why, and the two candidate explanations imply different futures.
-
Whatever the cause, the measured advantage is smaller after publication, smaller again after costs and taxes, and absent for stretches long enough to end a career.
-
Book value, the measure most of the record is built on, describes some companies well and others badly, and the share it describes badly has grown.
The case against, stated at full strength
A reader should be able to make this argument, and here it is.
The premium was found by searching a fixed historical dataset that everybody uses. It has decayed since publication, exactly as a data-mined or arbitraged result would. It rests on an accounting measure that no longer captures how a large part of the economy invests. It vanished for fourteen years in the most recent test. And it sits in a literature whose own statisticians say most published findings are probably false.
That argument is not a straw man. It is the reason this course exists, and a person who cannot state it has not understood the method they are using.
What is left, honestly
Two things.
The first is narrow and well evidenced: price paid is one of the few variables an investor controls, and the evidence that paying less for the same stream of earnings improves the outcome does not depend on any factor premium surviving. It is arithmetic about what you pay against what you get.
The second is a practice rather than a premium. Reading the filings, writing down why you bought, and being told when that reason breaks are activities whose value does not rest on a published anomaly. They reduce the chance of the specific mistakes the rest of this academy documents: paying for a growth rate nobody can sustain, mistaking an accrual for cash, missing a maturity wall.
Check yourself
Which statement is supported by the evidence in this course?
- The value premium is guaranteed over a long enough horizon
- A long record shows a positive average premium, with periods long enough to look like failure
- The premium has been proved to be a data mining artefact
Show the answer
Answer: B.
Both stronger claims go past the evidence. The measured record shows a positive average and a fourteen-year stretch inside it that was indistinguishable from failure at the time.What is the strongest form of the case against value investing?
- That cheap companies are unpleasant to own
- That the result was found by searching a shared dataset, has decayed since publication, rests on an accounting measure that fits fewer companies, and disappeared for fourteen years
- That the studies used the wrong countries
Show the answer
Answer: B.
Those four points are each measured rather than asserted, which is what makes the argument serious and worth being able to state.What does this course say survives regardless of whether the premium does?
- The ability to time the market
- The arithmetic of price paid against what is received, and the practice of documenting and monitoring a reason
- The guarantee of beating an index fund
Show the answer
Answer: B.
Neither claim depends on a factor premium. Both are about what an investor controls rather than what the market pays for a characteristic.Sources
- Fama and French three-factor monthly series, Kenneth R. French data library, file built from the July 2026 CRSP database, downloaded 2026-09-13, for the long-run and drawdown figures computed in this course.
- McLean and Pontiff, "Does Academic Research Destroy Stock Return Predictability?", Journal of Finance, 2016.
- Harvey, Liu and Zhu, "... and the Cross-Section of Expected Returns", Review of Financial Studies, 2016.
- Fama, "Two Pillars of Asset Pricing", Prize Lecture, 8 December 2013, published 2014.
- Novy-Marx, "The Other Side of Value: The Gross Profitability Premium", Journal of Financial Economics, 2013.
What this leaves out
- Every figure restated here is sourced in the lesson it came from, and none of them is recomputed on this page.
- A summary of a contested literature is itself a judgement. The four statements above are the ones this academy is prepared to defend, not a consensus.
- Nothing here forecasts any return, and none of it is advice about any security.