The Devil in the Details: How You Build a Factor Changes What It Is

This paper is about a plumbing detail, and it is one of the most important practical papers in factor investing, because the plumbing detail turns out to change what the factor is.

The famous Fama-French value factor, HML, sorts stocks on book-to-price. That means dividing a company's accounting book value by its stock price. Two numbers. Simple.

Except: which price?

The problem: a value signal that does not know the current price

Book value comes from the annual report, which is published with a delay. To avoid using information before it was actually available (a look-ahead bias, which is the cardinal sin of backtesting), the standard academic convention lags the book value by six months or more.

Fine. But the standard convention then does something else, and this is the part Asness and Frazzini fixed: it lags the price by the same amount. The value ratio is computed using the accounting book value from last year and the stock price from last year.

Think about what that means. It is now July. You are deciding whether a stock is cheap. The standard factor tells you the answer using the price from a year and a half ago. If the stock has doubled since then, the standard factor still thinks it is cheap.

That is bizarre. You do not need to lag the price, because today's price is not private information. It is on the screen. Everyone can see it. Lagging the book value is necessary; lagging the price is a mistake that got frozen into convention and then copied for twenty years.

The key idea via analogy: last year's price tag on this year's shelf

Imagine walking into a shop where every item is labelled with the price it had eighteen months ago. You would make some very strange purchasing decisions. You would buy things that have quietly tripled in cost, believing they were bargains.

That is what the standard HML does. It identifies "cheap" companies using a price that may be badly out of date.

Asness and Frazzini's fix is obvious once stated: use the most recent book value you legitimately have, and divide it by today's price. Keep the necessary lag on the accounting data, remove the unnecessary lag on the price. They call the result HML Devil.

The consequences of this small change are surprisingly large.

One: the value factor becomes cleaner. The updated version reflects what an actual investor would do, which is to look at the current price when deciding if something is cheap.

Two, and this is the interesting one: it changes the relationship between value and momentum. Here is why. If a stock has risen sharply over the past year, then the stale-price version of HML has not noticed, and may still classify it as cheap. So the old HML accidentally holds some recent winners. The updated version sees the price rise, notices the stock is no longer cheap, and drops it.

That means the old HML has a hidden, accidental momentum tilt buried inside it. Removing the stale price removes that tilt, and the updated value factor becomes much more negatively correlated with momentum, which is what the economic logic says it should be. Value buys what has fallen; momentum buys what has risen. They should oppose each other, and with the correct price they do so much more cleanly.

Three: combining value and momentum gets better. Because the corrected value factor is a purer opposite of momentum, a portfolio holding both becomes better diversified. The improvement is not marginal. It is one of the most useful practical results in the factor literature.

There is also a caveat the authors are careful about, and it is important: on a standalone basis, the updated value factor is not automatically better in every test. Its virtue shows up most clearly when it is used alongside momentum, which is how any serious investor actually uses it.

Why it mattered

• It showed that factor construction is not a detail. Two people can both say "I run a value factor" and hold materially different portfolios with different correlations to everything else. The label tells you far less than the construction does.
• It exposed a convention that nobody had questioned. The lagged price was not a considered choice, it was an artefact that got copied from paper to paper because that is how the data came. Twenty years of research inherited it.
• It improved multi-factor investing directly. The finding that a properly-priced value factor hedges momentum better is worth real money to anyone running both, which is most systematic equity managers.
• It is a lesson in reading papers carefully. If you are testing a strategy against a published factor, you had better know exactly how that factor was built, because the answer depends on it.

The honest limitations

• The authors sell this stuff. Both are principals at AQR, which runs value and momentum strategies together, and the paper's conclusion is that value and momentum work better together when built the way AQR builds them. The argument stands on its merits and the logic is hard to fault, but the reader should know the source.
• Fresher prices mean more turnover. A value signal that updates with today's price changes its mind more often than one anchored to a stale price. Turnover costs money, and the paper's improvement is measured before that cost is fully accounted for in practice.
• It does not settle what value is. Book-to-price is still book-to-price, with all its known problems: it undervalues companies whose worth is in intangibles, brands and software rather than factories. Fixing the price does nothing about the fact that the book value denominator has become less meaningful over time.
• The result is a comparison, not a proof. The claim that the timelier version is better rests on tests over particular samples. Others have argued the standard version has its own defensible logic, since it holds the signal constant between rebalances.

The one-line takeaway

The standard academic value factor divides last year's book value by last year's price, which is an unnecessary and distorting choice; Asness and Frazzini showed that using today's price makes the value factor a cleaner opposite of momentum, and makes the two work far better together.