The Rival Model: Hou, Xue and Zhang's q-Factors

Most factor models were built the same way: someone noticed a pattern in the data, and a factor was invented to capture it. Value, size, momentum, all discovered empirically first and explained afterwards. That process works, but it leaves the field open to an uncomfortable accusation: you are just fitting the past.

Kewei Hou, Chen Xue and Lu Zhang took a different route. They started with a piece of standard corporate-finance theory that has nothing to do with stock returns, asked what it implies about stock returns, and then went and checked. The result, the q-factor model, is the most serious rival to the Fama-French framework, and their paper argues it digests most of the anomaly zoo.

The problem: hundreds of anomalies, no organizing principle

By 2015 the academic literature had accumulated an alarming number of published stock-return "anomalies," patterns that supposedly predicted returns and that no standard model explained. Some were plausible, many were suspicious, and there was no coherent framework saying which ones should exist and which should not.

What the field needed was not another factor. It needed a reason.

The key idea via analogy: think like the company, not like the investor

Here is the shift in perspective, and it is the heart of the paper.

Everyone else asks what an investor demands as a return. Hou, Xue and Zhang ask what a company does when deciding whether to build a new factory.

A firm invests up to the point where the last dollar of investment is just barely worth it. This is old, uncontroversial theory, known as q-theory, from Tobin's q. Think of a company as deciding how many new stores to open. It will open a store if the value the store creates exceeds what it costs to build. So the amount a company chooses to invest is a signal about two things at once:

The cost of capital. If a company faces a low required return (investors are willing to fund it cheaply), then lots of projects clear the bar, and the company invests heavily. If it faces a high required return, few projects clear the bar and it invests little. So heavy investment reveals a low expected return, and stingy investment reveals a high one. That flips the intuition: a company on a building spree is not exciting, it is a company whose investors are demanding little.
Expected profitability. Now hold investment fixed. Between two firms that invest the same amount, the one expected to be more profitable must be facing a higher discount rate, because otherwise its high expected profits would have made it want to invest more. So more profitable firms should have higher expected returns.

Those two predictions fall directly out of the theory, before you look at a single stock return. And they are exactly the two effects the data shows. That is a much stronger scientific position than noticing a pattern and naming it.

So the q-factor model has four ingredients:

The market factor, as always.
A size factor, small versus large.
An investment factor, buy low-investment firms, short high-investment firms.
A profitability factor, buy high return-on-equity firms, short low return-on-equity firms.

Notably, there is no value factor and no momentum factor. The claim is that you do not need them: once you control for investment and profitability, the value and momentum patterns largely fall out as consequences rather than causes.

The finding: half the zoo isn't there, and most of the rest is explained

The authors assembled nearly 80 published anomalies and tested them carefully, using value-weighted returns and breakpoints that stop tiny illiquid stocks from dominating the results.

Two blows landed:

About half the anomalies were not statistically significant to begin with in the broad cross-section, once measured with sensible methods. Many published "effects" were artifacts of giving huge weight to microcaps that no real investor could trade.
Of the anomalies that survived, the q-factor model explained most of them, generally at least as well as, and often better than, the Fama-French three-factor and Carhart four-factor models.

That is the meaning of "digesting anomalies." A long list of separate, puzzling phenomena turns out to be, largely, two ideas wearing different hats.

Why it mattered

It gave asset pricing a theory-first factor model. The factors were not chosen because they fit; they were predicted, then found. That distinction matters enormously in a field constantly accused of data mining.
It changed the standards of evidence. The insistence on value-weighted returns and NYSE breakpoints, to stop microcaps from doing all the work, has become a baseline expectation for anomaly research. A lot of the literature simply does not survive that filter.
It set up the great model rivalry. Fama and French's five-factor model, published the same year, uses profitability and investment too. The two camps have argued ever since about priority, construction, and which model wins on which test. The argument has been unusually productive.
It seeded a bigger project. The same team went on to publish "Replicating Anomalies," a far larger replication effort that hit the literature even harder.

The honest limitations

Momentum is still a problem. The q-factor model does not have a momentum factor, and while the profitability factor picks up some of it (winners tend to be profitable), the model struggles to fully price momentum-sorted portfolios. Momentum remains the anomaly that refuses to be digested.
The theory is elegant but the measurement is not. Real firms invest for tax reasons, agency reasons and empire-building reasons that q-theory does not model. The clean derivation and the messy accounting data are not the same object.
The construction choices are load-bearing. Later work has argued that some of the model's advantage comes from specific decisions about how the factors are built, particularly the use of quarterly rather than annual earnings data. Reasonable people build these factors differently and get different results.
"Explaining" an anomaly is a low bar. Saying the q-factor model prices an anomaly means the anomaly's return is captured by exposure to investment and profitability. It does not prove the underlying economics; a correlated factor can absorb an effect without explaining it.

The one-line takeaway

Hou, Xue and Zhang started from the theory of how firms decide to invest, derived that low-investment and high-profitability companies should earn higher returns, and showed that a four-factor model built on those two predictions absorbs most of the published anomaly zoo, while roughly half that zoo turns out not to have been there in the first place.