The q-Factor Model

Fama and French built their factors by empirically sorting stocks on ratios that seemed to predict returns and calling the resulting long-short portfolios "risk factors." Hou, Xue, and Zhang took a different route: start from a firm's investment decision, use basic corporate-finance theory to derive what expected returns should look like, and only then check the data. The result, the q-factor model, competes with Fama-French using fewer, theoretically-motivated ingredients and, on many tests, fits at least as well.

The analogy

Two engineers are asked to design a bridge's safety margin. One studies a hundred existing bridges, notices the ones with thicker cables tend to survive storms better, and writes down "thicker cables, higher margin" as an empirical rule. The other starts from the physics of tension and load, derives what cable thickness should predict survival, and only then checks it against the hundred bridges. Fama-French is the first engineer; the q-factor model is the second. Both can end up with similar rules, but one is grounded in a causal mechanism, the other in pattern-matching.

Building the model

Investment theory (specifically, a firm's first-order condition for optimal investment, "q-theory") implies that a firm's expected stock return should rise with its profitability and fall with its investment rate, holding the other fixed: a highly profitable firm that reinvests little is creating value faster than the market is pricing in, while an unprofitable firm expanding aggressively is a red flag. That theory motivates exactly four factors:

$R_i - R_f = \alpha_i + \beta_{\text{Mkt}}(R_m - R_f) + \beta_{\text{Me}}\,r_{\text{Me}} + \beta_{\text{Ia}}\,r_{\text{Ia}} + \beta_{\text{Roe}}\,r_{\text{Roe}} + \epsilon_i.$

Here $r_{\text{Me}}$ is a size factor (small minus big market equity), $r_{\text{Ia}}$ is an investment factor (low minus high asset growth), and $r_{\text{Roe}}$ is a profitability factor (high minus low return on equity), each sorted and rebalanced monthly rather than annually. In words: this model claims a stock's expected return is fully pinned down by four things, market exposure, size, how disciplined the firm's reinvestment is, and how efficiently it turns equity into profit, with no separate role for a value factor at all.

Think of this curve as a stand-in for q-theory's core prediction: expected return rising with profitability at a decreasing rate, the theoretical shape the ROE factor is built to capture, rather than an empirically-discovered pattern with no underlying mechanism.

Worked example 1: a "value trap" reclassified

A stock has a high book-to-market ratio (looks cheap by Fama-French's value sort) but is expanding assets rapidly with weak ROE. Under Fama-French, its HML loading is strongly positive, so it inherits value's premium in the model's prediction. Under the q-factor model, its high investment loading and low ROE loading both push its predicted return down: $\beta_{\text{Ia}} = 0.6$ against a factor realization of $-2\%$ contributes $-1.2\%$, and $\beta_{\text{Roe}} = -0.4$ against $1.5\%$ contributes $-0.6\%$, a combined $-1.8\%$ drag Fama-French's HML-only view misses. This is exactly the "value trap" pattern, cheap on price, weak on fundamentals, q-theory is built to flag.

Worked example 2: comparing predicted returns

A profitable, conservative firm has ROE factor loading $\beta_{\text{Roe}} = 0.5$ and investment factor loading $\beta_{\text{Ia}} = 0.4$ (low investment, so a positive loading on "low minus high"). In a month where $r_{\text{Roe}} = 2\%$ and $r_{\text{Ia}} = 1\%$: contribution $= 0.5(2\%) + 0.4(1\%) = 1.0\% + 0.4\% = 1.4\%$, on top of its market and size exposure. A comparably-sized but unprofitable, aggressively-expanding firm with $\beta_{\text{Roe}} = -0.3$, $\beta_{\text{Ia}} = -0.5$ gets $-0.3(2\%) - 0.5(1\%) = -0.6\% - 0.5\% = -1.1\%$. The 2.5-percentage-point predicted gap between the two firms comes entirely from profitability and investment, no value factor was needed.

What this means in practice

Quant desks running factor attribution increasingly test both models side by side; when a q-factor regression's alpha for a strategy is near zero but Fama-French's is not (or vice versa), the discrepancy is diagnostic of which underlying mechanism, empirical pattern or investment-theory-driven pricing, is actually driving the strategy's returns.