What Really Drives the Asset Growth Anomaly? New Evidence Points to Mispricing, Not Risk

One of the most well-documented patterns in the cross-section of stock returns is that firms with high asset growth subsequently underperform firms with low asset growth. This “asset growth anomaly” is so well established that it now sits at the core of two of the most widely used benchmark factor models: the Fama-French five-factor model (via the conservative-minus-aggressive, or CMA, factor) and the Hou-Xue-Zhang q-factor model (via its investment factor).



Both models offer a risk-based, rational explanation. Fama and French argue that asset growth proxies for expected book equity growth, and the dividend discount model implies that higher expected growth (holding other things equal) should come with lower expected returns. Hou, Xue, and Zhang ground their version in q-theory: firms invest less when the cost of capital is high, so low investment signals a high discount rate and vice versa.



Zhuo Cheng, Daniel Cohen, and Jing Fang, authors of the July 2026 study “What drives the Asset Growth-return Relation: Mispricing or Risk?” challenge this rational, risk-based story directly. The paper is motivated by a critique from the 2024 study “The Use of Asset Growth in Empirical Asset Pricing Models,” by Michael Cooper,  Huseyin Gulen, and Mahai Ion who found that both benchmark models lose much of their explanatory power once the investment factor is built from more conventional or intangibles-inclusive measures of investment— factors based on growth in inventory and accounts receivable contain the bulk of the pricing information in the asset growth factor. That finding undercuts the premise that a negative link between investment and expected returns is really doing the work — and motivated the authors to investigate further.



The Core Idea: Mispricing at Both Extremes



The authors’ argument starts with the observation that firms at both extremes of the asset growth distribution — those shrinking sharply and those growing explosively — share a set of characteristics that make them hard to value. They tend to be smaller, younger, more R&D-intensive, and more operationally levered than firms with moderate asset growth. That combination produces more volatile earnings and cash flows, less predictable accounting information, and larger accrual estimation errors.



The paper backs this up with an extensive set of firm-characteristic comparisons across asset growth quintiles. Firms in the top and bottom quintiles show wider analyst forecast dispersion, larger forecast errors, and higher return volatility than firms in the middle. They are also harder to arbitrage and have less institutional ownership, thinner analyst coverage, and wider bid-ask spreads. Together, this is a recipe for mispricing that is both more severe and slower to correct — a U-shaped relation between asset growth and mispricing severity, which the authors confirm directly using price-to-value estimates built from two well-established valuation methodologies.



If mispricing is more severe at both tails of the asset growth distribution, and realized returns contain a meaningful mispricing-correction component, then the asset growth-return relation shouldn’t be a simple straight line. It should look different depending on whether a firm is undervalued or overvalued: U-shaped for undervalued firms (returns snap back up at both extremes as undervaluation corrects) and inverted U-shaped for overvalued firms (returns fall further at both extremes as overvaluation corrects).



The Methodological Wrinkle — and How They Solved It



Testing this hypothesis runs into the problem that you can’t observe which firms are undervalued and which are overvalued. The authors get around this by using finite mixture normal regression (FMNR), an unsupervised statistical technique that simultaneously sorts firms into latent groups (a collection that shares a common characteristic, or underlying pattern but are not directly observable) and estimates a separate return relation within each group — without requiring the researcher to specify in advance who belongs where. Using data on more than 172,000 firm-year observations from 1963 to 2022, they let the model find its own structure.



The result: the data clearly prefer two latent groups over one. Model fit — measured by log-likelihood, Akaike Information Criterion (a simple score that helps you pick the best model from a set of candidates by balancing how well the model fits the data against how complicated the model is), and Bayesian Information Criterion (similar in spirit to AIC but with a stronger penalty for complexity, especially when you have a lot of data) — improves dramatically when moving from a single group to two groups, with the BIC declining by more than 10,000, far past the threshold that constitutes decisive statistical evidence for the more complex model.



What the Two Groups Look Like



In the first latent group, the relation between asset growth and next-year returns is clearly U-shaped: returns are highest for the most extreme asset-shrinkage and asset-growth quintiles, and lowest in the middle. In the second latent group, the pattern flips into an inverted U: returns are lowest at the extremes and least negative in the middle.



Critically, in both groups, the pattern is asymmetric. The downward-sloping segment is steeper than the upward-sloping segment in both the U-shaped and inverted U-shaped groups. That asymmetry is what generates a net negative asset growth-return relation overall — the familiar anomaly — even though the underlying relation within each group is nonlinear rather than a simple downward slope. Run the whole sample through a standard single-group regression (the way most of the existing literature does it) and you get exactly the well-known monotonic decline in returns across asset growth quintiles. The nonlinearity and the two-regime structure are invisible unless you go looking for them.



Ruling Out the Risk Explanation



The authors run three additional tests designed to distinguish a mispricing story from a risk-based one:



Risk adjustment. They strip out each firm’s exposure to 10 established risk factors — including CMA itself, along with momentum and four factors explicitly built to capture mispricing and behavioral effects. If the pattern were really compensation for risk, this adjustment should make it disappear. It doesn’t: the U-shaped and inverted U-shaped patterns, and their asymmetry, survive largely intact.



Price-to-value validation. Using the same price-to-value estimates from their earlier analysis, they show that firms with higher price-to-value ratios (more likely overvalued) are significantly more likely to be assigned by the model to the inverted-U group — exactly what the mispricing story predicts, and a useful external check on whether the statistically-identified groups actually correspond to over- and under-valued firms.



Arbitrage risk. Using idiosyncratic volatility as a proxy for how costly a stock is to arbitrage, they show the nonlinear patterns are more pronounced among high-volatility, harder-to-arbitrage stocks — again consistent with mispricing that persists because it’s expensive to correct.



The findings also hold up across a long list of robustness checks: decile instead of quintile sorts, alternative and more conventional measures of asset growth (including several used in the Cooper, Gulen, and Ion critique), trimming extreme return observations, excluding firms without a full 12-month return, and splitting the sample into three 20-year subperiods. The pattern doesn’t fade over time, which matters given evidence that many anomalies have weakened as markets have become more efficient.



Why This Matters



The authors’ broader point cuts to the heart of a long-running debate in asset pricing: you cannot reliably infer the relation between a characteristic and expected returns just by looking at its relationship with realized returns. Realized returns bundle together an expected-return component with a mispricing-correction component, and this paper argues that for asset growth, the mispricing component dominates and is itself a nonlinear (U-shaped or inverted-U) function of asset growth. That means the well-known negative asset growth factor may be capturing return predictability that comes from price correction rather than from a genuine risk premium.



Investor Takeaways




The “asset growth effect” your factor tilt is capturing may not be a risk premium. If this paper’s interpretation holds up, part or all of the return associated with low-asset-growth stocks (as captured by factors like CMA) reflects the correction of mispricing in hard-to-value, hard-to-arbitrage stocks rather than compensation for bearing a priced risk. That’s a meaningfully different story for why the premium should be expected to persist, and for how confident investors should be in it repeating going forward.



Extreme asset growth or contraction is itself a signal of valuation uncertainty — in both directions. Firms shrinking rapidly and firms growing rapidly share a common profile: small, young, R&D-heavy, thinly covered by analysts, and expensive to trade. That’s worth remembering both when evaluating a factor strategy that leans hard into these names and when evaluating a specific stock at either extreme.



Don’t assume a single linear relation tells the whole story. The standard one-group regression used throughout the literature — and probably in most practitioner models — obscures the U-shaped and inverted-U patterns this paper uncovers. It’s a good reminder that a single average slope coefficient can mask sharply different behavior in different subgroups of the data, and that the “well-known” version of an anomaly is sometimes a statistical average of two very different underlying stories.



Costly arbitrage is a recurring theme across return anomalies, not just this one. The finding that mispricing patterns are strongest among high-idiosyncratic-volatility, low-institutional-ownership stocks echoes similar evidence found in the value, momentum, and other anomaly literatures. It reinforces a broader lesson for investors: pockets of apparent inefficiency tend to survive precisely where they are hardest and most expensive to exploit — which is also where transaction costs and implementation frictions are likely to eat into any strategy trying to capture them.




Larry Swedroe is the author or co-author of 18 books on investing, including his latest Enrich Your Future. He is also a consultant to RIAs as an educator on investment strategies. He has spent decades helping advisors and investors apply the evidence from academic finance to real-world portfolios. This article is for informational and educational purposes only and should not be construed as specific investment, accounting, legal, or tax advice.
What Really Drives the Asset Growth Anomaly? New Evidence Points to Mispricing, Not Risk was originally published at Alpha Architect. Please read the Alpha Architect disclosures at your convenience.