Stocks

Asset Growth Anomaly Study Finds Nonlinear Return Patterns Linked to Mispricing

Oct 02, 2026 • 5 min read • views

The familiar pattern of low asset-growth firms outperforming high asset-growth firms may conceal two different return relationships. A July 2026 study by Zhuo Cheng, Daniel Cohen, and Jing Fang finds that asset growth and subsequent returns form a U shape for one group of firms and an inverted U for another. The authors argue that this structure fits a mispricing explanation better than a simple risk premium.

The finding matters because asset growth features in two major asset-pricing frameworks: the conservative-minus-aggressive (CMA) factor in the Fama-French five-factor model and the investment factor in the Hou-Xue-Zhang q-factor model. Both frameworks interpret investment through a risk-based lens. The new study asks whether the observed return pattern instead reflects prices correcting in stocks that are difficult to value and trade.

A challenge to the investment-factor interpretation

In the Fama-French account, asset growth serves as a proxy for expected growth in book equity. Under the dividend discount model, stronger expected growth, all else equal, corresponds to lower expected returns. Hou, Xue, and Zhang draw on q-theory: firms invest less when their cost of capital is high, so investment can signal discount rates.

Cheng, Cohen, and Fang build on a challenge raised in a 2024 study, “The Use of Asset Growth in Empirical Asset Pricing Models,” by Michael Cooper, Huseyin Gulen, and Mahai Ion. That research reported that the benchmark models lost much of their explanatory power when investment factors used more conventional or intangibles-inclusive measures. Growth in inventory and accounts receivable, in particular, carried much of the pricing information associated with the asset-growth factor.

That result raises a basic question: does investment itself explain the returns, or do other features of firms with extreme asset growth drive the pattern? The July study, “What drives the Asset Growth-return Relation: Mispricing or Risk?”, examines that question by looking beyond the average slope across all firms.

Why the relationship may bend at both ends

Firms with very high or very low asset growth share traits that can make them hard to assess. Compared with firms in the middle of the asset-growth distribution, they tend to be smaller, younger, more research-and-development-intensive, and more operationally levered. Their earnings and cash flows are more volatile, their accounting information is less predictable, and their accrual estimates are more prone to error.

The study’s comparisons across asset-growth quintiles add evidence of valuation and trading frictions. Firms at both extremes show greater analyst forecast dispersion, larger forecast errors, and higher return volatility. They also have lower institutional ownership, thinner analyst coverage, and wider bid-ask spreads. Those features can make mispricing harder to identify and more costly to trade against.

The authors also report a U-shaped pattern in price-to-value estimates from two valuation methods: estimated mispricing is more pronounced at both ends of the asset-growth distribution than in the middle. Their interpretation leads to a testable prediction. Among undervalued firms, returns could be higher at both extremes as prices correct upward, producing a U shape. Among overvalued firms, returns could be lower at both extremes as prices adjust downward, producing an inverted U.

Separating two return patterns

The main empirical hurdle is that researchers cannot directly observe which firms are undervalued and which are overvalued. Cheng, Cohen, and Fang address this with finite mixture normal regression, a statistical method that identifies groups with distinct patterns in the data while estimating a separate return relationship for each group.

They apply the method to more than 172,000 firm-year observations spanning 1963 to 2022. The model favors two groups over one: measures of fit improve substantially, and the Bayesian Information Criterion falls by more than 10,000. That result indicates strong statistical support for a two-group structure under the criteria the authors report.

In the first group, next-year returns follow a U-shaped relationship with asset growth. Returns are highest among firms with the sharpest asset contraction and the fastest asset expansion, and lowest in the middle quintiles. In the second group, the relationship is inverted: returns are lowest at the extremes and least negative in the middle.

Both patterns are asymmetric. The downward-sloping portion is steeper than the upward-sloping portion. When the groups are combined, that imbalance produces the familiar negative relationship between asset growth and returns. A standard single-group regression therefore shows a monotonic decline across asset-growth quintiles, even though the separate group-level patterns curve in opposite directions.

Evidence against a simple risk account

The authors test whether the curved relationships disappear after controlling for risk. They adjust returns for exposure to 10 established factors, including CMA, momentum, and four factors designed to capture mispricing and behavioral effects. The U-shaped and inverted-U patterns, along with their asymmetry, remain largely intact after this adjustment.

A second check connects the model’s groups to valuation. Firms with higher price-to-value ratios, which are more likely to be overvalued under the study’s measure, are significantly more likely to fall into the inverted-U group. That association supports the authors’ interpretation that the two statistical groups reflect different valuation conditions.

The results also vary with the cost of arbitrage. The nonlinear patterns are stronger among stocks with higher idiosyncratic volatility, which the authors use as a proxy for greater difficulty in arbitraging mispricing. This aligns with the reported links between extreme asset growth, thin analyst coverage, low institutional ownership, and wider bid-ask spreads.

Additional checks produce similar patterns when the authors use deciles rather than quintiles, try alternative asset-growth measures—including measures discussed in the Cooper, Gulen, and Ion study—trim extreme returns, exclude firms without a full 12 months of returns, and divide the sample into three 20-year periods. The authors report that the relationship does not fade across those subperiods.

What the findings change

The paper’s central implication is about interpretation. Realized returns combine the return investors expect with gains or losses that arise as market prices move toward estimated value. If mispricing correction accounts for much of the asset-growth pattern, the historical returns associated with low asset growth may not represent compensation for bearing a priced risk.

That distinction affects how factor evidence should be read. A factor can summarize a persistent return pattern without establishing why it exists or whether the same source of return will persist. The study suggests that a single average slope may hide distinct behavior among firms whose valuations are especially uncertain.

For market participants, the practical signal is that extreme expansion and sharp contraction both coincide with characteristics that complicate valuation and trading. The study doesn’t establish that every company at either end is mispriced, nor does it show that the patterns can be captured after trading costs. It does show why asset-growth exposure may carry more than one economic interpretation—and why the composition of a factor strategy deserves close attention.

The authors’ evidence leaves a broader empirical lesson: return patterns can look linear in aggregate while arising from sharply different relationships within subgroups. For asset growth, the next question is how reliably those groups can be identified and whether the resulting patterns remain after real-world trading frictions.

Background reading: Enrich Your Future. The source article was originally published at Alpha Architect; see its disclosures.

Source: Larry Swedroe · alphaarchitect.com