By Andreas in momentum — 24 Aug 2026

Why Forecast Dispersion Sharpens a Revision Signal Rather Than Replaces It

A Research Note on forecast dispersion and its role in systematic momentum investing.

Two companies can carry the same analyst consensus and report the same upward revision, and yet the information behind that revision may differ substantially. In one, the individual estimates sit closely together. In the other, they are widely scattered. Attention usually falls on the direction and magnitude of a revision, while the question of how far the analysts agree receives less. That agreement is a distinct layer of information, but its value does not lie in signalling a price direction on its own. It lies in qualifying a signal that is already present.

What Dispersion Measures

Forecast dispersion is the spread of individual analyst estimates for the same figure and the same forecast horizon, most often measured as the standard deviation of the estimates around the consensus and scaled by price or by the level of earnings. It is a cross-sectional quantity at a point in time, and therefore structurally different from the direction and magnitude of a revision, both of which describe how the consensus moves over time. It is equally important to be clear about what dispersion is not. It is not the same as general uncertainty, not the same as the disagreement of all market participants, and not the same as the forecast error that is later realised. It is an observable proxy for one part of that disagreement, no more. Wide dispersion may reflect genuine differences of opinion, but it may equally reflect differing information sets, differing models, or a handful of outliers.

The quantity was established as a return predictor above all by Diether, Malloy and Scherbina, who documented a negative relationship between the dispersion of analyst estimates and subsequent stock returns. In their study, stocks with higher dispersion earned lower average returns than those with lower dispersion. The appropriate summary, however, is not that high dispersion is always bad and low dispersion always good. A negative relationship has been documented across several studies, but its strength and stability depend on the measurement definition, the market, the period, analyst coverage and the controls that are applied. It is therefore better described as a documented dispersion effect than as a law of nature.

Risk versus Mispricing

Two explanations contest the direction of this relationship. A risk-based explanation expects high disagreement to be accompanied by higher expected returns. Its reasoning is that dispersion stands for uncertainty about the true value, and investors who bear that uncertainty require compensation for it. This prediction runs against what is observed on average, though that does not refute the risk view, since dispersion may simply be an incomplete or poorly specified proxy for risk. The mispricing-based explanation goes back to Miller and predicts the opposite. On his argument, the price reflects a more optimistic valuation when pessimistic investors cannot fully act because of short-selling constraints, so that high disagreement can be associated with an overvaluation that later corrects through lower returns. This direction, lower returns where disagreement is higher, is consistent with Miller's mechanism but does not prove it. The same empirical direction can also arise from forecast errors, differences in liquidity, or unobserved firm characteristics. Miller is therefore best treated as a plausible economic interpretation rather than a conclusively identified cause.

The Limits of Dispersion as a Standalone Signal

The most common misunderstanding is to turn this finding into a fixed rule, of the form buy low dispersion and avoid high dispersion. The relationship is not stable but state-dependent, and it varies with investor sentiment, short-selling constraints and the specific construction of the measure. Raw dispersion also mixes several economically distinct things at once, namely genuine disagreement, systematic analyst bias and simple noise, which need not bear the same relationship to future returns. A recent review by Goulding, Harvey and Kurtović from June 2026 shows that the leading disagreement measures are only weakly correlated with one another in the cross section of firms, so that empirical conclusions can depend noticeably on which measure is chosen, and it surveys efforts to combine these measures into a more disciplined quantity. The conclusion is not that dispersion is worthless, but that it is unreliable as a standalone signal. An unconditional rule would therefore overstate what the evidence supports.

Dispersion as a Conditioning Variable

The workable use follows from this. Dispersion realises its value where it qualifies another signal rather than standing on its own. The relevant question is not whether dispersion predicts returns in isolation, but whether it changes the informational content of a revision signal. An upward revision at low dispersion may indicate that the move in the consensus is not carried by a few extreme estimates alone, whereas the same move at high dispersion may be more fragile, because the individual estimates remain far apart. This reading should be stated with care. Low dispersion does not prove that a revision is correct, since a narrow consensus can be collectively wrong. It merely indicates that the consensus move is less shaped by outliers, and so provides an additional input for assessing the quality of the signal rather than proof of forecast accuracy. For that reason the central claim concerns the interaction between revision and dispersion, not the dispersion effect on its own. Put this way, it is an empirical hypothesis that can be tested, not a property that follows from the definition of the measure.

Dische offers an early indication in this direction, showing that an earnings momentum strategy improved in his study when it was applied to stocks with low dispersion. Whether that effect is stable across other periods and universes, causal, and investable after costs remains a separate question that a single finding does not settle.

Conclusion

Forecast dispersion is an economically plausible and empirically tractable layer of information, but its content is conditional rather than absolute. A negative relationship between high analyst dispersion and subsequent returns has been documented more than once, yet its strength, stability and interpretation depend on the measurement method, the market environment and the controls for alternative explanations. The more convincing use lies not in an isolated dispersion effect but in whether dispersion alters the informational content of another signal, so that an upward revision at low dispersion may prove more consistent than the same revision at high dispersion. That remains an empirical hypothesis, best examined through the interaction between revision and dispersion rather than asserted. The disciplined systematic investor therefore treats dispersion neither as mere noise nor as a standalone criterion for buying or avoiding a security, but as a context variable that calibrates the strength of an existing signal, which in the end is a matter of process discipline.

Academic References

Diether, K. B., Malloy, C. J., & Scherbina, A. (2002). Differences of opinion and the cross section of stock returns. The Journal of Finance, 57(5), 2113–2141.
Miller, E. M. (1977). Risk, uncertainty, and divergence of opinion. The Journal of Finance, 32(4), 1151–1168.
Yu, J. (2011). Disagreement and return predictability of stock portfolios. Journal of Financial Economics, 99(1), 162–183.
Goulding, C. L., Harvey, C. R., & Kurtović, H. (2026). Measuring investor disagreement: Proxies, pitfalls, and a path forward. SSRN Working Paper No. 6915539. https://doi.org/10.2139/ssrn.6915539

Practitioner References

Dische, A. P. (2001). Dispersion in analyst forecasts and the profitability of earnings momentum strategies. SSRN working paper.
AQR Capital Management. (2014). Fact, fiction, and momentum investing.