Pooling Comparable Asset Betas

Why Pool Comparable Asset Betas?

A project discount rate should reflect the risk of the project’s operating cash flows, not simply the risk of the company that happens to undertake it. A company’s observed equity beta also reflects its financing. The earlier material on unlevering beta and screening pure-play comparables provides the ingredients for addressing that distinction: convert suitable comparable-company estimates to a common asset-beta basis, then combine them to estimate the project’s operating risk.

Pooling does not make imperfect comparables identical. The companies may differ in products, customers, geography, operating leverage, or the quality of their regression estimates. The purpose of pooling is to summarize relevant evidence, not to hide those differences behind one precise-looking number. In this tutorial, beta estimates are hypothetical, and all asset betas used in the examples have already been unlevered using consistent assumptions.

Pooling should come after comparability and measurement checks. For a given peer set, the simple mean is \( \bar{\beta}_A = \frac{1}{n}\sum_{j=1}^{n}\beta_{A,j} \), where \(n\) is the number of selected companies and \(\beta_{A,j}\) is the asset beta for company \(j\). To find the median, sort the estimates from smallest to largest and identify the central value or values. Neither measure is automatically the correct answer in every setting.

Make the Estimates Comparable Before Pooling

An arithmetic operation is only useful if the inputs are on a comparable basis. A peer set should therefore be screened for both business relevance and measurement consistency. The operating businesses should have enough in common with the project for their systematic-risk estimates to be informative. In addition, the betas should use compatible estimation periods, market benchmarks, and unlevering conventions.

Earlier material covers regression diagnostics and the adjustment from equity beta to asset beta. Here the practical implication is that an analyst should not pool raw equity betas when the companies have different leverage. Nor should the analyst silently combine asset betas calculated using materially different assumptions about debt risk or taxes. If inconsistent inputs cannot be reconciled reliably, report the limitation and use a better-supported set rather than implying that the estimates are directly comparable.

A small peer group is not necessarily better than a larger one. A peer with a materially different business can add noise, while excluding a relevant peer solely because its beta is inconvenient can bias the result. Assess each candidate using evidence such as its revenue sources, operating activities, customer base, and the stability and precision of its beta estimate. A high or low estimate is a reason to investigate; by itself, it is not proof that the company is unsuitable.

Mean, Median, and the Effect of an Extreme Estimate

The mean uses every estimate and is sensitive to extreme values. The median is less affected by an estimate far from the rest, but it also does not reflect how far that estimate lies from the center. Thus, the mean can be informative when all peers are reasonably comparable and no estimate dominates; the median can be a useful robust summary when a few estimates are unusually distant and the peer set remains meaningful.

The reason for the mean’s sensitivity can be seen directly: if one estimate increases by \(c\), while all others remain unchanged, the mean increases by \(c/n\). The median may not change at all if that estimate remains outside the central position or positions. This robustness is useful, but it does not establish that the median is the true project beta. A poorly selected peer group can make either summary misleading.

Why the Two Summaries Behave Differently

The mean and median have distinct mathematical roles. For the mean, let \(m=\bar{\beta}_A\). The sum of squared deviations from any proposed center \(a\) can be written as \( \sum_{j=1}^{n}(\beta_{A,j}-a)^2 \). It is minimized at \(a=m\): differentiating with respect to \(a\) gives \( -2\sum_{j=1}^{n}(\beta_{A,j}-a)=0 \), so \(a=\sum_j\beta_{A,j}/n=m\). This is why the mean is a natural center when squared deviations are the chosen criterion, but large deviations receive a large squared penalty.

For absolute deviations, the corresponding criterion is \( \sum_{j=1}^{n}|\beta_{A,j}-a| \). A median minimizes this total. If the proposed center moves away from a middle observation, the number of observations on each side determines whether total absolute distance increases; a median has at least half the observations on either side. This makes the median less sensitive to the magnitude of extreme observations than the mean. These properties explain the summaries; they do not determine which peer estimates belong in the analysis.

Investigate Outliers Before Excluding Them

A value that is distant from the others is an outlier relative to the observed peer set, not automatically an invalid observation. Investigate whether it reflects a different operating business, an unusual period, a short or unstable return history, a weak regression fit, or another measurement issue. Also check for data or calculation errors. If the estimate is valid and the company is relevant, excluding it solely to obtain a preferred beta is not defensible.

When there is a clear, documented comparability or data-quality reason to exclude a peer, show the estimate both with and without it if the decision has a material effect. This makes the judgment visible and allows a reader to assess whether the project conclusion depends on a disputed observation. If exclusion is not justified, report the effect on the mean and median as sensitivity information rather than disguising the estimate’s influence.

Peer Selection Is an Economic Judgment

The peer set defines the evidence being summarized. Consider whether the project resembles each company’s core operating exposure, not merely whether the company has a similar name or industry label. For example, two businesses classified in the same broad sector may face different demand drivers or have materially different operating models. Conversely, a company in a different formal industry may have a closely comparable operating activity.

Apply the same screening standard to every candidate. Record the reason for inclusion or exclusion, and distinguish a business-comparability decision from a statistical-estimate concern. A useful analysis may show a core peer set and a separate sensitivity case with plausible adjacent peers. It should not combine the two sets without explaining what changed.

Equal weighting is a transparent default for a simple peer mean: each comparable contributes one estimate. Weighting companies by size is not automatically better. A size-weighted estimate answers a different question and can allow the largest company to dominate, even if it is less comparable to the project. Use a different weighting rule only when the objective and rationale are clear, and show how it affects the estimate.

Reporting a Pooled Estimate Responsibly

A pooled beta is an estimate, not a guarantee that project risk has been measured without error. A concise, reproducible analysis should state the peer-selection rule, the number of peers, the pooling statistic, and any material exclusions. When the mean and median differ meaningfully, explain whether the difference comes from an extreme estimate or a skewed peer distribution. If a reasonable alternative peer set changes the result, report that sensitivity rather than implying undue precision.

The pooled asset beta can be used as an estimate of project operating risk in a later cost-of-capital calculation, subject to the assumptions and limitations of the comparable-company analysis. It should not be confused with the project’s equity beta after financing. The appropriate capital structure and the steps for building a project-specific discount rate are separate decisions.