Why Segment Exposure Matters
A diversified company’s beta reflects the combined market risk of its operating businesses. That companywide beta is not automatically a suitable discount-rate input for any one division. A division concentrated in a stable line of business can have materially different operating risk from another division in the same firm. The estimation task is to separate the operating exposures embedded in multi-segment comparables before assigning a rate to the division.
The key is to use each comparable’s business-line weights as well as its aggregate asset beta. A comparable with several segments supplies a weighted observation of segment betas, not a pure observation of any one segment. Across firms with different segment mixes, these weighted observations can identify the underlying exposures—provided the mixes vary enough and the segment definitions are consistent.
The weights should represent relative operating-asset values, not merely revenue shares. Revenue can be a practical proxy when segment values are unavailable, but it is not equivalent: segments with different margins, asset intensity, or growth prospects may have very different values per dollar of revenue. Use market values when they can be estimated credibly, allocate corporate assets consistently, and document any proxy used.
The firm-level asset betas used as observations must also be comparable. They should reflect operating assets rather than a mixture of operating and nonoperating assets, and financing risk should be removed using an appropriate unlevering method. The earlier material on beta unlevering and cash adjustments provides the relevant tools. Do not treat a raw equity beta as an asset beta simply because a firm reports multiple segments.
From Segment Mixes to Segment Betas
For two segments, X and Y, the model for firm \(j\) is \(\beta_{A,j}=w_{jX}\beta_{A,X}+w_{jY}\beta_{A,Y}\). With several firms, the corresponding equations form a system. If there are more firms than unknown segment betas, the system will usually not fit every observed firm beta exactly. Least squares estimates the segment betas that make the total squared discrepancy as small as possible.
Here \(\boldsymbol{\beta}_A\) is the vector of comparable firms’ asset betas, \(W\) is the matrix of segment weights, \(\boldsymbol{\beta}_S\) is the vector of unknown segment betas, and \(\boldsymbol{\varepsilon}\) is the residual vector. The displayed ordinary least-squares formula assumes the model has no intercept, the weights are measured without material error, the error terms have mean zero and equal variance, and \(W\) has full column rank. Full column rank means that the segment mixes contain enough independent variation to distinguish the segment exposures.
When firms have unequal estimation precision, weighted least squares can give more influence to more precise observations. If \(Q\) is a diagonal matrix of observation-precision weights, the estimate is \((W^\mathsf{T}QW)^{-1}W^\mathsf{T}Q\boldsymbol{\beta}_A\). These precision weights are not the segment operating-asset weights: \(W\) describes each firm’s business mix, while \(Q\) describes the reliability assigned to its beta observation.
Why Least Squares Gives This Estimate
The method follows from minimizing the squared residuals. For any proposed segment-beta vector \(\boldsymbol{b}\), the residual vector is \(\boldsymbol{\beta}_A-W\boldsymbol{b}\). The objective is the sum of its squared entries. Differentiating that objective with respect to \(\boldsymbol{b}\), setting the derivative equal to zero, and solving gives the least-squares estimate:
If \(W\) has full column rank, \(W^\mathsf{T}W\) is invertible, so the normal equations have the unique solution shown above. If every comparable has the same segment proportions, the columns of \(W\) are not independently informative. The data may still identify a weighted average beta, but cannot separate each segment beta without additional information.
Worked Example: Solving for Two Segment Betas
Four hypothetical diversified firms have been adjusted to comparable asset betas. Each operates in segments X and Y. Their segment X operating-asset weights are 0.80, 0.60, 0.40, and 0.20; segment Y makes up the remainder. Their asset betas are 0.90, 1.10, 1.30, and 1.50. Estimate the two segment betas and check the unused observations.
Method and assumptions. Use the two-segment weighted-beta model, treating the stated weights and asset betas as comparable inputs. First solve using the firms with the most different mixes. For those firms, the equations are \(0.80\beta_{A,X}+0.20\beta_{A,Y}=0.90\) and \(0.20\beta_{A,X}+0.80\beta_{A,Y}=1.50\).
Multiply both equations by 5 to make the coefficients whole numbers: \(4\beta_{A,X}+\beta_{A,Y}=4.50\) and \(\beta_{A,X}+4\beta_{A,Y}=7.50\). The first equation gives \(\beta_{A,Y}=4.50-4\beta_{A,X}\). Substitute into the second:
Then \(\beta_{A,Y}=4.50-4(0.70)=1.70\). Check the two other firms. At weights 0.60 and 0.40, the predicted beta is \(0.60(0.70)+0.40(1.70)=0.42+0.68=1.10\). At weights 0.40 and 0.60, it is \(0.40(0.70)+0.60(1.70)=0.28+1.02=1.30\). Both predictions match the stated observations.
Result and trap avoided. The estimates are 0.70 for X and 1.70 for Y. A common mistake is to assign the average of the four company betas, 1.20, to either division. That average describes neither segment here; each firm’s beta reflects its particular mix.
Residuals and Estimation Error
A close fit does not make the estimates certain. Residuals measure the discrepancies between observed firm betas and the values predicted by the fitted segment model. They can arise from beta-estimation noise, inaccurate segment weights, inconsistent segment classifications, omitted businesses, or a failure of the weighted-beta model. A residual is therefore a diagnostic, not a direct measure of the standard error of an estimated segment beta.
Under the ordinary least-squares assumptions, an estimate of the error variance is the residual sum of squares divided by the residual degrees of freedom. With \(n\) comparable firms and \(K\) estimated segment betas, the degrees of freedom are \(n-K\), provided the model has full column rank and no intercept. The coefficient covariance estimate is this residual variance times \((W^\mathsf{T}W)^{-1}\). These calculations quantify uncertainty under the model; they do not capture every possible error in the input weights or model specification.
Worked Example: Reading Residuals and Standard Errors
Use the same four segment mixes as in the preceding example, but suppose the four observed asset betas are 0.92, 1.08, 1.28, and 1.52. Fit the no-intercept two-segment model by ordinary least squares, then estimate the residual standard deviation and the standard errors of the two segment-beta estimates. Assume independent residuals with a common variance.
The fitted betas remain 0.70 for X and 1.70 for Y. To verify this, the predictions from those coefficients are 0.90, 1.10, 1.30, and 1.50. The residuals, observed minus predicted, are therefore \(0.02,-0.02,-0.02,0.02\). Their squared values are all \(0.0004\), so the residual sum of squares is \(4(0.0004)=0.0016\).
There are four observations and two fitted segment betas, leaving \(4-2=2\) residual degrees of freedom. The estimated residual variance is \(0.0016/2=0.0008\), and the residual standard deviation is \(\sqrt{0.0008}=0.0283\), rounded to four decimals.
For these weights, \(W^\mathsf{T}W\) has diagonal entries \(0.80^2+0.60^2+0.40^2+0.20^2=1.20\) and off-diagonal entries \(0.80(0.20)+0.60(0.40)+0.40(0.60)+0.20(0.80)=0.80\). Thus its inverse is:
The diagonal entries of the estimated coefficient covariance matrix are \(0.0008(1.50)=0.0012\). Each segment-beta standard error is therefore \(\sqrt{0.0012}=0.0346\), rounded to four decimals. The estimates are still 0.70 and 1.70, but the residual-based standard errors indicate uncertainty around them. The method avoids the mistake of interpreting a small residual standard deviation as zero uncertainty in the estimated segment betas.
Turning Segment Exposures into a Division Rate
If a division contains more than one operating segment, combine their asset betas using the division’s own operating-asset value weights. This produces an operating beta for that division. Use the division’s weights, not a comparable firm’s mix: the purpose of the comparable analysis is to estimate segment exposures that can then be applied to the division being valued.
Worked Example: Building a Levered Division Discount Rate
Suppose a hypothetical division is composed of 30% segment X and 70% segment Y by operating-asset value. Use the estimated segment asset betas 0.70 and 1.70. The division will maintain debt-to-equity \(D/E=0.50\); debt has beta zero and a pre-tax cost of 4.00%. The tax rate is 25%, the risk-free rate is 4.00%, and the market risk premium is 5.00%. Assume the CAPM is appropriate, interest is tax deductible, and the target leverage is maintained.
Step 1: Combine operating exposures. The division asset beta is \(0.30(0.70)+0.70(1.70)=0.21+1.19=1.40\). This is the division’s operating exposure before financing risk.
Step 2: Relever for the division’s financing. With zero-beta debt and tax-deductible interest, the relevering relation is \(\beta_E=\beta_A[1+(1-T)D/E]\). Therefore, \(\beta_E=1.40[1+(1-0.25)(0.50)]=1.40(1.375)=1.925\).
Step 3: Calculate the cost of equity. CAPM gives \(k_E=4.00\%+1.925(5.00\%)=4.00\%+9.625\%=13.625\%\).
Step 4: Calculate WACC. Since \(D/E=0.50\), choose \(E=2\) and \(D=1\) as proportional market values, giving \(E/(D+E)=2/3\) and \(D/(D+E)=1/3\). Then \(WACC=(2/3)(13.625\%)+(1/3)(4.00\%)(1-0.25)=9.0833\%+1.0000\%=10.0833\%\), or 10.08% to two decimal places.
Check and trap avoided. The division asset beta of 1.40 is between the segment betas 0.70 and 1.70, as it must be for positive weights summing to one. Applying CAPM directly to 1.40 would give an asset return of \(4.00\%+1.40(5.00\%)=11.00\%\), not the levered cost of equity. The financing step is necessary before computing the stated WACC.
Practical Checks Before Using the Estimate
A mathematically solvable system is not automatically a credible one. Before relying on a division rate, check both the structure of the data and the stability of the result.
Use asset betas on a consistent basis, with financing and nonoperating-asset adjustments made as appropriate. Do not combine raw equity betas with asset betas.
Map business lines consistently across firms. A segment label that groups different economic activities can make its estimated beta difficult to interpret.
Confirm that comparable firms have sufficiently different segment weights. Nearly identical mixes can make the system technically solvable but highly sensitive to small input changes.
Look for unusually large or patterned residuals and report coefficient uncertainty where the model supports it. Revisit the inputs or model if the fit is poor.
Use operating risk to estimate the division exposure, then apply the appropriate financing and tax assumptions for the discount rate being used.
A frequent exam trap is to confuse three distinct weights: a comparable’s segment weights, the target division’s segment weights, and the capital structure weights used in WACC. The first set helps estimate segment betas; the second combines those estimates for the target division; the third combines debt and equity costs. Each answers a different question and should be applied only at its corresponding step. This distinction is aligned with the SOA VEE Accounting and Finance syllabus.
Check Your Understanding
For each problem, state the relevant model and keep the segment-mix weights distinct from financing weights.
- A hypothetical comparable has segment X weight 0.35, segment Y weight 0.65, and asset beta 1.32. If the segment betas are 0.80 and 1.60 respectively, verify the aggregate beta. (Answer: \(0.35(0.80)+0.65(1.60)=1.32\).)
- Two diversified firms have X/Y weights 0.75/0.25 and 0.25/0.75, with asset betas 0.95 and 1.25. Solve for the segment betas. (Answer: \(\beta_{A,X}=0.80\), \(\beta_{A,Y}=1.40\).)
- A firm’s asset beta is 1.06 and its X/Y operating-asset weights are 0.60/0.40. If segment X beta is 0.70, solve for segment Y beta and check by substitution. (Answer: \(\beta_{A,Y}=1.60\); \(0.60(0.70)+0.40(1.60)=1.06\).)
- A division has 40% of its operating-asset value in a segment with beta 0.80 and 60% in a segment with beta 1.50. With risk-free rate 3.50% and market risk premium 6.00%, find its asset beta and CAPM asset return. (Answer: \(\beta_A=1.22\); asset return \(=3.50\%+1.22(6.00\%)=10.82\%\).)
- Three comparables all have X/Y weights 0.60/0.40, and each has asset beta 1.10. Can the two segment betas be estimated separately from these observations? State what the data do identify. (Answer: No; the identical mix gives rank one, so only \(0.60\beta_{A,X}+0.40\beta_{A,Y}=1.10\) is identified.)