Table IX is the equivalent of Table VIII, except that it is constructed for the utilities industry and based on the utility-specific prior. The results for Bay State Gas in Panel A are quite similar to those obtained with the all-stock prior. In Panel B, which reports averages across the 135 utility stocks, both model uncertainty and overall uncertainty about the expected excess return are smaller for utilities than for the whole cross-section.

In particular, even with aQ — 0, the model uncertainty for the pairing of CAPM and CK is only 0.74%, which is consistent with the close correspondence between the estimates of expected excess returns from those models displayed in Figure 9. Despite some differences in magnitude, the relative proportions of model uncertainty and overall uncertainty are similar to those observed in Table VIII. Thus, in the utilities industry, uncertainty about which model to use again appears to be less important than uncertainty about the parameters within a given model.


Costs of equity capital implied by factor-based pricing models can be estimated in a Bayesian setting. After using the available data, a decision maker possesses uncertainty about a firm’s cost of equity that is characterized by the posterior standard deviation of fi, the expected excess return on the firm’s stock. The posterior standard deviation of ц is typically at least 3% per year in a one-factor model and 4% per year in a three-factor model, even if the possibility that the model might misprice the stock is ruled out a priori.

For utilities, this standard deviation is smaller but generally at least 2% per year. Uncertainty about a pricing model’s potential mispricing of the stock (a) increases the uncertainty about ц, but the posterior mean of ц—the decision maker’s estimate of the expected excess return—is generally not affected greatly by uncertainty about a. When a decision maker is uncertain about which factor-based model to use, the estimate of the stock’s expected excess return is then a weighted average of estimates from different models.

The model uncertainty associated with that estimate is nontrivial, typically adding another 0.7% to the overall posterior standard deviation of ц, but the model uncertainty on average is less than the within-model parameter uncertainty. Electronic Payday Loans Online