Since the privatisation of BT in the early 1980s the CAPM has been the preeminent approach to determining a regulated company’s cost of capital. Like any model it has well recognised limitations but still allows an objective quantification that captures differences in the systematic risk across different sectors and companies.
Over the course of subsequent regulatory reviews various reformulations have been proposed: Fama French three factor models, Arbitrage Pricing Theory (APT) models, and extensions to take account of asymmetric risk and long tail risk (3rd and 4th moments). None of these have been adopted, essentially because of the trade-off between the challenges of model calibration and the importance (or lack of it) that these additional factors would contribute.
Whilst there's a danger in always seeing the current situation as unique, the current long term economic prognosis exhibits significant emerging systematic risks – notably around climate change and future pandemics - that would be difficult, if not impossible, for investors to diversify.
Forward looking applications of the CAPM capture investor expectations of how these evolving risks impact the cost of capital on a single dimension of market risk. However exposure to climate change (for example) cuts across industries and geographies in different ways to the overall market risk that is the focus of the CAPM.
The important point is that in the same way as CAPM market risk cannot be diversified, neither can the specific risk of climate change or pandemic. Exposure to these three fundamental risks will vary by sector and geography but none of them can be diversified away.
The standard CAPM will capture these other sources of systematic risk to the extent that they correlate with market risk, but will do this incompletely and so other non-diversifiable systematic risks will remain. The consequence will be that the true cost of capital will be over-estimated in sectors that are under-exposed to the additional underlying risk factors and under-estimated in sectors that are over-exposed.
Calibrating multifactor models will always be difficult because of the lack of relevant empirical data. But at least, qualitatively, it may be worth recognising the direction of these possible biases.