
While the measurement of risk is not an objective of the fairvalue framework, from the point of view of finance theory the starting point of any assetpricing approach is to postulate the existence of a tradeoff between risk and returns. From this initial postulate, the main question is to determine the appropriate measure of risk, from which a corresponding measure of expected returns, that is, the appropriate, riskadjusted discount rate, can be derived.
A second tenet of modern finance theory is that idiosyncratic or companyspecific risks cancel each other out in a welldiversified portfolio, and therefore only systematic risks should enter into pricing equations.
Hence, academic finance provides a framework for estimating investors' expected rates of return given common sources of risk found in financial assets. Expected returns predicted by statistically robust riskfactor models, taking into account current market conditions, provide a basis for the 'fair' discount rate that should prevail in the principal market.
Below, we briefly discuss standard factor models of expected returns, academic guidance to determine which factors should be used in such models, how they are estimated for assets traded in liquid markets where times series of prices and returns are readily available, and how this approach may be applied to seldomtraded private assets like unlisted infrastructure.
In other, we also discuss the relevance of risk factors identified in the academic literature to the valuation of infrastructure companies, and the statistical approach taken to estimate the value of factor prices in unlisted infrastructure.
The simplest model of expected returns is the wellknown capital asset pricing model (CAPM), according to which higher returns can only be achieved by exposing a welldiversified portfolio to higher market risk, also known as a higher market beta. This model serves as the basis for the discountrate formula described in equation .
Technically, the market beta is the coefficient of a single independent variable (the market return) in an ordinary leastsquare (OLS) regression explaining the variance of excess returns for a single asset (the dependent variable).
E(R_{i})R_f= \beta_{i}\left(E(R_{m})R_f\right)+\epsilon_{i} 
with as the market return, the riskfree rate, and the specific risk of investment , which by definition is not correlated with .
The CAPM implies that the 'market index' to which any asset relates is 'meanvariance efficient', that is, fully captures the asset's potential exposure to all systematic sources of risk. Empirically this prediction is not robust and has been proven inaccurate by multiple studies for listed equities.
Multifactor models have been developed to address the lack of statistical robustness of the singlefactor CAPM. Important work by Fama and French established multifactor models of asset returns as standard tools to create measures of expected returns.
Wellknown industry versions of these ideas that relate expected returns in the next period () to current exposures to risk factors known today (at time ) include the BARRA® multifactor models and can be written (omitting the time subscript for simplicity):
E(R_{i})R_f=\beta_{i,1}F_1+\dots+\beta_{i,K}F_K+\epsilon_{i} 
where is the factor exposure or loading of asset to factor (with being the correlation between asset and factor returns, and and being the volatility of asset and factor returns, respectively); is the return to factor during the period from time to time ; and is the asset's specific or idiosyncratic return during that period that cannot be explained by the factors.
The factors used in such models can include industrial and geographic segmentations of the data or various economic mechanisms that can be expected to have a systematic effect on average returns such as the tendency of wellperforming firms to continue to perform (the socalled momentum factor), as well as factors sometimes identified as 'anomalies' (like the outperformance of a 'low volatility' factor).
Beyond single or multifactor CAPMs, the second building block of modern asset pricing is the arbitrage pricing theory (APT). Under arbitrage pricing, pure arbitrage profits are not possible and differences between expected returns (or prices) and actual returns are the result of individual assets' relative exposure (betas) to a combination of zeromean risk factors (or 'surprises') . Hence,
R_{i}=E(R_{i})+\beta_{i,1}f_1+\dots+\beta_{i,K}f_K+\epsilon_{i}=E(R_{i})+\sum_{k=1}^K \beta_{i,k}f_k+\epsilon_i 
where .
All available information at the time of pricing is already incorporated in expected returns and the impact of factors on returns ex post is the combination of individual assets' exposure to these factors (the ) and any unexpected change or 'surprise' in the realisation of the , which have an expected value of zero.
Here again, expected excess returns are written as the linear combination of multiple riskfactor exposures and their risk premia or prices on the measurement date, that is,
E(R_{i})R_f=\lambda_{1} \tilde{\beta}_{i,1}+\dots+\lambda_{K}\beta_{i,K} 
where is the price of risk or premium of the risk factor with , and is the redefined measure of asset's exposure to the risk factor, which is equal to if factors are standardised to have a standard deviation of .
That is, expected (excess) returns are the sum of the riskfactor exposures of asset at time times the respective price of each risk factor to which asset is exposed. Realised excess returns are thus written:
R_{i}R_f = \sum_{k=1}^K \tilde{\beta}_{i,k}(f_k+\lambda_k)+\epsilon_i 
Thus, realised excess return for asset are generated by its exposure to risk factors and their unexpected realisation, as well as firmspecific risk. Because we will work only with standardised factors, there is no distinction between and , and for notational convenience we will only use to write expected returns, so that:
R_{i}R_f = \sum_{k=1}^K \beta_{i,k}(f_k+\lambda_k)+\epsilon_i 
where .
The price of asset today should thus equal the sum of all future cash flows discounted at the APT rate defined by equation, in which expected returns for asset are a linear function of various factors and the sensitivity to changes in each factor, as represented by each assetspecific beta coefficient.
Multifactor models focus on finding robust and persistent statistical effects that can explain and predict the price preferences of buyers and sellers over time. Modeling discount factors reflecting the fair value of financial assets is thus a matter of identifying relevant priced risk factors and determining both individual assets' exposure to each factor as well as the market price or required premia of a pure exposure to each such factor taken independently.
In effect, CAPM and APT are the only two theoretical frameworks that provide a solid foundation for computing the tradeoff between risk and returns . APT in particular leaves the identification of the relevant factors open and justifies the need to explicitly identify and test the factors impacting returns for different types of financial assets.