Why CAPM Needs Assumptions
CAPM is an equilibrium model: it describes the expected returns that make investors willing to hold the available assets at market-clearing prices. Its central prediction is that an asset’s expected return depends on its beta with the market portfolio, not on its total volatility considered in isolation. That result follows only under a set of assumptions about how investors choose portfolios, what they know, and how markets operate.
The earlier material on mean-variance efficiency and systematic risk establishes the role of portfolio risk and beta. Here the emphasis is on the conditions that connect individual decisions to an equilibrium prediction. The assumptions are not claims that actual investors or markets behave perfectly; they define the setting in which the standard CAPM result follows.
This objective is a model of preferences, not a universal description of human behavior. It gives a precise reason for treating expected return and variance as the relevant portfolio summaries. That treatment is justified directly if investors have mean-variance preferences. It can also be justified under specific distribution and utility assumptions, as discussed below.
Investor Behavior and Return Distributions
The standard CAPM assumes investors are risk averse and make portfolio choices over a common, single investment horizon. They compare the expected returns and risks of portfolios available over that horizon. A one-period model does not by itself describe how investors respond to changing opportunities across many periods.
For a portfolio with normally distributed returns, its distribution is determined by its mean and variance. Thus, if investors have preferences that rank portfolios by their return distributions, mean and variance can summarize the relevant comparison. Alternatively, quadratic utility makes expected utility depend on the first two moments. These are alternative conditions supporting mean-variance analysis; normality is not required when mean-variance preferences are simply assumed.
Outside these conditions, mean and variance need not describe all features investors care about. Skewness, extreme losses, or other distributional characteristics can affect expected utility. In that setting, two investments with identical means and variances may not be equivalent to an investor.
Worked Example: Same Mean and Variance, Different Distribution
Two hypothetical one-period investments have returns \(R_A\) and \(R_B\). Investment A returns either \(-10\%\) or \(10\%\), each with probability 0.5. Investment B returns \(-5\%\) with probability 0.8 and \(20\%\) with probability 0.2. Compare their means and variances, then consider the utility function \(U(R)=R+10R^3\), with returns expressed as decimals.
Method. Calculate each distribution’s first two moments. Then compare expected utility as a check on what happens when preferences depend on more than those moments.
For A, the expected return is \(0.5(-0.10)+0.5(0.10)=0\). Its variance is \(0.5(-0.10)^2+0.5(0.10)^2=0.01\).
For B, the expected return is \(0.8(-0.05)+0.2(0.20)=-0.04+0.04=0\). Its variance is \(0.8(-0.05)^2+0.2(0.20)^2=0.8(0.0025)+0.2(0.04)=0.002+0.008=0.01\).
Both investments therefore have the same expected return, 0, and the same variance, 0.01. A mean-variance investor is indifferent between them. But their third moments differ: A’s third moment is 0, while B’s is \(0.8(-0.05)^3+0.2(0.20)^3=-0.0001+0.0016=0.0015\).
Under the stated utility, \(E[U(R_A)]=0+10(0)=0\), while \(E[U(R_B)]=0+10(0.0015)=0.015\). The utility is increasing over both return ranges because \(U'(R)=1+30R^2>0\). This investor prefers B despite the matching means and variances. Check: the result illustrates why a mean-variance description needs an assumption about preferences or return distributions. The common mistake is to infer that matching means and variances must make any two investments equivalent.
Information and Common Expectations
The standard model assumes investors have homogeneous expectations: they use the same expected returns, variances, and covariances when evaluating the available assets. Common information can help motivate this assumption, but the two ideas are not identical. Investors might observe the same public data and still form different forecasts or estimate different covariances. CAPM’s portfolio-demand argument requires common expectations, not merely equal access to information.
With the same return estimates and risk estimates, investors identify the same efficient risky-asset mix. Different levels of risk aversion can change how much each investor allocates to that mix, but not its composition. The next example shows this distinction using a two-asset calculation.
Worked Example: Common Beliefs, Different Risk Tolerance
Two risky assets have expected excess returns of 8% and 6%. Their covariance matrix, using decimal returns, is \(\begin{pmatrix}0.04&0.01\\0.01&0.0225\end{pmatrix}\). Investors also have access to a risk-free asset. Investor 1 has risk-aversion coefficient \(A=4\), and Investor 2 has \(A=8\). Find their risky-asset allocations under the mean-variance model.
Method. For risky-asset allocation vector \(x\), maximize \(x^\mathsf{T}\mu_e-\frac{A}{2}x^\mathsf{T}\Sigma x\), where \(\mu_e=(0.08,0.06)^\mathsf{T}\) is the vector of expected excess returns and \(\Sigma\) is the stated covariance matrix. The first-order condition gives \(x=\frac{1}{A}\Sigma^{-1}\mu_e\).
First solve \(\Sigma y=\mu_e\). The determinant is \(0.04(0.0225)-(0.01)^2=0.0009-0.0001=0.0008\). Thus, \(y_1=(0.0225(0.08)-0.01(0.06))/0.0008=(0.0018-0.0006)/0.0008=1.5\), and \(y_2=(0.04(0.06)-0.01(0.08))/0.0008=(0.0024-0.0008)/0.0008=2\).
For Investor 1, \(x=(1/4)(1.5,2)=(0.375,0.5)\). The total risky allocation is \(0.875\), leaving \(0.125\) in the risk-free asset. For Investor 2, \(x=(1/8)(1.5,2)=(0.1875,0.25)\). The total risky allocation is \(0.4375\), leaving \(0.5625\) in the risk-free asset.
Both investors’ risky allocations have the same composition: asset 1’s share is \(1.5/(1.5+2)=3/7\), and asset 2’s share is \(2/3.5=4/7\). The more risk-averse investor holds half as much of each risky asset. Check: common beliefs preserve the risky mix while risk aversion changes its scale. The common mistake is to conclude that different risk tolerance necessarily means different risky-asset compositions.
Market Assumptions and the Equilibrium Link
The standard CAPM also idealizes the market. Investors are price takers, so no individual investor can set prices. Assets are divisible, and investors can trade the relevant assets without taxes or transaction costs. The model assumes a common risk-free rate and allows investors to lend or borrow at that rate. It also assumes the relevant risky assets are available to investors, so their aggregate holdings can represent the market portfolio. These conditions simplify the connection between individual portfolio choices and market-wide supply and demand.
The common risky-asset mix matters because markets must clear: in aggregate, investors’ holdings of risky assets must equal the available supply. If all investors demand the same risky mix, possibly in different amounts, the portfolio formed from the aggregate supply has that same mix. In the standard setting this is the market portfolio. The assumptions therefore connect investors’ individual efficient choices to an efficient market portfolio.
The equilibrium relation is commonly written as
Here \(R_i\) is the return on asset \(i\), \(R_m\) is the market-portfolio return, and \(r_f\) is the risk-free rate, all measured over the same period. Beta measures the asset’s covariance with the market relative to the market variance.
This argument explains how the assumptions support the equilibrium prediction: preferences and common expectations align investors’ risky choices; market clearing identifies the aggregate risky portfolio with the market; and efficiency imposes the covariance relation that becomes the beta equation. If one of the links fails, the standard prediction need not follow as stated.
Worked Example: Different Forecasts Change Risky-Asset Demand
At a quoted price, a hypothetical stock has variance of return \(0.04\). The risk-free rate is 3%. Two investors have the same risk-aversion coefficient, \(A=4\), but Investor 1 expects a 7% stock return and Investor 2 expects a 5% return. Find each investor’s optimal share in the stock using the mean-variance model with one risky asset and one risk-free asset.
Method. With one risky asset, the first-order condition reduces to \(x=(E[R]-r_f)/(A\operatorname{Var}(R))\), where \(x\) is the fraction invested in the stock.
Investor 1’s expected excess return is \(0.07-0.03=0.04\), so \(x_1=0.04/(4(0.04))=0.04/0.16=0.25\). Investor 2’s expected excess return is \(0.05-0.03=0.02\), so \(x_2=0.02/0.16=0.125\).
At this quoted price, the investors demand different stock allocations even though their risk tolerances and variance estimates match. The difference comes from their forecasts. This contradicts homogeneous expectations and breaks the simple argument that all investors choose the same risky mix. In a full equilibrium, prices can adjust; these calculations are conditional demands at the stated price. Check: a common-information statement alone would not resolve the disagreement in expected returns. The common mistake is to treat access to the same information as proof of identical expectations.
What the Assumptions Do—and Do Not—Claim
CAPM’s assumptions are useful because they isolate a clear pricing mechanism, not because every assumption is literally true in practice. Their implications should be kept distinct:
- Mean-variance behavior or suitable distribution and utility conditions makes expected return and variance sufficient for portfolio choice.
- Homogeneous expectations gives investors a common efficient risky-asset mix, even if they choose different amounts of it.
- Competitive, frictionless trading and market clearing connect individual demand to the aggregate portfolio of risky assets.
- A common risk-free rate and the standard one-period setting complete the textbook equilibrium framework.
In applications, these assumptions are a framework for interpreting a required return, not a guarantee that observed returns will equal the prediction in every period. For example, heterogeneous forecasts, trading frictions, or returns with important features beyond variance can limit the model’s fit. This distinction is especially important when using CAPM inputs: the equation is a conditional model result, and its interpretation depends on the assumptions that support it. This treatment is aligned with the SOA VEE Accounting and Finance syllabus.
Check Your Understanding
For each problem, identify the relevant assumption or apply the stated mean-variance model.
- An investor compares two portfolios with identical expected returns and variances. State one preference condition under which the investor must be indifferent between them. (Answer: Mean-variance preferences, where the ranking depends only on expected return and variance.)
- Two investors have the same expected excess return of 3%, variance of 0.09, and risk-aversion coefficients \(A=2\) and \(A=6\). Using \(x=(E[R]-r_f)/(A\operatorname{Var}(R))\), find their risky-asset shares and compare them. (Answer: \(x_1=0.03/(2(0.09))=1/6\approx0.1667\); \(x_2=0.03/(6(0.09))=1/18\approx0.0556\); Investor 1 holds three times the risky share.)
- Investors have identical expected returns and variances but different covariance estimates between two assets. Which CAPM assumption fails, and why does it matter for portfolio choice? (Answer: Homogeneous expectations fails; different covariance estimates can produce different efficient risky-asset mixes.)
- Suppose \(r_f=2\%\), \(E[R_m]=8\%\), and an asset has \(\beta=1.25\). Calculate its CAPM expected return. (Answer: \(2\%+1.25(8\%-2\%)=9.5\%\).)
- Returns are not normally distributed, and an investor’s utility responds to large losses beyond what variance captures. Does equal mean and variance alone establish indifference? Explain. (Answer: No; the investor may care about distributional features beyond mean and variance, so a mean-variance or other suitable utility/distribution assumption is needed.)