Asset Pricing and Factor Risk: What Investors Are Paid to Bear

An expected return is meaningful only in relation to the risks that investors are being asked to bear. Asset-pricing models organise that relationship. They can help estimate required returns and interpret performance, but a fitted risk model is not a guarantee of future returns or a recommendation for an individual security.

This reference consolidates the capital asset pricing model, arbitrage pricing theory and multifactor interpretation. It keeps the distinction between equilibrium statements, empirical estimates and realised outcomes explicit.

Risk compensation is different from total uncertainty

An investment can have a volatile return because of broad market exposure, sector shocks or events specific to the issuer. A diversified investor may be able to reduce the last component by combining many holdings, while broad systematic exposure remains.

The capital asset pricing model links expected excess return to market beta:

$$\mathbb E[R_i]-r_f=\beta_i(\mathbb E[R_m]-r_f).$$

Here $R_i$ is an asset return, $R_m$ the market-portfolio return, $r_f$ the risk-free return over the same horizon, and $\beta_i$ the asset’s exposure to market fluctuations. The expression is a conditional equilibrium relationship under the model’s assumptions. It is not a statement that the realised return over the next period will equal the right side.

Those assumptions include a particular investment opportunity set, investor preferences and access to borrowing or lending. Real-world constraints, heterogeneous beliefs and incomplete diversification can complicate the simple result.

Beta measures covariance rather than standalone volatility

Beta is

$$\beta_i=\frac{\operatorname{Cov}(R_i,R_m)} {\operatorname{Var}(R_m)} =\rho_{im}\frac{\sigma_i}{\sigma_m},$$

In words: Beta measures shared movement with the market relative to how much the market itself varies.

where $\rho_{im}$ is correlation and $\sigma_i,\sigma_m$ are return volatilities. An asset can have high standalone volatility and low beta if much of its variation is unrelated to the market.

A beta of 1.2 means a fitted or model-implied sensitivity of 1.2 units of asset return per unit of market return in the relevant linear relationship. It does not mean that the asset is exactly 20% more volatile in total, nor that every market move produces a mechanically proportional realised move.

Inputs to the retained beta teaching example

Its beta is

$$\beta=0.8\frac{0.20}{0.15}=1.066666\ldots.$$

Using the unrounded beta, the required return is 10.40%. The source rounds beta to 1.07 before multiplication and reports 10.42%. Both calculations are explained by the rounding convention; the extra two basis points are not additional economic information.

Figure 01 · Conceptual framework

Exposure and reward are separate inputs to required return

How the parts connect

1. ARisk exposuressensitivity inputsRequired return
2. BExpected factor premiareward assumptionsRequired return
3. CRequired returndiscounting inputValuation judgement
4. DMatched cash flowsvaluation objectValuation judgement

The security market line is a model benchmark

The security market line plots expected return against beta. Its intercept is the risk-free return and its slope is the expected market risk premium. Under the model, a higher beta requires more expected return when that premium is positive.

An estimated expected return above the line may be described as positive alpha relative to the model. That description depends on both the expected-return estimate and the benchmark. It is not proof of mispricing: the model may omit relevant risks, the estimate may be noisy, or implementation costs may absorb the apparent difference.

The capital market line is different. It relates expected return to total volatility for combinations of a risk-free asset and the tangency portfolio under the standard assumptions. Confusing it with the security market line mixes a portfolio-efficiency relationship with an individual asset’s systematic-risk relationship.

Estimating beta introduces choices

A common empirical regression is

$$R_{i,t}-r_{f,t}=\alpha_i+\beta_i(R_{m,t}-r_{f,t})+\varepsilon_{i,t}.$$

In words: The asset’s excess return is divided into a constant component, market-linked movement and an unexplained remainder.

The residual $\varepsilon_{i,t}$ is the part not explained by the fitted market factor. A regression coefficient is conditional on the sample, return frequency, market proxy and treatment of missing observations.

Daily data provide more observations but can contain nonsynchronous trading and microstructure effects. Monthly data reduce some of those problems but leave fewer observations. A long window improves precision under stability and can obscure a genuine change in the asset’s business or leverage.

Beta estimates should therefore be accompanied by uncertainty and stability checks. Rolling estimates, subperiod comparisons and exposure decomposition help distinguish a persistent relationship from an accidental sample fit.

An investment with nonlinear payoffs can have state-dependent exposure. A single linear beta estimated over a quiet period may not describe its response during a large move. Option delta, factor beta and full revaluation are different tools, each with its own units and approximation limits.

Figure 02 · Qualitative comparison

Beta, volatility and alpha answer different questions

Beta, volatility and alpha answer different questions

Measure	Question answered	Main limitation
Standalone volatility	Question answeredHow much does the asset vary?	Main limitationDoes not distinguish common from diversifiable risk
Market beta	Question answeredHow does it co-move with the chosen market proxy?	Main limitationDepends on benchmark, sample and changing business exposure
Historical alpha	Question answeredWhat average return is unexplained by the fitted model?	Main limitationDoes not establish future skill or persistent mispricing

Factor models distinguish exposures from premia

A multifactor expected-return expression is

$$\mathbb E[R_i]-r_f=\sum_{k=1}^K\beta_{ik}\lambda_k.$$

The exposure $\beta_{ik}$ says how asset $i$ responds to factor $k$. The premium $\lambda_k$ says what expected compensation is associated with that exposure under the model. A large exposure does not imply a high expected return unless the corresponding premium is positive and relevant.

Factors may be macroeconomic variables, traded factor portfolios or statistical components. Their interpretation differs. A principal component that explains return covariance is not automatically a priced risk factor. A statistical association with past returns does not by itself establish an economic reason for compensation.

Arbitrage pricing theory derives restrictions from a factor structure and the absence of suitable arbitrage opportunities in sufficiently diversified portfolios. It does not, by itself, identify the correct factors or promise that a finite, concentrated portfolio eliminates all residual risk.

Historical anomalies should remain historical evidence

The teaching materials discuss size, value, momentum and low-beta patterns as motivations for moving beyond a single market factor. These are historical empirical themes, not unconditional promises that one style always outperforms another.

A factor can experience long periods of weak performance, become crowded, depend on portfolio construction or lose much of its apparent return after costs. Searching many definitions and publishing only the strongest historical result creates selection bias.

Evaluation should specify formation rules, rebalancing dates, universe membership, treatment of delisted securities and transaction costs. A factor return series built using today’s surviving securities can overstate historical opportunity even if its regression arithmetic is correct.

Required return and performance attribution are separate uses

A required-return estimate can be used as a discount rate in an appropriately matched valuation. The cash-flow currency, horizon, leverage and risk definition should agree with the rate. Adding a risk premium twice—once in conservative cash flows and again in an excessive discount rate—can distort value.

Performance attribution asks how realised return decomposed into factor exposures and residual performance. A high realised return may come from large exposure to a rewarded factor rather than superior selection. Conversely, negative alpha over a short noisy sample is weak evidence of a persistent lack of skill.

Statistical significance, economic magnitude and persistence are separate questions. A coefficient can be precisely estimated but too small to matter after fees; a large estimate can be too uncertain to support a decision.

A useful interpretation framework

What asset-pricing models can and cannot establish

Asset-pricing models are most useful as explicit benchmarks. They discipline the link between risk and expected compensation while leaving room to examine estimation error, missing factors and regime change. The appropriate conclusion is an interpretable required-return framework, not a mechanically certain forecast.

Figure 03 · Conceptual framework

From an attractive backtest to a defensible alpha claim

How the parts connect

1. AObserved outperformancedecompose returnBenchmark attribution
2. BBenchmark attributionquestion apparent skillSelection review
3. CBenchmark attributiontest tradabilityImplementation test
4. DSelection reviewadjust for search biasQualified forward view
5. EImplementation testretain feasible economicsQualified forward view

Portfolio beta can conceal offsetting exposures

For fixed portfolio weights and a common market benchmark, portfolio beta is the weighted sum of constituent betas. An illustrative portfolio with 60% in an exposure of beta 1.5 and 40% in an exposure of beta 0.25 has beta one.

That aggregate matches the market beta, but the portfolio need not have market-like volatility, liquidity or tail risk. Its residual risks may be concentrated, its sector exposures different and its weights unstable as prices move. Matching beta is one hedge objective, not a complete replication of the market portfolio.

A portfolio with derivative positions requires consistent sensitivity units. A local option delta converts underlying changes into option-value changes; beta converts a specified underlying return exposure into a market-factor exposure. The transformations should be applied to the correct base and then reconciled with full portfolio revaluation for large moves.

Separate realised alpha from expected alpha

An ex-post regression intercept describes average unexplained return over the estimation sample under the fitted model. An expected alpha is a forward-looking claim about future unexplained return. The first does not establish the second.

Selection creates an additional problem. If many strategies are tested and only the highest historical intercept is reported, its estimate is likely to be optimistic. Honest assessment should account for the search process, preserve an untouched evaluation period and examine whether the apparent advantage survives realistic costs.

Changing the benchmark can also change alpha. A strategy with positive alpha relative to a single market factor may have little residual return after adding relevant style exposures. This is not necessarily evidence that the original arithmetic was wrong; it shows that attribution depends on the risk model.

Currency and horizon need to match

A risk-free rate is defined in a currency and over a horizon. Using a short-term rate in one currency with long-horizon cash flows in another can introduce a mismatch. Exchange-rate risk and cross-currency discounting require a coherent treatment rather than a simple addition of unrelated premia.

Expected market premiums also depend on whether returns are arithmetic or geometric and on the horizon of the decision. The arithmetic average is useful for an expected one-period return under suitable conditions; a compounded growth rate answers a different question. Mixing them can make a required-return estimate look more precise while changing its meaning.

For a leveraged equity exposure, beta can change with financing structure. Comparing two companies or portfolios on observed equity beta alone may conflate business risk and leverage. Any adjustment requires explicit assumptions about debt risk, taxes and the capital structure rather than an unexplained formula.

Factor diversification is not a free expected return

Combining factors can reduce dependence when their returns are imperfectly correlated. It does not guarantee that all factors have positive future premia or that their correlations remain stable. Crowded implementation can also make distinct factor definitions share liquidity risk.

A useful factor allocation records the economic rationale, construction rule, expected premium estimate and uncertainty for each exposure. It then examines how the combination behaves when several historical relationships weaken together.

The resulting framework supports a measured investment judgement. It can identify which risks a portfolio is taking and what compensation is assumed. The investor still needs to decide whether those assumptions are credible and whether the implementation can survive the periods when realised returns differ sharply from the model’s expectations.

That judgement should distinguish a historically estimated premium from a premium required to justify the current price. The former summarises a selected sample; the latter is an assumption in today’s valuation. They can inform one another, but substituting one without explanation changes the argument being made.