Why There Is More Than One Market Risk Premium Estimate
The market risk premium is an input to the CAPM required-return calculation. The CAPM uses an expected market risk premium: the expected market return above the risk-free rate over the relevant horizon. That expectation cannot be observed directly. An analyst therefore estimates it, often by looking backward at realized returns or forward at current market prices and expected cash flows.
These approaches do not measure the same thing in the same way. A historical estimate summarizes what happened in a chosen sample of periods. A forward-looking estimate infers what return is embedded in current prices, given a model and assumptions about future cash flows. Both can be useful, but neither is a mechanical guarantee of the premium investors will earn.
Historical-Average Estimates
A historical approach collects returns on a chosen market benchmark and a risk-free investment for the same periods. For each period, it calculates the market's excess return, then summarizes those excess returns. The arithmetic average is commonly used when the target is an expected one-period return.
The market return should be a total return if the estimate is intended to include both price changes and distributions. The risk-free return should match the market observation period as closely as possible. For example, annual market returns should be paired with a risk-free return earned over those same annual periods, rather than with a rate for a different horizon.
The historical estimate depends on choices that should be reported: the market benchmark, the start and end dates, the return frequency, whether returns include distributions, and the risk-free proxy. Using a different country, sample period, or risk-free series can produce a different estimate. A long sample may include market conditions unlike those expected in the future; a short sample may be dominated by a few unusual years.
Worked Example: Calculate an Arithmetic Historical Premium
A hypothetical fund uses a broad equity-market index as its benchmark. Over three successive years, the index total returns are 12%, −4%, and 9%. The matched annual risk-free returns are 3%, 2%, and 2%. Calculate the arithmetic historical market risk premium. State the method and avoid the mistake of averaging market returns before subtracting an unrelated risk-free average.
Step 1: Match each market return to its year's risk-free return. The annual excess returns are:
Step 2: Average the three excess returns. Their sum is \(9\%-6\%+7\%=10\%\). Divide by three:
Rounded to two decimal places, the arithmetic historical estimate is 3.33% per year. As a check, the market-return average is \((12\%-4\%+9\%)/3=5.67\%\), and the risk-free average is \((3\%+2\%+2\%)/3=2.33\%\); their difference is also \(3.34\%\) using those rounded intermediate figures. Using the unrounded values gives \(5.666\ldots\%-2.333\ldots\%=3.333\ldots\%\), consistent with the direct calculation. The method avoids rounding the intermediate averages too early.
Why the Arithmetic Average Fits a One-Period Expectation
The arithmetic average has a direct relationship to an expected one-period excess return. Let \(X_t=R_{m,t}-R_{f,t}\), and suppose the observations are independent draws from a stable distribution with finite mean \(\mu_X\). Each observation has \(E[X_t]=\mu_X\). By linearity of expectation:
Thus, under the stated stability assumption, the sample arithmetic average is an unbiased estimator of the mean one-period excess return. This justification does not say that a historical average is certainly the right forecast: the distribution may change, and the sample may be imprecise. Independence and a stable mean are simplifying assumptions, not facts ensured by collecting data.
A geometric average answers a different question. It describes a compound growth rate over multiple periods. Compounding depends on the sequence of returns: a gain followed by a loss does not generally restore the starting value, even if their arithmetic average is zero. A geometric market return less a geometric risk-free return can summarize relative compound growth over a sample, but it is not interchangeable with the arithmetic estimate of a one-period expected premium.
Forward-Looking Estimates from Market Inputs
A forward-looking method begins with the prices investors pay now and estimates the return consistent with those prices and expected future cash flows. One simplified approach uses a constant-growth payout model. Let \(P_0\) be the current market value, \(D_1\) the expected total shareholder payout over the next period, and \(g\) the assumed constant growth rate of payouts after that. The model assumes payouts grow at rate \(g\) indefinitely and that the required return \(k\) exceeds \(g\).
The payout yield \(D_1/P_0\) is the next-period expected payout divided by today's market value. For a broad market, a payout measure can include dividends and repurchases, provided both the payout forecast and the model are defined consistently. The growth assumption is a major input: a small change in a perpetual growth rate can materially change the implied return. Analysts may also use more detailed forecasts with several growth stages, but those require additional assumptions about cash flows and their eventual growth.
Worked Example: Infer a Forward-Looking Premium
Suppose a hypothetical market index is valued at $2,500 per unit. Its expected total shareholder payout over the next year is $80 per unit, and an analyst assumes those payouts will grow at 4.0% per year indefinitely. The current annual risk-free rate is 2.5%. Under the constant-growth model, estimate the forward-looking market risk premium. The method is to infer the market return from payout yield plus growth; do not treat the payout yield alone as the full return.
Step 1: Calculate the payout yield.
Step 2: Add the assumed growth rate to obtain the implied market return.
Step 3: Subtract the matched risk-free rate.
The estimate is 4.7% per year. The calculation assumes the stated payout growth persists indefinitely and that \(k>g\), which holds here because 7.2% exceeds 4.0%. The common mistake avoided is omitting growth and reporting 3.2% minus 2.5%, or 0.7%, as the premium. That would omit part of the model-implied market return.
Comparing the Approaches and Their Uncertainty
| Feature | Historical-average approach | Forward-looking approach |
|---|---|---|
| Primary inputs | Past market total returns and matched risk-free returns | Current market value, expected payouts, growth assumptions, and current risk-free rate |
| What it summarizes | Realized excess returns in a selected sample | A return implied by current prices and a cash-flow model |
| Main uncertainty | Sampling variation and whether past conditions represent the future | Forecast and model uncertainty, especially payout and growth assumptions |
| Common limitation | Results can depend strongly on the sample dates and benchmark | Different reasonable models or assumptions can imply different premiums |
For a simple illustration of historical sampling uncertainty, if \(X_t\) denotes each observed annual excess return, the sample standard error of its arithmetic mean is \(s_X/\sqrt{n}\), where \(s_X\) is the sample standard deviation and \(n\) is the number of observations. This expression relies on independent observations with a common variance. It measures sampling uncertainty under that model; it does not capture structural change, a mistaken market proxy, or uncertainty about whether the historical mean remains relevant.
Worked Example: Compare Estimates and Quantify a Historical Standard Error
A hypothetical analyst has four annual historical market premiums, measured in percentage points: 2, 4, 6, and 8. A separate forward-looking model produces an estimate of 4.5%. Calculate the historical average and its standard error under an independent, stable-return assumption, then compare the two estimates. The method is to calculate dispersion around the historical mean before dividing by the square root of the sample size; the common mistake is to divide the variance, rather than the standard deviation, by that square root.
Step 1: Find the historical arithmetic mean. The sum is \(2+4+6+8=20\) percentage points, so the mean is \(20/4=5\) percentage points.
Step 2: Calculate the sample standard deviation. Deviations from 5 are −3, −1, 1, and 3. Their squared deviations sum to \(9+1+1+9=20\). With \(n-1=3\) in the sample variance denominator:
Step 3: Calculate the standard error.
The historical estimate is 5.0%, which is 0.5 percentage points above the forward-looking estimate of 4.5%. The estimated historical standard error is 1.291 percentage points under the stated assumptions. This comparison does not establish that either estimate is correct: four observations provide limited historical information, while the forward estimate depends on its model inputs.
Interpreting Differences Without False Precision
A difference between estimates is a prompt to inspect their construction, not a reason to select one automatically. Check whether both are nominal or real, whether their periods are aligned, whether the historical return includes distributions, and whether the risk-free rate is consistent with the market-return horizon. Also ask whether the historical period and the forward cash-flow assumptions are plausible for the forecast being made.
Historical estimates can be affected by sample length, the chosen beginning and ending dates, unusual economic episodes, and changes in the market benchmark. Forward estimates avoid relying solely on past realized returns, but they exchange that dependence for forecast uncertainty: payout yields may be measured incorrectly, growth assumptions may be optimistic or conservative, and a constant-growth structure may not fit the market. Neither method removes the need for judgment.
For a CAPM application, an analyst should state which estimate is being used, its horizon and terms, and the key assumptions. If several defensible estimates are available, it can be useful to explain the range of inputs considered rather than report a number with unjustified precision. The purpose is to estimate an expected premium coherently, not to imply that the market's future excess return is known.
Check Your Understanding
For each question, distinguish the calculation from the assumptions behind the estimate.
- A hypothetical index has annual total returns of 9%, −3%, 11%, and 7%. Its matched risk-free returns are 2%, 1%, 3%, and 2%. Calculate the arithmetic historical market risk premium. (Answer: 4.00%)
- A broad market has a current value of $4,000 per unit, expected next-year total payouts of $120 per unit, and an assumed constant payout growth rate of 3.5%. If the matched risk-free rate is 2.25%, calculate the forward-looking premium under the constant-growth model. (Answer: 4.25%)
- Five annual historical premiums, in percentage points, are 1, 3, 5, 7, and 9. Calculate their arithmetic mean and the standard error of the mean, using the sample standard deviation and assuming independent observations with a stable variance. (Answer: Mean = 5.00%; standard error = 1.414 percentage points)
- Over two years, a hypothetical market earns 10% and then −10%; the matched risk-free return is 2% in each year. Calculate the arithmetic average historical premium and compare it with the difference between the market and risk-free geometric average returns. Round the geometric comparison to two decimal places. (Answer: Arithmetic premium = −2.00%; difference in geometric average returns ≈ −2.50%)
- A historical estimate is above a forward-looking estimate. Name one source of uncertainty specific to each method and explain why the difference alone does not identify which estimate is the better forecast. (Answer: Historical uncertainty can arise from sample-period choice or sampling variation; forward-looking uncertainty can arise from payout or growth forecasts and model choice. The difference alone cannot establish forecast accuracy because the methods rely on different data and assumptions.)