Equity Risk Premium: Definition, Estimation, and the Puzzle

Every valuation model in finance depends on one critical input: the equity risk premium. Whether you are estimating a company’s cost of equity through the CAPM, calculating WACC, or building a discounted cash flow model, the equity risk premium is the single variable that drives the largest swings in your output. A one-percentage-point change in the ERP can shift a company’s implied value by 20–30%. Despite its importance, there is no consensus on the “correct” ERP — and the debate over how to estimate it remains one of the most active in academic finance.

What is the Equity Risk Premium?

The equity risk premium (ERP) is the expected excess return that investors demand for holding equities over risk-free government bonds. It compensates investors for bearing the systematic risk of the stock market — the risk that cannot be eliminated through diversification.

The ERP also serves as the market factor in multifactor models like the Fama-French three-factor model, where it is one of three systematic risk premiums used to explain portfolio returns.

The terms “equity risk premium” and “market risk premium” are used interchangeably in most finance textbooks and practice. Both refer to the expected excess return of the equity market over the risk-free rate. When researchers refer to the realized (historical) excess return in a specific period, that is more precisely called the ex post market excess return, whereas ERP refers to the ex ante (forward-looking) expectation.

The Equity Risk Premium Formula

Where:

• E(R_M) — expected return on the broad equity market (e.g., S&P 500)
• R_f — risk-free rate, typically the yield on a government bond matching the investment horizon (10-year Treasury for long-term valuation, 3-month T-bill for short-term analysis)

Because E(R_M) is unobservable — no one knows future stock returns with certainty — the ERP must always be estimated. The two primary approaches are historical estimation and forward-looking (implied) estimation:

This implied approach derives the ERP from current market prices using a constant-growth dividend discount model. In practice, analysts often substitute total payout yield (dividends plus buybacks) or use earnings-based models to account for the fact that modern corporations return cash through repurchases as well as dividends. The risk-free rate used should match the currency and horizon of the cash flows being valued — using a U.S. Treasury yield to discount euro-denominated cash flows, for instance, introduces a mismatch.

Historical Equity Risk Premium

The most straightforward approach to estimating the ERP is to measure the historical average excess return of stocks over government bonds. Using the S&P 500 Total Return Index (or the broader CRSP value-weighted index for pre-1957 data), the U.S. has the longest and most studied dataset:

The choice between arithmetic and geometric means matters. The arithmetic mean (~8% over T-bills for the full sample) is the better estimate of the expected return in any single future period, while the geometric mean (~6%) reflects the compound growth rate actually earned by investors over long holding periods. Most academic models use arithmetic means; most practitioners and long-term planners prefer geometric means.

A critical caveat: the U.S. stock market was the outstanding success story of the 20th century. Using only U.S. data may overstate the ERP due to survivorship bias. The Dimson-Marsh-Staunton global dataset, covering 23 countries from 1900 to the present, shows a worldwide geometric equity premium of approximately 3–5% over government bonds — meaningfully lower than the U.S. figure alone.

Forward-Looking ERP Estimation

Many practitioners argue that forward-looking estimates are more relevant than historical averages because they reflect current market conditions rather than a century of past data:

1. Implied ERP from market prices. Using a dividend discount model or total payout model, analysts can “back out” the expected return embedded in current stock prices. If the S&P 500 has a dividend yield of 1.5%, an expected buyback yield of 1.5%, and expected real earnings growth of 3%, the implied real ERP is approximately 6% minus the current real risk-free rate. Aswath Damodaran of NYU Stern publishes a widely-cited annual implied ERP estimate, which has ranged between approximately 4–6% in recent years.

2. Survey-based estimates. CFO surveys and analyst polls typically produce ERP estimates in the 5–6% range, though surveys are sensitive to framing effects and recency bias.

3. Building-blocks approach. Start with the historical average, adjust for current conditions (dividend yield, earnings growth, inflation), and arrive at an estimate. This approach often produces figures of 4–6%.

Forward-looking estimates tend to cluster in the 4–6% range, somewhat lower than the long-run historical arithmetic mean but consistent with geometric historical returns. The advantage is that they incorporate current valuation levels — when stock prices are high relative to earnings, the implied ERP is lower, and vice versa.

Equity Risk Premium Example

The Equity Premium Puzzle

In 1985, economists Rajnish Mehra and Edward Prescott published a landmark paper showing that the historical equity risk premium in the United States was far too large to be explained by standard economic models of investor behavior. This became known as the equity premium puzzle.

The problem is straightforward: standard expected utility theory predicts that investors with reasonable levels of risk aversion (a coefficient of roughly 1–10) should demand an equity premium of only about 1–2%. Yet the observed historical premium has been 6–8%. To rationalize the actual premium, the model requires a risk aversion coefficient of 30–40 — implying investors are implausibly terrified of risk.

Several explanations have been proposed over the past four decades:

• Rare disaster risk (Barro, 2006) — Investors demand a high premium because they fear low-probability but catastrophic events (wars, depressions, pandemics) that could wipe out wealth. The probability-weighted impact of these tail events may justify the observed premium even with moderate risk aversion.
• Myopic loss aversion (Benartzi & Thaler, 1995) — Drawing on behavioral finance, this explanation argues that investors evaluate their portfolios too frequently (annually rather than over decades) and feel losses roughly twice as intensely as equivalent gains. This combination of short evaluation horizons and loss aversion produces an ERP consistent with historical observations.
• Habit formation — Risk aversion is not constant but increases during economic downturns when consumption falls relative to a reference “habit” level. This time-varying risk aversion can generate a higher average premium.
• Liquidity premium — Equities are less liquid than Treasury bonds in crisis periods, and investors demand compensation for the risk of being unable to sell at fair value when they need cash most.

The equity premium puzzle remains unresolved. It has spawned hundreds of academic papers and has been called “the most important unresolved puzzle in financial economics.” Its persistence suggests that either our models of investor behavior are incomplete, or that the historical premium overstates the true ex ante risk premium investors required.

ERP Across Countries

The equity risk premium varies significantly across countries, reflecting differences in economic stability, market development, and political risk:

For non-U.S. valuations, practitioners often start with a base mature-market ERP (typically the U.S. or global average) and add a country risk premium (CRP) for additional risks. Damodaran’s widely-used approach derives the CRP from a country’s sovereign credit rating or CDS spread, scaled by the relative volatility of its equity market versus its bond market. This prevents the common error of simply using a developed-market ERP for an emerging-market company.

Equity Risk Premium vs Risk-Free Rate

The ERP and the risk-free rate are the two building blocks of the expected market return (E(R_M) = R_f + ERP), but they behave very differently:

The relationship between the two inputs is important for valuation. When risk-free rates rise, the ERP does not necessarily stay constant — it may decline if higher rates reflect improved economic confidence, or it may increase if higher rates reflect uncertainty. During the low-rate environment of 2010–2021, implied ERPs were often elevated (5–6%), partly compensating for near-zero risk-free rates. As rates have normalized, the implied ERP has adjusted accordingly.

Common Mistakes

1. Using a single year’s market return as the ERP. The S&P 500 returned 26% in 2023, but that does not make the ERP 22% (26% minus a 4% risk-free rate). The ERP is a long-term expected premium requiring at least 50–100 years of data or a forward-looking estimation model. Single-year returns are dominated by noise.

2. Conflating arithmetic and geometric means. The arithmetic mean is always higher than the geometric mean due to the variance drag effect. Using the arithmetic mean (~8%) in a multi-period compound return model overstates expected wealth. Match your mean to your purpose: arithmetic for single-period expected returns, geometric for long-horizon compound projections.

3. Assuming the ERP is constant. The ERP varies over time. It tends to be higher after market crashes (when fear is elevated and prices are depressed) and lower after extended bull markets (when optimism pushes prices up). Using a fixed ERP regardless of market conditions ignores this variation.

4. Ignoring survivorship bias. Using only U.S. data to estimate the global ERP overstates the premium because the U.S. had the most successful equity market of the 20th century. Markets that experienced wars, expropriations, or prolonged stagnation (Russia, Japan, Argentina) show much lower premiums. Global datasets provide a more balanced picture.

5. Mixing inconsistent inputs. A common practitioner error is mismatching the ERP’s building blocks: using a nominal ERP with real cash flows, a T-bill-based ERP with a T-bond risk-free rate, or a U.S. ERP for a euro-denominated valuation. All inputs must be internally consistent in currency, inflation basis, and maturity horizon.

Limitations of the Equity Risk Premium

1. The ERP is unobservable. Unlike the risk-free rate, which can be read from government bond yields, the ERP exists only as an expectation about the future. It can never be directly measured — only estimated through historical data or models, each with their own assumptions and limitations.

2. Historical data may not predict the future. Structural changes in markets — higher retail participation, passive investing, central bank intervention, improved information technology — may have permanently altered the risk-return tradeoff. The premium earned over the past century may not repeat.

3. Forward-looking models require assumptions. Implied ERP models depend on forecasts of earnings growth, dividend yields, and buyback rates. If these inputs are wrong, the implied ERP will be wrong. Models are only as good as the assumptions feeding them.

4. Small errors cascade into large valuation mistakes. Because the ERP feeds into cost of equity, which feeds into WACC, which feeds into DCF valuations, a seemingly small estimation error compounds at each stage. There is no way to eliminate this uncertainty — only to be transparent about the sensitivity of your conclusions to the ERP assumption.