The equity premium puzzle asks why stocks historically outperformed safer government debt by more than standard consumption-based models can explain.
The equity premium puzzle is the difficulty standard consumption-based asset-pricing models have in explaining why broad equities historically earned such a large average return above safer government securities. The puzzle is not simply that stocks earned more; it is that the observed premium appears too large relative to measured consumption risk and plausible investor risk aversion in the basic model.
The equity risk premium can be written as:
where E(R_e) is the expected equity return and R_f is the return on the selected lower-risk asset. Researchers can also calculate a historical realized premium from observed returns. Expected and realized premiums are not the same quantity.
| Concept | Question being asked |
|---|---|
| Expected equity premium | What additional return do investors currently require for holding equities? |
| Historical realized premium | How much did equities actually outperform the selected benchmark over a past sample? |
| Equity premium puzzle | Why does a standard economic model struggle to explain the historical premium using plausible risk and preference assumptions? |
A positive premium is not itself puzzling. Risky claims should generally offer compensation in expectation. The puzzle concerns the magnitude of the historical premium relative to what the model predicts.
Consumption-based asset pricing starts from the idea that money is more valuable in economically difficult states. Let m represent the stochastic discount factor, which is high when the investor places greater value on an additional future payoff. A basic pricing condition is:
For a risky asset, its expected excess return is related to how its return covaries with that discount factor:
Here r_i and r_f are net returns expressed on the same basis. The equivalent pricing equation uses 1+r_i as the asset’s gross return.
An asset deserves a high expected premium when it tends to perform badly precisely when m is high, such as when consumption is low and losses are especially painful. If equity returns have only a modest relationship with aggregate consumption growth in the measured data, the basic model generates only a modest premium unless risk aversion is set very high.
That creates the central tension:
The puzzle is therefore a test of the model, its inputs, and its assumptions. It is not evidence that equity risk is imaginary.
Assume a researcher uses a hypothetical 40-year sample and calculates:
9%3%The sample’s realized equity premium is:
Suppose a basic consumption model, calibrated with the same economy’s measured consumption growth and a risk-aversion assumption the researcher considers plausible, implies a premium of only 1.5%.
The 4.5% gap is the puzzle in this simplified example. It does not mean equities are guaranteed to earn 6% above bills in the future. A different period, country, benchmark, averaging method, inflation convention, or model could produce a different gap. The numbers are hypothetical and illustrate the reasoning rather than estimate today’s premium.
The puzzle asks whether a model captures the risks investors actually care about. A model that matches average returns only by assuming extreme preferences may fit one statistic while failing to explain consumption behavior, interest rates, return volatility, or other asset prices.
Equity-premium estimates feed into discount rates and cost of equity. Historical outperformance cannot be inserted mechanically into a valuation as a forward-looking required premium. The puzzle reinforces that expected returns are model-dependent and unobservable.
The historical premium does not establish that an all-equity portfolio is suitable. Investors experience drawdowns, liquidity needs, taxes, fees, sequence risk, and different horizons. A high long-run sample average can coexist with severe losses and long periods of weak realized performance.
Researchers have changed different parts of the standard model rather than relying on one settled solution.
| Explanation | Core idea | What still requires scrutiny |
|---|---|---|
| Rare disasters | Investors price the possibility of infrequent economic collapses that may be underrepresented in ordinary samples | Disaster probability, severity, recovery, and whether prices reflect changing disaster risk |
| Long-run risks | Small but persistent changes in consumption growth or uncertainty can matter greatly for long-lived assets | Identification of persistent risk and sensitivity to preference assumptions |
| Habit formation | Risk aversion effectively rises when consumption approaches a reference or habit level | How the habit is specified and whether the model also fits other evidence |
| Incomplete markets | Aggregate consumption can hide large risks borne by particular households | Household heterogeneity, insurance, data quality, and market participation |
| Borrowing constraints | Investors who would buy equities may be unable to borrow or participate fully | Which constraints bind, for whom, and over what period |
| Market frictions | Taxes, transaction costs, illiquidity, and participation costs can support a larger premium | Whether measured frictions are large and persistent enough |
| Behavioral preferences | Loss aversion, ambiguity, or non-standard beliefs can change required compensation | Stability of behavior and consistency across decisions and markets |
| Data and sample effects | Survivorship, benchmark selection, disasters, and estimation uncertainty can overstate or destabilize the measured premium | Results across countries, periods, asset definitions, and statistical methods |
These explanations are not mutually exclusive. A model can combine several mechanisms, but adding flexibility also raises the risk of fitting historical data without producing reliable out-of-sample implications.
Before accepting a chart, estimate, or claimed solution, check:
This article provides financial and economic education, not personalized investment, valuation, tax, or legal advice. Historical premiums and model estimates do not predict or guarantee future returns.