Why Stocks Pay So Much More Than the Math Predicts
For close to a century, stocks have beaten government bonds by roughly 4 to 7 percentage points a year on average, a gap standard economic models cannot fully explain without assuming investors are almost irrationally fearful. This is the equity premium puzzle, one of the longest-standing open questions in empirical finance, and it matters directly for how much of the historical stock return you should actually expect to see repeat going forward.
Pricing assets through consumption
Consumption-based asset pricing starts from a different question than most textbook models: not "how volatile is this asset's price," but "does this asset pay off when I actually need the money." An asset is risky, in this framework, not because it bounces around, but because its returns are low exactly when your own consumption is under pressure, typically during recessions, when income is threatened, jobs are less secure, and an extra dollar matters more than it does in good times. A dollar earned during a downturn is worth more to your wellbeing than a dollar earned during a boom, so any asset that fails you precisely during downturns should have to pay a premium to get held at all.
Stocks fit that description closely: corporate profits and stock prices both fall sharply in recessions, the same periods when consumption growth slows. Government bonds, by contrast, often hold their value or even rise during recessions, as investors flee to safety and interest rates fall. This is the same intuition behind ordinary insurance: a policy that pays out precisely when a house burns down is valuable specifically because the payout arrives when it is needed most, and buyers will pay a premium for that timing even though the policy's expected payout, averaged across all outcomes, might be modest. An asset priced under consumption-based logic is really being asked the analogous question: does it behave like insurance against a bad economic outcome, or does it behave like the opposite, an asset that adds to the pain of a bad outcome by losing value at the same time. Consumption-based pricing therefore predicts stocks should carry a meaningful premium over bonds, sized in proportion to how strongly their returns move together with the pain of a bad consumption year. The strength of that co-movement, formally the covariance between an asset's return and consumption growth, is the entire engine of the model.
This framing has a genuinely useful implication that a purely price-volatility-based view of risk misses entirely. Two assets can have identical price volatility and still deserve very different risk premiums, if one tends to fall during recessions and the other tends to rise or hold steady. A commodity or an asset class whose returns are largely unrelated to the broader economic cycle, for instance, could in principle be quite volatile on its own terms while still requiring little consumption-based risk premium, because its swings do not systematically coincide with the periods when investors' own consumption is under the most pressure. This is a meaningfully different, and in some ways more demanding, definition of risk than the simple standard deviation used throughout most of introductory portfolio theory.
The math of the puzzle
The model's pricing relationship can be written, in simplified form, as equity risk premium ≈ risk aversion coefficient (γ) × standard deviation of consumption growth (σc) × standard deviation of stock returns (σs) × correlation between the two (ρ). This lets you work backward from an observed historical premium to figure out how risk-averse investors would have to be for the model to explain it.
Take realistic inputs: annual consumption growth has historically had a standard deviation of roughly 1.5%, a genuinely smooth series, since households and the broader economy work hard to avoid letting spending swing much even when income does. Stock returns, by contrast, have a standard deviation closer to 18% a year. The correlation between the two has historically run fairly low, around 0.20, since consumption is smoothed through saving, borrowing, and social insurance even when markets are volatile. Multiplying these together: covariance ≈ 0.015 × 0.18 × 0.20 = 0.00054 (0.054%). If the observed historical equity premium is 6.0% a year, solving for the risk aversion coefficient that would justify it gives γ = 0.06 ÷ 0.00054 = 111. A risk aversion coefficient near 111 implies an investor so averse to consumption risk that they would refuse extremely favorable coin-flip bets over trivial sums of money, a degree of caution far beyond what economists consider a reasonable description of how people actually behave in ordinary decisions.
The second worked example runs the calculation the other direction, using a risk aversion coefficient of 5, a figure closer to what studies of ordinary financial and gambling decisions suggest is plausible. Using the same covariance of 0.00054, the model-predicted premium is γ × covariance = 5 × 0.00054 = 0.0027, or 0.27% a year. Compared with the actual historical premium of roughly 6%, this leaves a shortfall of about 5.7 percentage points a year that a plausible level of risk aversion, applied to the actual smoothness of consumption growth, simply cannot account for. That gap, sitting between what reasonable risk aversion predicts and what markets have actually delivered, is the equity premium puzzle in numbers.
What the evidence shows
The puzzle has proven durable across different sample periods, different countries, and different ways of measuring consumption, which is part of why it remains a live area of research rather than a solved textbook exercise. Several explanations have been proposed, each addressing a different possible flaw in the simple version of the model above. One line of reasoning points to habit formation: investors may care not just about the level of consumption but about consumption relative to a recently adapted-to standard, making a given drop in spending feel much more painful than the raw consumption-growth numbers suggest, which would justify a larger premium without requiring an implausibly high, constant risk aversion coefficient. A second line points to rare, severe disaster risk, the possibility of infrequent but catastrophic economic collapses, worse than anything in the twentieth-century U.S. sample, that investors rationally price in even though none occurred often enough to show up clearly in the covariance calculation above. A third points to survivorship and selection: the historical U.S. record reflects one of the most successful capital markets and economies of the past century, and other countries' twentieth-century stock markets, including several that suffered wars, hyperinflations, or expropriations, delivered far weaker, and in some documented cases negative, long-run real returns, meaning the U.S. sample used in most textbook premium calculations may simply be a lucky draw rather than a representative one.
A fourth line of explanation focuses on market frictions rather than investor preferences at all: borrowing constraints, limits on how much leverage ordinary investors and even many institutions can actually take on, may prevent capital from flowing into stocks aggressively enough to compress the premium down to the level a frictionless model would predict. If some investors who would otherwise like to hold more stock, financed with borrowed money, at a lower expected premium, are simply unable to borrow enough to do so, the market-clearing premium can stay elevated for reasons that have nothing to do with how risk-averse the marginal investor actually is. This explanation, unlike habit formation or disaster risk, points toward a genuinely structural, persistent reason the premium might stay wide rather than a description of investor psychology or rare tail events.
None of these explanations has fully closed the gap on its own, and researchers continue to debate how much weight each deserves. What is well established, however, is the practical implication: the historical U.S. equity premium of roughly 6% real, measured against short-term government debt, sits well above what any standard, internally consistent model of investor risk aversion can justify using observed consumption behavior. That should make you cautious about assuming the next 40 years reproduce the last 100.
Using this in a real portfolio
For financial planning purposes, the practical lesson is to treat the historical equity premium as an upper bound on what to expect, not a baseline. Many long-run planning models today use a forward-looking equity premium in the range of 3% to 5% over government bonds rather than the full historical 6%, partly because valuations have risen over the sample period (mechanically lowering the return going forward for any given level of future earnings growth) and partly in direct acknowledgment of the puzzle: some of the historical premium may reflect a one-time re-rating or survivorship luck rather than a permanent structural feature of markets.
For a high-earning professional building a retirement plan or evaluating how much of a portfolio to hold in stocks, this argues for building in a margin of safety rather than assuming decade after decade of 10% annualized returns. A plan that only works if stocks deliver their full historical average is a plan built on the least well-explained number in empirical finance.
This has a specific, quantifiable implication for retirement modeling. A savings plan calibrated to a 10% annualized return needs a materially lower savings rate today than one calibrated to a 7% or 8% forward return, since the compounding gap between those assumptions widens enormously over a multi-decade career. Building a plan around the more conservative, evidence-consistent forward premium costs little if returns turn out to be generous, since the surplus simply arrives sooner than planned, but it can be the difference between a secure and an insecure retirement if returns come in closer to what plausible risk-aversion models, rather than the twentieth-century U.S. record, would predict.
Actionable breakdown
- Setting return assumptions
- Use a forward equity premium closer to 3 to 5%.
- Avoid extrapolating the full historical 6% forward.
- Stress test plans against a lower realized premium.
- Understanding what you're paid for
- Remember stocks pay a premium for recession risk specifically.
- Expect the premium to show up unevenly, not smoothly.
- Treat drawdowns as the mechanism, not a bug.
- Avoiding survivorship bias
- Do not assume every country's market behaves like the U.S.
- Diversify internationally rather than betting on one record.
- Read "always wins over 30 years" claims skeptically.
Common pitfalls
Extrapolating the historical premium forever: valuations have risen over the past century, mechanically lowering the forward-looking premium; assuming the past decade's return repeats overstates likely future results.
Confusing volatility with the actual risk being priced: the theory says the danger is falling during recessions specifically, not day-to-day price swings; investors who panic-sell during downturns destroy the very premium they were meant to earn.
Ignoring survivorship bias: long-run U.S. stock data reflects one of history's most successful economies; several other developed markets have delivered far weaker, even negative, long-run real returns over comparable stretches.
The bottom line
The equity premium is real compensation for bearing recession risk, but its historical size is larger than any standard model of reasonable risk aversion can justify, so plan around a smaller, more conservative premium than the history books suggest.
Risk and risk premiums · Bills and inflation, 1926 to 2012 · The Capital Asset Pricing Model · Market history · Understanding risk