The Equity Premium Puzzle: Stocks Pay Too Much, and Nobody Knows Why

Most famous economics papers announce a solution. This one announces a failure, and it is far more celebrated for the failure than most papers are for their successes. Mehra and Prescott set out to check whether the standard, respectable, consensus model of how people trade off risk and reward could explain the single most basic fact about financial markets: that stocks pay more than bonds.

It could not. Not by a little. By a factor of roughly ten.

The problem: the numbers do not remotely add up

The fact is uncontroversial. Over the long US historical record they examined, running from 1889 to 1978, stocks returned around six percentage points a year more than short-term government debt. Everybody knew this. The standard explanation was equally familiar: stocks are riskier, so investors demand compensation for bearing that risk.

Mehra and Prescott did something nobody had bothered to do carefully. They asked: exactly how risk averse would people have to be for that story to work?

To answer, they used the consumption-based framework that Lucas and Breeden had built. In that framework, stocks are risky in one specific sense: they tend to do badly when the economy does badly, that is, when people's consumption is falling and every dollar matters most. The size of the premium you demand for holding stocks should therefore depend on two things: how much consumption actually fluctuates, and how much you hate those fluctuations.

They plugged in the real numbers.

And here is the crux of the whole paper. Aggregate consumption is remarkably smooth. Real per-capita consumption growth barely moves. In good years people consume a bit more; in bad years a bit less. Even in genuinely bad recessions, aggregate consumption dips by a small percentage. It is not a rollercoaster; it is a gentle slope with small bumps.

So stocks, from the point of view of this theory, are not actually very dangerous. They do not threaten to plunge you into starvation. They correlate with a variable that hardly moves. And an asset that is correlated with something that hardly moves should command only a tiny risk premium.

The model says the premium should be a fraction of one percent. The data says six percent. To close that gap using risk aversion alone, you would have to assume a level of aversion so extreme that it implies absurd behavior elsewhere: a person who would rather give up a large fraction of their wealth for certain than accept a modest coin-flip gamble. Nobody is like that.

The key idea, via analogy

Imagine you are offered a job that pays a huge premium over an identical job, with the only difference being that the risky one occasionally makes you eat a slightly smaller lunch.

If the downside is that mild, no rational person would demand a large premium to accept it. A big premium for a small annoyance means one of three things: you are absurdly fussy about lunch, or the risk is much worse than it looks, or something else entirely is going on.

That trilemma is exactly the state of the equity premium literature. Either investors are implausibly risk averse, or stocks carry some danger the smooth consumption data is failing to reveal, or the whole framework is wrong.

The authors did not claim to solve it. They named it a puzzle and left it on the table.

Why it mattered

• It exposed the emptiness at the center of asset pricing. Finance's most fundamental empirical fact, the equity premium, has no accepted explanation. That is a startling admission for a mature field, and this paper is what forced it into the open.
• It became the benchmark that every new model must clear. Habit formation (Campbell and Cochrane), long-run risk (Bansal and Yaron), rare disasters (Rietz, and later Barro), heterogeneous agents, limited participation, ambiguity aversion, and prospect-theory preferences are all, in essence, attempts to answer Mehra and Prescott. An entire subfield exists because of this paper.
• It has a direct, practical implication. If the premium is unexplained, then nobody actually knows whether it will persist. Every retirement plan, every endowment model, every pension projection assumes stocks will keep out-earning bonds by a healthy margin. That assumption rests on history, not on theory.
• It is a model of how to write a negative result. Precise, honest, quantitative, and framed as an open question rather than a demolition. It made everyone want to solve it.

The honest limitations

• The premium may simply be smaller than it looked. The historical US premium is measured on the most successful stock market in history, which is a textbook case of survivorship bias. Markets that were wiped out, in Russia, in China, in interwar Europe, do not appear in the sample. Global evidence suggests the true forward-looking premium is likely lower than the US number, and later work with fuller data has trimmed the puzzle without eliminating it.
• Aggregate consumption may be the wrong measure. Only some people own stocks. The consumption of stockholders is more volatile than the national average, and if you use it, the required risk aversion falls. Whether it falls far enough remains contested.
• Rare disasters are missing from the data. If investors fear a catastrophic event that simply did not happen during 1889 to 1978, they would rationally demand a large premium for a risk we never observe in the sample. This is an elegant answer, and also a slightly unfalsifiable one, since the whole point is that the disaster is not in your data.
• The model's preferences are restrictive. Standard time-separable utility ties risk aversion to willingness to substitute consumption over time, forcing a single parameter to do two jobs. Epstein-Zin preferences break that link, and much of the modern literature depends on doing so.
• It also created a second puzzle. Trying to raise the equity premium by raising risk aversion drives the model's predicted risk-free rate absurdly high, contradicting the low real rates we observe. This is the risk-free rate puzzle, and it means you cannot fix the first problem by simply turning one dial.

The one-line takeaway

Mehra and Prescott showed that consumption is far too smooth for stocks to be as frightening as their returns imply, so either investors are irrationally terrified, or the risk we fear never showed up in the data, and forty years on the profession still cannot agree which.