Efficient Market Hypothesis

The core claim

Suppose someone discovered that stocks ending in vowels rise on Mondays. The discovery would be self-defeating: traders would buy vowel-stocks Friday afternoon, the price would rise immediately, and the Monday gain would vanish. The pattern existed only because no one knew it. Once known, it gets arbitraged away.

EMH generalizes this argument to all publicly available information. Anything anyone could exploit, someone already exploited; anything left in prices is statistical noise. New information arrives unpredictably, so price changes are unpredictable. The market becomes a fair game — you earn the risk-adjusted required return on average, but you can't beat it systematically with public data.

Fama's 1970 paper distinguished three nested forms of the hypothesis, varying by what information set "available" includes:

The three forms

Form	Information already in prices	What it rules out	Empirical status	Practical implication
Weak	Past prices and volume	Technical analysis based purely on charts	Mostly supported — but momentum (Jegadeesh-Titman 1993) is a partial counterexample	Don't pay for chart-reading services
Semi-strong	All public information (prices + filings + news + analyst reports)	Fundamental analysis using public data	Mixed — earnings drift, drift around analyst upgrades, M&A announcements all contradict it	Active management adds little value after fees
Strong	All information including private/inside	Profitable insider trading	Rejected — insiders consistently beat the market (which is why insider trading is illegal)	Even with private information, beating the market in expectation is hard if it's already illegal to use it
Adaptive markets (Lo 2004)	Information varies with environment	Static efficiency in changing markets	Behavioral synthesis, debated	Markets are sometimes efficient, sometimes not — depends on capital flows
Grossman-Stiglitz (1980)	"Efficiently inefficient"	Costless information	Theoretically rigorous, widely accepted	Active research must earn enough to cover its costs, or no one will do it
Behavioral finance	Rejects EMH; prices reflect crowd psychology	Rational pricing as default	Strong on bubbles and anomalies; weaker on stable equilibria	Bubbles are real; sentiment matters; arbitrage is risky

Each form is strictly stronger than the one above it. If strong holds, semi-strong holds; if semi-strong holds, weak holds. Most modern academic finance treats weak-form as essentially supported, semi-strong as supported with caveats, and strong-form as falsified.

The clearest implication: passive beats active on average

If markets are even semi-strong efficient, the average active fund — which charges 0.5-1.5% in fees and trades to incorporate public information — must underperform a low-cost index fund (0.03% fees) by roughly its fee differential. This is what William Sharpe (1991) called the "arithmetic of active management": before fees, the average dollar in active management earns the market return by definition; after fees, it must underperform.

The empirical record supports this overwhelmingly. SPIVA (S&P Indices Versus Active) reports show that over 15-year horizons, roughly 85-90% of active U.S. large-cap equity funds underperform the S&P 500. Similar patterns hold globally. Vanguard founder Jack Bogle built a $9 trillion firm on this single insight.

Shiller's excess volatility critique

Robert Shiller's 1981 paper asked a deceptively simple question: if stock prices are rational forecasts of future dividends discounted at a constant rate, how volatile should they be? He computed the present value of actual realized dividends from 1871 to 1979 (treating future dividends as known) and compared its volatility to the volatility of actual stock prices over the same period.

The actual price was 5-13 times more volatile than the ex-post rational price. There is no constant-discount-rate forecast of future dividends that is volatile enough to justify the historical volatility of the S&P. Either rational expectations are wrong, or discount rates vary enormously over time — and if the latter, EMH loses much of its empirical bite because almost any price movement can be rationalized as a change in expected returns.

Shiller's CAPE (cyclically adjusted P/E ratio), now followed widely, makes the same point in cross-section. CAPE has ranged from 5 (1920) to 44 (1999) — bigger swings than fundamentals can justify. Shiller's Irrational Exuberance (2000) called the dot-com bubble months before it burst, and the second edition (2005) called the housing bubble.

Behavioral finance and the Mandelbrot critique

Behavioral finance — Shiller, Richard Thaler, Daniel Kahneman, Andrei Shleifer — argues that systematic cognitive biases prevent prices from being rational averages of expectations. Investors anchor on recent prices, extrapolate recent trends, herd, panic, and over-react to news and under-react to less-salient information. The 1987 Black Monday crash (22.6% S&P drop in one day with no news catalyst) is a frequently cited counterexample to rational pricing.

Benoit Mandelbrot's critique runs deeper. Standard finance assumes returns are roughly Gaussian (or at worst log-normal). Mandelbrot showed in the 1960s that cotton prices, then stocks, exhibit fat tails — extreme moves happen orders of magnitude more often than a normal distribution predicts. The 1987 crash was a 22-sigma event under Gaussian assumptions — should occur once in the lifetime of the universe. Under Mandelbrot's stable Paretian distributions, with fractal scaling, such moves are rare but expected. The implication: not only are markets potentially irrational, but the volatility of irrationality itself is wildly unpredictable.

Documented anomalies

• Post-earnings announcement drift. Stocks that beat earnings continue rising for 60 days after the announcement. Documented by Bernard and Thomas (1989), it has not disappeared despite four decades of publicity.
• Momentum. Past 3-12 month winners outperform losers over the next 3-12 months. Jegadeesh and Titman (1993). Persistent across 30+ years and most asset classes.
• Long-horizon reversals. Past 3-5 year winners underperform over the next 3-5 years. De Bondt and Thaler (1985).
• Closed-end fund discounts. Funds trade at persistent discounts to their NAV — sometimes 10-20% — that EMH should arbitrage away.
• Equity premium puzzle. Stocks have earned 5-7% over bonds, far more than standard utility models can rationalize. Mehra and Prescott (1985).
• January effect, Monday effect, turn-of-month effect. Calendar anomalies that persist or fade unevenly.

Why EMH survives anyway

Despite documented anomalies, EMH remains the default null hypothesis in academic finance for three reasons:

• Joint hypothesis problem. A test of EMH is jointly a test of EMH and a model of expected returns. Reject the joint test, and you can't tell which side failed. Anomalies might be real or might be measurement artifacts.
• Most professionals still lose to indexes. Whatever inefficiencies exist, they're hard enough to exploit that the median active manager doesn't.
• Limits to arbitrage. Shleifer and Vishny (1997) showed that even known mispricings can persist — short-selling is risky, capital is finite, and arbitrageurs can be liquidated before being right.

Practical takeaways

• Default to index funds. The cost-after-fees math is overwhelming.
• Treat any "edge" you think you've found with extreme suspicion. The likely candidates: data-mining, survivorship bias, or compensating for unmeasured risk.
• Be wary of bubbles. Shiller's CAPE-and-narrative framework has called several major turning points; rational efficient markets did not.
• Distinguish "market is efficient" (mostly true on average) from "this specific price is right" (often false at extreme valuations).
• Risk premia vary over time — efficient markets do not require constant expected returns. Predictability of returns at the index level (e.g., from CAPE) is consistent with EMH if it reflects time-varying discount rates.

Common pitfalls

• Confusing EMH with random walk. Random walk is a particular form of EMH (constant expected returns). Modern EMH allows time-varying expected returns; predictability per se doesn't violate it.
• Citing one anomaly as a refutation. Anomalies in a non-anomalous-on-average market are consistent with efficiently priced risk you haven't measured.
• Conflating EMH with utility maximization. EMH is about prices reflecting information; rational expectations and rational behavior are separate (and stronger) assumptions.
• Believing a famous investor's success refutes EMH. With millions of investors, some will produce 30-year track records by luck. Buffett-style success is consistent with EMH if the population of attempts is large enough — though Buffett's specific record stretches that defense.
• Trading on EMH. "Markets are efficient, so I'll buy index funds" is the safest application; "Markets are inefficient, so I'll trade more" usually loses to fees and bid-ask spreads.