The Three Forms
Fama defined three nested versions of the hypothesis. Each makes a strictly stronger claim about which information set is impounded in price.
Weak Form
Price reflects all past trading data — historical prices, volume, and any function of them. Under this version, technical analysis cannot produce consistent excess returns, because any signal extractable from price-and-volume history is, by assumption, already in the current price. Moving averages, RSI, MACD, chart patterns — all are uninformative in expectation, modulo transaction costs.
Implication: if weak-form efficiency holds, technical strategies should produce zero risk-adjusted alpha after costs.
Semi-Strong Form
Price reflects all publicly available information — past prices plus financial statements, news, analyst reports, and macroeconomic releases. Under this version, fundamental analysis based on balance sheets, income statements, and P/E ratios also cannot produce consistent excess returns. By the time the information is public, it is priced.
Implication: if semi-strong efficiency holds, the long-horizon record of Warren Buffett is either statistical noise across a large pool of managers, or it is compensation for a risk factor the standard EMH tests have not modelled correctly.
Strong Form
Price reflects all information, including non-public (insider) information. This version is almost universally rejected on direct evidence: regulatory case studies of insider trades repeatedly document profitable execution before the relevant disclosure. The existence of profitable insider trading is sufficient to falsify the strong form.
Evidence Consistent With EMH
Active manager underperformance. Over rolling 15-year windows, roughly 85–90% of actively managed mutual funds underperform their benchmark net of fees. This is the largest single piece of evidence for semi-strong efficiency: if exploitable inefficiencies were widespread, professional managers should, on average, capture them.
Speed of price adjustment. Earnings surprises, M&A announcements, and macroeconomic releases are typically incorporated into price within seconds. By the time a retail trader reads the headline, most of the move has occurred.
Random-walk properties at short horizons. Statistical tests on daily and intraday returns show low serial correlation. Past short-horizon price changes have minimal predictive power for future short-horizon changes, consistent with weak-form efficiency.
Index-fund dominance. Since the launch of Bogle’s Vanguard 500 Index Fund in 1976, the fund has outperformed the majority of actively managed peers net of fees over multi-decade horizons. If active analysis added value in aggregate, this ordering would be reversed.
Evidence Inconsistent With EMH
Bubbles and crashes. The Tulip Mania, dot-com bubble, and 2008 housing bubble saw prices diverge sharply from any plausible fundamental anchor. A semi-strong-efficient market does not, in principle, support equity prices at 200× revenue for issuers without earnings.
Persistent outperformers. Buffett, Soros, Druckenmiller, Lynch, and Jones each produced multi-decade records whose joint probability under a pure-luck null is small. The standard EMH response — that ex-post selection from a large manager population guarantees survivors — is partial; it does not fully account for the magnitude and consistency of the records.
Documented anomalies. Empirical research has identified return patterns that should not persist under semi-strong efficiency: the value premium (low price-to-book outperforming high), the size premium (small caps outperforming large), the momentum effect (12-month winners continuing to outperform on average), and post-earnings-announcement drift. Some are now interpreted as risk factors and absorbed into multi-factor models; others remain unresolved.
Behavioural finance. Work by Daniel Kahneman and Amos Tversky, extended by Richard Thaler, documents systematic cognitive biases — overconfidence, anchoring, loss aversion, herding — that produce predictable deviations from rational pricing. If pricing errors are systematic rather than independent, they need not cancel out at the aggregate level.
The Operational Synthesis
The position consistent with the body of evidence is intermediate. Markets are approximately efficient, not perfectly so. The approximation is good enough that the average active strategy fails net of costs, and bad enough that specific, identifiable inefficiencies persist in measurable subsets of the market.
The subsets where inefficiencies are most defensible:
Small and micro-cap names. Lower analyst coverage, lower institutional ownership, and thinner information flow widen the dispersion of pricing errors. This is the segment where post-cost edges, when they exist, are easiest to document.
Extreme-sentiment regimes. During bubbles and panics, fear and greed dominate marginal pricing decisions. Mispricings of meaningful magnitude can persist for months.
Complex corporate events. Mergers, spin-offs, bankruptcies, and recapitalisations create temporary information-processing frictions that specialist analysts can sometimes exploit.
Information-arrival windows. Price adjustment is fast but not instantaneous. The window is short and crowded, but non-zero.
Practical Consequences
For non-specialist investors: the hypothesis is sufficiently close to true that the dominant strategy is a low-cost index fund, in line with Bogle’s argument. Over a 15-year horizon, this approach is expected to outperform roughly 85% of actively managed alternatives.
For active traders: the operational requirement is a specifiable edge. An edge is a documented statistical advantage with stated conditions: the universe in which it holds, the regime in which it holds, the entry and exit rules, the position-sizing convention, and the risk/reward discipline that keeps the average loss bounded relative to the average win. “I am skilled” is not an edge. A trader who cannot write the edge down in a paragraph that another trader could replicate probably does not have one.
Assumptions, to be explicit: the empirical results above (manager underperformance rates, anomaly magnitudes, random-walk test statistics) assume return distributions are sufficiently well-behaved for the standard tests to apply. Tail-driven regimes — the second-order phenomenon that EMH treats as exogenous — routinely violate that assumption. In Go terms, EMH describes the steady-state board; it does not describe the moments when shape collapses. Both readings are needed.
See also: John Bogle · Technical Analysis · Fundamental Analysis · Trading Psychology: Fear and Greed · Value Investing
