The Efficient Market Hypothesis — and Its Critics

Reading-order note. Numbered #1 and drafted fifth, after the machinery. The debate is far easier to hold once a reader knows what a risk-adjusted expected return is, because, as the joint-hypothesis section below explains, the argument cannot even be stated without a model of expected return, and that is not a rhetorical point but the reason the question remains open.

It is the most consequential idea in modern finance and one of the most argued-about. This article gives both sides at full strength and resolves neither, because the honest position is that the question is not settled and, for a reason that turns out to be structural rather than temporary, may not be settleable. This portal takes no position on whether markets are efficient and recommends no strategy on either view. What a reader should take away is not a verdict but a working understanding of what each side actually claims and what follows either way.

What the claim is, and what makes it plausible

The mechanism is competition, not rationality. The hypothesis does not require investors to be sensible; it requires that enough informed, profit-seeking participants exist that any identifiable mispricing gets traded away. If a security is obviously cheap given public information, buying pressure removes the discount, and the very act of exploiting an inefficiency destroys it. Prices are efficient because people are trying to prove they are not. There are three forms, by how much information is claimed to be in the price. Weak form: prices already reflect all past price and volume data, which implies that patterns in charts alone cannot produce systematic excess returns. Semi-strong: prices reflect all publicly available information, filings, news, announcements, implying that analysis of public data cannot systematically produce excess returns. Strong form: prices reflect all information including private, which essentially nobody defends, since the existence of insider-trading law rests on the assumption that private information is valuable and profitable. Most serious argument concerns the semi-strong form. The supporting evidence is substantial. Prices adjust to major public news very quickly, often in minutes. Simple mechanical trading rules based on past prices have generally failed to survive out of sample. Most active funds underperform their benchmarks over long periods after costs. And persistent, identifiable manager skill is extraordinarily hard to demonstrate; the alpha article puts a number on that: a modest genuine edge needs decades of record before it can be distinguished from luck.

The critics, at full strength

The case against is not a fringe position and includes work that has been recognised at the highest level of the discipline. Documented anomalies. Return patterns associated with characteristics such as company size, valuation ratios, and price momentum have persisted across long periods and many markets, patterns that a strictly efficient market should not display, and which spawned the factor literature covered in Pillar 28. Excess volatility. Research has argued that prices move far more than the underlying stream of fundamental news can justify, which is difficult to reconcile with prices as sober information aggregators. Episodes without news. The most vivid counterexample is a market falling more than a fifth in a single day with no proportionate change in fundamentals, as in October 1987; asset-price booms and collapses more generally are hard to describe as continuous fair pricing. Behavioural evidence. Systematic, repeatable errors in how people process information, the subject of Pillar 32, undermine the assumption that mistakes are random and cancel out. And limits to arbitrage, which is the strongest structural criticism: correcting a mispricing requires capital, carries risk, and can lose money for a long time before being proved right, and professional managers who lose money for a long time lose their clients first. Mispricings can therefore persist precisely because exploiting them is dangerous; the corrective mechanism the hypothesis depends on is real but bounded.

Two problems that keep the debate open

First, the hypothesis contains a genuine paradox. If prices already reflected all information perfectly, nobody would be paid for gathering information, and if nobody gathered information, prices could not reflect it. Perfect efficiency is self-defeating: markets can only be efficient if they are inefficient enough to pay the people who make them efficient. The implication is that the useful question was never "is the market efficient?" but "how efficient, in what securities, and at what cost to whom?", a question about degree, in which large liquid markets are demonstrably harder to beat than small neglected ones.

Second, and more fundamentally: the hypothesis cannot be tested on its own. To say a return was "abnormal" you need a model of what a normal return would have been, and that model, as CAPM demonstrated, is itself contested. So every apparent anomaly admits two readings: the market was inefficient, or the model of expected return was wrong and the return was fair compensation for a risk the model failed to capture. No test can distinguish these, because any test evaluates the hypothesis and the model together. This is why decades of evidence have not resolved the argument, and why the same finding is cited by both camps as support. It is also the most intellectually honest thing this pillar has to say about its own subject matter: the disagreement is not a failure of effort but a property of the question.

What follows regardless of who is right

The remarkable thing about this argument is how little a reader's sensible conduct depends on its outcome. Costs are certain while any edge is speculative, true whichever side prevails, and the reason expense ratios matter more than forecasts. Information that reached you through a public channel is already in the price, since the price moved when the information became public, not when you read it; both camps agree on this and only disagree about whether the resulting price is right. Track records shorter than a decade carry very little evidential weight, per part one above. And to the extent inefficiency exists, it is concentrated where competition is thin, in obscure, small, or hard-to-analyse securities, which is also where trading costs are highest and information is worst, so the opportunity and the difficulty arrive together. The one posture the debate genuinely rules out is confidence: the efficient view says you cannot reliably beat the market, and the inefficient view says you can only do so by being right where a great many well-resourced professionals are wrong. Neither is an argument for certainty about any particular security.