Efficient Market Hypothesis: Forms, Tests, and Practical Limits

Distinguish weak, semi-strong, and strong market efficiency; interpret event studies and abnormal returns; and test whether an apparent edge survives risk, costs, timing, and out-of-sample evidence.

For educational purposes only; not investment advice. Investing may result in loss.

Direct answer

The efficient market hypothesis (EMH) is a family of hypotheses about how information is reflected in security prices. Its practical claim is not that every price is correct. It is that a strategy using a specified information set should not earn reliably positive risk-adjusted returns after realistic costs merely because that information is already available to competing market participants.

The claim must identify the market, information set, horizon, expected-return model, costs, and decision timestamp. Evidence for or against efficiency is therefore conditional rather than a universal verdict on every asset and period.

How it works

The conventional forms are nested information sets:

Weak form: Current prices reflect information in past prices and trading volume. A repeatable rule based only on that history should not produce net risk-adjusted excess returns.
Semi-strong form: Current prices reflect publicly available information, including filings, releases, and broadly disseminated news. This does not require literally instantaneous or costless adjustment.
Strong form: Prices reflect public and private information. This is the strongest claim and is contradicted by the premise that material nonpublic information can confer an advantage. Insider-trading law and Regulation FD govern conduct and disclosure; they are not proofs of strong-form efficiency.

Prices can change sharply in an efficient market because new information changes expected cash flows, risk, discount rates, or the distribution of possible outcomes. Efficiency also does not require every investor to be rational. Competition, arbitrage, and the losses faced by systematically wrong traders can still constrain predictable errors, although funding, short-sale, liquidity, mandate, and horizon limits can prevent immediate correction.

Information is costly to collect and interpret. If prices were perfectly informative without compensating information production, investors would have little incentive to acquire information. A realistic view therefore allows some gross reward for research while asking whether that reward survives data, labor, trading, financing, taxes, capacity, and model risk.

An event study often begins with:

abnormal return on day t = security return on day t - expected return on day t

and aggregates the selected window as:

cumulative abnormal return = sum of abnormal returns over the event window

The expected return may come from a market-adjusted benchmark, factor model, or another prespecified model. This creates the joint-hypothesis problem: rejecting zero abnormal return may indicate market inefficiency, a poor expected-return model, bad data, or both. A statistically significant result is not automatically tradable because the public release may precede the observable trade, quotes may move before execution, and costs or capacity may absorb the edge.

Long-horizon tests are especially sensitive to benchmark choice, overlapping observations, delisting treatment, survivorship, multiple testing, and small-sample inference. A credible strategy should be specified before evaluation, use point-in-time data, reserve an untouched out-of-sample period, and report both failed and successful variants.

Example

Assume a company releases earnings publicly at 4:05 p.m. ET after a regular-session close of US$50.00. The next official opening price is US$54.00. Measured close-to-open, the stock return is:

stock event return = (US$54.00 / US$50.00) - 1 = 8.0000%

Suppose a prespecified broad-market benchmark returns 1.0000% over the same interval. The simple market-adjusted result is:

abnormal event return = 8.0000% - 1.0000% = 7.0000 percentage points

The price gap is consistent with rapid incorporation of public news, but one observation neither proves semi-strong efficiency nor proves that US$54.00 equals intrinsic value. A trader learning the news at 4:05 p.m. ET could not assume execution at the stale US$50.00 close.

Now assume the stock ends the next regular session at US$52.00, a close-to-close return of:

stock close-to-close return = (US$52.00 / US$50.00) - 1 = 4.0000%

If the benchmark’s close-to-close return is 1.5000%, the one-day market-adjusted return is 2.5000%. This does not contradict the 7.0000 percentage-point close-to-open result: the event windows and executable prices differ.

Next consider a rule tested across many public announcements. Its portfolio earns a gross return of 9.2000% while its matched benchmark earns 7.5000%. Estimated trading costs are 0.9000%, and data plus financing costs are 0.3000%:

net pre-tax excess return = 9.2000% - 7.5000% - 0.9000% - 0.3000% = 0.5000%

If live slippage is 0.4000% worse than the backtest assumption, the remaining excess return is only 0.1000% before tax and before any correction for factor exposures. The strategy is not established as an exploitable inefficiency until it survives risk adjustment, point-in-time reconstruction, multiple-testing correction, out-of-sample evidence, and realistic execution.

Risks and verification checklist

State the hypothesis: Name the weak, semi-strong, strong, or narrower information-set claim being tested.
Define availability: Record when each price, filing, release, estimate, and dataset was actually observable.
Use point-in-time data: Preserve original values rather than revised, restated, or backfilled histories unavailable then.
Choose the event clock: Identify time zone, release timestamp, trading session, and exact start and end prices.
Predefine the window: Avoid selecting an event window after seeing which interval produces the desired result.
Model expected return: State the benchmark or factor model and why it matches the security and horizon.
Recognize the joint test: Separate possible mispricing from expected-return-model error.
Include distributions: Adjust consistently for dividends, splits, rights, spin-offs, and other corporate actions.
Include failed securities: Preserve delistings, bankruptcies, and inactive names to prevent survivorship bias.
Control selection: Explain exclusions, missing observations, liquidity screens, and changing index membership.
Correct multiple testing: Count discarded signals, parameters, universes, windows, and model variants.
Reserve holdout data: Keep a genuinely untouched period or market for out-of-sample evaluation.
Estimate all costs: Include spread, commissions, fees, market impact, borrow, financing, data, labor, and taxes.
Use executable prices: Do not trade retrospectively at a quote or close that preceded public information.
Test capacity: Measure how returns change with order size, turnover, crowding, and liquidity.
Adjust for risk: Check market, size, value, momentum, quality, duration, volatility, and downside exposures as relevant.
Report uncertainty: Provide sample size, dispersion, confidence intervals, and economic as well as statistical significance.
Stress regimes: Test calm and stressed markets without treating one favorable subperiod as universal evidence.
Separate law from theory: Do not infer strong-form efficiency from disclosure rules or lawful access to information.
Demand a mechanism: Explain why the effect exists, why competition has not removed it, and what could eliminate it.

Common misconceptions

“Efficient prices are always equal to intrinsic value.” Efficiency concerns predictable profits from an information set; intrinsic value is unobservable and model-dependent.
“A price crash disproves EMH.” A large move can reflect new cash-flow or discount-rate information; distinguishing overreaction requires a prespecified model and evidence.
“Any predictable return proves mispricing.” Expected returns may compensate for risk, illiquidity, constraints, or other exposures, and the benchmark may be wrong.
“Strong-form efficiency is securities law.” Strong-form EMH is an information claim; insider-trading and selective-disclosure rules are legal standards with defined scope.
“Research is useless in an efficient market.” Research helps information enter prices and may earn a gross reward, but a claimed edge must survive competition, risk, and total cost.

Sources

Fama, Efficient Capital Markets: A Review of Theory and Empirical Work.
Grossman and Stiglitz, On the Impossibility of Informationally Efficient Markets.
Fama, Market Efficiency, Long-Term Returns, and Behavioral Finance.
Malkiel, The Efficient Market Hypothesis and Its Critics.
SEC, Selective Disclosure and Insider Trading.
Investor.gov, Executing an Order.