Efficient Markets: Information, Price Discovery, Behavioral Finance, and Adaptive Markets

If a piece of information seems to imply an easy investment profit, ask why the current price has not already adjusted. That question is the practical starting point for market efficiency. The answer may involve competition, risk, information costs, trading constraints, or a mistaken forecast—not necessarily a free opportunity.

Learning goals

You should finish able to distinguish no-arbitrage from informational efficiency, explain why efficiency does not require a literal random walk, calculate an event-study abnormal return, evaluate active performance after costs, and describe behavioral and adaptive explanations without turning them into guaranteed trading rules.

Use total returns and information genuinely available at the decision date. Expected returns, risks, and the appropriate benchmark must be specified before calling a result “abnormal.” Numerical examples ignore taxes unless stated otherwise.

1. What does market efficiency mean?

The Efficient Market Hypothesis, or EMH, concerns how prices incorporate a specified information set. Its economic implication is that investors should not be able to systematically earn abnormal risk-adjusted returns net of relevant costs merely by trading on information already reflected in prices.

Three qualifications are essential:

• Information set: past prices, public disclosures, and private information are different claims.
• Risk adjustment: a higher return may compensate for risk rather than reveal mispricing.
• Costs and feasibility: research, turnover, spreads, shorting, financing, and capacity affect achievable profits.

“Fair” in this context does not mean morally fair, socially optimal, or known with certainty. A competitive price can change sharply when new evidence changes expected cash flows or discount rates. Investors can suffer losses in an efficient market.

The three familiar forms

Form	Information claimed to be reflected	A relevant empirical question
Weak	Past price and trading information	Can a rule using only historical market data earn persistent net abnormal returns?
Semi-strong	Publicly available information	Do prices adjust to public news without an exploitable delayed response?
Strong	Public and private information	Can even privileged private information produce abnormal profits?

The statement “prices reflect all public information” describes the semi-strong form. Strong-form efficiency is a much more demanding claim. Treating all forms as interchangeable makes evidence hard to interpret.

2. No-arbitrage, market clearing, and efficiency are different ideas

An arbitrage is, in the standard frictionless definition, a feasible zero-net-investment trade with no possible loss and a positive payoff in at least one relevant state. No-arbitrage rules out such opportunities.

A strategy with positive expected profit but possible losses is not automatically an arbitrage. A stock believed to be undervalued can remain risky because its future payoff, sale price, or the estimate of fundamental value can be wrong.

Market clearing means quantities supplied and demanded balance at prevailing prices. It does not, by itself, prove that all information is incorporated or that beliefs are correct. No-arbitrage can hold while information processing is costly and prices are only partially revealing.

Worked distinction

Two claims have identical legally enforceable future payments, but one can be bought for 99 while the other can be shorted for 101. If shorting, settlement, and all costs allow exact offset, the initial difference is an arbitrage profit.

By contrast, buying a stock at 99 because your valuation is 101 exposes you to valuation error and adverse outcomes. Even a sensible expected gain is not the same logical claim as the matched-payment trade.

3. Unpredictable news does not require a literal random walk

A basic random walk in a log price $p_t$ can be written

$$p_{t+1}=p_t+a+\varepsilon_{t+1},$$

with independent identically distributed innovations of mean zero. This is a particular statistical model, stronger than a general claim of informational efficiency.

A more flexible return representation is

$$R_{t+1}=\mu_t+\varepsilon_{t+1},\qquad \mathbb E[\varepsilon_{t+1}\mid\mathcal I_t]=0.$$

Here $\mathcal I_t$ is the current information set. Expected returns $\mu_t$ can vary with risk and economic conditions, and conditional volatility can change even when innovations are not predictably profitable.

Unpredictable risk-adjusted surprises are the relevant intuition. Efficiency does not imply independent returns, constant volatility, zero expected returns, or a flat price forecast.

Under appropriate no-arbitrage conditions, discounted cum-dividend prices can be martingales under a suitable pricing measure. That is different from claiming the raw stock price is a martingale under real-world probabilities. Do not transfer risk-neutral pricing statements directly into return forecasts.

Worked news example

A stock trades at 50. An earnings announcement causes investors to revise its value to 55, and it immediately trades there. A 10% jump is compatible with rapid information incorporation. After seeing the public news, buying at 55 does not allow you to collect the earlier jump from 50.

Even a good earnings result can produce a price fall if investors expected something better. The market responds to the surprise relative to prior expectations, not merely to whether an announcement sounds positive.

4. How event studies investigate price discovery

An event study compares observed returns around an event with a benchmark for normal returns. For a simple market model:

$$R_{i,t}=\alpha_i+\beta_iR_{M,t}+\varepsilon_{i,t}.$$

Estimate normal-return parameters in a pre-event window. During the event window, calculate

$$AR_{i,t}=R_{i,t}-(\widehat\alpha_i+\widehat\beta_iR_{M,t}), \qquad CAR_i=\sum_{t\in\text{window}}AR_{i,t}.$$

This CAR is a sum of abnormal returns, not an exact compounded wealth return. Other designs use excess-return models, factor benchmarks, or buy-and-hold comparisons.

Worked calculation

A stock returns -8% on the event day. The market returns -1%, estimated beta is 1.2, and the daily intercept is zero. Normal predicted return is -1.2%, so

$$AR=-8\%-(-1.2\%)=-6.8\%.$$

If the next day’s abnormal return is +0.3%, the two-day CAR is -6.5%. This describes a benchmark-adjusted price movement; it does not by itself prove legal liability, irrationality, or a profitable trade available after observing the event.

Check announcement time, trading halts, concurrent news, model uncertainty, and statistical significance. A credible study defines the event and comparison method without selecting a window solely because the result looks dramatic.

5. Challenger: a striking case with an unresolved mechanism

On January 28, 1986, Challenger was lost shortly after launch. Maloney and Mulherin (2003) examine the market responses of four major shuttle contractors. They report that within roughly an hour, prices appeared to distinguish Morton Thiokol from the others. Its stock fell about 12% over the day, while the other contractors’ initial losses largely reversed.

The authors interpret the episode as rapid price discovery consistent with the later identification of the faulty component. They could not establish precisely how particular informed traders caused the adjustment; a trading halt also complicates a simple trading-by-trading story.

NASA’s Rogers Commission report is dated June 6, 1986, a little over four months after the accident. Avoid converting the story into a precise “six-month” lag or claiming the market itself conducted a technical investigation.

The case illustrates information aggregation, not universal market infallibility. Price evidence is neither a substitute for engineering evidence nor a judicial finding. One successful event study cannot prove efficiency across all markets, events, and information sets.

6. Active management: competition and arithmetic

Efficiency predicts difficulty earning persistent net abnormal returns from already incorporated information. That does not mean every active manager loses, that no skill exists, or that everyone should have identical portfolios. Constraints, tax positions, liabilities, and service needs can differ.

A separate argument is Sharpe’s arithmetic of active management. For a consistently defined market, passive holdings replicate the market, and active investors collectively hold the remainder. Before costs, the average dollar of active investment must earn the same market return; higher aggregate active costs lower its average net return. This accounting result does not require every price to be informationally perfect.

Worked cost example

Suppose the market earns 8% gross. A representative passive portfolio costs 0.10%, while aggregate active management costs 1.00%. With equal aggregate gross returns under the accounting setup, net returns are 7.90% and 7.00%.

An individual active fund might beat both. The difficult prospective question is whether skill can be identified before investing and whether its benefit exceeds fees, turnover, taxes, and the risk of benchmark mismatch. Selecting yesterday’s winners after observing outcomes introduces hindsight and survivorship problems.

Index replication also has costs and tracking differences. Passive investing is an implementation choice, not proof that the chosen index perfectly represents all risky wealth.

7. Why prices need not reveal everything for free

If obtaining information is costly and prices instantly revealed everything, paying to acquire information could have no reward. But without information production, what would keep prices informative?

Grossman and Stiglitz (1980) formalize this tension in a model with costly information and noisy prices. The lesson is not that research always pays. It is that perfectly revealing prices and incentives to pay for information need not coexist in that setting.

An economically useful view compares attainable trading benefits with the resources needed to obtain and use information. Some gross predictability can remain without an easy net profit available to everyone.

Worked net-edge calculation

Suppose a strategy’s estimated gross annual alpha is 2%. Expected trading and implementation cost is 0.8%, financing and stock-borrow cost is 0.7%, and the management charge is 0.6%. On these nonoverlapping assumptions, estimated net alpha is -0.1%, before considering uncertainty in the estimate.

The observed gross pattern is not equivalent to an economically attractive investment. Costs must be measured consistently to avoid either omitting or double counting them.

8. Behavioral finance: mechanisms, not automatic signals

Behavioral finance studies how attention, beliefs, framing, and preferences influence decisions. Efficient outcomes do not require every individual to be perfectly rational, so documenting a bias alone does not establish exploitable market mispricing.

Tendency	Meaning	Hypothetical investment manifestation
Loss aversion	Losses relative to a reference point weigh especially heavily	Refusing a sensible sale because it would realize a loss
Overconfidence	Excess certainty about ability or information	Trading more than the evidence justifies
Anchoring	Insufficient adjustment from an initial number	Treating a past share price as a reliable fair-value target
Framing	Equivalent choices appear different when described differently	Changing risk choices when outcomes are described as gains rather than losses
Mental accounting	Separating economically related resources into mental buckets	Taking excessive risk with money labeled “winnings”
Herding	Following others’ actions	Buying because a crowd is buying, with little independent evaluation

Averaging judgments works best when errors are not all driven by the same source. Shared leverage, common incentives, correlated beliefs, and attention can make a crowd less informative. Herding can also reflect rational responses to information or incentives; observing similar trades does not uniquely reveal psychological irrationality.

Arbitrageurs face constraints too. A perceived mispricing may worsen before correcting, forcing margin calls or investor redemptions. Uncertain convergence, shorting restrictions, and funding risk can prevent informed traders from immediately correcting it. A risky correction trade is not the frictionless arbitrage of Section 2.

9. Anomalies and the joint-hypothesis problem

Historically studied patterns include size, value, momentum, and accrual-related return differences. High book-to-market corresponds to low price-to-book; momentum commonly ranks intermediate past returns, while accrual studies compare accounting components of earnings. The Risk and Return lesson develops these definitions further.

These patterns vary across samples, markets, and implementation methods. They may reflect missing risk factors, behavior, frictions, measurement error, or chance. “Persistent” in a historical study does not mean a guaranteed permanent premium.

The joint-hypothesis problem is central: an abnormal return is abnormal only relative to a model of normal expected return. Rejecting zero CAPM alpha jointly challenges market efficiency and the benchmark; it does not uniquely establish inefficiency.

Worked benchmark example

A strategy averages 10% a year. CAPM predicts 7%, giving estimated CAPM alpha of 3 percentage points. A justified multifactor benchmark predicts 9.5%, leaving 0.5 percentage points before costs. The observations did not change; their interpretation did.

Adding factors merely to remove an inconvenient alpha does not validate them. A defensible test needs economically motivated exposures, information available at the time, uncertainty estimates, out-of-sample evidence, and feasible costs. Searching hundreds of rules and reporting only the strongest one exaggerates apparent evidence.

10. Adaptive Markets: efficiency as an evolving outcome

In Lo’s Adaptive Markets Hypothesis (2004), efficiency depends on competition, opportunities, and participants’ adaptation. People pursue their interests but make mistakes, use heuristics, and learn. Herbert Simon’s satisficing describes seeking an adequate feasible choice rather than solving an exhaustive optimization every time.

A rule can work in one environment and fail in another. Entry and imitation can erode an opportunity; changing conditions can create new ones. Risk premia and strategy effectiveness can vary, and innovation and survival shape which participants remain. This evolutionary account does not predict a regular calendar cycle or make adaptation costless.

AMH offers a framework rather than a universally calibrated trading equation. Institutional sophistication does not guarantee efficiency, and retail participation alone does not prove irrational pricing. Crisis behavior can combine incentives, rational liquidity demands, and emotion; these are not a simple logic-versus-emotion dichotomy.

An illustrative ecology of a strategy

Consider a hypothetical signal discovered in a small market. A few traders first have enough capacity to implement it. More competitors copy the signal, narrowing the opportunity. Leverage and crowded positions grow. A funding shock then forces simultaneous exits, changing both liquidity and the signal’s behavior.

This thought experiment motivates testable questions: does measured performance decline with capital chasing it? Does funding availability alter execution costs? Does a rule trained in one regime survive another? A persuasive adaptive explanation should make predictions that can be checked, rather than relabel every gain and loss as adaptation after the fact.

11. Time-varying risk and survival

A changing expected return is not automatically inefficiency: rational required premia can vary when risks or risk-bearing capacity change. An adaptive account can additionally investigate learning, competitive entry, and changing heuristics. Separating those channels requires evidence.

For a hypothetical position with annual expected excess return 4% and volatility 10%, Sharpe ratio is 0.4. If conditions change to expected excess return 2% and volatility 20%, it falls to 0.1. A constant allocation based on the old estimates now represents a different trade-off.

Even a positive expected return does not ensure survival. If an investor has 100 of equity, borrows 100, and holds 200 of risky assets, a 50% asset loss wipes out the equity before interest and costs. Funding or margin constraints can force exit much earlier. This is simple balance-sheet arithmetic, not a proof that a particular dynamic strategy is optimal.

Adaptive reasoning encourages monitoring assumptions, capacity, leverage, and liquidity. It does not provide advance knowledge of turning points or guarantee that frequent tactical changes improve performance after costs.

12. A disciplined way to evaluate an efficiency claim

1. Identify the information set and when it became tradable.
2. State the expected-return benchmark and relevant risks.
3. Distinguish a theoretical arbitrage from a risky expected profit.
4. Include research, financing, turnover, and implementation costs.
5. Check sample selection, multiple testing, and out-of-sample performance.
6. Consider alternative explanations and the conditions under which the claim should fail.

The useful question is not simply whether markets are “efficient” or “irrational.” It is how quickly a particular market incorporates particular information, at what cost, and under which conditions.

13. Practice questions

1. A trading rule uses only historical prices. Which form of efficiency is most directly relevant? What changes if it uses published annual reports?
2. A stock jumps immediately from 40 to 44 after unexpected good news. Does the jump disprove efficiency? Can an investor buying at 44 earn the already realized jump?
3. An event-day stock return is -6%, market return is -2%, estimated beta is 0.8, and daily intercept is zero. Find abnormal return. The next day’s AR is +0.4%; find two-day CAR.
4. A strategy has estimated gross alpha 1.5%, trading costs 0.6%, borrowing costs 0.5%, and a separate fee of 0.7%. Find estimated net alpha under these assumptions.
5. A fund averages 12%, while CAPM predicts 8% and an economically justified multifactor benchmark predicts 11%. Find both estimated alphas. Which conclusion follows about inefficiency?
6. An investor has equity 100 and debt 150 invested in assets worth 250. Ignoring interest, what asset-price decline wipes out equity?
7. Why does observing loss aversion among individuals not by itself prove an exploitable pricing error? Why is predictable volatility compatible with efficiency?
8. Give one testable prediction of an adaptive explanation for a crowded trading strategy, and one result that would not follow automatically from AMH.

14. Answers and reasoning

1. Historical-price rules test weak-form efficiency. Public annual reports bring the claim into the semi-strong information set. Both tests still need risk and cost adjustments.
2. No. Immediate adjustment to a surprise is compatible with efficiency. Buying at 44 occurs after the gain from 40 has been incorporated.
3. Normal predicted return is $0.8(-2\%)=-1.6\%$. AR is $-6\%-(-1.6\%)=-4.4\%$. CAR is $-4.4\%+0.4\%=-4.0\%$.
4. Net alpha is $1.5-0.6-0.5-0.7=-0.3\%$. The gross estimate alone does not establish a positive net opportunity.
5. Estimated CAPM alpha is 4 percentage points; multifactor alpha is 1 percentage point. Neither by itself identifies inefficiency, because model validity, estimation error, costs, and sample design still matter.
6. Assets must fall from 250 to debt value 150: a 40% decline. Equity bears the residual loss, so a smaller asset move can produce a much larger equity percentage loss.
7. Biases can offset, be anticipated, or be constrained by competition; the proposed trade can also be risky or too costly. Predictable conditional variance does not imply predictably profitable risk-adjusted return innovations.
8. A possible prediction is that net signal performance declines as competing capital and execution impact increase. AMH does not automatically imply that institutions always price correctly, that retail trading creates mispricing, or that a trader can forecast the next regime change.

Further reading

• Andrew W. Lo, MIT OpenCourseWare, Efficient Markets: the course lectures and slides.
• Michael T. Maloney and J. Harold Mulherin (2003), The Complexity of Price Discovery in an Efficient Market: The Stock Market Reaction to the Challenger Crash, Journal of Corporate Finance 9, 453–479.
• NASA, Report of the Presidential Commission on the Space Shuttle Challenger Accident: official investigation source, dated June 6, 1986.
• William F. Sharpe (1991), The Arithmetic of Active Management: aggregate gross and net performance accounting.
• Sanford J. Grossman and Joseph E. Stiglitz (1980), On the Impossibility of Informationally Efficient Markets: costly information and the incentives to acquire it.
• Andrew W. Lo (2004), The Adaptive Markets Hypothesis: Market Efficiency from an Evolutionary Perspective: the original adaptive framework.

Feynman study lab

Explain it simply

Market efficiency asks how much information prices already reflect and whether it offers usable excess returns.

Concrete example. A public earnings announcement may change price before a reader can place a trade.

Teach it back without looking

Close the lesson and explain the idea aloud to a beginner in two minutes. Define each technical term you use, walk through the example, and state one assumption or limitation. Then answer:

Does an efficient market require prices to stay constant or investors to predict every future event?

Check your explanation after you try

No. Prices can change with news. Efficiency concerns information incorporation and opportunities relative to risk and trading costs.

Find the gap and try again

If you hesitated, relied on a formula you could not explain, or missed the self-check, write down the exact gap. Revisit the relevant section of this lesson and work through its example. Close the notes, explain it again in simpler language, then change one number or assumption and predict what changes. Recalculate or check the lesson to test your prediction. Try once more tomorrow from memory.