Efficient market hypothesis and random walk theory
Efficient market hypothesis: which chart is the real one?
The efficient market hypothesis says that prices already reflect available information, so they move only on news, and new information cannot be predicted. Its close relative, the random walk theory, says that tomorrow’s step does not depend on yesterday’s. There are two charts below. One is real. The other is built from the same returns, shuffled.
InteractiveSame start, same end, same returns: only the order changes. Which of the two looks like the market?
The three forms of the efficient market hypothesis
Eugene Fama gave the efficient market hypothesis its classic form in his 1970 review, Efficient Capital Markets: A Review of Theory and Empirical Work, in the Journal of Finance. There is one question: what information is already in the price? The three answers are the three forms.
Fig. 1 · What is already in the price
- Weak form: past prices are already in the price. If it holds, studying the chart gives no lasting edge.
- Semi-strong form: all public information is in too. Earnings and news would enter the price as soon as they come out.
- Strong form: private information is in as well. Fama himself treats it as a limiting case: a benchmark for measuring deviations, not an exact description.
The debate is not closed. In 2013 the Nobel Prize in economics went jointly to Fama and Robert Shiller, with Lars Peter Hansen: two scholars on opposite sides of the same question.
What is the random walk theory?
In a random walk each step is independent of the ones before. That does not mean the price stands still: it means the road traveled says nothing about the road ahead. The idea is old: Louis Bachelier used it in his 1900 thesis, Théorie de la spéculation. Burton Malkiel brought it to a wide audience with A Random Walk Down Wall Street, first published in 1973.
Fig. 2 · The same cards in a different order
- Each session becomes a card. On it is that day’s return, real, taken from the series.
- The deck gets shuffled. The cards stay the same: the three largest, in amber, only change place.
- The final price does not change. Returns multiply, and the order of the factors does not change the product.
Malkiel tells the idea with two images. Charts built by flipping a coin, which look like stock charts. And the blindfolded monkey throwing darts at the financial pages, picking a portfolio that, he writes, would do just as well as one carefully selected by experts. That is his thesis: those who dispute it come further down.
Chance draws trends and patterns
A random walk is not flat. It rises, falls, bounces where it has bounced before. The charts below are shuffles: a written rule searched for trends and patterns inside chance, and found them.
Fig. 3 · Two “trends” born of chance
- A rise of +99% in 89 sessions, a fall of −21% in 41. Neither has a cause: it is only the order in which the cards fell.
- The eye looks for a direction even where there is none. That is why a line drawn afterwards proves nothing on its own.
Fig. 4 · Head and shoulders and double bottom, by chance
- The patterns are there, in proportion. Shoulders at the same height, a neckline, a break. A rule found them, not a hand.
- It does not prove that patterns are useless. It proves that seeing one is not enough: what counts is how often it shows up by chance, and how often because something is behind it.
Deeper dives: the trend · head and shoulders · inverse head and shoulders · support and resistance · the flag.
Does technical analysis work? What the efficient market hypothesis implies
If even the weak form holds, studying past prices gives no lasting edge. Technical analysts start from the opposite idea. There are studies on both sides, and none has settled the question.
What the hypothesis says
- The past is already in the price.
- A rule that reads it should stop giving an edge as soon as many people use it.
What technicians reply
- News spreads little by little and people imitate each other: that is why trends form.
- Chart readers answer that “random” may only mean that whoever measures has not found the order.
A study in favor
- Brock, Lakonishok and LeBaron (1992) test moving averages and trading range breaks on the Dow Jones from 1897 to 1986.
- They find that returns after signals are not explained by the models they test, the random walk included.
A study of caution
- Sullivan, Timmermann and White (1999) take up those rules and account for how many have been tried.
- In the period after the original research, the edge of the best rule does not hold.
Patterns put to the test
- Lo, Mamaysky and Wang (2000) recognize patterns with an algorithm, with no human eye.
- Some carry information about the distribution of returns: that is a measure of information, not of profit.
Where it stands
- The debate is open: conclusions change with market, period, costs and method.
- This page gives no verdict: it gives the tools to read other people’s verdicts.
Deeper dives: what technical analysis is · Dow theory · the simple moving average.
Criticisms of the random walk theory
The simplest random walk wants independent steps, all with the same distribution. On real series things go partly otherwise. Below are the measures for JPMorgan Chase: 3,016 daily returns from to , calculated by the page.
Fig. 5 · Fat tails: more extreme days than the bell curve
- Beyond 3 sigma, three standard deviations, there are 41 sessions; the normal curve would expect 8. Kurtosis, which measures the weight of the tails, is 16.3; the normal has 3. Benoit Mandelbrot pointed this out in 1963, on cotton prices.
- The worst session: −15.0%, , 9.8 sigma. The best: +18.0%, . Tails say how large moves are, not where they go.
Fig. 6 · Clusters: wide sessions arrive together
- In the real series wide sessions sit close together. In the shuffled one they are the same sessions, scattered. In amber, those beyond 3 sigma.
- Autocorrelation of the sizes: 0.51 against 0.05. Size is the return without its sign; autocorrelation measures how much it resembles the size of the session before. Shuffling the deck loses the link.
Fig. 7 · Autocorrelation at different distances
- Returns stay close to zero: -0.08 at one day, with the band at ±0.04. Yesterday’s direction says little about today’s. Small, though, does not mean nil.
- Sizes do not: 0.35 at one day, and they stay above the band for a long time. How much it will move can be measured better than where it will go.
Fig. 8 · The variance ratio of Lo and MacKinlay
- In a random walk variance grows in proportion to time. The ratio is 1 at every horizon. This is the Lo and MacKinlay test, from 1988.
- Here, at 10 sessions, it is 0.89; the chance band runs from 0.88 to 1.12. This is the simplest version, on one stock: the original study works on indexes and portfolios.
Who disputes the random walk and efficiency
Lo and MacKinlay
- In 1988 they reject the random walk on weekly returns of American indexes and portfolios, from 1962 to 1985.
- In 1999 they make it a book: A Non-Random Walk Down Wall Street.
Robert Shiller
- In 1981 he asks whether prices move too much compared with the dividends that should justify them.
- His answer is yes: the excess volatility thesis.
Kahneman and Tversky
- Prospect theory, 1979: a loss weighs more than an equal gain.
- It is one of the foundations of behavioral finance: traders are not the rational deciders of the models.
De Bondt and Thaler
- In 1985 they ask whether the market overreacts.
- They compare the biggest losers of the previous years with the biggest winners: afterwards, the losers do better.
Jegadeesh and Titman
- In 1993 they measure the opposite over shorter horizons: winners tend to keep winning for some months.
- This is momentum, and it matters to anyone who follows trends.
Grossman and Stiglitz
- In 1980: if the price already held everything, nobody would pay to get informed, and then the price would hold nothing.
- A perfectly efficient market, they argue, is impossible.
Defenders of efficiency answer. Fama, in 1998, observes that anomalies split between overreaction and underreaction, and that many vanish when the method changes.
Measuring the problem: selection, costs, sample
Telling whether there is something beyond chance is hard for three reasons. People try a lot and report the best. Every trade costs. There are never enough sessions.
Fig. 9 · The best of many walks with no direction
- A typical walk makes 0%; the best of 1,000 makes +138%. None of them knows anything: real returns drawn at random, with the mean removed.
- The more you try, the smarter the best one looks. Median of 21 runs, walks of 250 sessions. This is not a strategy: it is what chance produces.
Fig. 10 · How many sessions you need
- With 100 sessions the chance band reaches ±0.20; with 3,016, ±0.04. Inside the band, an autocorrelation cannot be told from chance.
- The dots are three estimates on the same stock. Only the number of sessions changes: on few, a measure can say almost anything.
1Hindsight selection
- You try a hundred rules and show the one that did best: this is data snooping.
- The right question: how many tries came before this one?
2Costs
- Commissions, spread, taxes and slippage are paid on every trade.
- A small, frequent edge can vanish: do the math net of costs.
3Sample
- A few cases, one period, one market are not enough.
- A measure that holds in one period can change sign in the next.
4Out of sample
- You set the rule on one period and judge it on another, never seen.
- If you look at it and then adjust the rule, it is no longer out of sample.
5Survivorship
- Vanished stocks must stay in the data.
- Studying only what is still listed measures only the survivors.
6Joint hypothesis
- Every test of efficiency is also a test of the model that says which price would be normal.
- If the test fails, you do not know which of the two was wrong.
Deeper dives: the trading journal · drawdown · the stop loss · how to choose stocks.
Common mistakes about the efficient market hypothesis
Efficient does not mean right
- A price can be wrong and correct itself later.
- The hypothesis says you cannot predict which way it is wrong, not that it is never wrong.
Random walk and efficiency are not the same
- The first concerns the statistics of prices, the second concerns information.
- Rejecting the random walk is not enough to reject efficiency.
Seeing a pattern is not proof
- Chance draws them: Fig. 4 is made that way.
- You have to count how often they appear, and how often nothing happens.
One example does not close the debate
- A winning trade does not prove that technical analysis works.
- A failed test does not prove that it never works.
Random does not mean flat
- A random walk makes rises, falls and bounces.
- A trend on a chart, by itself, does not tell chance from a cause.
How much is not where
- Sizes can be measured better than directions.
- Knowing that a session will be wide does not say whether it will rise or fall.
Interactive exerciseExercise: which series is real?
Three charts, one real. Chance gets it once in 3, and all 3 cases once in 27.Same start, same end, same scale. One is the real series, one is shuffled, one has steps from a normal curve. Click the one you think is real.
Where to go from here
The basicsThe other lessons in the «The basics» section
What is technical analysis
- the study of price and volume to recognise the trend and the levels
- three premises: price discounts everything, trends, history repeats
- exercise: uptrend, downtrend or sideways on three real charts
Candlestick chart
- the four prices of a session: body and shadows
- green and red, long and short bodies, long shadows, the doji
- find the candle on a real chart: the judge says yes or no with the numbers
Uptrend: higher highs and higher lows
- higher highs and higher lows: how to recognize an uptrend
- swing points, and the two signals that end a trend
- mark the higher lows yourself on a real chart
Support and resistance
- the level where the decline stopped, and the one where the rise stopped
- touches, time and volume: how much a level counts; once broken, it switches roles
- find the level yourself on a real chart: the judge counts the touches
Frequently asked questions
What is the efficient market hypothesis?
It is the idea that prices already reflect available information, so they move only on news that cannot be predicted. Eugene Fama gave it its classic form in his 1970 review.
What are the three forms of the efficient market hypothesis?
Weak: past prices are already in the price. Semi-strong: all public information is in too. Strong: private information is in as well. The classification is in Eugene Fama’s 1970 review.
What is the random walk theory?
It is the idea that each price step is independent of the ones before, so the road traveled says nothing about the road ahead. Burton Malkiel made it famous with A Random Walk Down Wall Street. It is related to the efficient market hypothesis but not the same: rejecting the random walk is not enough to reject efficiency.
Does technical analysis work?
If the weak form held, past prices would give no lasting edge. Whether it holds is still debated: there are studies on both sides, and conclusions change with period, market, costs and method. This page sets out the debate and gives no verdict.
Method and sources
| Source |
|---|
| Eugene F. Fama, Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance, 1970 |
| Burton G. Malkiel, A Random Walk Down Wall Street (W. W. Norton, first edition 1973) |
| Andrew W. Lo and A. Craig MacKinlay, Stock Market Prices Do Not Follow Random Walks, Review of Financial Studies, 1988; A Non-Random Walk Down Wall Street (Princeton University Press, 1999) |
| Robert J. Shiller, Do Stock Prices Move Too Much to Be Justified by Subsequent Changes in Dividends?, American Economic Review, 1981; Daniel Kahneman and Amos Tversky, Prospect Theory, Econometrica, 1979 |
| Brock, Lakonishok and LeBaron, Journal of Finance, 1992; Sullivan, Timmermann and White, Journal of Finance, 1999; Lo, Mamaysky and Wang, Foundations of Technical Analysis, Journal of Finance, 2000 |
| Mandelbrot, Journal of Business, 1963; Grossman and Stiglitz, American Economic Review, 1980; De Bondt and Thaler, Journal of Finance, 1985; Jegadeesh and Titman, Journal of Finance, 1993; Fama, Journal of Financial Economics, 1998 |
| John J. Murphy, Technical Analysis of the Financial Markets, ch. 1 |
Real series: JPMorgan Chase, from to , adjusted prices. The page builds the walks with fixed seeds: by shuffling the real returns, or by drawing steps from a normal curve with the same mean and sigma. Every measure is recalculated at each generation. The window was chosen on purpose because it holds a crash and the contrast shows: this is a declared selection, not a proof. This guide explains how the tools work: it does not contain investment recommendations and promises no returns. Our method · license · report an error.
DOMINA’s stock pages, indices and commodities are rebuilt every night. Leave your email: we write to you when the picture changes.
No third-party advertising. Unsubscribe with one click. Go to the platform →With your consent we use Google Analytics cookies to count visits. If you decline, the page works the same and no measurement cookie is set. Cookie policy