Why Stock Prices Move Like a Random Walk
Every new investor eventually studies a price chart looking for a pattern that predicts what comes next. The random walk model explains, in precise statistical terms, why that search usually fails, and why the failure is not an accident of bad luck but a direct consequence of how competitive markets process information.
The core mechanism: why prices should be unpredictable
The efficient market hypothesis holds that a security's price at any moment reflects all information currently available about it. If that is true, then the only thing that can move the price going forward is new information, information that has not yet arrived and, by definition, cannot be known in advance. A price series driven purely by the arrival of unpredictable news should therefore look statistically like a random walk: each step, or price change, is independent of the step before it, so knowing yesterday's move tells you nothing useful about tomorrow's.
The mechanism that produces this outcome is competition, not coincidence. Suppose a stock reliably rose 1 percent every Monday because of some recurring, identifiable pattern in order flow. Professional traders who noticed this would buy on Friday afternoon and sell Monday, and their buying would push the Friday price up until the Monday gain disappeared, arbitraged away by the very act of exploiting it. Any pattern predictable enough to trade on profitably tends to be predictable enough for someone else to discover and eliminate. What remains, once the easily found patterns have been traded away, is a price series whose short-term changes carry very little residual predictability, which is what a random walk describes mathematically.
It is worth being precise about which form of efficiency this describes. The weak form of the efficient market hypothesis specifically claims that past prices and trading volume, the raw material of technical chart analysis, contain no information useful for predicting future price changes beyond what is already reflected in the current price. This is a narrower and more testable claim than the broader idea that markets are always right, and it is the form most directly supported by the statistical evidence on random walks. Stronger forms of the hypothesis, concerning whether public fundamental information or even private insider information is fully reflected in price, are separate questions with separate, weaker evidence behind them.
A closely related concept from probability theory, the martingale, sharpens this idea further. A martingale is a sequence in which the best forecast of tomorrow's value, given everything known today, is simply today's value. A random walk with no drift is a martingale; a random walk with a small positive drift, which is closer to how most economists actually describe long-run stock prices given that equities carry a positive expected risk premium over time, is a "submartingale," where the best forecast of tomorrow is today's price plus a small, steady upward tilt reflecting compensation for bearing risk. This distinction matters in practice: it explains why "prices are unpredictable in the short run" is fully compatible with "stocks have a positive expected return over the long run." The unpredictability applies to the deviations around the trend, not to the existence of the trend itself, which is why a long-horizon investor can rationally expect positive returns while still having no ability to predict next Tuesday's close.
The math: variance scaling and the odds of a false pattern
Worked example 1: how volatility should scale under a true random walk. If daily returns are independent draws from the same distribution, a foundational property of random walks is that variance is additive across time, which means standard deviation scales with the square root of the number of periods, not with the number of periods itself. Suppose a stock has a daily return standard deviation of 1 percent. Over a five-day trading week, the predicted standard deviation is 1% x sqrt(5) = 1% x 2.236 = 2.24%. Over a full trading year of 252 sessions, it is 1% x sqrt(252) = 1% x 15.87 = 15.87%, a figure that lines up closely with the annualized volatility level, roughly in the mid-teens to low twenties, that broad U.S. stock indexes have actually exhibited over long historical periods. That correspondence between a simple theoretical prediction and real observed data is one of the quieter but more persuasive pieces of evidence for the random walk description: if returns were strongly trending rather than close to independent, volatility would scale faster than the square-root rule, and if they were strongly mean-reverting, it would scale slower.
Worked example 2: how often a false "hot streak" appears by pure chance. This example addresses a different and equally important question: even if daily price moves really are close to independent coin flips, how often should an investor expect to see what looks like a meaningful winning streak purely by accident? Model an up day as a fair coin flip, probability 0.5, over a 10-trading-day window (two calendar weeks). Using the binomial distribution, the number of ways to get exactly 7, 8, 9, or 10 up days out of 10 is C(10,7) + C(10,8) + C(10,9) + C(10,10) = 120 + 45 + 10 + 1 = 176, out of 2^10 = 1,024 equally likely sequences. The probability of seeing 7 or more up days in a random two-week stretch is therefore 176 / 1,024 = 17.2%, meaning roughly one out of every six two-week periods should produce what looks, on a chart, like a strong and meaningful winning streak, purely from chance alone, with no underlying pattern present at all.
It is worth extending this second example one step further, because the practical failure of technical trading rules is easier to see once transaction costs enter the picture. Suppose an investor adopts a rule that buys after any streak of 7 or more up days within a 10-day window, reasoning that momentum is confirmed. Because roughly 17.2 percent of random 10-day windows will trigger this rule by chance alone, and because a genuine random walk has no memory, the expected return following the signal is the same as the market's unconditional expected return, not higher, before costs. If each such trade incurs even a modest 0.15 percent in combined bid-ask spread and commission on both entry and exit, that is 0.15% x 2 = 0.30% of guaranteed drag per round trip, subtracted from a signal that carried no genuine predictive edge in the first place. Trade this rule ten times a year and the cumulative cost is roughly 0.30% x 10 = 3.0% of the portfolio annually, a real and certain loss purely from trading friction layered on top of a pattern that was statistical noise to begin with.
What decades of testing actually show
Formal statistical tests of the random walk model, most commonly measuring the autocorrelation of daily and weekly returns, the correlation between today's return and returns some number of days earlier, have consistently found autocorrelations very close to zero at short horizons across a wide range of markets and time periods, which is broadly consistent with the weak-form efficient market hypothesis. Related tests, including runs tests that check whether the sequence of up and down days looks statistically distinguishable from a random coin-flip sequence, and variance-ratio tests that check whether volatility actually scales with the square root of time as worked example 1 predicts, have produced similar conclusions: short-horizon price changes are difficult to distinguish from statistical noise using only past price and volume data.
The picture is not perfectly clean, and an honest account has to say so. Researchers have documented small, statistically detectable deviations from a pure random walk at certain horizons, including short-term momentum, where recent winners continue outperforming for a period of months, and longer-horizon mean reversion, where stocks and broad indexes that have done unusually well or badly over three to five years show some tendency to partially reverse. These findings are real and have been replicated across multiple datasets, but two things limit their practical relevance for most investors: the effects are small relative to a stock's total volatility, and exploiting them requires frequent trading that generates transaction costs and taxes large enough, in most documented tests, to consume most or all of the theoretical edge before it reaches an investor's actual account.
A separate strand of evidence comes from studying professional forecasters directly rather than testing statistical properties of price series. Surveys tracking the short-term price targets and directional calls made by professional equity strategists and technical analysts have generally found accuracy rates for near-term market direction not meaningfully better than chance, alongside wide dispersion of opinion among forecasters looking at the identical set of public information. If short-term price direction were reliably predictable from available data, this professional community, with access to institutional research budgets and decades of collective experience, would be among the first groups expected to demonstrate it consistently, and the absence of a persistent, identifiable subset of forecasters who reliably outperform is itself informative.
What this means for a real portfolio
The practical implication for an individual investor is not that price patterns are meaningless in some abstract philosophical sense, but that the specific, actionable patterns visible on a price chart, head-and-shoulders formations, support and resistance lines, moving-average crossovers, have not held up as reliable, tradable sources of profit once transaction costs, taxes, and the sheer number of other market participants searching for the same patterns are accounted for. Any trading rule based purely on past price and volume is competing against thousands of professional quantitative researchers running the same class of tests on the same public data, and a rule simple enough for an individual investor to compute by eye is, almost by construction, simple enough to have already been found and arbitraged by that competition.
This does not mean an investor should ignore price entirely. Valuation still matters over long horizons, since prices well above or below fundamentals do eventually tend to correct, and the random walk description applies most cleanly to short-horizon price changes rather than to the multi-year relationship between price and underlying business value. The practical takeaway is narrower and more useful: build a portfolio and a savings plan around a long time horizon and a fixed asset allocation rather than around an expectation of correctly predicting next week's or next month's direction, since the statistical evidence gives very little reason to believe that short-term prediction, from chart patterns alone, is a repeatable skill for anyone without a genuine informational or structural edge.
Actionable breakdown
- Treat any chart-based "signal" as likely noise until proven otherwise.
- Remember a 17 percent chance exists for a false hot streak by luck alone.
- Use a fixed savings and allocation plan instead of short-term timing.
- Reserve pattern-based trading, if any, for a small, defined portion of capital.
- Separate short-term price noise from long-term valuation, which still matters.
Common pitfalls
- Mistaking ordinary statistical variation for a meaningful, tradable pattern.
- Believing "random" means prices never trend over long, fundamentals-driven horizons.
- Backtesting a rule on the same data used to discover it, which inflates results.
- Ignoring transaction costs and taxes when judging whether a pattern is profitable.
The bottom line
Short-term price changes behave close enough to a random walk that chart-based prediction rarely survives contact with real trading costs, which is the strongest practical argument for a long-horizon, low-turnover approach.
All articles · How markets work · Implications of the EMH · Are markets efficient · Efficient market hypothesis