Investor psychology Lesson 11 of 15
Articles · Behavioral finance
Gambler’s fallacy and the hot hand: streaks in investing
Why a streak feels as if it must end, or must go on, what the research on streaks found, and what 33 years of SPY months show about the month after one.
On this page
The gambler's fallacy is the belief that a random process corrects itself: after five reds on a roulette wheel, black is "due". The hot-hand fallacy is the opposite belief, that a streak makes its own continuation more likely: the player who has made three shots will make the fourth. Both are mistakes about the same thing, which is how much a short run of results says about what comes next.
Investors hold both beliefs, often in the same week. A market that has risen five months in a row is "due for a correction". A fund manager who has beaten the index three years running "has it figured out". This lesson covers where each belief comes from, what the research found (including a famous result that was partly overturned in 2018), and what the monthly record of the S&P 500 says about the month after a streak.
Two mistakes with one cause
In 1971 Amos Tversky and Daniel Kahneman surveyed research psychologists, people trained in statistics, and found they expected small samples to resemble the population far more closely than chance allows. The pair named the belief the law of small numbers, a joke on the real law of large numbers, which says only that averages settle down over very many trials.
A fair coin lands heads about half the time over 10,000 flips. Over six flips it does nothing of the kind. Yet people judge the sequence H-T-H-T-T-H to be more likely than H-H-H-T-T-T, because the first "looks random". Both have exactly the same probability, 1 in 64.
If you believe short runs must look balanced, a streak creates a debt. Five reds must be "paid back" with blacks, soon. That is the gambler's fallacy. The wheel has no memory, so the chance of red on the next spin is what it always was.
The hot hand grows from the same root when the observer is unsure what kind of process they are watching. A run of five heads is too unbalanced to be a fair coin under the law of small numbers, so the observer concludes that the coin is not fair: the player is "hot", the manager is skilled. Matthew Rabin and Dimitri Vayanos formalized this link in 2010. In their model a person who believes in the gambler's fallacy, on seeing a long streak, switches to believing in a hot hand, because the streak is the only way left to explain what they saw.
| You believe the process is… | After a streak you expect… | Typical investing version |
|---|---|---|
| Random (a wheel, the index) | Reversal | "Five up months. It has to pull back now." |
| Skilled (a manager, a stock story) | Continuation | "Three great years. This fund knows something." |
| Unknown | Whichever fits the mood | Selling a winner as "due", buying another as hot |
What the research found
Casinos
Rachel Croson and James Sundali coded 18 hours of security video of roulette at a Reno casino: 139 players and about 24,000 bets. After a streak of one outcome (red or black, odd or even), they counted bets with and against the streak. After a streak of one or two the bets were split about evenly (48% and 51% against). After a streak of five, about 65% of bets were placed against it, and after streaks of six or more, 85%.
Judges, loan officers and umpires
Daniel Chen, Tobias Moskowitz and Kelly Shue looked for the gambler's fallacy where the stakes are higher than a casino chip. In US asylum courts, cases reach a judge in an order that is effectively random, so one decision should not predict the next. They found that a judge was up to 3.3 percentage points more likely to deny asylum after granting it in the previous case. Loan officers in a field experiment and Major League Baseball umpires calling pitches showed the same negative pattern. The authors estimated that it reversed about 2% of asylum decisions and up to 9% of the loan decisions.
The mechanism matters for investors. These were experienced professionals who expected their own sequence of decisions to "look random", and bent single decisions to make it so.
The hot hand, and the correction
In 1985 Thomas Gilovich, Robert Vallone and Amos Tversky asked whether basketball players really get hot. Of the fans they surveyed, 91% believed a player has a better chance of making a shot after making his last two or three than after missing them. The shooting records of the Philadelphia 76ers, free throws by the Boston Celtics and a controlled shooting test with Cornell players showed no such thing. For three decades the paper was the textbook example of people seeing patterns in noise.
In 2018 Joshua Miller and Adam Sanjurjo showed that the test itself was biased. If you take a finite sequence and look only at the shots that follow a hit, you are more likely to find a miss than the true rate suggests, even when shots are independent.
Corrected for this bias, the Cornell data from the 1985 paper show players hitting about 13 percentage points more often after a run of hits than after a run of misses, and 19 of the 25 shooters were better after a hit streak. The fans were not entirely wrong: there is a hot hand in controlled shooting, though it is smaller than fans believe. The belief still counts as a bias when people act on a streak far more than its real size justifies.
The episode is worth remembering for a second reason. A result that was taught as settled for 33 years had a statistical flaw that anyone could have checked with a coin. Findings about streaks in markets deserve the same caution.
What it looks like in an account
Consider an illustrative investor, Sam, with $40,000 in an S&P 500 index fund and a habit of reading streaks.
- In month five of a rally the fund is up 14%. Sam sells half, $22,800, because "it can't go up six months in a row", and plans to buy back after the pullback.
- The index rises another 3% over the next two months. Sam's cash earns 0.6%. Being half out has cost about $550 against holding, and now buying back means paying more than the sale price, which feels like admitting a mistake.
- Meanwhile Sam moves $10,000 into a fund that topped its category three years running, because the manager "is on a roll".
The first move is the gambler's fallacy applied to the market. The third is the hot hand applied to a manager. Neither used any information except the streak itself.
Among funds, the evidence on streaks is well documented. Mark Carhart's 1997 study of US stock funds found that last year's winners did tend to do somewhat better the next year, but that almost all of it was explained by the stocks they happened to hold (the momentum effect) and by costs. The persistence that could not be explained was in the worst funds, which kept underperforming, largely because of high expenses. Why most active funds underperform covers how rarely top results repeat.
A real test: SPY after a streak
The figure counts every month from March 1993 to October 2026 for the SPDR S&P 500 ETF, with dividends included. Each bar answers one question: given the streak that had just happened, how often was the next month up?
SPY rose in 64.6% of the 404 months. After a streak the figure hardly moved:
| What had just happened | Cases | Next month up |
|---|---|---|
| Any month | 404 | 64.6% |
| Exactly 3 up months | 36 | 75.0% |
| 4 or more up months | 80 | 66.3% |
| Exactly 2 down months | 32 | 62.5% |
| 3 or more down months | 20 | 60.0% |
Two readings of this table are wrong. The first is "three up months is a buy signal, 75%". There are 36 cases; if four of them had gone the other way the figure would be 64%, the same as any month. The second is "after three down months the market is due". It rose in 60% of those 20 cases, slightly less often than usual. The longest run of up months in the sample was 15 and the longest run of down months was 5, and nothing about the month after a streak stood out.
The daily record teaches the same thing by accident. In two years of daily closes (500 days to November 17, 2026) SPY rose on 54% of days. After four or more down days in a row it rose 83% of the time, which sounds like a strategy until you see that it is 10 of 12 cases. After five or more up days it rose 43% of the time, 9 of 21. Believing either number is the law of small numbers at work.
This does not mean prices have no memory at all. Over horizons of 3 to 12 months, stocks that have outperformed tend to keep doing so on average (see 52-week-high momentum), and over several years expensive markets tend to return less. Those are weak statistical tendencies measured over thousands of observations, and they are a different claim from "this streak must end" or "this streak will go on".
What helps
No study shows that learning about these fallacies makes an investor immune. What the evidence supports is narrower.
- Ask what would have to be true. For a reversal to be "due", the market would need a memory of its own streak. For a manager to be "hot", skill would have to show up in so few observations. Naming the mechanism often dissolves the feeling.
- Count the cases. Any claim of the form "after X, the market does Y" comes with a number of cases. Under about a hundred, treat it as an anecdote. The Finabase backtest page shows the number of trades beside every win rate.
- Decide by rule, not by streak. A rebalancing rule that trims stocks when they exceed a set share of the portfolio sells after rises for a reason that has nothing to do with streaks. See rebalancing methods.
- Judge managers on long records and costs. Three years is a short sample. Costs are known in advance and persist. Process over outcome shows how many years it takes to tell skill from luck.
How Finabase helps
The SPY price history tab lists each day's close and percentage change for up to two years, so you can count streaks yourself rather than trust a remembered one. The backtest page runs indicator and candlestick signals over real prices and reports the number of trades, wins, losses and the win rate, which is the honest way to check an "after X comes Y" belief. The performers page lists the day's biggest winners and losers. It is a good place to notice the pull of both fallacies at once: the urge to buy what is running and to call a top in it.
Frequently asked questions
What is the gambler's fallacy in simple terms?
It is the belief that after a run of one outcome, the opposite outcome becomes more likely, even though each event is independent. A roulette wheel that has landed on red five times is no more likely to land on black next time than it ever was.
Is the hot hand real?
In basketball, partly. The 1985 study that found no hot hand used a method biased against finding one. Corrected analyses of controlled shooting show a real but modest effect. Fans and investors still overestimate how much a streak predicts.
Is the stock market due for a correction after a long rally?
Not because of the rally's length. In SPY's monthly record since 1993, the month after four or more straight up months was up 66% of the time, about the same as any month. Declines happen, but a streak does not schedule them.
How is the gambler's fallacy different from mean reversion?
The gambler's fallacy expects a random process to balance out in the short run. Mean reversion is a measured tendency of some quantities, such as valuation ratios, to drift back toward an average over long periods. One is a mistake about independence; the other is an empirical claim that has to be tested on data. See mean reversion.
Summary
The gambler's fallacy expects streaks to reverse and the hot-hand fallacy expects them to continue. Both come from the law of small numbers, the expectation that a short run should look like the long-run average. Casino bets, asylum rulings and umpire calls all show people leaning against their own recent streaks, while the hot hand turned out to be a real but small effect that a flawed test had hidden. In 33 years of SPY months, what happened after a streak was nearly indistinguishable from what happened in any month. The practical defense is to count cases, name the mechanism and let rules, not runs, trigger decisions.
Check your understanding
5 questions on this lesson. Each answer comes with the reason, and nothing is scored.
A stock index has risen six months in a row. What does the streak itself tell you about next month?
- A fall is more likely than usual, because the rise must be paid back
- A rise is more likely than usual, because the market is hot
- Very little: in SPY’s record since 1993 the month after a long run was up about as often as any other month
Show the answer
C. Very little: in SPY’s record since 1993 the month after a long run was up about as often as any other month Expecting a reversal is the gambler’s fallacy and expecting continuation is the hot hand. After four or more up months SPY rose 66% of the time, against 65% for all months.
An investor says a roulette wheel is “due” for black but that a fund manager with three good years is “on a roll”. What explains holding both beliefs?
- People expect reversal from processes they see as random and continuation from ones they see as skilled
- Roulette and fund returns follow different laws of probability
- Three years is a long enough sample to prove skill
Show the answer
A. People expect reversal from processes they see as random and continuation from ones they see as skilled Both beliefs come from expecting a short run to look like the long-run average. A streak in a “random” process must be balanced out; a streak that is too long to be random is read as skill.
In two years of daily data, SPY rose after 10 of the 12 runs of four or more down days. What is the right conclusion?
- Buying after four down days works 83% of the time
- Twelve cases are too few to conclude anything; two different outcomes would change the figure by 17 points
- The market always rebounds after four down days
Show the answer
B. Twelve cases are too few to conclude anything; two different outcomes would change the figure by 17 points A percentage from a dozen cases is an anecdote. Treating it as a rule is the law of small numbers, the same error the lesson is about.
Why did the 1985 basketball study find no hot hand when later analysis of the same data did?
- The players shot worse in the later analysis
- Looking only at shots that follow a hit, in a finite sequence, understates the hit rate, so equal rates actually meant better shooting after hits
- The original data were lost and replaced
Show the answer
B. Looking only at shots that follow a hit, in a finite sequence, understates the hit rate, so equal rates actually meant better shooting after hits Miller and Sanjurjo showed the selection bias with coin flips: in three flips, the average share of heads among flips that follow a head is 5/12, not 1/2.
Which rule sells part of a holding after a long rise without relying on either fallacy?
- Selling because the price has risen five months in a row
- Selling when the holding exceeds a target share of the portfolio set in advance
- Selling because the chart looks tired
Show the answer
B. Selling when the holding exceeds a target share of the portfolio set in advance A rebalancing rule responds to the portfolio’s weights, not to the streak. It needs no belief about what the next month will do.
Exercise
One task to do in Finabase. Nothing is graded or saved.
Open SPY’s Price history tab and choose the 3M period. Read down the Change % column and mark every run of three or more closes in the same direction. For each run, note whether the next day continued it or reversed it. Before you count, write down your guess: how many of the runs were followed by a reversal?
Open SPY price historyWhat you should find
Three months is about 63 trading days, so you will find only a handful of runs, often fewer than ten. Whatever share reversed, it rests on too few cases to mean anything, and it will be different next quarter. If your result felt like a pattern, that feeling is the law of small numbers.
Sources and further reading
- Tversky and Kahneman — Belief in the law of small numbers, Psychological Bulletin 76(2), 1971
- Gilovich, Vallone and Tversky — The hot hand in basketball: on the misperception of random sequences, Cognitive Psychology 17(3), 1985
- Miller and Sanjurjo — Surprised by the hot hand fallacy? A truth in the law of small numbers, Econometrica 86(6), 2018
- Croson and Sundali — The gambler’s fallacy and the hot hand: empirical data from casinos, Journal of Risk and Uncertainty 30(3), 2005
- Chen, Moskowitz and Shue — Decision making under the gambler’s fallacy: evidence from asylum judges, loan officers, and baseball umpires, Quarterly Journal of Economics 131(3), 2016
- Rabin and Vayanos — The gambler’s and hot-hand fallacies: theory and applications, Review of Economic Studies 77(2), 2010
- Carhart — On persistence in mutual fund performance, Journal of Finance 52(1), 1997
- Finabase chart data — SPDR S&P 500 ETF (SPY) monthly closes and dividends, February 1993 to October 2026, and daily closes, November 2024 to November 2026
