Law of Small Numbers

The law of small numbers is the cognitive error of assuming that small samples are representative of their underlying population. An investor observing a stock’s 12% return over three months might conclude it is a “growth stock” and allocate heavily based on that fragment of data. Statistically, three months of returns contain almost no signal about long-term prospects; the law of small numbers causes investors to treat noise as a meaningful pattern.

Why the three-month sample deceives

Imagine a new ETF launches and returns 15% in its first quarter. The law of small numbers causes many investors to assume this reflects the fund manager’s skill. Statistically, a single quarter tells almost nothing. Even a mediocre strategy or pure luck can deliver 15% in a good quarter. To distinguish skill from random walk, an investor would need to observe 5–10 years of historical-volatility data and risk-adjusted returns. A single quarter is barely more informative than a coin toss.

Yet small samples feel real in a way that large theoretical samples do not. If I show you a stock that rose 30% in two months, that tangible performance feels more convincing than a regression showing that the stock’s long-term alpha is indistinguishable from zero. The human brain readily builds a narrative from a small dataset: “This manager found a mispricing,” “This sector is turning,” “This company nailed its pivot.” The narrative feels true, so the investor acts on it—buying the fund or the stock in hopes of riding the trend further.

The law of small numbers is Tversky and Kahneman’s term for this systematic underestimation of how much data you need before drawing confident conclusions. A small sample can be wildly unrepresentative by pure chance. Yet investors habitually treat small samples as if they were large, generating overconfident forecasts and portfolio rotations that lag diversified alternatives.

The mechanics of spurious trends

Here’s how the bias operates in practice. An investor notices that growth-fund outperformed value-fund in the last 18 months. The recent evidence feels compelling. Based on this small sample, the investor sells value positions and buys growth, expecting the trend to continue.

But 18 months is a blink in market history. Over 50 years, growth and value alternate dominance in multiyear cycles. A manager who sees 18 months of outperformance and assumes it reflects a regime shift is committing the law of small numbers. The true state—whether growth or value genuinely has an edge—requires decades of data to discern. An investor acting on 18 months has likely bought the recent winner at an inflated valuation, just as the cycle turns.

The same applies to sector-rotation. Energy stocks rise 40% in one year on oil prices; an investor concludes “energy is the place to be” and overweights it. But one year of price movement reflects cyclical factors, geopolitical shocks, and leverage dynamics that may reverse in the next year. An allocation based on a single year’s performance is premature. Ten years of data would show that energy performs cyclically, with no persistent outperformance edge—just correlation patterns that are unpredictable year to year.

Small-sample bias in manager selection

Institutional investors and retail alike use small-sample logic when evaluating fund managers. A manager with three years of 20% annual returns seems exceptional. Investors flock to the fund, pushing assets to billions. Then performance normalizes or deteriorates, and the fund falls out of favour. The initial enthusiasm was the law of small numbers at work.

Academic studies show that three years of mutual fund performance-fee data is statistically useless for predicting future returns. The top-performing managers of one three-year period are only marginally more likely to outperform in the next period than the bottom quartile. Yet fund flows chase recent winners obsessively, driven by investor belief that the small sample was informative.

The bias is particularly damaging in hedge-fund investing. A private equity strategy shows 15% returns over five years; an institutional investor concludes the manager is skilled and commits capital. But five years with eight or ten data points (annual or quarterly returns) is still a small sample. The manager might have had one extraordinary exit (winning a court settlement, selling into a hot market) that drove the alpha. Strip that out, and the manager’s baseline return-on-equity is unremarkable. The law of small numbers led the investor to extrapolate from an atypical five-year window.

Downward extrapolation: the reverse mistake

The law of small numbers also works in reverse. A manager underperforms for two years and investors yank their capital, assuming structural underperformance. But two years can contain many random headwinds—sector rotations, macro shocks, temporary liquidity issues. A truly mediocre manager might outperform for two years by coincidence; a skilled manager might underperform. Neither small sample is conclusive.

Value investors fell prey to this in the 2010s, when value underperformed growth for nearly a decade. Many exited value strategies, convinced the approach was “broken.” They conflated a single lengthy period of relative underperformance (a still-small sample in market-history terms) with evidence that the strategy no longer worked. In reality, a decade is a single market regime, not enough to overturn a century of evidence showing that cheap stocks eventually outperform. Investors who abandoned value at the trough in 2020 missed the strong rebound that followed.

The base-rate blindness

Part of the problem is that investors often forget the base rate when interpreting a small sample. There are thousands of stocks. By pure chance, some will have great runs. An investor sees a stock that rose 50% in six months and buys it, forgetting that in a universe of 5,000 equities, a handful will hit 50% gains in any given six-month period, regardless of quality. The base rate of random 50% gains is actually fairly high; the base rate of these gains persisting is close to zero.

Professional investors have better discipline. They track the base rate: “In our universe, X% of managers outperform in a given year by pure luck.” They then ask whether an observed outperformance margin exceeds what luck alone would predict. This requires large samples and careful statistical testing. Most fund marketing skips this rigor and instead highlights the small sample that looks good.

The cost in turnover and taxation

The law of small numbers drives excess turnover. An investor chasing recent winners rotates her portfolio quarterly or semiannually based on recent momentum. Each rotation incurs transaction costs, market-impact costs, and—in taxable accounts—capital-gains-tax-investor liabilities. Over time, this activity costs 1–3% of annual returns. A disciplined investor using longer lookback windows (5+ years) and rebalancing only annually avoids the churn.

Institutional investors waste billions on this bias. Active managers launch new strategies based on recent outperformance, collect a year or two of impressive returns, then face normal mean reversion. Investors who jumped in on the small-sample evidence get poor risk-adjusted returns after fees. A passive index-fund or ETF that ignores the short-term sample beats most active rotating strategies over time.

Overcoming the bias

The first antidote is statistical literacy. Understand that one year of data is basically noise. Two years is weak evidence. Five years is respectable but still admits large sampling error. Only at ten-plus years does a sample become truly informative. A casual investor who commits to reviewing portfolio decisions only on a five-year or longer cycle automatically eliminates much of the small-sample bias.

Second, use rules-based strategies that ignore short-term performance. A mechanical rebalancing rule—“Sell the asset class that has risen fastest and buy the one that has fallen most, rebalancing annually”—removes the temptation to extrapolate from recent months. Ironically, this contrarian discipline—buying recent losers and selling recent winners—often outperforms momentum strategies that chase the small-sample winners.

Third, lean on base rates. Before buying a stock because it rose 30% this year, check: How often do individual stocks rise 30% in a year? (Answer: fairly often, in a universe of thousands.) How often do those stocks outperform in the following year? (Answer: barely better than chance.) Base-rate thinking is humbling and antidote to the small-sample fallacy.

See also