# Effective Sample Size > The number of independent observations a dependent sample is actually worth — the count that governs a statistic's standard error once autocorrelation and overlapping windows are accounted for. - Canonical: https://hadalinstruments.com/glossary/effective-sample-size/ - Term set: https://hadalinstruments.com/glossary/ --- A sample of ten thousand observations sounds decisive. Whether it is depends entirely on how many independent pieces of information those ten thousand rows contain, and in market data the answer is almost always far fewer. Two mechanisms do most of the damage. The first is autocorrelation: consecutive returns, volatility estimates and feature values are related to their neighbours, so each new row adds less than one row's worth of evidence. For a stationary series the effective count is approximately the nominal count divided by a factor built from the autocorrelation function — one plus twice the sum of the autocorrelations — and for a persistently correlated series that divisor is large. The second is overlap. Any label defined over a forward horizon — the return over the next twenty bars, a barrier outcome, a rolling z-score — shares most of its underlying data with its neighbours. A million overlapping twenty-bar labels carry closer to fifty thousand non-overlapping observations, and treating them as a million is not a rounding error but an order-of-magnitude misstatement of the evidence. Clustering does the same at portfolio level. Positions taken simultaneously across correlated instruments are one bet wearing several tickets; breadth, in the sense that matters for a [Sharpe ratio](/glossary/sharpe-ratio/) or an [information coefficient](/glossary/information-coefficient/), counts independent bets rather than fills. Standard errors scale with the square root of the effective count, so every t-statistic, confidence interval and significance claim computed on the nominal count is too narrow — sometimes by a large factor. It is the same dependence that forces financial cross-validation to [purge](/glossary/purged-cross-validation/) and [embargo](/glossary/embargo-period/) rather than shuffle. ## Why it matters Sample size is the quantity most often reported and least often earned. A finding backed by a hundred thousand rows and an effective count in the hundreds is not strong evidence weakly stated; it is weak evidence stated with unearned confidence. Publishing the effective count beside the nominal one costs nothing and changes how the result should be read. The same shortfall decides a stake. A win rate is a proportion measured on a finite record, so it supports an interval rather than a point, and the [Kelly criterion calculator](/tools/kelly-criterion-calculator/) carries that interval through to the sizing it implies — reporting the range beside the point figure, and saying so plainly where the record is too short to rule out an edge of zero. ## Commonly confused with Neighbouring concepts that get used interchangeably, and the distinction that actually separates them. - **Nominal sample size** — The row count is what gets reported; the effective count is what the statistics are entitled to. A finding backed by a hundred thousand rows and an effective count in the hundreds is not strong evidence weakly stated — it is weak evidence stated with unearned confidence. - **Statistical power** — Power is what you can detect; effective sample size is one of the inputs that decides it. Computing power from the nominal count is how a study convinces itself it can see something it cannot. - **Breadth** — Breadth in the sense that matters for a Sharpe ratio or an information coefficient counts independent bets, not fills. Positions taken simultaneously across correlated instruments are one bet wearing several tickets, however many order confirmations arrive. - **[Purging and embargoing](https://hadalinstruments.com/glossary/purged-cross-validation/)** — Purging and embargoing are the remedy applied to cross-validation folds; effective sample size is the measurement of the underlying problem. The same dependence that shrinks the effective count is what forces financial cross-validation to purge and embargo rather than shuffle. ## How to measure it in your own data A definition you cannot test is a definition you have to take on trust. This is the shortest honest route from the concept to a number you computed yourself. - **Records you need** — The series itself, its autocorrelation function, the forward horizon of any label defined over a window, and the correlation structure across simultaneously held positions. - **What you compute** — For a stationary series, approximately the nominal count divided by one plus twice the sum of the autocorrelations. For overlapping labels, divide by the overlap: a million overlapping twenty-bar labels carry closer to fifty thousand non-overlapping observations. At portfolio level, count independent bets rather than fills. - **What the answer tells you** — Standard errors scale with the square root of the effective count, so every t-statistic, confidence interval and significance claim computed on the nominal count is too narrow — sometimes by a large factor. Publishing the effective count beside the nominal one costs nothing and changes how the result should be read, which is why its absence is worth noticing in someone else's work. ## Questions and answers ### Why do overlapping labels inflate significance? Because neighbouring labels share most of their underlying data. A return measured over the next twenty bars overlaps almost entirely with the one starting a bar later, so consecutive observations are not independent evidence. Treating a million overlapping twenty-bar labels as a million observations is not a rounding error; it is an order-of-magnitude misstatement of how much you know. ### How do I compute the effective sample size? For a stationary series it is approximately the nominal count divided by a factor built from the autocorrelation function — one plus twice the sum of the autocorrelations — and for a persistently correlated series that divisor is large. For labels defined over a forward horizon, the overlap itself is the divisor. ### Does this apply to a portfolio as well as a series? Yes, through clustering. Positions opened simultaneously across correlated instruments behave as one bet however many tickets they generate, so the number of independent bets — not the number of trades — is what a Sharpe ratio or an information coefficient is entitled to treat as breadth. ### What should I report? Both counts, side by side. The nominal count describes the dataset and the effective count describes the evidence, and a reader cannot judge the second from the first. It costs nothing to publish and it changes how every significance claim in the work should be read. ## Work it out yourself Free calculators that take this concept as an input. Each shows its working, so the number it gives you can be checked rather than taken on trust. - [Sample-size check](https://hadalinstruments.com/tools/sample-size-check/) Luck horizon, in trades · Your record against it ## Related terms Derived from the links this entry makes and the entries that link back to it. - [Combinatorial Purged Cross-Validation (CPCV)](https://hadalinstruments.com/glossary/combinatorial-purged-cross-validation/) A backtest protocol that partitions a history into groups, holds out every combination of them in turn with purging and an embargo, and so produces many out-of-sample paths instead of a single one. - [Embargo Period](https://hadalinstruments.com/glossary/embargo-period/) A span of observations discarded immediately after a test window, so that serial correlation cannot carry information from the tested period into the data used to train. - [Information Coefficient (IC)](https://hadalinstruments.com/glossary/information-coefficient/) The correlation between a forecast and the outcome it predicted — the standard scalar summary of how accurate a signal was, measured across a cross-section or through time. - [Purged Cross-Validation](https://hadalinstruments.com/glossary/purged-cross-validation/) A cross-validation scheme for time-dependent data in which any training observation whose information horizon overlaps the test window is removed, so that no label can leak across the split. - [Sharpe Ratio](https://hadalinstruments.com/glossary/sharpe-ratio/) The ratio of a return series' mean excess return to its standard deviation — how much return was earned per unit of volatility borne, expressed as a single scalar. - [SMT divergence](https://hadalinstruments.com/glossary/smt-divergence/) A disagreement between correlated instruments at time-aligned swings: one takes out its prior high or low and the other does not, a refusal the ICT method reads as a sign that price is about to turn. ## In the research Effective Sample Size comes up in nine research notes on this site, and this entry lists three of them. - [Is my volatility regime just telling me the time?](https://hadalinstruments.com/research/is-my-volatility-regime-just-telling-me-the-time/) I removed the time-of-day pattern from one FX pair and a standard volatility classifier stopped finding compression at all. Six features, all six moved. - [The floor that never fired](https://hadalinstruments.com/research/the-floor-that-never-fired/) A guard set at 0.90 never fired in years of use. Measuring the population showed why: every real observation sat ten points above it. - [How many trades prove a trading edge?](https://hadalinstruments.com/research/how-many-trades-prove-a-trading-edge/) There is no universal number: the trades required scale with the square of your edge's dispersion-to-size ratio. How to compute your own, from your own log. ## Cite This Definition APA BibTeX HTML Hadal Instruments. (2026). Effective Sample Size. Hadal Glossary. https://hadalinstruments.com/glossary/effective-sample-size/ Version c050a1c, 2026-08-30. @misc{hadal_2026_effective-sample-size, author = {Hadal Instruments}, title = {Effective Sample Size}, year = {2026}, url = {https://hadalinstruments.com/glossary/effective-sample-size/}, howpublished = {Hadal Glossary}, version = {c050a1c}, note = {Pre-launch publication; version dated 2026-08-30} } Source: Hadal Instruments, Effective Sample Size. Original Research Copy Citation **Version c050a1c** identifies the commit that last changed this page in Hadal's content repository. That repository is not public, so the identifier does not resolve externally — it is published so a citation pins one specific state rather than a moving page. To obtain the exact version cited, use the [press and research route](https://hadalinstruments.com/press/).