Direct Answer
Autocorrelation is a statistical & quantitative measure used to quantify statistical properties, relationships, normalization, or model residuals. The production page must define the exact calculation convention before interpreting higher or lower readings, because platform and provider implementations can differ.
Autocorrelation: Formula, Meaning, Signals, Examples and How to Use It
How Is Autocorrelation Calculated?
Autocorrelation is the correlation of a series with a lagged copy of itself. It shows whether past values help describe present values, such as whether up moves tend to follow up moves.
Formula: r_k = sum over t of (x_t - mean)*(x_(t+k) - mean) / sum over t of (x_t - mean)^2, for lag k
Inputs:
- Series values, often returns
- Lag k
- Window length
Worked example. For the five values 2, 4, 3, 5 and 6, the lag-1 autocorrelation is 0.100. The inputs are illustrative numbers, not market data.
Reading it. Positive values mean moves tend to persist, negative values mean they tend to reverse, and zero means no linear dependence at that lag. A rough significance band is plus or minus 1.96 divided by the square root of the sample size.
Limits. Estimates in short windows are noisy and biased toward zero. It captures only linear dependence and can be created by trends or volatility clustering.
Indicator overview
| Attribute | Value |
|---|---|
| Category | Statistical & Quantitative |
| Asset classes | Cross-asset, Portfolios, Pairs, Time series |
| Output type | Line/scalar |
| Parameters | Varies by implementation; document exact settings on use |
What is Autocorrelation?
Autocorrelation is a statistical & quantitative measure used to quantify statistical properties, relationships, normalization, or model residuals. The production page must define the exact calculation convention before interpreting higher or lower readings, because platform and provider implementations can differ.
For Swoopr, the canonical profile separates definition, calculation, interpretation and decision use. That keeps readers from collapsing a descriptive metric into a trading strategy. An indicator can describe trend, momentum, volatility, participation, positioning or risk without providing a complete entry, exit, sizing or portfolio decision.
What does it measure?
Autocorrelation measures specific market properties in the Statistical & Quantitative domain. The practical value is repeatability: the same inputs and formula can be applied across a defined dataset, allowing users to compare readings through time or across instruments when the calculation convention is held constant.
How to read Autocorrelation
Interpretation depends on the market property being measured. A descriptive reading is not automatically predictive. The same numerical state can lead to different outcomes in a strong trend, quiet range, event-driven gap or illiquid market. Always combine with context from price structure, volume, and broader market regime before drawing conclusions.
Strengths
- Computing the correlation at lag k shows directly whether values k periods apart tend to move in the same direction.
- Plotting values against lag gives a profile of the series' memory, from short-term persistence to periodic cycles.
- Reproducible when the formula and data convention are fixed.
Weaknesses and limitations
- A single lag is one slice of dependence, so a pattern that spans several lags can be missed when only lag 1 is examined.
- Bid-ask bounce in transaction prices can produce negative first-order autocorrelation unrelated to any broader pattern.
- Correlated indicators can create false confidence by repeating the same underlying information.
Editorial and risk note
This page is educational content, not individualized investment advice. It should not imply guaranteed prediction, accuracy or outperformance. Where multiple valid definitions exist, Swoopr states the alternatives and the reason for selecting its primary convention.
Frequently Asked Questions
Is Autocorrelation a buy or sell signal?
No. Positive values mean moves tend to persist, negative values mean they tend to reverse, and zero means no linear dependence at that lag. Estimates in short windows are noisy and biased toward zero. A trade built on it still needs a trigger, an invalidation rule, a position size and realistic execution assumptions.
What is the best setting for Autocorrelation?
Autocorrelation takes these inputs: series values, often returns; lag k; window length. There is no universal best setting for the tunable ones, so test them across instruments, regimes and out-of-sample periods and favor stable parameter regions over one historical winner.
Can Autocorrelation be used by itself?
Estimates in short windows are noisy and biased toward zero. It captures only linear dependence and can be created by trends or volatility clustering. For that reason it is read next to measures of direction, participation, valuation, liquidity and risk, not alone.
Does Autocorrelation work on every timeframe?
The calculation (r_k = sum over t of (x_t - mean)*(x_(t+k) - mean) / sum over t of (x_t - mean)^2, for lag k) can be run on any bar size, but behavior is not portable: a reading on a five-minute chart describes a very different market window from the same reading on a daily chart. Validate the exact timeframe before relying on it.
Why does Autocorrelation give false signals?
A single lag is one slice of dependence, so a pattern that spans several lags can be missed when only lag 1 is examined. Bid-ask bounce in transaction prices can produce negative first-order autocorrelation unrelated to any broader pattern. Lag, noise and regime changes add to this, because market participants react to new information after the reading is calculated.
How should Autocorrelation be backtested?
Rebuild Autocorrelation from series values, often returns using point-in-time data, write the rules before testing, include delisted securities where relevant, model realistic fills and costs, reserve a validation sample and compare with a simpler baseline.