Definition
Autocorrelation, also known as serial correlation, is a statistical measure of the similarity between a given time series and a lagged version of itself over successive time intervals. It quantifies the degree to which current values of the series are related to past values at specified lag lengths.
Overview
In time‑series analysis, autocorrelation functions (ACF) are employed to detect patterns such as trends, cycles, and seasonality. Positive autocorrelation indicates that high (or low) values tend to be followed by similar values, while negative autocorrelation suggests that high values are likely to be followed by low values. Autocorrelation is fundamental in model identification for autoregressive (AR), moving‑average (MA), and ARIMA models, and it also serves as a diagnostic tool for assessing the adequacy of fitted models. In signal processing, the concept extends to measuring similarity between a signal and a delayed copy of itself, aiding in tasks such as spectral analysis and pattern detection.
Etymology/Origin
The term combines the Greek prefix “auto‑” meaning “self” and “correlation,” derived from the Latin “correlatio,” meaning “a relationship.” The concept emerged in the early 20th century within the fields of statistics and stochastic processes, with early formalizations appearing in works on random functions and time‑series by scholars such as G. U. Yule (1927) and R. H. Whittle.
Characteristics
| Characteristic | Description |
|---|---|
| Lag | The displacement (usually measured in time steps) between the original series and its shifted copy. Autocorrelation is computed for multiple lags to produce the autocorrelation function. |
| Range | Values lie between –1 and +1. A value of +1 denotes perfect positive correlation, –1 denotes perfect negative correlation, and 0 indicates no linear relationship at the given lag. |
| Stationarity Requirement | Interpretation of autocorrelation assumes weak (second‑order) stationarity: constant mean, variance, and autocovariance that depends only on lag, not on absolute time. |
| Statistical Significance | Confidence intervals (often ±1.96/√N for large N) are used to judge whether observed autocorrelations differ significantly from zero. |
| Partial Autocorrelation | The partial autocorrelation function (PACF) isolates the correlation at a specific lag after removing the influence of intervening lags, useful for identifying AR order. |
| Applications | Forecasting (economics, finance, climatology), signal detection, quality control, neuroscience (e.g., spike train analysis), and image processing (texture analysis). |
Related Topics
- Correlation – General measure of linear relationship between two distinct variables.
- Cross‑Correlation – Correlation between two different time series as a function of lag.
- Stationarity – Property of a stochastic process where statistical moments are invariant over time.
- Autoregressive (AR) Models – Time‑series models that express current values as linear combinations of past values, characterized by the autocorrelation structure.
- Moving‑Average (MA) Models – Models that express current values as linear combinations of past error terms; autocorrelation patterns help to identify MA order.
- ARIMA (AutoRegressive Integrated Moving Average) – Integrated framework combining AR and MA components with differencing to handle non‑stationarity.
- Spectral Density – Fourier transform of the autocorrelation function, describing the distribution of variance across frequencies.
Autocorrelation remains a central tool for analyzing temporal dependence across a broad spectrum of scientific and engineering disciplines.