📈Autocorrelation Function

NEW

Also known as: ACF

A function describing correlation of a time series with lagged versions of itself across different time shifts.

DESCRIPTION

The autocorrelation function quantifies linear dependence between a time series and its lagged values. It is used to detect seasonality, choose AR terms, and assess residual independence. Correlograms visualize ACF values across lags with confidence bounds. Proper interpretation assumes stationarity or applies differencing and detrending first.

ALTERNATIVE DEFINITIONS

Wikipedia

Autocorrelation, also known as serial correlation, is the correlation of a signal with a delayed copy of itself as a function of delay.

General signal-processing definition emphasizing correlation with delayed copies.

Source

NIST e-Handbook of Statistical Methods

The autocorrelation function (ACF) is a plot of the autocorrelation coefficient r as a function of the lag k.

Focuses on the correlogram representation used in exploratory analysis.

Source

RELATED TERMS

Temporal Data Partial Autocorrelation Stationarity ARIMA Lag Cross-Correlation Seasonality White Noise Spectral Density Correlogram

COMPETING TERMS

Partial Autocorrelation Function

RELATIONS

• Broader: Autocorrelation
• Narrower: Partial Autocorrelation Function

KNOWLEDGE AREAS

Mathematics Data Science AI

TARGET AUDIENCE

Data Scientist Data Analyst MLOps Engineer AI Engineer

EXTERNAL SOURCES

• Wikipedia - Autocorrelation
• Forecasting: Principles and Practice - ACF explanation and examples
• statsmodels - API for computing the ACF

TOP QUESTIONS

How is the Autocorrelation Function used in model selection?

ACF helps identify AR terms and seasonal lags by inspecting significant spikes. Combine with PACF to choose appropriate ARIMA orders.

Category: How-To

What assumptions underlie the Autocorrelation Function?

Stationarity is typically assumed for meaningful interpretation. Differencing or detrending can help stabilize the mean.

Category: Assumptions

How do I interpret ACF confidence bands?

Spikes outside bands indicate significant autocorrelation at that lag. Bands are approximate and depend on sample size.

Category: Interpretation

What’s the difference between ACF and PACF?

ACF measures total correlation with lagged values, while PACF removes effects of intermediate lags to show direct relationships.

Category: Comparison

How does seasonality appear in the ACF?

Seasonality manifests as repeating spikes at multiples of the seasonal period, indicating cyclical structure.

Category: Concepts

Can the Autocorrelation Function detect white noise?

Yes. For white noise, ACF values are near zero for all non-zero lags within confidence bounds.

Category: Detection

How does sample size affect ACF estimation?

Smaller samples yield noisier estimates and wider bands. Larger samples improve stability and inference.

Category: Limitations

How is ACF used in residual diagnostics?

Residual ACF should show no significant autocorrelation. Remaining structure suggests model misspecification.

Category: Diagnostics

What are common mistakes using ACF?

Interpreting non-stationary series directly and overfitting based on random spikes are common pitfalls.

Category: Mistakes