ACF and PACF autocorrelation calculator
Measure how a series relates to earlier observations across successive lags.
ACF and PACF by lag
The dashed ±1.96/√n lines are approximate white-noise reference bounds rather than a complete model test.
Coefficients by lag
| Lag | ACF | PACF | Lower bound | Upper bound |
|---|---|---|---|---|
| Fill in the fields above — the result is computed on this page. | ||||
How do I use this page?
- Paste an evenly ordered time series.
- Set a maximum lag below the number of observations.
- Compare ACF and PACF peaks with the reference bounds.
What this page does not do: The result describes the supplied series. Trend, seasonality, non-stationarity and a small sample can change the interpretation of ACF and PACF.
Further reading
Theory — what are ACF and PACF autocorrelation?
Autocorrelation (ACF) measures how similar a time series is to its shifted copy at successive lags. Partial autocorrelation (PACF) shows the dependence at a given lag after removing the influence of shorter lags, which the calculator computes using the Durbin–Levinson method. A slowly decaying ACF can signal a trend or non-stationarity in the series. PACF at lag k removes the linear influence of shorter lags, while ACF also contains their indirect contribution.
Frequently asked questions
How does PACF differ from ACF?
Does crossing a bound prove seasonality?
See also
A/B test calculator
Compare two proportions in an A/B test. Calculate conversion rates, difference, uplift, z score, p value and c
A/B test planning — sample size, MDE and duration
Plan a two-proportion experiment, estimate detectable effect and traffic duration.
Adjusted R-squared calculator
Convert R² to adjusted R² using the number of observations and predictors in a regression model.