Course contentsShow
← Machine Learning and Deep Learning
Lesson 2387 of 3,53852. Time Series and ForecastingFree lesson

Autocorrelation and Partial Autocorrelation

Measuring correlation between a series and its lagged values; interpreting ACF and PACF plots for pattern identification.

Autocorrelation and Partial Autocorrelation

You'll learn how to measure and interpret the relationship between a time series and its own past values to identify patterns.

What Is Autocorrelation?

Autocorrelation (ACF) measures how correlated a time series is with itself at different time lags. Think of it like asking: "If I know the value today, how well can I predict the value 1 day ago, 2 days ago, or k days ago?" It's correlation, but with your own past.

For a time series at time t, autocorrelation at lag k measures the correlation between values at time t and values at time t-k. A high positive ACF at lag 7 in daily data suggests a strong weekly pattern.

What Is Partial Autocorrelation?

Partial autocorrelation (PACF) goes deeper—it measures the direct relationship between observations at lag k, removing the influence of all shorter lags. Imagine you're measuring the correlation between today and 3 days ago, but you "control for" the effects of 1 and 2 days ago. This isolates the pure connection.

Why Both Matter

ACF plots show you overall patterns: gradual decay suggests trend or non-stationarity, sharp cutoffs suggest moving average processes, and periodic spikes reveal seasonality.

PACF plots help identify autoregressive order: if PACF cuts off after lag p while ACF decays gradually, you likely have an AR(p) process—meaning the series depends directly on its past p values.

Together, these plots are your diagnostic tools for understanding temporal dependencies before building forecasting models.

Key Takeaway: ACF reveals total correlation with lagged values (including indirect effects), while PACF isolates direct relationships—both are essential for identifying time series patterns and choosing appropriate model structures.