Using Stationarity and Correlograms to Interpret Serial Correlation
Summary
The article introduces serial correlation, also called autocorrelation, as dependence between observations at different times. It reviews expectation, variance, covariance, and correlation, then explains why correlation is a normalized measure of linear association. For time series, it discusses stationarity in the mean and variance as assumptions that help estimate population properties from a single observed history.
It describes autocorrelation at different lags and uses correlograms to inspect the resulting pattern. Examples show that a steadily declining autocorrelation pattern can arise from a linear trend, while recurring peaks can reveal periodic structure. Confidence bands provide evidence against zero autocorrelation at individual lags, but some exceedances are expected by chance and adjacent lags are dependent. These plots are diagnostic rather than proof of a predictive strategy; trends, seasonality, and stationarity assumptions need attention before using the patterns for forecasting or simulation.
Key ideas
- Serial correlation measures dependence between observations separated by time lags.
- Correlation normalizes covariance by the variables' spreads and ranges from negative to positive linear association.
- Stationarity in the mean and variance supports estimating time series properties from a single history.
- A correlogram displays sample autocorrelation across lags and can help reveal trends or periodic effects.
- Confidence bands require cautious interpretation because chance exceedances and dependence across nearby lags are possible.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.