Autocorrelation

Also known as: Serial correlation, ACF dependence

Autocorrelation (serial correlation) is correlation of a series with its own lags: $\mathrm{Corr}(\varepsilon_t,\varepsilon_{t-k})\neq 0$. In regression errors it violates the uncorrelated-errors part of Gauss–Markov.

Detect with residual plots, ACF/PACF, or Durbin–Watson. Consequences: inefficient OLS and invalid classical inference. Remedies include lagged models, ARIMA errors, and HAC standard errors.

← Back to CS Glossary