Skip to main content

Autocorrelation

Autocorrelation is the correlation of a variable with its own past values, or of regression residuals across time. In regression it violates the independence assumption, making standard errors unreliable though coefficients remain unbiased.

ByHoang TruongUpdated

FrameworkClassical linear regression model

See it move

Loading infographic...

Autocorrelation occurs when a regression's residuals are correlated with their own past values, most often in time-series data. Positive first-order autocorrelation (ρ₁ > 0) means consecutive errors cluster in the same direction, which understates OLS standard errors and inflates t-statistics. Coefficients remain unbiased, but hypothesis tests built on them become unreliable.

Where it fits
SubjectData Analysis & StatisticsAdvancedTopicRegression Diagnostics & ProblemsAdvanced

The formula

LaTeX
ρ1=Cov(εt,εt1)Var(ε)\rho_1 = \frac{\text{Cov}(\varepsilon_t,\, \varepsilon_{t-1})}{\text{Var}(\varepsilon)}

Variables

First-order autocorrelation coefficient (ranges from −1 to +1)
Regression error term at time t
Regression error term one period earlier

ρ₁ = 0 indicates no first-order autocorrelation. Positive ρ₁ means consecutive errors tend to share the same sign; negative ρ₁ means they alternate.