Autocorrelation in the regression errors
Serial correlation in the errors: AR(1), why OLS stays unbiased but the standard errors break, the Durbin-Watson test, and the fixes.
A 3 minute 35 second animated lesson on serial correlation, also called autocorrelation, in regression errors. Built for ECON3006 Economic & Financial Modelling at Western Sydney University.
When the error term is correlated with its own past, captured by the AR(1) model where a shock lingers, the consequences are specific: under strict exogeneity the OLS coefficients stay unbiased and consistent, but the standard errors are biased, usually too small, so the t and F tests become unreliable. The video shows how to spot it in a residual plot, how the Durbin-Watson statistic maps onto the degree of autocorrelation, and how to respond with Newey-West robust standard errors, a richer dynamic model, or feasible GLS.
The key message is that autocorrelation breaks your inference, not your point estimates, so diagnose a missing lag before patching the standard errors. Pair it with the Atlas concept page for the formulas, a worked Stata example, and a quick quiz.