Durbin–watson test
A test for first-order serial correlation in regression residuals; a statistic near 2 suggests none. It is invalid for autoregressive models, whose residual autocorrelations are tested instead.
Where you will meet it
Durbin–watson test appears in 3 learning outcomes on the CFA syllabus. A few of them:
- Say what serial correlation is and what it does to statistical inference
“The Durbin–Watson statistic tests first-order correlation, while the Breusch–Godfrey test regresses residuals on the regressors and p lagged residuals and handles several lags.”
- Explain what decides the choice between a linear and a log-linear trend for a given series, and assess where trend models fall short
“Both models share limits: they assume the past pattern continues indefinitely, and their residuals are often serially correlated, which a Durbin–Watson test detects.”
- Explain how residual autocorrelations let you check whether an autoregressive model suits a time series
“To check that an AR model is correctly specified, examine the autocorrelations of its residuals rather than a Durbin–Watson statistic, which is invalid when a lagged dependent variable is a regressor.”
See also
Autocorrelation · Autoregressive model · Correlation · Serial correlation
Practise it
Practise durbin–watson test until it is automatic: questions that explain every wrong option.
Take the free diagnostic →