<p>Portmanteau tests have drawn much interest in economics and finance because of their strong relationship to model specification. The majority of current testing, however, concentrates on stationary time series. This article proposes an empirical likelihood-based portmanteau test for the autoregressive model, no matter if it is stationary, nearly integrated, or unit root, and with or without an intercept. It turns out that the final statistic is always asymptotically chi-squared distributed. A simulation study confirms the good finite sample performance of the proposed test before illustrating its practical merit in analyzing real data sets.</p>