In this paper, we propose a data-driven Portmanteu test for conditional goodness-of-fit in dynamic models. Our method uses the well-known fact that under the correct specification of the conditional distribution the generalized "errors" obtained after the conditional probability integral transformation are iid U[0,1]. The proposed test is a modified Box-Pierce statistic applied to the generalized errors, with a data-driven choice for the number of autocorrelations used. The test explicitly takes into account of the parameter estimation effect, and as a result it has a convenient standard chi-squared limit distribution. Hence, the main distinctive feature of our approach is its simplicity. The basic methodology is extended to conditional models for the tail, conditional hazard models and diffusion models. It is shown that, unlike existing approaches, our approach is applicable to a wide class of models, including ARMA-GARCH models with time varying higher order moments, such as Hansen's (1994) skewed t model. A simulation study shows that our test has a satisfactory size and power performance. Finally, an empirical application to the Nikkei Index data highlights the merits of the proposed test over competing alternatives.