A marginal beta regression model with autoregressive and moving average errors is developed for the analysis of time series of values in the standard unit interval (0,1), such as proportions and rates. The dependence structure is conveniently related to the marginal model through a Gaussian copula specification. Likelihood inference, model validation via residual analysis, and prediction are briefly discussed. The methodology is applied to the time series of the rate of hidden unemployment in S ̃ao Paulo, Brazil.