This paper discusses mixture periodic GARCH (M-PGARCH) models that constitute very flexible class of nonlinear time series models of the conditional variance. It turns out that they are more parsimonious comparatively to high-order MPARCH models. We first provide some probabilistic properties of this class of models. We thus propose an estimation method based on the Expectation-Maximization (EM) algorithm. Finally, we apply this methodology to model the spot rates of the Algerian dinar against euro and U.S. dollar. This empirical analysis shows that M-PGARCH models yield the best performance among the competing models.