This letter proposes a new model for preferential attachment in dynamic directed networks. This model consists of a linear time-invariant system that uses past observations to predict future attachment rates, and an innovation noise process that induces growth on vertices that previously had no attachments. Analyzing a large citation network in this context, we show that the proposed model fits the data better than existing preferential attachment models. An analysis of the noise in the dataset reveals power-law degree distributions often seen in large networks, and polynomial decay with respect to age in the probability of citing yet-uncited documents.