We study implicit systems of linear time-varying (LTV) difference equations with rational coefficients of arbitrary order and their solution spaces, called discrete LTV-behaviors. The signals are sequences, i.e., functions from the discrete time set of natural numbers into the complex numbers. The difference field of rational functions with complex coefficients gives rise to a noncommutative skew-polynomial algebra of difference operators that act on sequences via left shift. For this paper it is decisive that the ring of operators is a principal ideal domain and that nonzero rational functions have only finitely many poles and zeros and grow at most polynomially. Due to the poles a new definition of behaviors is required. For the latter we derive the important categorical duality between finitely generated left modules over the ring of operators and behaviors. The duality theorem implies the usual consequences for Willems' elimination, the fundamental principle, input/output decompositions, and controllability. The generalization to autonomous discrete LTV-behaviors of the standard definition of uniformly exponentially stable (u.e.s.) state space systems is unsuitable since u.e.s. is not preserved by behavior isomorphisms. We define exponentially stable (e.s.) discrete LTV-behaviors by a new analytic condition on its trajectories. These e.s. behaviors are autonomous and asymptotically stable. Our principal result states that e.s. behaviors form a Serre category, i.e., are closed under isomorphisms, subbehaviors, factor behaviors, and extensions, or, equivalently, that the series connection of two e.s. input/output behaviors is e.s. if and only the two blocks are. As corollaries we conclude various stability and instability results for autonomous behaviors. There is presently no algebraic characterization and test for e.s. of behaviors, but otherwise the results are constructive.