We improve thepolicy iteration-based algorithm for value set analysis by giving a new heuristic for policy selection based on a local static analysis. In particular, we detect loops in the program and perform an analysis to discover therelative changesof variables in the loop, that is, whether a variable is constant or whether its value rises, falls or both. We use these relative changes to improve the old heuristic, achieving better (that is, smaller) fixed points than the original approach.