A PDE approach to asymptotic estimates for optimal exit probabilities