On Dynamic Programming Decompositions of Static Risk Measures in Markov Decision Processes