An Adjustment to the Norman-White Approach to Approximating Dynamic Programs

This note considers an adjustment to the Norman-White approach of approximating stochastic dynamic programs by using expectations: The expectations are used only in approximating the transition mechanism. The original expected returns are left intact. The adjustment also involves eliminating the policy improvement step proposed by Norman and White. The approach appears to work surprisingly well on the examples considered, It may turn out to be a competitive practical approach to approximating solutions to the (s,S) inventory model.