We team up the modeling flexibility of simulation with
sample path optimization to address the multi-location transshipment
problem. Under a modified base stock policy, we
determine a myopically optimal transshipment policy using an
LP/Network flow framework. Then with SAA, we determine the
optimal values of base stock levels, using random search and
gradient search. We analyze the impact of finite supplier capacity
on system behavior and performance measures of the stocking
locations. With SAA, we also investigate scenarios with nonidentical
costs. Albeit our gradient search technique which
requires us to propagate the gradient information throughout a
regenerative cyclis is complex, the quality of solutions attained
by gradient search was significantly better than those obtained
through random search.