Policies for lot sizing with setup learning

We address the issue of determining lot sizes in finite horizon environments when learning effects and/or emphasis on continuous improvement result in decreasing setup costs. Estimation of optimal lot sizes under these circumstances requires exact cost information on all future setup costs. In practice, this information is difficult to forecast. We analyze and evaluate a myopic policy which is intuitively appealing, easy to implement, and requires no information on future setup costs. We derive analytical measures for its effectiveness. Our analytical results, and computational experiments show that the policy is a good choice for machine intensive environments which are characterized by high learning rates. Using an optimal lot sizing method developed by us, we further show that the myopic policy yields good results even when setup cost changes cannot be completely modeled by stylized learning curves used in earlier research studies. Managerial implications are discussed, and future research directions are indicated.