Last few days I have been working on the TYM revision. All the ingredients are there. It’s just a matter of how to sell it.

The model has three unique features. First, customers differ in lead time requirement. This is common in practice, but few models take it into account. We find that lead time heterogeneity leads to fulfillment flexibility, which can be exploited to improve operational efficiency. Mathematically, heterogeneous lead times relax the constraints, thereby increasing optimal value.

Second, supply is random. Most dynamic allocation problems assume deterministic supply, and the control policy is driven by demand risk only. In contrast, we deal with both random supply and demand, and the control policy accounts for both risks. Depending on the nature of the supply risk, the refined threshold maybe higher or lower than the classical one.

Third, We deal with both intertemporal and interspacial interactions. Classic models treat only intertemporal interactions. The key trade-off is between using the unit now and saving it for the future; the policy boils down to comparing the revenue margin and option value of that unit.

When interspacial interactions also matter, we need to strengthen the classic model in two ways. (a)  The benefit of using the unit now includes not the revenue in the current location, but also the marginal value generated in the destination a few periods later. (b) Coordination mechanisms are needed to synchronize decisions across the network. Because of the complexity, the grand, centralized control mechanisms are impossible; coordination must be achieve in a decentralized fashion. Using the dual price, we develop a novel price coordination scheme. The price vector serves two purposes: (1) signal of location congestion, (2) incentive for voluntary repositioning.

If we cook well, it should sell.


[Napa, CA, 7/27/2009]