Supermodularity And Complementarity PDF Download

During the past two decades firms have adopted many types of functional improvement programs. A goal of this paper is to study interactions of operations programs with improvement programs in other functions. An important issue is whether these programs are complements (or substitutes): that is, whether implementation of a pair of programs adds more value (or loses value) than the sum of the individual programs. Recent advances based upon Topkis's (1998) theory of supermodularity have created a framework to establish sufficient conditions to guarantee that two interacting activities or programs are complementary. But not all programs can be characterized by supermodularity. In particular, I show that programs that involve uncertainty reduction do not fit the supermodularity framework with respect to other programs. In the absence of supermodularity, complementarity properties can sometimes be established, but some assumption about the production function must be assumed. Assuming that production is based upon queuing technology, three programs are studied and I show that complement/substitute properties can be established. The three programs are uncertainty reduction about a key demand parameter, elimination of biased estimates of cost and direct cost savings. I show that uncertainty reduction and bias removal can be complements or substitutes; but, uncertainty reduction and cost estimation bias elimination are both complementary to direct cost savings. Also, several operations improvement programs are shown to be complementary to each other. The fact that the operations improvement programs are complementary to all other programs is significant for the organizational design of capital budgeting, as this assures that the true project value is higher than its standalone net present value. Results are generalized by presenting conditions that assure the complementarity or substitutability of these three types of programs. Because batching/lot sizing and any fixed charge problem (such as facility location) have cost structures similar to queuing, the results are shown to apply to a broad range of technologies.