Özet:
For a number of commodity markets, including many agricultural and energy products, a typical problem faced by decision makers is to determine the optimal spot and forward contracting quantities when both the product demand and the spot price are random. While satisfying uncertain demand, the lack of storability possibilities for certain commodities {electricity for example{ adds another layer of complexity to the procurement policy optimization as the non-storable feature is ruling out the classical buy-and-hold argument for them [1]. Under the circumstances, rms increasingly engage in forward contracts for future deliveries, as well as, to avoid price uctuations and bene t from lower transaction costs. However, along with the increasingly volatile business environments, there is the risk of suppliers not delivering the agreed quantities. Delivery risks particularly hold a great importance in the pro tability of rms as they have no perfect guarantee that the quality or the quantity of the delivered goods absolutely meets their expectations. Still, `uncertain deliveries' is not discussed much in the literature and most of the time, other drivers mentioned above is held individually rather than a multivariate setting. Recognizing the imperfect delivery aspect of forward procurement, integrated with stochastic price and demand environment, this thesis presents a new trivariate approach for examining policy-optimization models regarding non-storable commodities. For each of the variables {price, demand and delivery of supplies{ evolving in a geometric Brownian motion, an optimal forward contracting policy is proposed and algorithmically characterized by means of stochastic optimization. The most signi cant result is that pro t improvements can be achieved by introducing uncertain delivery notion into procurement decision making. We also demonstrate that it can be valuable to model delivery interactions with price and demand into the procurement models, and we present several possible ways to model these interactions. We present structure, algorithms, and heuristics for our models.
