| Summary: | As markets are increasingly volatile today, this thesis aims to study two strategies, process flexibility and real options, that can help hedge against demand uncertainty which has posed big challenges for many manufacturing systems. We consider the process flexibility design problem for balanced but non-identical systems. While several methods and indices have been proposed in the literature, they are mostly time-consuming or do not explicitly take into account the demand variance information. In this thesis, we analytically study the effect of demand variance on the system performance and the long chain efficiency. We then develop a variance-based Hub-and-Chain method (VHC) to generate flexibility designs for the balanced but non-identical case. Numerical tests show that VHC outperforms the long chain by 45% on average and outperforms the constraint sampling method by 25% on average. Moreover, it only utilizes 1% of the computational requirements of the constraint sampling method. We then extend VHC to the unbalanced case. We benchmark VHC against the constraint sampling method and the expansion heuristic. We find that our method produces better flexibility designs than both existing methods in 99% of all the scenarios considered. Finally, we test the distributional robustness of VHC and show it is robust for the lognormal distribution and the gamma distribution. In the second essay, we prove in a multistage system the existence of a sparse process flexibility structure that is able to achieve most benefits of full flexibility and identify the corresponding conditions which show the number of links needed to accomplish such benefits is related to the number of stages. We also prove the dual problem of our model is a 0 -1 integer program. Assuming only the moment information known, we then solve the dual problem by a completely positive cross-moment model (CPCMM). In the third essay, we consider a two-period newsvendor problem in a B2C setting. To mitigate the demand risk, the manufacturer introduces a real-option-selling scheme in period 1 despite the spot market in period 2. Customers are uncertain about their future valuations for the product and may prefer to buy the option, the right to purchase one unit product at a fixed price in period 2. By inducing customers to buy the option, the manufacturer obtains more accurate demand information for period 2. We study the joint option pricing and production planning decision faced by the manufacturer. In the first part of this essay, we introduce a fluid model where the number of customers who exercise the option in period 2 is deterministic given a specific option. We first show in the single-option and single-market-segment setting the option-selling scheme is always optimal in comparison with the newsvendor. Moreover, surprisingly we find the optimal strike price constantly equals to the unit production cost. We then extend the fluid model to the two independent market segments setting with three subcases as follows, single option for high segment, single option for both segments, one option for each segment. The findings of interest are subcase 1 is equivalent to a single-option and single-market-segment problem along with a separate newsvendor problem and subcase 2 is a classic Kalman filtering problem while subcase 3 is similar to a mechanism design problem. The dominance of option selling over the newsvendor is again confirmed. In addition the condition of subcase 3 dominating subcase 1 can be characterized by the ratios of segment sizes and marginal profit and cost. In the second part of this essay, we concentrate on a binomial model where the number of customers who will exercise the option is binomial but can be approximated by the normal distribution. The market has correlated informed and uninformed segments with homogeneous valuation distribution. The purpose of this part is threefold, to incorporate the advance selling into the option-selling framework and show the dominance of the latter, to show the dominance of option selling over the newsvendor in the expedite setting and a counterintuitive result - having more information is not always optimal.
|
|---|