Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WC36

2 - Weighted Stochastic Approximation for Large-scale Network Computation Jingchen Liu, Columbia University, Department of Statistics, 1255 Amsterdam Avenue, New York, NY, 10027, United States, Xueying Tang, Zhi Wang We consider the parameter estimation and computation via stochastic approximation for large-scale multi-relational network and propose a weighted sampling scheme biased towards observations with more information regarding the parameters. It is shown to be computationally more efficient if the weights are properly chosen. Besides the computational efficiency, the biased scheme is observed to contribute to the estimation quality and substantially improves the out of sample prediction performance. 3 - Credit Risk: Closed Form Approximate Maximum Likelihood Estimation and Fast Simulation Anand Deo, Tata Institute of Fundamental Research, Homi Bhabha Road, Navy Nagar, Colaba, Mumbai, India, Sandeep Juneja We consider discrete default intensity based and logit type reduced form models for conditional default probabilities for corporate loans where we develop simple closed form approximations to the maximum likelihood estimator (MLE) when the underlying covariates follow a stationary Gaussian process. In a practically reasonable asymptotic regime where the default probabilities are small, the number of firms and the time period of data available is reasonably large, show that the proposed estimator behaves similarly to the MLE. We also derive large deviations asymptotics for large losses. These have interesting geometric properties, which we exploit to develop efficient simulation techniques. 4 - Efficient Rare-event Probability Computation of Functional Exceedance Raghu Pasupathy, Purdue University, West Lafayette, IN, 47907, United States We present methods for efficiently calculating the probability of the function f of an elliptical random vector X exceeding a given threshold. Elliptical random vectors are general in that they subsume a variety of random vectors that are commonly in use, e.g, multivariate normal, multivariate T, and multivariate logistic. The proposed method uses importance sampling and actively exploits the local structure of f to construct estimators that exhibit bounded relative error. We discuss extensions to functions of appropriate stochastic processes. n WC40 North Bldg 226B Combinatorial Optimization Contributed Session Chair: Hadi Farhangi, Savannah State University, Savannah, GA, United States 1 - Client Selection for a Risk-sensitive Commodity Options Underwriter with Poisson Demand Belleh Fontem, University of Mary Washington, Fredericksburg, VA, 22407, United States, Megan Price We consider a maximization problem for a risk-sensitive underwriter of an option contract on a commodity with geometric Brownian motion spot price trajectories. Firms hoping to enter into service agreement with the underwriter each face Poisson demands that are the underwriter’s responsibility to satisfy. While considering payoff risk, the underwriter aims to select the optimal combination of client firms to privilege with its option contract. Using payoff variance as our risk measure, we derive for a special case, the optimal solution algorithm for the variance-constrained maximization problem. Results from the special case analysis inspire the design of two heuristics for the general case. 2 - An Evolutionary Approach to Constrained Many-objective Combinatorial Optimization Hayrullah Mert Sahinkoc, Bogazici University, Istanbul, Turkey, Umit Bilge Many-objective optimization tries to characterize and overcome the challenges posed by the high number of objectives. Most of the existing studies work on well-defined continuous mathematical functions with designed Pareto front characteristics whereas combinatorial and constrained problems are rarely addressed. Many-objective 0-1 knapsack problem with multiple constraints is chosen in our study and our proposed algorithm combines the effective features of the existing many-objective approaches with several other prominent evolutionary strategies in an innovative fashion. Numerical results show the success of the proposed algorithm compared to some existing approaches.

n WC36 North Bldg 224B Inventory Management III Contributed Session Chair: Stefan Minner, Technische Universitaat München, Munich, 80333, Germany 1 - A Supply Planning and Inventory Management Problem for a Single Item with Random Demand and Substitutable Components Zehra Melis Teksan, Assistant Professor, Ozyegin University, Cekmekoy Kampusu Nisantepe Mah Orman Sk, Cekmekoy, Istanbul, 34794, Turkey, Joseph Geunes We consider a supply planning and inventory management problem for an item within a single planning period with random demand. The production of the end- item requires a particular component whose supply availability depends on the price the producer offers to suppliers. The amount of supply and the production capacity are determined by the response of suppliers to the producer’s price offer. The goal is to determine optimal supply pricing and production decisions when the component has a number of potential substitutes with different supply availabilities. We analyze the optimal supply-pricing policy with various pricing options, as well as policies when supplier-specific fixed charges exist. 2 - An Inventory System with Reserved Stock We consider an infinite horizon periodic-review base stock inventory system with partial backorders and lost sales. For the long-run average cost problem, we determine structural properties of a proposed (k, R) policy and discuss how to find the optimal base stock R and reservation or holdback stock k. We also describe how two supply modes, of the type due to Barankin (1961), can be incorporated into our model by using a (k, Z, R) policy while generating further insights into the policy structure. 3 - The Joint Replenishment Problem under Cycle Time Constraints Yu-Liang Lin, North Carolina State University, Raleigh, NC, 27606, United States, Chi-Yi Chen The Joint Replenishment Problem (JRP) is to determine the replenishment cycle times of a group of items in single facility system over an infinite horizon. We investigated the JRP with cycle time constraints that are popular in inventory systems with restrictions such as max/min order quantity and storage capacity, etc. Based on the optimality structures, we categorize the items into four types and propose an efficient algorithm for solving this constrained JRP. Our random experiments demonstrate the proposed algorithm is effective as compared with the existing approaches in the literature. 4 - Integrated Location-inventory Optimization in Slow Moving Spare Parts Networks Using Benders Decomposition Stefan Minner, Technical University of Munich, Munich, 80333, Germany, Patrick Zech, Zuo-Jun Max Shen Virtual inventory sharing is an effective method to pool inventories in spare parts networks. We propose a novel location-inventory model which integrates strategic facility choice, tactical base-stock level setting and operational sourcing decisions. The mixed integer linear program combines a set-covering problem and a semi-Markov decision process. The model’s special structure suggests Benders decomposition as an effective method to solve the problem. Numerical experiments confirm the efficiency of Benders decomposition and emphasize the value of an integrated model compared to a sequential location first, inventory and sourcing second approach. n WC37 North Bldg 225A Applied Probability and Simulation Sponsored: Applied Probability Sponsored Session Chair: Sandeep Juneja, Tata Institute of Fundamental Research, Mumbai, 400005, India 1 - An Asymptotically Optimal Index Policy for the Finite Horizon Restless Bandit Peter Frazier, Cornell, Weici Hu We consider the restless bandit, a generalization of the multi-armed bandit in which arms may change state when they are not pulled. In the stochastic infinite- horizon setting, Whittle proposed the Whittle index policy and conjectured the per-arm optimality gap vanishes as the number of arms grows to infinity while holding fixed the fraction of arms that can be pulled per period. Weber and Weiss showed this conjecture is true when arms have only 3 states, or when the fluid limit has a globally stable equilibrium point. We propose a novel but related index policy for the finite-horizon setting, and show it is asymptotically optimal in the same sense without restrictions on the fluid limit. Arnab Bisi, Johns Hopkins Carey Business School, 100 International Drive, Baltimore, MD, 21202, United States, Yanyi Xu, Maqbool Dada

482