Informs Annual Meeting 2017

SA56

INFORMS Houston – 2017

SA54

subset of the N stocks only. We propose a hybrid approach to this problem based on an iterated greedy heuristic and quadratic programming. In a computational experiment with real-world data, we demonstrate that the proposed approach leads to a high tracking accuracy. 3 - Pricing of Cyber Insurance Contracts in a Network Model Kerstin Weske, Gottfried Wilhelm Leibniz Universität Hannover, Welfengarten 1, Hannover, 30167, Germany, weske@stochastik.uni-hannover.de, Matthias A. Fahrenwaldt, Stefan Weber We investigate a novel approach for pricing cyber insurance. The considered cyber threats, such as viruses and worms, diffuse in a data network. The spread of the cyber infection is modeled by an interacting Markov chain. Conditional on the underlying infection, the occurrence and size of claims are described by a marked point process. We develop a novel mean-field approach to compute prices of cyber insurance contracts. Several numerical case studies illustrate the impact of the network topology. 4 - Regulation of Financial Networks Dan Mitchell, University of Minnesota, 111 Church Street SE, Minneapolis, MN, 55455, United States, damitche@umn.edu, Stathis Tompaidis We consider a financial system in which individual institutions are organized in a directed network representing random payoffs from financial instruments and a regulator is able to set capital requirements for each firm in the network. The regulator sets these capital requirements to reduce the risk of systemic failure to the whole system. 362C Advances in Location Theory Sponsored: Location Analysis Sponsored Session Chair: Oded Berman, University of Toronto, Toronto, ON, M5S 3E6, Canada, berman@rotman.utoronto.ca 1 - Maximal Collaborative Cover Problems with an Application to Hazardous Materials Emergency Response and a Case Study from China Jiahong Zhao, McMaster Unviersity, Hamilton, ON, Canada, zhaoj77@mcmaster.ca, Peng Hu, Vedat Verter An efficient emergency system can reduce the risk derived from hazmat network. We restrict our attention to reducing the waiting time and propose a methodology of optimizing multiple emergency sources. We generate the hazmat network based on urban road, and calculate the response time for different hospitals and fire stations. A new model is modified to decide the locations of multiple sources, where the locations are optimized according to minimizing the gap of each hospital and fire station. To solve this model, a solution procedure is designed from generic algorithm. This methodology is applied to the problem in Chengdu, China, and the results provides a significant reduction in waiting time. 2 - Location and Pricing of a Mobile Food Vending Facility using Public Data Rajan Batta, University at Buffalo, Buffalo, NY, United States, batta@buffalo.edu, Rishabh Bhandawat The aim of this talk is to formulate a location problem for a market entrant mobile food vendor to maximize profits based only on public data. Specifically, the model suggests dynamic location and service quality decisions (cuisine, price and other service criteria) from online yelp reviews, customer’s proximity to the facility along with geographic and demographic data. The error in prediction is reduced by including other facilities and reviews. 3 - Generalised Facility Location and Design Problem Dmitry Krass, University of Toronto, Rotman School of Management, 105 St George Street, Toronto, ON, M5S.3E6, Canada, krass@rotman.utoronto.ca, Robert Aboolian, Oded Berman We present a general problem that can be used to represent a wide variety of known location and design models, including models with elastic demand, competitive facilities, different customer allocation rules, etc. While the general model has non-linearities in both the objective and the constraint, these can be removed by exploiting the special structure. This leads to a solvable MIP formulation. SA56

362A Agriculture Analytics at IBM Research Invited: Agricultural Analytics Invited Session Chair: Robin Lougee, IBM Research, Yorktown Heights, NY, 10598, United States, rlougee@us.ibm.com 1 - Precision Agriculture leveraging IoT and Curated Satellite Data Analytics Ranjini Guruprasad, IBM.Research India, Manyata Embassy Business Park, Bangalore, Karnataka, 560045, India, ranjinib@in.ibm.com, Jagabondhu Hazra Increasing agricultural productivity is key to raising GDP growth in India, as well as lifting millions of people out of poverty. Digitization, mobile and cognitive technologies have the opportunity to play a pivotal role to enable this future. At IBM Research - India, we are combining multiple global satellite based information sources to compute agronomic insights at sub-acre level. Farmers can now tap the sophisticated output of cognitive analytics to get predictive insight at the right time to manage scarce groundwater-based irrigation, optimize the timing/amount of fertilizers & pesticides, minimize wastage, etc. This presentation will cover few use cases specific to India/SA. 2 - Modeling of Expert Knowledge for Crop Cultivation Tatsuya Ishikawa, IBM Research - Tokyo, 19-21, Nihonbashi Hakozaki-cho, Chuo-ku, Tokyo, 103-8510, Japan, Ishikawa@jp.ibm.com, Yasuteru Kohda, Hironobu Takagi We are facing a risk of losing skills of crop cultivation due to the retirements of aging expert farmers in Japan. Their key knowledge is adaptive decision-making such as appropriate temperature settings in greenhouses depending on stages of development in crops and environmental conditions. In this talk, we present a framework which combines (1) measurements of crop growth and environment, (2) simulation models of these dynamics with data assimilation, and (3) model- based adaptive control. We discuss the capability of our framework to model farmer’s decision-making. 3 - Geospatial Analytics Powered by IBM Pairs Big-data Platform Siyuan Lu, IBM, 1101 Kitchawan Road, Yorktown Heights, NY, 10598, United States, lus@us.ibm.com, Marcus Freitag The PAIRS geospatial analytics platform is a big-data service of IBM Research that aims at reducing the barrier for analytics or machine-learning on large-scale and disparate spatio-temporal data sources such as weather, satellite, land- classification, lidar, census, sensors, etc. Incoming data is curated on the unified nested PAIRS grid, and derived analytics layers, such as evapotranspiration, are automatically populated as new timestamps of the underlying data arrive. The users may upload their own data in additional layers of the same grid, and perform their analytics on a mix of open and proprietary layers. In this talk the PAIRS architecture and analytics on top of it will be discussed. 362B Multi-Firm Decision Making Sponsored: Financial Services Sponsored Session Chair: Daniel Mitchell, University of Minnesota, damitche@umn.edu 1 - Robust Portfolio Selection with Transaction Costs Jingnan Chen, Singapore University of Technology and Design, 20 Dover Drive, Singapore, 138682, Singapore, jingnan_chen@sutd.edu.sg, Ning Zhang We propose a robust and sparse portfolio selection model to obtain portfolios that are insensitive to parameter estimation errors and have low transaction costs. Two components of transaction costs, i.e., the broker’scommission and the market maker’s spread, are both considered. We adopt a distributional robust approach to maximize the worst-caseexpected utility over a set of distributions with the same first- and second-order moment information, and impose l0 and l1 regularization to account for transaction costs. The inclusion of l0 norm results in a non-convex and non-smooth optimization program and we develop an efficient algorithm to find the optimal portfolio. 2 - A Hybrid Approach to the 1/n Portfolio Tracking Problem Oliver Strub, University of Bern, Schuetzenmattstrasse 14, FM.Quantitative Methoden, Bern, 3012, Switzerland, oliver.strub@pqm.unibe.ch, Norbert Trautmann The 1/N portfolio strategy corresponds to investing an equal amount in all N stocks from a given investment universe. This strategy exhibits desirable empirical risk-return properties, but causes substantial management costs if N is large. We consider the problem of tracking the 1/N portfolio’s returns by investing in a small SA55

21