Informs Annual Meeting 2017

WC61

INFORMS Houston – 2017

2 - Conditional Value at Risk as SVR Minimization Emre Eryigit, Middle East Technical University-Ankara, Ankara, 06800, Turkey, emreryigit@gmail.com, Cem Iyigun, Ozlem Cavus Iyigun We study Support Vector Regression (SVR) to forecast electricity load in short term. Our algorithm includes Conditional Value at Risk measure and the optimization model behind the forecasting algorithm minimizes the risk measure while minimizing the forecasting result. Our preliminary results are promising in comparison to SVR without risk parameter. 3 - BBQ Time Series and Optimal Control Joshua Woodruff, University of Texas, 4509 Rio Robles Dr, Austin, TX, 78746, United States, jwoodruff@optimizedfinancialsystems.com, Dragan Djurdjanovic Cooking truly exemplary Texas pit BBQ requires maintaining the perfect temperature over 8 to 12 hours. Making this type of BBQ in an offset smoker requires constant attention to the wood added, vent openings, and outside temperature. In this session we present a tutorial on time series modeling in R and optimal control using CPLEX. 4 - Forecasting the Automobile Market Equilibrium in Consideration of the Externalities of Transport Sector YoungJun Park, Seoul National University, Seoul, Korea, Republic of, pyjoon90@snu.ac.kr, HyungBin Moon Countries are implementing different political measures to manage and alleviate the greenhouse gas emissions from the transportation sector. The present study forecast the automobile market equilibrium by considering the fuel cost and the external cost related to the transportation sector. In particular, the study constructs a linkage among the changes in fuel costs, market shares of different vehicle types, and the external costs to estimate the market equilibrium according to the changes in consumer behavior. Based on the estimation, the study derives the principle rules on how to reflect the external costs of transportation sector that satisfy the goal of greenhouse gas emission reduction. 370B Finance, Banking and Insurance Contributed Session Chair: Aein Khabazian, University of Houston, Houston, TX, United States, akhabazian2@uh.edu 1 - Data Mining and Machine Learning Role in Insurance Business Xiaoguang Tian, University of North Texas, 1155 Union Circle, Denton, TX, 76203-5017, United States, xiaoguang.tian@unt.edu The insurance industry is facing unprecedented challenges and opportunities in the data-driven paradigm. The incoming big data and technologies are transforming the traditional insurance business model and pushing the new generation of insurers to rethink their business strategies and then to make better decisions. In this study, the author will discuss challenges that insurance industry is encountering, and review the existing data mining and machine learning techniques that can be applied in insurance management across the business value chain. Two ongoing data mining projects will be introduced by using real- world data. 2 - Bank Risk Assessment using Unstructured Text Disclosures Xiaodi Zhu, PhD Candidate, Stevens Institute of Technology, Hoboken, NJ, 07030, United States, xzhu@stevens.edu, Aparna Gupta, Steve Yang Bank risk has become an important focus of financial regulators and investors after the 08-09 financial crisis. Banking sector risks are usually measured by quantitative information from structured financial disclosures. This study analyzes the textual risk disclosures that presumably provide forward looking information about firms’ future risks. We extract risk factors from banks’ risk disclosures using a sentence based dynamic topic modeling approach. Using data from both FDIC and the SEC, we find significant correlation between the unstructured risk topics and the conventional risk measures. Moreover, we demonstrate the predictive power of risk factors on bank’s future risk. 3 - Modeling Psycho-financial Profiles of Italian Bank Customers Galina Andreeva, Associate Professor, University of Edinburgh, 29 Buccleuch Place, Edinburgh, EH8 9JS, United Kingdom, galina.andreeva@ed.ac.uk, Caterina Liberati We investigate the relationship between psychometric measures obtained through the survey, personal/demographic and financial characteristics of the account holders of an Italian bank. The main objective of the analysis is to explore the latent structure of different types of variables that the dataset contains via factorial representation. We also detect presence of the natural groups by means of explorative unsupervised classification, and comment on different types of customers’ profiles from psychological and financial behavior perspectives. WC61

4 - Assessment and Control of Systemic Risk Aein Khabazian, University of Houston, 5465 Braesvalley Dr. APT.566, Houston, TX, 77096, United States, akhabazian2@uh.edu, Jiming Peng In this work, we assess the systemic risk based on the optimization model introduced by Eisenberg and Noe (2001), where only partial information regarding the liability matrix is revealed. We develop an algorithm to identify the most vulnerable structure in the network. Numerical experiments illustrate that the contagious risk in the identified vulnerable network is more significant than what underestimated in the literature. Second, we develop an algorithm to identify the most stable network structure, and use it to develop a strategy for systemic risk mitigation. We evaluate the performance of the new strategy. 370C Reliability Contributed Session Chair: Douglas King, University of Illinois at Urbana-Champaign, Urbana, IL, United States, dmking@illinois.edu 1 - Enforcing Contiguity in Branch and Bound Algorithms to Solve Geographical Districting Problems Rahul Swamy, PhD Student, University of Illinois at Urbana- Champaign, 205 E.Clark St, Apt 303, Champaign, IL, 61820, United States, rahulswamy91@gmail.com, Sheldon H. Jacobson, Douglas M. King The class of Geographical Districting Problems has been modeled in the literature as a combinatorial graph partitioning problem. Enforcing contiguity is typically the bottleneck in exact methods, since it is hard to do so without avoiding an exponential set of constraints. We present a method to enforce contiguity using Geo-Graphs in branch and bound algorithms, where the special planar structure of the districting problem is used for improved efficiency. This is done so dynamically while evaluating potential contiguity of partial solutions. Theoretical and computational results are presented. 2 - Practical Mathematical Methods for Political Redistricting and Competitive and Fair Elections Hsiao-Shen Jacob Tsao, Professor, San Jose State University, 1 Washington Square, San Jose, CA, 95192-0085, United States, jacob.tsao@sjsu.edu, Royce Lin A reason for the current political polarization in the US is partisan political redistricting or “gerrymandering” for congressional elections. Population equality, contiguity and compactness for any district are required by the Constitution. Political reforms for decreasing partisan influence will not suffice; the commission staff needs software tools to help achieve these requirements and to explore or achieve fair representation, competitiveness, etc. However, the mathematical solution processes produced so far have not been practical for such use. We propose a new approach that is mathematically rigorous, easier for the staff to understand and practical for software implementation. 3 - Scalable Contiguity Assessments in Practical Political Districting Douglas M.King, University of Illinois at Urbana-Champaign, 117 Transportation Building, 104 S.Mathews Avenue, Urbana, IL, 61801, United States, dmking@illinois.edu, Sheldon H. Jacobson, Edward C. Sewell Practical political districting involves the creation of contiguous districts from small discrete units. When the number of units is large, ensuring contiguity poses a substantial computational burden. This talk applies scalable contiguity algorithms based on the geo-graph framework to the creation of United States Congressional Districts in four states. WC62 370F Special Settings of Vehicle Routing Problems Sponsored: Transportation Science & Logistics Sponsored Session Chair: Xiangfei Meng, USC, 2801 Orchard Ave, Apt 10, Los Angeles, CA, 90007, United States, xiangfem@usc.edu 1 - An Adaptive Large Neighborhood Search for Real-world Vehicle Routing Problems Haoyuan Hu, Zhejiang Cainiao Supply Chain Management Co., Ltd., Hangzhou, China, haoyuan.huhy@cainiao.com, Ying Zhang, Wenjia Ma, Kunpeng Han, Guotao Wu Along with the rapid growth of online shopping, real-world vehicle routing problems (VRP) arise in city logistics. We propose an adaptive large neighborhood search heuristic for the VRP and some of its realistic variants. Problem specific WC65

524