Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD63

(i.e. retain, buy or sell shares). To identify CEOs’ emotions in more than 500 calls we developed a deep learning algorithm trained by experts annotating vocal data in a subset of calls. Our findings shed light on underlying emotional mechanisms of financial decision-making under uncertainty, thereby contributing to behavioral economic theory. 2 - Modeling Multivariate Time Series of Counts via Common Factors Fangfang Wang, University of Wisconsin, Madison, WI, United States We develop a new parameter-driven model for multivariate time series of counts. The mean process is modeled as the product of modulating factors and unobserved stationary processes. The former characterizes the long-run movement in the data, while the latter is responsible for rapid fluctuations and other unknown or unavailable covariates. The latent processes are governed by possibly low-dimensional factors. This model is applied to analyze the intraday trading activity of U.S. stocks from ten sectors in the first quarter of 2012. Dynamic relationship between common factors adjusted by their associated loading and intraday volatility is also investigated. 3 - Insample Tangency Portfolio Based Portfolio Forcasting Kyungchan Park, PhD Candidate, The University of Iowa, 21 E. Market St, PBB S221, Iowa City, IA, 52242-1994, United States, Hongseon Kim, Seongmoon Kim Representative portfolio selection model consists of two stages: 1) estimating input values, and 2) constructing portfolio by inputting the estimates. Though estimation errors on inputs are inevitable, optimizations construct portfolios sensitively to the fluctuation of inputs. To date, portfolio models have been designed to minimize the influence of estimation errors while maintaining the two-stage mechanism of optimization based on estimated. Portfolio forecasting method of this study was begun to find a method to utilize the most efficient in- sample portfolio outside of the two-stage mechanism by constructing portfolios using historical in-sample tangency portfolios as input values. 4 - Multivariate Bayesian Structural Time Series Model Ning Ning, University of Washington, Seattle, Seattle, WA, United States This paper deals with inference and prediction for multiple correlated time series, where one has also the choice of using a candidate pool of contemporaneous predictors for each target series. Starting with a structural model for the time series, Bayesian tools are used for model fitting, prediction, and feature selection, thus extending some recent work along these lines for the univariate case. We run an empirical study with one-step-ahead prediction on the max log return of a portfolio of stocks that involve four leading financial institutions. The extensive empirical study confirm that this multivariate model outperforms three other benchmark models. Joint Session DM/Practice Curated: Data Science and Deep Learning III Sponsored: Data Mining Sponsored Session Chair: Muye Wang, Columbia Business School, 3022 Broadway, 4th Floor West, New York, NY, 10027, United States 1 - Network Intrusion Detection Using Partially Labeled Cybersecurity Data Sandjai Bhulai, Vrije Universiteit Amsterdam, Faculty of Science, Network intrusion detection systems (NIDS) have been used to defend computer networks against malicious cyber activities. The lack of training data hinders discovery of new, previously unknown, intrusions. We use recent cybersecurity partially labeled data based on the 2017 Locked Shields exercise. We show how autoencoder and gradient boosting algorithms deal with missing labels on partially labeled data. The results were analyzed by a cybersecurity expert and confirmed that the models were able to classify the known intrusions as malicious and that they were also able to discover new intrusions. This shows how well an unlabeled dataset can be used to construct and evaluate a NIDS. De Boelelaan 1081a, Amsterdam, 1081 HV, Netherlands, Jan Klein, Mark Hoogendoorn, Robert van der Mei, Raymond Hinfelaar n TD65 West Bldg 104B

n TD63 West Bldg 103B Joint Session DM/Practice Curated: Data Science for Engineering and Facilities Design Sponsored: Data Mining Sponsored Session Chair: Reza Alizadeh, University of Oklahoma, Norman, OK, United States 1 - Parameter Design Optimization of Products with an Ordinal Categorical Response Using Random Forests Gulser Koksal, Professor, Middle East Technical University, Industrial Engineering Department, Ankara, 06800, Turkey, Secil Gulbudak Dil We propose an alternative method for finding optimal settings of product and process design parameters for the case of an ordinal categorical product/process response. The method utilizes Random Forests for modelling mean and variance of the response at a given set of parameter settings. The method uses different weighting strategies of Random Forest, and it is applied on three case problems. Results obtained are compared with those of previous studies that used the same data sets. 2 - Minimization of Cycle Time Variance in Smart Warehousing Mohamed El Tonbari, Georgia Institute of Technology, 755 Ferst Dr, Atlanta, GA, 30318, United States, Leon McGinnis We are interested in the operational decision in central fill pharmacies of assigning drugs to dispensers to be collected by vials traveling on conveyors. More specifically, we are concerned with finding the drug assignment which minimizes the cycle time variance. We show that the organ pipe arrangement attains the minimum under the assumption of constant travel times between adjacent dispensers. In the case of varying travel times between dispensers, we propose a heuristic which performs well empirically, achieving an optimality gap in the order of 0.1%. Our results can be generalized to other applications as well. 3 - Queueing Analysis of an MIAPP-AS/RS Order Picking Operation Jingming Liu, Research Assistant, University of Arkansas, Fayetteville, AR, 72701, United States, John A. White, Haitao Liao An M/G/1 queue is used to analyze an MIAPP-AS/RS system, in which case-level order picking occurs at multiple in-the-aisle picking positions (MIAPP) located on the floor and mezzanine levels. The S/R machine is treated as a server, and picking positions are treated as customers. Picking positions generate demands for replenishment performed by the S/R machine. Performance measures are obtained analytically for dedicated and random storage policies, as well as for finite and infinite populations. The results yield insights into the impact of the storage policies on the performance measures and the conditions where an infinite population approximation is reasonable. 4 - Using Multiple Surrogates for Metamodeling in Complex Engineering Systems Reza Alizadeh, Graduate Research Assistant, The University of Oklahoma, 202 W. Boyd Street, Room 219, Norman, OK, 73019- 1022, United States, Liagyue Jia, Guoxin Wang, Janet K. Allen, Farrokh Mistree Surrogate models are commonly used to replace expensive simulations of engineering problems. Frequently, a single surrogate is chosen based on past experience. However, we may use all or a subset of the accurate surrogates. We aim at achieving the least accuracy loss using all or the combination of surrogates using multiple surrogates, cross-validation, covariance matrix approximation, and random forest methods. n TD64 West Bldg 104A Joint Session DM/Practice Curated: Data Science in Finance Sponsored: Data Mining Sponsored Session Chair: Ning Ning, University of Washington, Seattle, WA, United States 1 - The Effect of Emotional Cues on Making Economic Decisions under Uncertainty Pieter Geelen, Maastricht University, Maastricht, Netherlands, Business Intelligence and Smart Service Institute, Heerlen, Netherlands, Stefano Bromuri, Stefano Bromuri, Mahdi Ebrahim, Mahdi Ebrahim, Deniz Iren, Deniz Iren We investigate the understudied role of emotions on investors’ decision. Specifically, we identify six basic human emotions (i.e. happiness, sadness, surprise, fear, anger, and disgust) expressed by S&P 500 CEOs during their earning conference calls and examine their effects on financial analysts’ decisions

370