Informs Annual Meeting 2017

WA26

INFORMS Houston – 2017

4 - Scope Advantage or Disadvantage? – Cost Spreading and Firm’s Innovation Response to Demand Shock Xiaoshu Bei, Duke University, 2514 Preston Ave, Durham, NC, 27705, United States, xiaoshu.bei@duke.edu In this paper, I study firms’ investment in innovation in response to exogenous shock in demand. I look at firms’ patenting activity and how that is correlated with US exportation to China after China joined WTO. I use the cost spreading theory to explain the relationship between firms’ structure/diversification strategy and their response to increase in demand. Consistent with early studies, both size and scope confer an advantage in innovation when facing demand pull. Vertically integrated firms have a much higher response in innovation, mostly due to their ability to scale up dynamically and capture the market opportunity. This enables them to attain a larger output to spread the cost of innovation. 5 - Dark Sides of Interpersonal Citizenship Behaviors for Innovation: A Social Network Approach Hongzhi Chen, PhD Candidate, Purdue University, 146 Arnold Dr Apt 3, West Lafayette, IN, 47906, United States, hzchen@purdue.edu, Lusi Wu, Chunhua Chen Assuming its desirable outcomes, researchers tend to focus on how to promote interpersonal citizenship behaviors (ICB), leaving its consequences untapped on. Using social network approach, this study examines dark sides of ICB in innovation process. Specifically, how performing and receiving ICB will affect an employee’s structural positions within intra-organizational collaboration networks, which in turn unintentionally inhibit innovation through network closure and collaborative overload. Based on time-lagged data from 200 employees in a pharmaceutical R&D center, our findings will provide novel and practical insights on ICB and network dynamics, especially in innovation process. 350B Simulation and Optimization Contributed Session Chair: Melike Meterelliyoz, TOBB University of Economics and Technology, Ankara, Turkey, mkuyzu@etu.edu.tr 1 - Winner Play Structures in Random Knockout Tournaments Yang Cao, University of Southern California, 909 W. Temple St, Apt 354, Los Angeles, CA, 90012, United States, cao573@usc.edu We consider a knockout tournament whose contestants are 1,2,...,n. Suppose that each player i has strength v_i, and a game between i and j is won by i with probability v_i/(v_i+v_j). The players are randomly assigned to positions 1,2,...,n. The players in the first two positions play a game in the first round, and then in round r, r = 2,3,...,n-1, the winner of the previous round plays with the player in position r+1. We call this tournament format the winner-play structure. We prove bounds on players’ tournament win probabilities, provide efficient simulation algorithms, and propose conjectures with evidence. Moreover, we study a continue play model and give the stationary probabilities. 2 - Compactness Approaches for Importance Sampling Alexander Shkolnik, Postdoctoral Scholar, University of California- Berkeley, Evans Hall 339, Berkeley, CA, 94720, United States, ads2@berkeley.edu A standard approach for optimal importance sampling (IS) relies on large deviations (LD) theory. That is, LD techniques are employed to derive upper and lower bounds on the second and first moments of the IS estimator. This requires the development of a LD principle, a program that is frequently too difficult to implement for practical problems. We provide an analysis based on the LD counterpart of Prohorov’s (relative compactness) theorem that circumvents many obstacles in this standard program. We demonstrate the effectiveness of the approach on several challenging problems involving Markov chains, affine processes and event timing models. 3 - Governance of Complex Transactions in Project Business Networks Jaakko Kujala, Professor, University of Oulu, Oulu, Finland, jaakko.kujala@oulu.fi The purpose of the study is to analyze the effectiveness of network form of governance to coordinate and safeguard complex transactions in inter- organizational project business network. The network form of governance is suggested to arise as a response for demand uncertainty, project uniqueness and project complexity. It is characterized by informal social systems rather than bureaucratic structures and formal contractual relationships. An agent-based simulation model is created to simulate the effectiveness of social mechanisms such as restricted access, macroculture, collective sanctions and reputation, and how structural embeddedness in a network facilitates use of these mechanisms. WA26

4 - Characterization and Prediction of Hyper-jump Diffusion in Consensus Dynamics on Networks Xin Su, PhD Student, Arizona State University, ASU - CIDSE, PO Box 878809, Room 553, Tempe, AZ, 85287-8809, United States, xinsu3@asu.edu, Theodore P. Pavlic Network consensus dynamics have been studied for: modeling opinion diffusion, robotic formation-control policies, and development of particle swarm optimization (PSO) metaheuristics. A recent discovery is that stochastic link failures can lead to Levy-like hyper-jump diffusion, which might be used for new PSO strategies or to better understand sharp changes in public opinion. Here, we: (a) quantify the emergence of these jumps; (b) demonstrate the role of parameters in the control of this behavior; and (c) use information-theoretic insights to find real-time predictors for impending jumps. 5 - Multiply Reflected Variance Estimators for Simulation Melike Meterelliyoz, Assistant Professor, TOBB University of Economics and Technology, Sogutozu Cad. No 43, Sogutozu, Ankara, 06560, Turkey, mkuyzu@etu.edu.tr, Kemal Dingeç, David Goldsman, Christos Alexopoulos We consider new estimators for the variance parameter of a steady-state simulation output process that are based on the Reflection Principle of Brownian motion. These estimators are computed by invoking multiple reflections of the standardized time series of the underlying process, and thus generalize previous reflected estimators that are each based on only a single reflection point. Our current scheme can be exploited to produce low-bias, low-variance estimators that are superior to their single-reflection-point predecessors. We illustrate the enhanced performance of the multiply reflected estimators via exact calculations and Monte Carlo experiments. 350C Crowd Wisdom Transformation using Social Media Analytics Invited: Social Media Analytics Invited Session Chair: Wenqi Shen, PhD, Virginia Tech, Blacksburg, VA, 24060, United States, shenw@vt.edu Co-Chair: Fang Jin, PhD, Texas Tech, Lubbock, TX, 43104, United States, fang.jin@ttu.edu Co-Chair: Zhilei Qiao, Virginia Polytechnic Institute and State University, Blacksburg, VA, 24061, United States, qzhilei@vt.edu 1 - Sensing Specific Hazards from Online Reviews across Diverse Product Categories Nohel Zaman, 258 Old Cedarfield Drive, Blacksburg, VA, 24060, United States, znohel@vt.edu, Alan Abrahams, Richard James Gruss We focus on online reviews that report a physical health and safety concerns of consumers from product-usage in diverse product categories. We distinguish between events that are pathways to injury, and injury outcomes, and between actual injuries and potential injuries. We develop event-specific, injury-specific, and body-part specific “smoke” term lists. These terms can be used for detecting, categorizing, and prioritizing hazards across diverse consumer product categories. We determine which terms are generic and which are hazard-specific. This study facilitates rapid sensing of specific potential hazard events and outcomes, and maps online reviews to standard medical coding schemes. 2 - Aggregating Crowd Opinions using Statistical Learning Qianzhou Du, Virginia Tech, Pamplin Business School, 100 Otey Street, Room 301, Blacksburg, VA, 24061, United States, qiand12@vt.edu Recent studies have shown that aggregating crowd opinions has been an effective method to improve the performance of decision making. Most existing crowd opinion aggregation studies employ heuristics to determine a weighting scheme for individual judges based on their expertise or past prediction performance. However, those heuristics based methods often fail to determine the optimal weights and achieve the optimal prediction performance in the aggregated crowd opinion. We propose a new crowd opinion aggregation model, namely aggregating crowd opinions using statistical learning (ASL), which has a fitting procedure for the weighting scheme and a mechanism of estimating the probabilities of event outcomes based on statistical learning. Statistical learning is expected to capture hidden patterns in crowd opinions and assign better weights to individuals than heuristic based weighting methods. We empirically evaluate ASL in comparison to four baseline methods using real data collected from an online stock prediction community, StockTwits. The results show that ASL significantly outperforms all the baseline methods in terms of both a quadratic prediction score and prediction accuracy. WA27

450