Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TD21

introduced a transformation of the likelihood function that significantly improved the algorithms’ performance. The method was used to develop a tool for managing vending machines, and results based on these data are presented. 2 - A Constant-factor Approximation Algorithm for Network Revenue Management Mika Sumida, Cornell Tech, New York, NY, 10128, United States, Yuhang Ma, Paat Rusmevichientong, Huseyin Topaloglu We provide a constant-factor approximation algorithm for network revenue management problems. In our approximation algorithm, we construct a policy using value function approximations that are expressed as linear combinations of basis functions. We use an efficient backward recursion to compute the coefficients of the basis functions. If each product uses at most L resources, then the total expected revenue obtained by our policy is at least 1/(1 + L) of the optimal total expected revenue. In many network revenue management settings, the number of resources used by a product remains bounded. In this case, our policy provides a constant-factor performance guarantee. 3 - CDLP-based Bid-prices for Network Revenue Management The primal solution of the linear programming formulation for choice-based revenue management (CDLP) is known to be effective from a revenue perspective. However, from a practical perspective, there is a representation issue since airline systems are in principle incompatible with those offer sets. In this work, we consider a variation of the CDLP that naturally generates bid-prices. Based on our numerical experiments, their performance is promising. 4 - On Reformulations of Approximate Linear Programs for Network Revenue Management Jiannan Ke, Shanghai Jiao Tong University, 1954 Huashan Road, Zhongyuan 111, Shanghai, 200030, China, Dan Zhang, Huan Zheng Approximate linear programming is a popular method to approximately solve dynamic programs that suffer from the curse of dimensionality. However, approximate linear programs (ALPs) still pose computational challenge due to the large number of variables or constraints. Vossen and Zhang (2015) show that ALPs for the network revenue management problem can be dramatically reduced in size for affine and separable piecewise linear approximations under either independent or discrete choice models of demand. In this paper, we give an alternative proof of the results based on reformulations of constraints in the ALPs and duality arguments. The result also applies to the dynamic pricing setting. Juan Jose Miranda Bront, Universidad Torcuato Di Tella / CONICET, School of Business, Av Figueroa Alcorta 7350, Buenos Aires, C1428BCW, Argentina, Isabel Mendez-Diaz, Gustavo J. Vulcano, Paula Zabala Sponsored: Finance Sponsored Session Chair: Xuedong He, The Chinese University of Hong Kong, Shatin, Hong Kong Co-Chair: Nan Chen, Chinese University of Hong Kong, Shatin N. T, Hong Kong 1 - Risk Adjustment Processes Sturm Stephan, Worcester Polytechnic Institute (WPI), 100 Institute Road, Worcester, MA, 01609, United States In the wake of the financial crisis factors commonly ignored in financial modeling got on the center stage: funding differentials and the absence of the postulated riskfree rate and counter party credit risk. These factors have been taken into account in derivatives pricing in the framework of value adjustments (also known as XVA). However, to the best of our knowledge this has so far not been accounted for in risk management. We are developing a theory of risk-adjustment processes incorporating funding spreads and credit risk to market risk evaluations based on BSDE representations of convex risk measures. 2 - Evidence from Risk Disclosures: Are Banks Transparent About Their Risks? Steve Yang, Stevens Institute of Technology, Hoboken, NJ, United States, Aparna Gupta, Xiaodi Zhu We address the risk information transparency of US banks by examining the banks’ risk disclosures in annual reports (10-K) and its relationship with bank risk-taking. We quantify the descriptive using the Sentence-based Dynamic Topic Model and propose an automatic topic labeling algorithm based on a predefined keyword dictionary of 16 risk factors. Comparing the risk factors against the CAMELS risk rating system shows that US banks’ risk disclosures provide useful information pertaining to specific bank risk-taking behavior. Large and small banks differ in emphases, with both lacking disclosing management and liquidity risks. n TD23 North Bldg 131A Topics in FinTech and Risk Management

n TD21 North Bldg 129B Information in Networks Sponsored: Revenue Management & Pricing Sponsored Session Chair: Ozan Candogan, University of Chicago, Chicago, IL, 27708, United States 1 - Fake News Propagation and Detection Yiangos Papanastasiou, University of California Berkeley, 2220 Piedmont Ave, Berkeley, CA, United States We consider the problem faced by a social media platform that is observing the sharing actions of a sequence of rational agents and is dynamically choosing whether to conduct an inspection (i.e., a “fact-checkö) of an article whose validity is ex ante unknown. We present results pertaining to: (i) the properties of the agents’ sharing behavior; (ii) the structure of the platform’s optimal inspection policy; and (iii) the impact of fake news on the society’s learning environment. 2 - Fact-checking Policies for Fake Content Detection in Social Networks Gizem Yilmaz, University of Chicago, 5807 S. Woodlawn Ave, Chicago, IL, United States, Ozan Candogan, Varun Gupta We study the spread of multiple possibly erroneous pieces of content on a network. For a given content the spread rate is time-varying, and the spread process is approximated using an inhomogeneous continuous time branching process. We compute the offspring mean and variance of this branching process and discuss its relationship to the degree distribution of the network. We address the problem of minimizing the total spread of erroneous content, by using simulation techniques together with the moment information of the approximating branching process. We pose the problem as a continuous time optimal control problem and suggest heuristics for an approximate solution. 3 - An Edge-based Model of Network Formation Hao Li, Hong Kong University of Science and Technology, Clear Water Bay, Academic Building, Room 5551, Kowloon, Hong Kong, Ying-Ju Chen, Jin Qi We present a dynamic model of network formation where new nodes find and connect old nodes based on edges: first, a new node finds an edge purely randomly; second, the new node connects to each endpoint of the selected edges with separated probabilities; third, the new node connects to the out-neighbours of the tail node of the selected edge with a probability. We checked the features of degree distribution and clustering coefficient exhibited by large random networks. Simulations are presented to illustrate the theoretic results of the model. This model could be applied to describe the network where cooperations are formed to produce a masterpiece, e.g. co-authorship network and movie actors network. 4 - Optimal Signaling of Content Accuracy: Engagement vs. Misinformation Ozan Candogan, University of Chicago, Booth School of Business, Chicago, IL, 27708, United States, Kimon Drakopoulos We study information design in social networks. We consider a setting, where agents’ actions exhibit positive local network externalities. There is uncertainty about the underlying state of the world, which impacts agents’ payoffs. The platform can choose a signaling mechanism that sends informative signals to agents upon realization of this uncertainty, thereby influencing their actions. We investigate how the platform should design its signaling mechanism to achieve a desired outcome. We discuss our findings in the context of increasing engagement or decreasing misinformation in social networks.

n TD22 North Bldg 130

Computational Methods in RMP Sponsored: Revenue Management & Pricing Sponsored Session

Chair: Gustavo J. Vulcano, Universidad Torcuato di Tella, Universidad Torcuato di Tella, Buenos Aires, 1428, Argentina 1 - Demand Model Estimation with Unobserved Stockouts Hongzhang Shao, Georgia Institute of Technology, Atlanta, GA, United States, Anton J. Kleywegt In many settings one wants to estimate discrete choice models, but customers’ choice sets are not known. In one such setting stock levels are observed periodically, and stockouts occur between successive observations, with stockout times not being observed. Therefore different customers had different unobserved choice sets. We developed a choice model estimation method that works with data with both unobserved customer arrivals and unobserved stock-outs. We

354