Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SA23

2 - A Probabilistic Modeling Framework to Optimize Volume Pricing Settings Jose Luis P. Arreola, Home Depot (Blacklocus), Austin, TX, United States, Joseph Nipko We optimize volume pricing settings in terms of tier quantity (Qi) and discount (di) over retail price in order to perform constrained optimization on performance metrics such as Margin, Revenue and Unit Sellthrough. Our inference approach is based on probabilistic modeling of the order size distribution of similar products, under an observed array of (Qi, di) settings. Further, our methodology allows for the overlaying of product-specific context parameters in a modular way. We apply our approach to three different groups of construction material products and we conduct market price tests on them. We observe that our framework predicts directionally correct revenue and margin lift estimates. 3 - DP Based Efficient Frontier in Revenue Management Wei Wang, PROS Inc, 3100 Main St, Suite #900, Houston, TX, 77002, United States, Fangzhou Sun, Darius Walczak, Subhash C. Sarin In this paper, we consider jointly optimizing expected profit contribution and resource utilization for revenue management problems. In particular we use a constrained Markov decision process-based approach to find the corresponding efficient frontier. 4 - Predicting Users’ Forecast Influences via Machine Learning Ang Li, PROS, Inc., 3100 Main Street, Suite #900, Houston, TX, 77002, United States, Stephanie Zipkin Airline RM analysts spend a significant amount of time monitoring and influencing forecasted demand. In this innovation study, we developed several supervised learning models to predict demand influences based on historical data. In particular, we derived features which mimic a variety of influence criteria and rules the analysts apply through surveying domain experts. Our numerical experiments show that a medium-complexity decision tree model reliably predicts user influences with low error rate. We discuss how the RM system may be augmented with our model, and outline our next steps of study. Sponsored: Finance Sponsored Session Chair: Justin Sirignano, University of Illinois at Urbana-Champaign, Irvine, CA, 92617, United States Co-Chair: Alexandra Chronopoulou, University of Illinois, Urbana-Champaign, Urbana, IL 1 - Optimal Kernel Estimation of Spot Volatility of Stochastic Differential Equations Jose E. Figueroa-Lopez, Washington University-St Louis, One Brookings Drive, St Louis, MO, 63130, United States A feasible method of bandwidth and kernel selection for spot volatility kernel estimators is proposed, under some mild conditions on the volatility process, which not only cover classical Brownian motion driven dynamics but also some processes driven by long-memory fractional Brownian motions. The optimal selection of the kernel function is also investigated. For Brownian Motion type volatilities, the optimal kernel turns out to be an exponential function, while, for fractional Brownian motion type volatilities, numerical results to compute the optimal kernel are devised. Simulation studies further confirm the good performance of the proposed methods. This is based on joint work with Cheng Li. 2 - Delta-hedging in Fractional Volatility Models Alexandra Chronopoulou, University of Illinois, Urbana- Champaign, 104 South Mathews Avenue, 117 Transportation Building, Urbana, IL, 61801, United States, Qi Zhao In this talk, we propose a delta-hedging strategy for a long memory stochastic volatility model (LMSV). This is a model in which the volatility is driven by a fractional Ornstein-Uhlenbeck process with long-memory parameter H. We compute the so-called hedging bias, i.e. the difference between the Black-Scholes delta and the LMSV delta as a function of H, and we determine when a European-type option is over-hedged or under-hedged. Finally, we apply our approach to SP 500 data. 3 - Mean Field Analysis of Neural Networks in Machine Learning Alexandra Chronopoulou, University of Illinois, Urbana- Champaign, 104 South Mathews Avenue, 117 Transportation Building, Urbana, IL, 61801, United States, Justin Sirignano, University of Illinois at Urbana-Champaign, Irvine, CA, USA. Neural network models in machine learning have revolutionized fields such as image, text, and speech recognition. There’s also growing interest in using neural n SA23 North Bldg 131A Joint Session FSS/Practice Curated: Financial Engineering Applications

networks for applications in science, engineering, medicine, and finance. Despite their immense success in practice, there is limited mathematical understanding of neural networks. We mathematically study neural networks in the asymptotic regime of simultaneously (A) large network sizes and (B) large numbers of stochastic gradient descent training iterations. We rigorously prove that the neural network satisfies a Law of Large Numbers (LLN) and a Central Limit Theorem (CLT). The LLN is the solution of a nonlinear partial differential equation while the CLT satisfies a stochastic partial differential equation. 4 - Deep Learning in Asset Pricing We estimate a general non-linear asset pricing model and optimal portfolios with a deep-neural network applied to all U.S. equity data combined with all relevant macroeconomic and firm-specific information. No-Arbitrage Pricing theory implies that the conditional expectation of the discounted excess returns are zero under the risk-neutral probability. Our data-driven approach estimates the stochastic discount factor that explains all asset prices from this basic assumption. Our model allows us to identify the key factors for asset prices, find mis-pricing of stocks and generate the optimal portfolio with the highest Sharpe ratio. n SA24 North Bldg 131B E-Business Sponsored: EBusiness Sponsored Session Chair: Gulver Karamemis, University of Rhode Island, Kingston, RI, 02881, United States 1 - Beauty Contest and Social Value of Fintech: An Economic Analysis With the advance of Fintech, traders in financial markets use the information available on social media to gauge investor sentiment and form higher order beliefs. Following the insight of Keynes (1936) on beauty contest, we develop an analytical model to analyze how higher order beliefs, driven by the Fintech revolution, affect market efficiency. We find that higher order beliefs tend to reduce market efficiency because public information is over weighted. Since accounting disclosure is a main source of public information, our results highlight that the optimal level of accounting disclosure (transparency) can be dramatically Markus Pelger, Stanford University, 312 Huang Engineering Center, 475 Via Ortega, Stanford, CA, 94305, United States Anurag Garg, University of Florida, Gainesville, FL, 32603, United States, Liangfei Qiu, Subhajyoti Bandyopadhyay Hong Guo, University of Notre Dame, 356 Mendoza College Of Business, University of Notre Dame, Notre Dame, IN, 46556-5646, United States, Chao Ding, Xuying Zhao, Jing-Sheng Jeannette Song In this paper, we investigate the clustering phenomenon of heterogeneous retailers at heterogeneous locations. Specifically, we develop a game theoretical model to analyze mainstream and niche retailers’ incentives to locate close to each other and their choices between central and peripheral locations. 3 - An Empirical Investigation of Online Crowd-funding Performance Lina Zhou, University of North Carolina-Charlotte, Charlotte, NC, USA, Kexin Zhao, Anamika Paul, Xia Zhao Online crowdfunding is an emerging platform that facilitates raising money for a for-profit or non-profit project from a large number of people. The determinants of online crowdfunding performance may depend on the type of a project. We conducted an empirical investigation of one type of crowdfunding project to gain insights into its performance. 4 - Search Engine Advertising; Optimal Contractual Strategies Between Firms and Their Affiliates Siddharth Bhattacharya, Fox School of Business,Temple University, Alter Hall, 1801 Liacouras Walk, Philadelphia, PA, 19122, United States, Subodha Kumar, Sunil Wattal Firms increasingly utilize third party affiliates to advertise on their behalf. The focus of our research is to find what optimal pricing and advertising strategies between firms and affiliates maximize profits and how does product quality, customer heterogeneity, affiliate type and type of ad contract (with Google) affect those strategies. affected by the use of Fintech in financial trading. 2 - Clustering of Heterogeneous Retailers at Heterogeneous Locations

10