2016 INFORMS Annual Meeting Program

WA44

INFORMS Nashville – 2016

WA41 207C-MCC Machine Learning for Finance Sponsored: Financial Services Sponsored Session

3 - Implementing Optimal Decisions In Business Processes Using 4Ps: Proof Of Concept, Prototyping, Production, Performance Sachin Sumant, Hertz Car Rental, sumantsachin@yahoo.com In the new era of analytics, there is abundance of data, analytical models, visualization tools and integration technologies. Corporations are spending millions of dollars in building analytical infrastructure in the hope of significant ROIs. Large teams are getting formed and everybody is talking about “Next Generation Systems” within analytics team and “Change Management” among business users. This paper discusses how 4Ps can be utilized to unify the analytics team and business users to achieve the optimal benefit from decision making systems and processes by reducing rework, providing well-calibrated solutions, increasing acceptability and guaranteeing positive results. 4 - Deriving Price Elasticity Estimates In The UK Cruise Market John Harvey, Carnival UK, johnandrewharvey@googlemail.com I shall outline the data-driven approach used in deriving a first set of price elasticities for the UK Cruise Market, using purely observations from booking history. By segmenting UK cruises based on their demand and price behaviour, I will show how we approximated elasticity estimates through constructing willingness-to-pay models and including a weighting factor based on the average expected demand in different time intervals; we can estimate elasticities that represent the impact of price at points in the booking curve versus an average assumed weekly demand across the booking horizon. Chair: Chris Spetzler, Decision Education Foundation, DEF, Palo Alto, CA, 00000, United States, chris.spetzler1@gmail.com 1 - Adding Social Impact To Research Efforts And Grants. Ali Abbas, University of Southern California, aliabbas@price.usc.edu 2 - Teaching Decision Skills In College And Career Readiness Frank Koch, Koch Decisions, frank@kochdecision.com During the 2015-2016 school year, Thurston High School in Springfield Oregon offered a College and Career Readiness class to juniors and seniors. The basic principles of decision quality were taught as well as how to write effective essays for college applications and how to plan to improve their college and career decisions during the rest of the school year. We learned that many of the same approaches used in business decision analysis are very effective with teenagers. The approach that has been used at Thurston should be easily adaptable to other schools as well as other organizations where the youth are starting to face significant life decisions. 3 - Recent Advances In Spreading Decision Skills Chris Spetzler, Decision Education Foundation, chris@decisioneducation.org The Decision Education Foundation has been working on spreading decision skills for more than a decade. Numerous opportunities exist for practitioners and academics to contribute and help spread the word. This talk will discuss possible collaboration scenarios. WA44 208B-MCC Environmental and Water Resources Decision Analysis Sponsored: Decision Analysis Sponsored Session Chair: Fengwei Hung, Johns Hopkins University, Baltimore, MD, United States, hfengwe1@jhu.edu Co-Chair: Liang Chen, Johns Hopkins University, Baltimore, MD, United States, chenliang1468@gmail.com 1 - Case Studies In Water Resources Management For Sustainability And Resilience Cate Fox-Lent, US Army Corps of Engineers, The US Army Corps of Engineers has several missions related to Environmental Restoration, Water Quality, and goals for Sustainability and Resilience. Meeting those goals requires life-cycle planning and decision making beyond the usual project time horizon. This presentation will present 3 case studies of the various ways in which formal decision analytics are integrated in to water resources management in a Federal agency context. 696 Virginia Rd, Concord, MA, 01742, United States, Catherine.Fox-Lent@usace.army.mil, Igor Linkov WA43 208A-MCC Spreading Decision Competencies Sponsored: Decision Analysis Sponsored Session

Chair: Justin Sirignano, University of Illinois at Urbana-Champaign, Champaign, Champaign, IL, 61801, United States, jasirign@illinois.edu 1 - Recurrent Neural Networks For Modeling Financial Data Justin Sirignano, University of Illinois at Urbana-Champaign, Champaign, IL, jasirign@illinois.edu We explore using recurrent neural networks for modeling financial time series. Recurrent neural networks depend upon the full history of the time series, allowing for modeling long-term correlations. In out-of-sample tests on financial data, we show recurrent neural networks can outperform standard feedforward neural networks. 2 - Deep Learning For Mortgage Risk Apaar Sadhwani, Stanford University, apaars@gmail.com We analyze mortgage risk at loan and pool levels using an unprecedented data set of over 120 million mortgages originated in United States, which includes the origination data, monthly updates on loan performance, and several time-varying economic variables. We develop, estimate, and test dynamic models for mortgage prepayment, delinquency, and foreclosure that capture loan-to-loan correlation. At heart of our model is a deep neural network trained using GPU-accelerated clusters. We develop several metrics to test model performance, which is a major improvement over existing models and highlights the importance of local factors. This is joint work with Justin Sirginano and Kay Giesecke. 3 - Background Subtraction For Pattern Recognition In High Frequency Financial Data Alex Papanicolaou, Integral Development Corporation, alex.papanic@gmail.com Financial markets produce massive amounts of complex data from multiple agents, and analyzing these data is important for building an understanding of markets, their formation, and the influence of different trading strategies. We apply low-rank plus sparse background subtraction methods to high frequency FX quote data. For prices in a single currency pair from many sources, we model the market as a low-rank structure with an additive sparse component representing transient market making behavior. We show case studies with real market data, showing both in-sample and online results, for how the model reveals pricing reactions that deviate from prevailing patterns. WA42 207D-MCC RM in Practice I Sponsored: Revenue Management & Pricing Sponsored Session Chair: Wei Wang, Pros, Inc, 3100 Main St, Ste 900, Houston, TX, 77002, United States, weiwang@pros.com 1 - Dynamic Pricing And Learning In Airline Revenue Management Ravi Kumar, PROS Inc, rkumar04@pros.com, Wei Wang Many airlines have been actively looking into class-free demand control structures, which requires demand models where price varies over a continuous interval. As evidenced both in literature and in practice one of the big challenges in this setting is the trade-off between policies that learn quickly and those that maximize expected revenue. We investigate applicability of recent advances in the area of optimal control with learning. We examine a demand model where customers maximum WTP is modeled as Gaussian and study approaches that generate sufficient variability in pricing to ensure discovery of the underlying customer behavior while providing appropriate level of expected revenue. 2 - Risk Management In Price And Revenue Optimization Yanqi Xu, Princess Cruises, yanqi6@yahoo.com Price and revenue optimization has been instrumental in delivering profit lift for companies across industries. Typical analytical models in this area involve providing a forecast, and then use price and revenue optimization model to balance supply and demand to extract the maximum profit for company’s assets. In many cases, critical factors in the model are treated as deterministic and their stochastic nature is frequently ignored, a trade-off for simplicity in models. The utility of such solutions may be doubtful at best in situations where the modeled factors have large variances. In this talk, we will discuss models that account for risks in optimization, and show why it can be productive to do so.

377