2016 INFORMS Annual Meeting Program

SB44

INFORMS Nashville – 2016

SB44 208B-MCC Decision Analysis, Game Theory, and Homeland Security Sponsored: Decision Analysis Sponsored Session” Chair: Jun Zhuang, University at Buffalo, SUNY, Buffalo, NY, United States, jzhuang@buffalo.edu Co-Chair: Jing Zhang, University at Buffalo, SUNY, University at Buffalo, SUNY, Buffalo, NY, 14228, United States, jzhang42@buffalo.edu 1 - Defensibility - A New Concept In Risk Analysis Vicki Bier, University of Wisconsin-Madison, vicki.bier@wisc.edu Alexander Gutfraind, Ziyang Lu We define a system as defensible if modest investment of resources can significantly improve the outcome to the defender. After quantifying defensibility, we use empirical examples and stylized examples to show that some systems that appear highly vulnerable are actually highly defensible. 2 - Using The Concept Of Multidimensional System Resilience In Decision And Risk Analysis Resilience is generally understood as the ability of an entity to recover from an external disruptive event. Systems such as cities, face the challenge of each of their subsystems being vulnerable to multiple threats. This work analyzes the compilation of subsystem and multiple measurements in order to have more accurate description of system resilience. The object of this work is to introduce the use of this multidimensional system resilience model in the disciplines of decision and risk analysis, showing how it allows creating more comprehensive and intuitive tools for decision makers. 3 - Behavioral Experiments On Deterrence Richard John, USC, richardj@usc.edu When evaluating potential countermeasures, emphasis is often placed on interdiction over deterrence because the impact of interdiction focused countermeasures are easier to identify and quantify compared to the impact of countermeasures designed to deter. Resource allocation decision often focus on measures of interdiction enhancement only, even though the involve countermeasures are expected to improve both interdiction and deterrence. I will focus on innovative methods to characterize and quantify the deterrent effects of countermeasures. I will also include methods and findings drawn from decision and risk analysis, game theory, and behavioral research on deterrence. 4 - A Robust Optimization Approach For Electric Power Grid Protection Alberto Costa, NUS, Singapore, Singapore, isealc@nus.edu.sg Alberto Costa, Future Resilient Systems (FRS) - ETH Zurich, Singapore, Singapore, isealc@nus.edu.sg We study the problem of the optimal allocation of protection resources in an electric power grid with the aim of maximizing its robustness against attacks to the lines. This problem can be seen as a game between two players, i.e., the system operator and the attacker. Given a budget for protecting the lines and a performance threshold, i.e., the maximum value of load shed tolerated by the system operator, the attacker wins the game if the load shed after the attack is above the threshold. We propose an algorithm to find the allocation of the system operator’s budget to the lines of the grid which maximizes the amount of budget needed by the attacker to win the game. Dante Gama Dessavre, Stevens Institute of Technology, dgamades@stevens.edu, Jose Emmanuel Ramirez-Marquez

missing which impairs regulatory ability to determine whether a financial institution provided fair access to its mortgage products. We use a multinomial logit with spatial data analysis coupled with a multiple imputation methodology to infer the missing HMDA data and mitigate the effect of model uncertainty. Our empirical analysis concerns varied institutions with different levels of missing protected class data including a large bank and a non-bank lender. 2 - Prudential Policies And Their Impact On Credit In The United States Paul Calem, FRB of Philadelphia, paul.calem@phil.frb.org We analyze impacts on bank lending of two supervisory policies. We find that banks reduced their share of jumbo mortgage originations following the stress test in 2011, but not in later years when they were better capitalized. We find little initial impact of the 2013 Leveraged Lending Guidance, but follow-up FAQs issued late in 2014 marked a significant drop in leveraged lending. Thus, measureable risk and capital appear to have a more immediate impact on lending. Model governance can still have compliance implications—exemplified by banks failing the stress tests on qualitative grounds. Our findings for the 2013 Guidance and FAQ suggest that clarity of regulatory communications also play a role. 3 - Forecast Combination Of Machine Learning Models With Application To Camels Early-warning Systems Lewis Gaul, Office of the Comptroller of the Currency, lewis.gaul@occ.treas.gov This paper uses forecast combination methods to predict future CAMELS bank ratings assigned by the Office of the Comptroller of the Currency. We use several individual algorithms and statistical models to forecast future CAMELS ratings with information on lagged financial statement ratios and macroeconomic variables. We then analyze whether combinations of multiple forecasts provide more accurate out-of-sample forecasts of future CAMELS ratings than any individual forecast model. Results indicate that the out-of-sample forecast performance of most individual models varies over time, and that combinations of forecasts generally perform better than any individual model. SB46 209B-MCC Sharing Economy, Mechanism Design and Networks I Sponsored: Revenue Management & Pricing Sponsored Session Chair: Santiago Balseiro, Duke University, Durham, NC, United States, sbalseiro@gmail.com Co-Chair: Ozan Candogan, University of Chicago, Durham, NC, Nicholas A. Arnosti, Stanford University, narnosti@stanford.edu We consider a model in which sellers compete by posting prices and buyers visit sellers sequentially. We show that there is a unique equilibrium outcome, which is constrained efficient. We then study the consequences of reducing search costs. This benefits buyers, but may either increase or decrease seller revenue. If there are sufficiently many buyers, sellers benefit from lower search costs. Otherwise, the effect on seller revenue depends on the shape of the distribution of buyer values. If it is heavy- tailed (has a decreasing hazard rate), then sellers benefit from lower search costs. If it is light-tailed (has an increasing hazard rate), then seller revenue falls as search becomes easier. 2 - Dynamic Mechanisms With Martingale Utilities Santiago Balseiro, Duke University, srb43@duke.edu, Vahab Mirrokni, Renato Paes Leme We study the dynamic mechanism design problem of a seller who repeatedly sells independent items to a buyer with private values under two practically relevant business constraints: (i) a periodic individual rationality constraint, which limits the mechanism to charge at most the buyer’s value in each period and (ii) a martingale utility constraint, which imposes that from the perspective of the buyer, the next item’s expected utility is equal to the present one’s. Our main contribution is the design of a dynamic auction that asymptotically achieves full extraction of buyer surplus as agents become more patient. 3 - Ridesharing Networks United States, ozan.candogan@chicagobooth.edu 1 - Matching Markets With Search Frictions

SB45 209A-MCC Model Uncertainty, Risk, & Compliance Invited: Risk and Compliance Invited Session

Chair: Ricky Rambharat, Lead Statistician, Office of the Comptroller of the Currency, 400 7th SW, Mail-stop 6E-2, Washington, DC, 20219, United States, ricky.rambharat@occ.treas.gov 1 - Missing Data Inference With Application To The Home Mortgage Disclosure Act Andrew Porter, Office of the Comptroller of the Currency, Washington, DC, United States, andrew.porter@occ.treas.gov Tong-yob Nam The Home Mortgage Disclosure Act (HMDA) mandates financial institutions to report protected class information such as race and ethnicity for each mortgage applicant when available. However, a significant proportion of these data is

Ozan Candogan, University of Chicago, 7449 9th Street, Unit 472, Durham, NC, 27705-1084, United States, ozan.candogan@chicagobooth.edu, Daniela Saban, Konstantinos Bimpikis

We consider the problem faced by a revenue optimizing ride-sharing platform, which must decide on how to price the rides as well as how to compensate the drivers. These decisions will impact both the entry of customers and the actions of the drivers. We study the impact that the underlying network structure has on the pricing strategy.

58