Informs 2015

270 / 552

INFORMS Philadelphia – 2015
268
2 - Optimization of Resource use in Massively Multiplayer
Online Games
Betty Love, University of Nebraska at Omaha, UNO Mathematics
Dept., 60th & Dodge Sts., Omaha, NE, 68182, United States of
America, Andrew Cockerill
With over 400 million players worldwide, massively multiplayer online games
(MMOs) continue to be a popular source of online recreation. MMOs frequently
involve resource management and virtual economies. This project demonstrates
the introduction of optimization strategies in the MMO game World of Warcraft.
A simulated annealing algorithm was implemented in a Lua script which runs in
the game’s user interface and determines how to use the player’s current
resources to maximize virtual profit.
3 - A Gravity Model for Tourist Forecasting at FIFA Soccer
World Cups
Ghaith Rabadi, Associate Professor, Old Dominion University,
2102 Eng Systems Build, Dep. of Eng.Mngt. and Systems Eng.,
Norfolk, VA, United States of America, ,
Mohammed Al-salem, Ahmed Ghoniem
FIFA Soccer World Cups are sport mega-events that enjoy tremendous popularity
worldwide. This paper analyzes historical bilateral tourist flows over the last two
decades to forecast the number of inbound tourists into future World Cup host
countries. Hosting sport mega-events will be considered as one of the input
factors to measure their impact on the number of tourists forecasted.
4 - Optimal Hiking: Bi-modal Variation of the Traveling
Salesperson Problem
Roger Grinde, Associate Professor, University of New Hampshire,
Paul College of Business & Economics, 10 Garrison Avenue,
Durham, NH, 03824, United States of America,
The problem addressed is motivated by a mountaneering problem where there is
a network a peaks (destinations) connected by trails and a network of parking
areas connected by roads. Various objectives are possible; generally one wishes to
construct a series of hikes that together visit all the destinations. A formulation
and solution approach is presented.
5 - Analysis of Potential Solutions to Competitive Imbalance
in the NBA
Stephen Hill, Assistant Professor, UNC Wilmington,
601 South College Road, Wilmington, NC, 28403-5611,
United States of America,
The National Basketball Association (NBA) is in the midst of an extended period
of competitive imbalance with teams in the Western Conference widely viewed as
being stronger than those in the Eastern Conference. In this work, we evaluate a
set of possible changes to the structure of the NBA. Each of these changes is
analyzed via Monte Carlo simulation with the impacts on competitive balance
and playoff participation described.
TA31
31-Room 408, Marriott
Financial Applications of Data Mining and Machine
Learning Techniques
Sponsor: Data Mining
Sponsored Session
Chair: John Guerard, Director Of Quantitative Research, McKinley
Capital Management, LLC, 3301 C Street, Suite 500, Anchorage, AK,
99503, United States of America,
1 - Optimal Global Efficient Portfolio with Emerging Markets using
Earning Forecasts
Shijie Deng, Georgia Inst of Tech, 755 Ferst Dr, Atlanta, GA,
United States of America,
We apply a multi-factor stock selection model which includes earning forecast to
analyze the performance of the optimal global portfolio which includes the
emerging markets. Under the Markowitz mean-variance framework, applied
optimization techniques are employed to address the practical issues of risk-
tolerance, turn-over, and tracking-error. The impacts of these practical constraints
on the portfolio performance are analyzed through extensive numerical
experiments.
2 - Data Mining Corrections Testing
John Guerard, Director of Quantitative Research,
McKinley Capital Management, LLC, 3301 C Street, Suite 500,
Anchorage, AK, 99503, United States of America,
Harry Markowitz, Ganlin Xu
Data mining corrections (DMC) tests of Global, Russell 3000, Non-U.S. stocks,
Emerging Markets, Japan-only, and China-only during the 2000-2014 period for
21 individual financial variables and two composite (robust, PCA-based)
regression models. We find that earnings forecasting models and regression-based
models emphasizing forecasted earnings acceleration and price momentum
models dominate the DMC tests which allow us to statistically dismiss Data
Mining as a potential source of modeling bias.
3 - Applications of Machine Learning over Alpha Signals to Improve
Stock Selection and Boost Returns
Abhishek Saxena, Quantitative Research Analyst, McKinley
Capital Management, LLC, Suite 500, 3301 C Street,
Anchorage, AK, 99503, United States of America,
, Sundaram Chettiappan
The paper explores the possibility of enhancing an alpha model through various
machine learning techniques. We show that these techniques can have
statistically significant additions to both raw returns and simulated returns in
various equity universes. These excess returns are mostly attributed to improved
stock selection as the risk profile doesn’t change significantly in terms of both
direct risk measurements (standard deviation based risk models) and exposures to
various fundamental factors.
4 - The Rise of the Machines: Machine Learning in Stock Selection
Rochester Cahan,
Models that attempt to forecast the cross-section of future stock returns are often
structured as linear multifactor models. In this research we study the efficacy of
non-linear modeling techniques in stock selection strategies. We use a range of
factors known to predict stock returns as raw ingredients and investigate whether
various non-linear and machine learning algorithms can combine those
ingredients into predictive alpha signals, using only information known ex ante.
We benchmark the predictive power of the non-linear models against traditional
linear regression models constructed using the same data and estimation
windows.
TA32
32-Room 409, Marriott
Principles in Applied Probability
Sponsor: Applied Probability
Sponsored Session
Chair: Josh Reed, Associate Professor, NYU, 44 W. 4th St., New York,
NY, 10012, United States of America,
1 - Relating Busy Period Duration and the Single Big Jump Principle
in Heavy Traffic
Bart Kamphorst, PhD Student, CWI, Science Park 123,
Amsterdam, 1098 XG, Netherlands, ,
Bert Zwart
Queueing literature shows many results for the M/G/1 queue with a fixed server
utilization. However, in practice the server utilization may be increasing due to a
growing number of jobs per time unit. This causes a significant increase in waiting
times and the busy period duration. I will present asymptotic relations for the tail
probabilities of the former characteristics. Moreover, I will illustrate a typical long
busy period and discuss its relation with the Principle of a Single Big Jump.
2 - Capacity Allocation in a Transient Queue
Britt Mathijsen, PhD Student, Eindhoven University of
Technology, P.O. Box 513, 5600 MB, Eindhoven, Netherlands,
, Bert Zwart
We consider an optimal capacity allocation problem of a two-period queueing
model, being in steady-state in the first time interval, but changing parameters at
the instance of the new period. The error in the objective function made by
disregarding the transient phase before reaching stationarity in this second
interval is quantified and approximated. Furthermore, we analyze the
consequence of staffing the system according to its steady-state behavior and
propose a corrected staffing rule.
3 - Analysis of Cascading Failures
Fiona Sloothaak, PhD Student, Eindhoven University of
Technology, P.O. Box 513, 5600 MB, Eindhoven, Netherlands,
, Bert Zwart
Inspired by analyzing the reliability of energy networks, particularly the
occurrence of large blackouts, we consider a stylized model of cascading failures.
By using connections with extreme value theory and Brownian bridge
approximations, we establish that the number of failed nodes follow a power law.
Time permitting, we also discuss connections with similar models and questions
from material science.
TA31