Informs Annual Meeting 2017

SD54

INFORMS Houston – 2017

SD54

2 - Machine Learning and Portfolio Optimization Gah-Yi Ban, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, gban@london.edu, Noureddine El Karoui, Andrew Lim We adapt two machine learning methods, regularization and cross-validation, for portfolio optimization. We introduce performance-based regularization (PBR), where the idea is to constrain the sample variances of the estimated portfolio risk and return, which steers the solution toward one associated with less estimation error in the performance. We show that the PBR models can be cast as robust optimization problems with novel uncertainty sets and establish asymptotic optimality of PBR solutions. We find PBR dominates all other benchmarks for two out of three Fama-French data sets where parameters are calibrated using a new, performance-based k-fold cross-validation algorithm. 3 - Interpreting Neural Networks Although neural networks can provide highly accurate predictions, they are often considered as opaque “black boxes”. The difficulty of interpreting their predictions often prevents their use in financial practice, where regulators and auditors often insist on model explainability. In this project, we formulate the objectives of interpretability and define it for neural networks. We then consider the interplay between network architecture, identifiability, and interpretability. Considering neural networks as nonparametric models, we show how to use nonparametric variable significance tests to assess the importance of covariates. Tests using mortgage data illustrate our results. Enguerrand Horel, Stanford University, Stanford, CA, United States, ehorel@stanford.edu, Kay Giesecke 362C Non-standard Applications of Location Methods Sponsored: Location Analysis Sponsored Session Chair: Dmitry Krass, University of Toronto, University of Toronto, Toronto, ON, M5S 3E6, Canada, krass@rotman.utoronto.ca 1 - Online Grocery Retail and its Environmental Impact Ekaterina Astashkina, INSEAD, Boulevard de Constance, Fontainebleau, 77305, France, ekaterina.astashkina@insead.edu, Elena Belavina, Simone Marinesi We build a stylized model for traditional and online grocery retail chains to understand the drivers of the consumer and retailer carbon footprint, including emissions that come from food waste and transportation. In our model, consumers make endogenous choices between different channels and the associated food-buying policies, while retailers optimally manage their inventory replenishment. We find that, in most cases, the availability of an online retailer reduces the emissions associated with the grocery sector in a city. We also consider the effectiveness of alternate policy instruments including sales and carbon taxes, and identify actions that improve the behavior of the worst offenders. 2 - Competitive Store Location Models Considering Comparison-shopping Vladimir Marianov, Professor, Pontificia Universidad Catolica de Chile, Department of Electrical Engineering, Vicuna Mackenna 4860, Santiago, 782-0436, Chile, marianov@ing.puc.cl, H.A. Eiselt, Armin Lüer-Villagra Consumers’ behavior strongly influences the location patterns of retail stores. In competitive settings, firms locate their stores to take advantage of this behavior to maximize their market share. A common behavior is comparison-shopping, by which consumers visit more than one store selling products that are mutual substitutes, before making their purchase decision. We include comparison- shopping into competitive facility location models, and compare the obtained results with those of current models. 3 - Optimal Scheduling of Bridge Maintenance using Geometric Brownian Motion Process Tsuyoshi Nobata, Tokyo University of Tokyo, Ce-408, 4-6-1, Komaba, Tokyo, Japan, t-nobata@iis.u-tokyo.ac.jp, Yudai Honma It is well known that there are many studies using Markov chain model in considering uncertainty in bridge asset management. However, there are only a few studies that consider the degradation process of bridges as a continuous model. In addition, considering the scheduling on multiple bridges is very important. In this paper, we propose a degradation model using the geometric Brownian motion and formulate a scheduling strategy for inspection and repair of multiple bridges. SD56

362A Agricultural Analytics Invited: Agricultural Analytics Invited Session Chair: Jack M Kloeber, Kromite, LLC, 82 Nelson Drive, Churchville, PA, 18966, United States, jkloeber@kromite.com Co-Chair: David Krahl, Kromite LLC, 243 N Union Street, Lambertville, NJ, 08530, United States, dkrahl@kromite.com 1 - Optimization of a Supply Chain for the Production of Biofuel from Corn Stover Hernan Chavez, University of Texas at San Antoni, San Antonio, TX, United States, chavezpaurhc@ornl.gov, Krystel Castillo-Villar, Erin Webb Second generation biomass exhibits more variability on its physical properties than first generation. Hence, the quality of feedstock does not always meet the specifications demanded by the conversion processes. This research aims to quantify the cost of imperfect feedstock quality in a biomass-to-biorefinery supply chain (SC) and to develop a model that supports the tactical decisions related to the design and execution of operations that have relevant effects on the physical properties of the biomass. The proposed simulation-based optimization model is based on an extension from the Integrated Biomass Supply and Logistics (IBSAL) simulation model. We present a case study in Ontario, Canada. 2 - Discrete Rate Simulation of Combine Harvester Operations David Krahl, Kromite LLC, 243 N.Union Street, Lambertville, NJ, 08530, United States, dkrahl@kromite.com A combine harvester is a moving factory. As the combine moves through a field harvesting grain, one or more external carts transfer the grain from the harvester’s internal buffer storage to a roadside semi-trailer. An ExtendSim model will be presented that uses discrete-rate simulation for the process of harvesting grain, filling the combine’s internal storage, and transferring the storage to the carts, and discrete-event simulation to represent the availability of carts. The goal of the model is to balance the desired operation of the combine with the number of carts. 3 - Smart Farming: Howthe Internet-of-Things will Spark a Data Revolution in Agriculture Jay Ham, Colorado State University, Ft. Collins, CO, United States, jay.ham@colostate.edu The Internet of Things (IoT) is rapidly migrating onto the farm and food supply chain. New sensor technology coupled with improved connectivity in rural areas (LoRa, cellular gateways) will stream farm data to the cloud at unprecedented rates. This presentation will highlight Ag-IoT projects at Colorado State University and show how innovations in data science are needed to fully profit from the coming revolution in Ag technology. 362B Machine Learning for Finance Sponsored: Financial Services Sponsored Session Chair: Chaithanya Bandi, Northwestern University, 2001 Sheridan Rd, 566, Evanston, IL, 60208, United States, c-bandi@kellogg.northwestern.edu 1 - Machine Learning Based Robust Optimization Models for Limit Order Books to Predict Price Movements Chaithanya Bandi, Northwestern University, 2001 Sheridan Rd, 566, Evanston, IL, 60208, United States, c-bandi@kellogg.northwestern.edu This paper develops a machine learning based robust optimization model to estimate structure of limit order books. The new architecture yields a low- dimensional model of price movements deep into the limit order book, allowing more effective use of information from deep in the limit order book (i.e., many levels beyond the best bid and best ask). Due to its more effective use of information deep in the limit order book, the spatial neural network especially outperforms the standard neural network in the tail of the distribution, which is important for risk management applications.Our data-driven approach offers new benefits for practical applications. SD55

120