The Big Data Newsvendor: Practical Insights from Machine Learning

Abstract

In Ban and Rudin’s (2018) “The Big Data Newsvendor: Practical Insights from Machine Learning,” the authors take an innovative machine-learning approach to a classic problem solved by almost every company, every day, for inventory management. By allowing companies to use large amounts of data to predict the correct answers to decisions directly, they avoid intermediate questions, such as “how many customers will we get tomorrow?” and instead can tell the company how much inventory to stock for these customers. This has implications for almost all other decision-making problems considered in operations research, which has traditionally considered data estimation separately from the decision optimization. Their proposed methods are shown to work both analytically and empirically with the latter explored in a hospital nurse staffing example in which the best one-step, feature-based newsvendor algorithm (the kernel-weights optimization method) is shown to beat the best-practice benchmark by 24% in the out-of-sample cost at a fraction of the speed.

Feature-Driven Improvement of Renewable Energy Forecasting and Trading

“…Nonetheless, problem (10)-(11) still requires further elaboration to become a fully data-driven model. Indeed, while in the technical literature on the data-driven newsvendor problem (see, for instance, [5] and [21]), the marginal opportunity costsψ − andψ + are assumed to be known with certainty, in our case, these costs are unknown to the renewable energy producer at the moment of bidding into the day-ahead market.…”

Dynamic Procurement of New Products with Covariate Information: The Residual Tree Method

“…There are now many studies on data-driven inventory management (e.g. Burnetas and Smith 2000, Huh and Rusmevichientong 2009, Kunnumkal and Topaloglu 2008, Godfrey and Powell 2001, Levi et al 2007, Ban and Rudin (2018, Ban 2018 and references therein). A number of recent studies are further motivated by the availability of large data sets that contain covariate data (also referred to as features, characteristics, attributes or explanatory variables), as is this paper.…”

“…autoregressive and martingale models such as the Martingale Model of Forecast Evolution (MMFE) of Heath andJackson 1994 andGraves et al 1986). We refer the readers to Ban and Rudin (2018) for details on how smooth nonlinear models can be approximated by linear ones through Taylor expansions and enlargement of the covariate dimension, and on how time-series models and MMFE models can be captured by the linear demand model.…”

Data Analytics in Operations Management: A Review

“…Using data from Car2Go, the paper applies the approach to designing the service region in San Diego, and shows how the approach leads to higher revenues than several simpler, managerially-motivated heuristic approaches to service region design. Ban and Rudin (2019) consider a data-driven approach to inventory management. The context that the paper studies is when one has observations of demand, together with features that may be predictive of demand, such as weather forecasts or economic indicators like the consumer price index.…”

“…To make inventory decisions in this setting, one might consider building a demand distribution that is feature-dependent, and then finding the optimal order quantity for the distribution corresponding to a given realization of the features. Instead, the paper of Ban and Rudin (2019) proposes two alternate approaches. The first, based on empirical risk minimization, involves finding the order quantity by solving a single problem, where the decision variables is the decision rule that maps the features to an order quantity, and the objective is to minimize a sample-based estimate of the cost.…”