Georgia Institute of Technology College of Sciences

Demand forecasting plays an important role in many inventory control problems. To mitigate the potential harms of model misspecification, various forms of distributionally robust optimization have been applied. Although many of these methodologies suffer from the problem of time-inconsistency, the work of Klabjan et al. established a general time-consistent framework for such problems by connecting to the literature on robust Markov decision processes. Motivated by the fact that many forecasting models exhibit very special structure, as well as a desire to understand the impact of positing different dependency structures, in this talk we formulate and solve a time-consistent distributionally robust multi-stage newsvendor model which naturally unifies and robustifies several inventory models with demand forecasting. In particular, many simple models of demand forecasting have the feature that demand evolves as a martingale (i.e. expected demand tomorrow equals realized demand today). We consider a robust variant of such models, in which the sequence of future demands may be any martingale with given mean and support. Under such a model, past realizations of demand are naturally incorporated into the structure of the uncertainty set going forwards. We explicitly compute the minimax optimal policy (and worst-case distribution) in closed form, by combining ideas from convex analysis, probability, and dynamic programming. We prove that at optimality the worst-case demand distribution corresponds to the setting in which inventory may become obsolete at a random time, a scenario of practical interest. To gain further insight, we prove weak convergence (as the time horizon grows large) to a simple and intuitive process. We also compare to the analogous setting in which demand is independent across periods (analyzed previously by Shapiro), and identify interesting differences between these models, in the spirit of the price of correlations studied by Agrawal et al. This is joint work with Linwei Xin, and the paper is available on arxiv at https://arxiv.org/abs/1511.09437v1

The long term behavior of dynamical systems can be understood by studying invariant manifolds that act as landmarks that anchor the orbits. It is important to understand which invariant manifolds persist under modifications of the system. A deep mathematical theory, developed in the 70's shows that invariant manifolds which persist under changes are those that have sharp expansion (in the future or in the past) in the the normal directions. A deep question is what happens in the boundary of these theorems of persistence. This question requires to understand the interplay between the geometric properties and the functional analysis of the functional equations involved.In this talk we present several mechanisms in which properties of normal hyperbolicity degenerate, so leading to the breakdown of the invariant manifold. Numerical studies lead to surprising conjectures relating the breakdown to phenomena in phase transitions. The results have been obtained combining numerical exploration and rigorous reasoning.

On the two-dimensional square grid, remove each nearest-neighbor edge independently with probability 1/2 and consider the graph induced by the remaining edges. What is the structure of its connected components? It is a famous theorem of Kesten that 1/2 is the ``critical value.'' In other words, if we remove edges with probability p \in [0,1], then for p < 1/2, there is an infinite component remaining, and for p > 1/2, there is no infinite component remaining. We will describe some of the differences in these phases in terms of crossings of large boxes: for p < 1/2, there are relatively straight crossings of large boxes, for p = 1/2, there are crossings, but they are very circuitous, and for p > 1/2, there are no crossings.