Optimal Transport and Applications to Economics, Statistics and Finance

Abstract

These lectures, based on my recent book Optimal Transport Methods in Economics (Princeton, 2016), will provide an introduction to the theory of optimal transport, with a focus on applications to economics, statistics and finance. The basic results in Optimal Transportation will be covered, as well as its relations to linear programming, network flow problems, convex analysis, and computational geometry. Applications to economics (labor markets), statistics (quantile methods) and finance (martingale optimal transport) will be given.


Lecture 1. MONGE-KANTOROVICH DUALITY; THE OPTIMAL ASSIGNEMENT PROBLEM. APPLICATION TO THE LABOR MARKET
(Tuesday March 6, 10:00 to 12:00, Seminar Room, Ground Floor, Weeks Building Google Map here)

In order to introduce the Monge-Kantorovich theorem of optimal transport theory, we consider a stylized assignment problem. Assume that a central planner (say, a plant manager) needs to assign workers to machines in order to maximize total output. Workers vary by their individual characteristics, and machines come in various sorts, where the set of characteristics of workers and firms may be either discrete or continuous. We shall assume that there is the same number of workers and machines, which is normalized to one; hence the distribution of workers and machines are described by probability distributions over workers and firms types respectively. The output of a worker assigned to a machine depends on both the worker's and the machine's characteristics, so some workers may be better with some machines, and worse with some others. The central planner's problem, which is the optimal transport problem, consists of assigning workers to machines in a way such that the total output is maximized. It will predict the equilibrium wages and the assignment of workers to machine. This model can be used to estimate the structural parameters of the matching market, i.e. workers' productivity and job amenity.


Lecture 2. POSITIVE ASSORTIVE MATCHING. APPLICATION: MULTIVARIATE QUANTILES AND RISK MEASURES
(Tuesday March 13, 10:00 to 12:00, Seminar Room, Ground Floor, Weeks Building Google Map here)

In dimension one, the quantile is the inverse of the cumulative distribution function. Quantiles are extremely useful objects because they fully characterize the distribution of an outcome, and they allow to provide directly a number of statistics of interest such as the median, the extremes, the deciles, etc. They allow to express maximal dependence between two random variables (comonotonicity) and are used in decision theory (rank-dependent expected utility); in finance (value-at-risk and Tail VaR); in microeconomic theory (efficient risksharing); in macroeconomics (inequality); in biometric (growth charts) among other disciplines. Quantile regression, pioneered by Roger Koenker, allows to model the dependence of a outcome with respect to a set of explanatory variables in a very flexible way, and has become extremely popular in econometrics. The classical definition of quantiles based on the cumulative distribution function, however, does not lend itself well to a multivariate extension, and given the number of applications of the notion of quantiles, many authors have suggested various proposals.

We have proposed a novel definition of multivariate quantiles called “Vector quantiles” based on optimal transport. The idea is that instead of viewing the quantile map as the inverse of the cumulative distribution function, it is more fruitful to view it as the map that rescales a distribution of interest to the uniform distribution over [0,1] in the least possible distortive way, in the sense that the average squared distance between an outcome and its preimage by the map should be minimized. While equivalent to the classical definition in the univariate case, this definition lends itself to a natural multivariate generalization using Monge-Kantorovich theory. We thus define the multivariate quantile map as the one that attains the L2-Wasserstein distance, which is known in the optimal transport literature as Brenier’s map. With Carlier and Santambrogio, we gave a precise connection between vector quantiles and a celebrated earlier proposal by Rosenblatt. We have shown that many of the desirable properties and uses of univariate quantiles extend to the multivariate case if using our definition. With Ekeland and Henry, we have shown that this notion is a multivariate analog of the notion of Tail VaR used in finance. With Henry, we have shown that it allows to construct an extension of Yaari’s rank-dependent utility in decision theory, and have provided an axiomatization for it. With Carlier and Dana, we have shown that this notion extends Landsberger and Meilijson’s celebrated characterization of efficient risksharing arrangements. With Charpentier and Henry, we have shown the connection with Machina’s theory of local utility. With Carlier and Chernozhukov, we have shown that Koenker’s quantile regression can be naturally extended to the case when the dependent variable is multivariate if one adopts our definition of multivariate quantile. With Chernozhukov, Hallin and Henry we have defined empirical vector quantiles and have studied their consistency.


Lecture 3. DYNAMIC OPTIMAL TRANSPORT AND MARTINGALE OPTIMAL TRANSPORT. APPLICATION: BOUNDS ON DERIVATIVE PRICES
(Thursday March 15, 10:00 to 12:00, Seminar Room, Ground Floor, Weeks Building Google Map here)

Optimal transport theory has important applications in finance, more specifically in option pricing theory. Financial derivatives may depend on several underlying assets; this is the case of spread options, for instance, or of basket options. The standard Black-Scholes-Merton theory of option pricing says that if there is a liquid market of vanilla options on a single underlying, then the risk-neutral distribution of the underlying can be recovered from the option prices; and we can therefore obtain a unique price associated with any more complicated single-underlying option. However, in the case of an option on two underlying, the market prices on the single-name options do not imply the joint distribution of two such assets, and one can then define no-arbitrage bounds, which corresponds to the cheapest and most expensive prices of the option that is consistent with the market. These bounds formulate as a Monge-Kantorovich problem, and the dual problem ensures that they correspond to the most expensive sub-replicating (lower bound) and the cheapest super-replicating portfolio (upper bound).

In a number of cases, the two underlying quantities are not the value of two assets at the same date in time, but the price of the same asset at two different dates in the future. There is then an important further restriction on the joint distribution of these assets: they should be the margins of a martingale. Computing the bounds of the option prices leads then to a variant of Monge-Kantorovich theory, where one looks the optimal coupling that is a martingale. This further constraint yields a supplementary term in the dual formulation, which has an interesting interpretation in terms of sub/super-replicating portfolio: the portfolio is not only made of calls and puts at the two maturities (static hedging), but also allows for rebalancing at the earlier maturity, allowing for dynamic hedging. Moving beyond the static problem, there are interesting dynamic formulation of the problem. In particular, one may consider among the set of semi-martingales that start at a given distribution and end up at a given distribution, those who minimize at the time integral of the expectation of a Lagrangian that depends on the drift and diffusion parameters. This nicely extends the Benamou-Brenier dynamic formulation of optimal transport, and can provide interesting insights on particular solutions to the Skorohod embedding problem.

Contact us

CFM-Imperial Institute of Quantitative Finance
Department of Mathematics,
Imperial College
London
SW7 1NE

Email: iqf-events@imperial.ac.uk