A Data-Driven Approach to Newsvendor Problems

We propose an approach to the classical newsvendor problem and its extensions subject to uncertain demand that: (a) works directly with data, i.e., combines historical data and optimization in a single framework, (b) yields robust solutions and incorporates risk preferences using one scalar parameter, rather than utility functions, (c) allows for tractable formulations, specifically, linear programming problems, and (d) leads to closed-form solutions based on the ranking of the historical demands, which provide key insights into the role of the cost parameters. Numerical results are very encouraging.


Technical Report, MIT, Cambridge, MA, May 2005. Also appears in Thiele's PhD Thesis, "A robust optimization approach to supply chains and revenue management", MIT, Cambridge, MA, May 2004. Revised version coming soon.



View A Data-Driven Approach to Newsvendor Problems