What is new in inventory models (part 3)?
Posted by Sridhar Tayur on Thu, Feb 03, 2011 @ 08:51 AM
A central problem in setting inventory targets is that historical data, even when available, may be limited. How to account for this? In a very charming paper published in Management Science, Bob Hayes (in 1969!) suggested the use of Expected Total Operating Cost (ETOC) framework to incorporate the "parameter uncertainty" in the distribution of demand within the news-vendor setting. However, he had to limit himself to a couple of distributions, such as Normal or Exponential. A larger conceptual problem, in my opinion, is how would do you -- the inventory manager -- know what the distribution is! Thus, this work, although cute, has been essentially useless for practice.
It turns out that the ETOC framework is worth keeping. To create something useful, we (Professor Bahar Biller and graduate student Alp Akcay, and myself) decided to imbed a flexible family of distributions, called the Johnson Translation System (JTS), that can fit any possible first four moments to any available data set, into the ETOC framework. With this approach, the need for the inventory manager to guess at a demand distribution (or arbitrarily choose one) is eliminated. Our experiments, several with actual data from SmartOps customers, show that we can improve the inventory productivity by over 15%, while guaranteeing the desired service levels with even greater confidence. One more practical hurdle ---- limited historical data -- has been overcome. That is the value of academic research. This work is forthcoming in Manufacturing & Services Operations Management Journal, the flagship journal of our research community.
PS: To honor Bob, and because we liked how it sounds, we decided to call this new inventory policy: HIP (Hayes Inventory Policy), although it was not originally suggested by him.