B0232
Title: Rough-probabilistic modelling for demand specification
Authors: Abhirup Banerjee - University of Oxford (United Kingdom) [presenting]
Sujay Mukhoti - Indian Institute of Management Indore (India)
Abstract: A longstanding problem in demand analysis is to identify an appropriate demand distribution from qualitative feedback obtained from field surveys. A typical survey from vendors would indicate a flat-top density curve tri-partitioned into a positive region consisting of the most likely set, the boundary set with the possibility of belonging to the class, and a negative region having the least likelihood of occurrence. Such flat-top density curves imply that multiple values are equally most likely to occur and hence, are the modes. However, the most popular probability models used in demand analysis are all unimodal, presenting a single point in the positive region with maximum likelihood. A new class of probability distributions is proposed, called the stomped family of distributions, that provides better model fitting for the flat-top demand densities. The statistical properties of a special stomped distribution are discussed, called the stomped normal distribution, as well as investigate its parameter estimation.
