Data enrichment for simulation

Abstract

Data enrichment in the present context is concerned with the creation of a stochastic time series whose parameters are consistent with a prior data base and whose cumulative values over a given period satisfy constraints dictated by the prior data base. The specific problem is to generate statistical data for shorter time intervals than the raw data provides. The need for this type of enrichment arises, for example, when historical data gives total demand in each quarter over a multiyear period but the output of a simulation is sensitive to the manner in which demand is distributed over each quarter.

The method utilizes a random fraction to select re cursively a portion of the remaining demand in a quarter. Formulae for the mean and variance of the random fraction are derived. This random fraction can be computer generated with a single random number using hybrid spike-uniform distributions with given means and variances. The properties of such dis tributions are derived, and may have applications in other contexts.