Statistical aspects of the evaluation of lot-sizing techniques by the use of simulation

An exploratory study was undertaken in order to determine appropriate procedures for statistically analysing output from a stochastic simulation experiment. Single level lot-sizing was studied in an environment with a fixed planning horizon, zero lead time, no forecast errors, and no initial inventories. Of special interest was the specification of factors such as length of the horizon, average time between orders, coefficient of variation related to demand, and number of observations. To achieve increased accuracy, three variance reduction techniques were investigated.