Measuring Noise in Inventory Models

This paper has two purposes. One is to assess different models of inventory behavior in terms of their ability to well approximate the realized data on inventories. We do this initially for the pure production smoothing model and then for a sequence of generalizations of the model. Our analysis both performs specification tests as well as measures the deviations of the data from each null model, which we refer to as model noise. This involves the introduction of a noise ratio which provides a metric for measuring the magnitude of the noise component of the data. A second purpose is to explore whether observed cost shocks, including in particular carefully measured series on raw materials prices, can be helpful in explaining inventory movements. We find that the basic production level smoothing model of inventories, augmented by buffer stock motives, observed cost shocks, properly measured, and to a lesser extent stockout avoidance motives, appears to well approximate monthly inventory data.