Data Revision, Reconstruction, and Prediction: An Application to Inventory Investment

A bstract-This paper uses state space methods to examine the extent to which forecast errors in inventory investment, especially at turning points, may be due to measurement error. It is found that while quarterly observations on both inventory and sales are subject to substantial revision, these revisions have on average only a modest impact on the accuracy of inventory investment forecasts. However, there is evidence that the variance of inventory measurement error increases with the rate of change of inventory stocks. This implies that substantial improvements in forecast accuracy, especially at turning points, are potentially available if the preliminary and revised data are used optimally.