Price Index Estimation Using Price Imputation for Unsold Items

Although scanner price quotes and expenditures have great promise in improving the Consumer Price Index (CPI), their use has introduced new problems. One major problem confronting the Bureau of Labor Statistics (BLS) is the treatment of an item that experiences no purchases during a particular time period. Either a price can be explicitly imputed for this item or it can be implicitly imputed by ignoring it in the construction of the price index. This paper examines and compares different imputation methods. Indexes are then constructed using these various imputation methods using price and expenditure data from scanner sales of cereal in the New York area. In scanner databases, when a particular item in a particular outlet does not sell in a certain time period, it is not possible to determine if this nonsale was the result of no inventories or of no consumer demand for the product, perhaps because the price was too high. Currently, under the BLS manual sampling system, if an item is still on the outlet shelf and experiences no purchase, the “list” price displayed on the shelf is sampled because the data collector can still observe the price on the shelf. In the scanner databases, when an item is not sold, it is not even possible to get a “list” price. Therefore, missing prices can be a result of at least one of two events in scanner data, but only one event in the current manual sampling system of the CPI. 11 Price Index Estimation Using Price Imputation for Unsold Items