High-Frequency Substitution and the Measurement of Price Indexes

This paper investigates the use of high-frequency scanner data to construct price indexes. In the presence of inventory behavior, purchases and consumption by individuals differ over time. Cost-of-living indexes can still be constructed using data on purchases. For weekly data on canned tuna, the paper contrast two different types of price indexes: fixed-base and chained indexes. Only the former are theoretically correct, and in fact, the chained indexes have a pronounced upward bias for most regions of the U.S. This upward bias can be caused by consumers purchasing goods for inventory. The paper presents some direct statistical support for inventory behavior being the cause of the upward bias, though advertising and special displays also have a very significant impact on shopping patterns.