Scanner Data, Elementary Price Indexes and the Chain Drift Problem

Statistical agencies increasingly are able to collect detailed price and quantity information from retailers on sales of consumer products. Thus elementary price indexes (which are indexes constructed at the first stage of aggregation for closely related products) can now be constructed using this price and quantity information, whereas previously, statistical agencies had to construct elementary indexes using just retail outlet collected information on prices alone. Thus superlative indexes can now be constructed at the elementary level, which in theory, should lead to more accurate Consumer Price Indexes. However, retailers frequently sell products at heavily discounted prices, which lead to large increases in purchases of these products. This volatility in prices and quantities will generally lead to a chain drift problem; i.e., when prices return to their “normal†levels, quantities purchased are frequently below their “normal†levels and this leads to a downward drift in a superlative price index. The paper addresses this problem and looks at the likely bias in various index number formulae that are commonly used. The bias estimates are illustrated using some scanner data on the sales of frozen juice products that are available online.