Scanner data generated from electronic paying devices, like cashing desks, are a rich data source on transactions.
More and more countries use scanner data for (segments of) their Consumer Price indices (CPIs), or consider to do so. It is generally agreed that multilateral price index methods are most appropriate for transaction data. A complication of many multilateral methods is the relaunch problem. Price changes due to the replacement of a product by a similar product are not properly measured. A well-known solution is to combine similar products into product clusters. This approach might however produce unit value bias when applied to inhomogeneous product strata. Other correction methods are imputation and product matching. This paper compares clustering, imputation and matching. The impact of these methods is empirically evaluated for different multilateral index methods using transaction data with simulated relaunches.
