Font Size: 

Scanner data arising from retail transactions have replaced survey of food price observations for the consumer price index (CPI) for more than a decade. The same data source can provide the expenditure weights needed for the CPI as well, when combined with population data using secure linkage and processing techniques that protect confidentiality. This would alleviate the most burdensome part of diary collection for the Consumer Expenditure Survey that collects expenditure data from households. Due to the sheer amount of transactions, automatic classification of the consumption subclasses of the goods requires natural language processing techniques, as long as there does not exist a catalogue that covers all the goods. Statistical theories pertaining to these big-data expenditure weights and classification are discussed.