Abstract： 

We examine how precisely one can reproduce the CPI constructed based on price
surveys using scanner data. Specifically, we closely follow the procedure adopted by the
Statistics Bureau of Japan when we sample outlets, products, and prices from our scanner data and aggregate them to construct a scanner databased price index. We show that the following holds the key to precise replication of the CPI. First, the scanner databased index crucially depends on how often one replaces the products sampled. The
scanner data index shows a substantial deviation from the actual CPI when one chooses
a value for the parameter associated with product replacement such that replacement
occurs frequently, but the deviation becomes much smaller if one picks a parameter value
such that product replacement occurs only infrequently. Second, even when products are
replaced only infrequently, the scanner data index differs significantly from the actual
CPI in terms of volatility. The standard deviation of the scanner databased monthly
inflation rate is 1.54 percent, which is more than three times as large as that for actual
CPI inflation. We decompose the difference in volatility between the two indexes into
various factors, showing that it mainly stems from the difference in price rigidity for
individual products. We propose a filtering technique to make individual prices in the
scanner data stickier, thereby making scanner databased inflation less volatile. 