Online and Official Price Indexes: Measuring Argentina’s Inflation

The first exercise uses a different approach to impute missing values within price series. The method used in the paper is to fill missing prices with the last available price for each product. This approach is reasonable because the gaps in the online data last only a few days. However, official statistical offices deal with missing values in their monthly series in different ways. The standard approach, also used in Argentina, is to impute missing prices with the average price change of similar products. Methods vary slightly across countries, so here I follow the “cell-relative” approach used by the Bureau of Labor Statistics (BLS): if a product is missing on a particular day, I do not use that product for the calculation of that day’s inflation, but I impute a price for it equal to the previous price times the average price change for products in the same category that day. In theory, different methods to fill in price gaps should not impact long-term inflation estimates, but they could yield differences in some of the short-run dynamics. Indeed, Figure B.1 shows that the monthly inflation estimates are nearly identical for most of the sample, with the exception of March and November 2010. The overall differences are small, and not nearly strong enough to explain any of the differences between online and official estimates. The second robustness test is to use an unweighted index, just as the one constructed for Venezuela in the paper. This index is a geometric mean applied to all price changes observed each day in the store, without any categories or relative weights. Naturally, the effect of a geometric mean is to reduce the impact of products that have relatively large price changes (either up or down). It is implemented by the BLS and other statistical offices at a sub-category level in order to approximate the effect of withincategory substitution (for example, when the price of ”McIntosh” apples goes up, consumers might tend to buy ”Fuji” apples). In this data, we are pooling together goods from different categories, effectively allowing cross-category substitutions as well. In theory, we should expect to see less inflation when it is rising, and more inflation when it is falling. That is precisely what Figure B.1 shows. In particular, this