Flaw of One Price; The International Comparison Program in Historical Perspective

When the Euro was introduced Denmark recommended abandonment of the PPP program in Eurostat precisely on the grounds the prices would be the same in the Euro zone. It was pointed out that within a single currency zone in the US prices varied by 30–40 % between cities of the South and Midwest compared to the Northeast or West Coast. And within the Euro zone price levels today have diverged, not converged after the Euro was introduced.

The partial title of this paper, The Flaw of One Price, is taken from an article in the Economist (October 18, 2003, p. 73) reporting on the Euro-price of a number of items in Euro-area countries. Clearly the expectation was that one price should prevail, but in fact very significant discrepancies remained 4 years after the Euro was introduced. The percentage difference between high and low prices across the countries was 160 % for a cinema ticket and 100 % for a cup of coffee (non-tradables), and 100 % for bottled, 75 % for milk and jeans and over 50 % for Pampers and Nurofen, some typical tradables. The consistent message of empirical studies is that the law of one price does not even prevail for tradable goods.

A very good survey of source materials is provided by Allen (1994).

For a fuller treatment, see Balk (2008). pp. 42–4.

Two aspects of this initial study stand out from my conversations with Kravis. First, in their work Gilbert always focused on the quantity comparisons and Kravis the price side of purchasing power comparisons. Necessarily they would end up with the same conclusion, but they always argued from a different starting point. A second recollection provides a sense in which Kravis always gave great attention to detail and to checking results, much due to Simon Kuznets, one of his teachers at Penn. When the initial results became available few days before Kravis was to return to Philadelphia to begin teaching, he was so surprised by the Paasche–Laspeyres spreads for Italy and the US of 75 % that he postponed his return from Paris several days to redo the calculations. Happily there were no mistakes, and the Italy-US spread turned out to be unremarkable for many pairs of countries. However, at that time Kravis was familiar with Paasche–Laspeyres spreads from time to time price indexes where differences were rarely over 5 %, hence his initial surprise at the Italy-US result.

Youri Ivanov, who currently works with CISTAT in Moscow where PPP comparisons are carrieds out across the Commonwealth of Independent States, kindly provided background on previous comparisons.

Mexico submitted data but in the end did not want to be included in the final report that includes 60 countries.

Unfortunately the tape containing the ECIEL data was damaged in transmission to ECLAC and the UNSO leaving some of the data on Producers Durables compromised. Further some of the price and rent data were not absolute currency figures but price levels relative to other Latin American countries. Some special collections were undertaken to make it possible to use the compromised data, but linking of Latin American countries was less satisfactory than for other regions.

Feenstra and Shapiro (2003) edited an NBER volume on uses of scanner data in a number of contexts. A study that explicitly used scanner data across countries in the EU was Heravi et al. (2003).

Evaluations of scanner versus survey data in the US have primarily been for temporal price indexes. An interesting use of web scraping is the Billion Price Project. Cavallo (2013).

The BLS totally revised the CPI in the 1970s. There was a new expenditure survey to obtain the weights to move the price relatives over time. And instead of pricing a given specification, collectors now filled out a check sheet of the outlet type and characteristics of the good or service priced. At first glance it appeared that it would be difficult to compare these prices across different areas of the US. However Kim Zieschang, then of BLS, adapted the CPD method of the ICP to a full hedonic estimate based on the characteristics of items to estimate PPPs across the 38 areas of the BLS.