James Hamilton on Shadowstats' and Niall Ferguson's Claim That "True" Inflation is More than 10% per Year
Econbrowser: Shadowstats debunked: I've yet to find someone who has been able to reproduce the claims made by Shadow Government Statistics about the extent to which government agencies are grossly misreporting the U.S. inflation rate. Apparently, neither has the Bureau of Labor Statistics, as detailed in an article by BLS economists John Greenlees and Robert McClelland in the latest issue of Monthly Labor Review.
First, some of the bolder claims by Shadowstats:
The Boskin/Greenspan argument was that when steak got too expensive, the consumer would substitute hamburger for the steak, and that the inflation measure should reflect the costs tied to buying hamburger versus steak, instead of steak versus steak. Of course, replacing hamburger for steak in the calculations would reduce the inflation rate, but it represented the rate of inflation in terms of maintaining a declining standard of living. Cost of living was being replaced by the cost of survival. The old system told you how much you had to increase your income in order to keep buying steak. The new system promised you hamburger, and then dog food, perhaps, after that....
The BLS initially did not institute a new CPI measurement using a variable-basket of goods that allowed substitution of hamburger for steak, but rather tried to approximate the effect by changing the weighting of goods in the CPI fixed basket. Over a period of several years, straight arithmetic weighting of the CPI components was shifted to a geometric weighting. The Boskin/Greenspan benefit of a geometric weighting was that it automatically gave a lower weighting to CPI components that were rising in price, and a higher weighting to those items dropping in price.
Once the system had been shifted fully to geometric weighting, the net effect was to reduce reported CPI on an annual, or year-over-year basis, by 2.7% from what it would have been based on the traditional weighting methodology. The results have been dramatic. The compounding effect since the early-1990s has reduced annual cost of living adjustments in social security by more than a third.
And here's the response by Greenlees and McClelland:
To begin, it must be stated unequivocally that the BLS does not assume that consumers substitute hamburger for steak. Neither the CPI-U, nor the CPI-W used for wage and benefit indexation, allows for substitution between steak and hamburger, which are in different CPI item categories. Instead, the BLS uses a formula that implicitly assumes a degree of substitution among the close substitutes within an item-area component of the index. As an example, consumers are assumed to respond to price variations among the different items found within the category "apples in Chicago." Other examples are "ground beef in Chicago," "beefsteaks in Chicago," and "eggs in Boston"....
The quantitative impact of the CPI's use of the geometric mean formula also has been grossly overstated by some, with one estimate exceeding 3 percent per year. It is difficult to identify real-world circumstances under which geometric mean and Laspeyres indexes could differ by such a large amount. The two index formulas will give the same answer whenever the prices used in an index all change by the same percentage. The bigger the differences in price changes, the more the Laspeyres index will tend to exceed the geometric mean. For the growth rate of the Laspeyres index to exceed the growth rate of a geometric mean index by 3 percentage points, however, the differences in individual price changes have to be quite large.
To see this point, consider another very simplified example. Suppose that the CPI sample for ice cream and related products in Boston consisted only of an equal number of prices for ice cream and frozen yogurt and that, between one year and the next, all the prices of ice cream in Boston rose by 8.6 percent while all the frozen yogurt prices fell by 4.2 percent. In that case, the geometric mean estimate of overall annual price change would be 2.0 percent, only slightly less than the Laspeyres estimate of about 2.2%. In order to come up with a difference of 3 index points, one has to assume a much more dramatic divergence between ice cream and frozen yogurt prices than the one hypothesized. For example, if ice cream prices rose 30 percent in one year, while frozen yogurt prices fell by 20 percent, the overall geometric mean index would still rise by 2 percent, but the Laspeyres index would rise 5 percent, for a difference of 3 index points. However, such a large annual divergence would be quite uncommon within CPI basic indexes-- between ice cream and yogurt, between types of candy and gum, between types of noncarbonated juices, or between varieties of ground beef. Moreover, for a 3-percentage-point divergence to continue year after year, the divergence between the individual component prices would have to continue to widen. For example, if, by contrast, during the next year ice cream prices increased by the same amount as frozen yogurt prices, then the two index formulas would give the same inflation estimate for that year. Although such a divergence might plausibly occur in one component for 1 year, it is beyond belief that such sharply divergent price behavior would continue year after year across the whole range of CPI item-area components.
Finally, and most importantly, there is rigorous empirical evidence on the actual quantitative impact of the geometric mean formula, because the BLS has continued to calculate Laspeyres indexes for all CPI basic indexes on an experimental basis for comparison with the official index. These experimental indexes show that the geometric mean led to an overall decrease in CPI growth of about 0.28 percentage point per year over the period from December 1999 to December 2004, close to the original BLS prediction that the impact would be approximately 0.20 percentage point per year.
There's much more in the BLS article on this and related questions such as hedonic price adjustment and owner's equivalent rent.
Why do people continue to give credibility to an operation like Shadowstats? Now that's something that I'd like to hear explained.