Beware the Shifting U.S. Data Sands
by Cynthia Latta
Although data revisions are common and announced in advance, they can still cause considerable confusion. Numbers pulled from a data set one day may be different the next day and different again the day after that. The reason is the difference between "benchmark" revisions, which incorporate updated information on the structure of the economy, and the normal monthly revisions, which result more from including late reports or new seasonal adjustments. Benchmark revisions update extended periods of history and are released separately from the regular monthly or quarterly data releases. The most recent example was the revision of the Federal Reserve's industrial production data. The benchmark revision was released on November 10 and covered the period through September. On November 14, the Fed released the regular monthly report with new data for October, but also the revisions for August and September stemming from the inclusion of late reports and other new information.
A shock wave is coming on December 10, when the Bureau of Economic Analysis will release its benchmark revisions to the national income accounts. This whammy will throw an additional curve—a base-year change. Data pulled either from Global Insight databanks or from the BEA's web site on December 10 will differ from what was there on December 9. The revised data will go only through the second quarter of 2003, so in effect, there will be no revised third-quarter data until the regular monthly GDP release on December 23. Trying to link the new historical data with the preliminary third-quarter GDP data that will be released on November 25 will be inadvisable because of the many conceptual changes in the detail data.
This benchmark revision has a further wrinkle. Real GDP will change in level because the base year will change from 1996 to 2000, not just for the total, but for all the pieces. Real GDP will rise in level from this change, while some pieces, like those for computer and software investment—where prices have been falling—will go down sharply. The use of chain weighting means that the base-year shift alone will not affect the growth rates. The revisions to the source data and updating of methodologies and added information will alter growth rates, but chain-weighting arithmetic reduces the impact of the pure base-year shift to zero.
We would advise clients to minimize data access for analytical purposes involving third-quarter national income data between December 10 and 23 and to keep a copy of the late-November GDP release handy.