NABE presidential address: building improved official statistics for decisionmaking - National Association of Business Economists - Transcript

Business Economics, Jan, 1994 by Joseph W. Duncan

The press also has a responsibility. They should seek out the experts in statistical agencies and they should challenge the knowledge of the commentators they are interviewing. In investment circles there is a tendency to seek a consensus before official statistics are released. Yet many of the sources of consensus have no special independent information; rather they simply offer "guesstimates."

Finally, the statistical agencies should take care to help present clearer indications of what the numbers mean. I will illustrate this with the many revisions that are published concerning the Gross Domestic Product (GDP). All members of NABE are familiar with the many revisions published as part of the GDP estimates. The key estimates are the preliminary (25-day estimate, i.e., 25 days after the end of the quarter), the revised (the 55-day estimate), the final current (85 days), the 1st July revision (the July following the year end), the 2nd July revision (one year later) and the 3rd July revision (an additional year later). There are also revisions as a result of comprehensive benchmarking after data from the quinquennial results from the Census of Business are available. These revisions have long been a subject of controversy. Many observers who do not understand statistical programs think that someone is "manipulating" the numbers, yet every analysis of the process reveals that these revisions are the final information. The classical study of the revisions was conducted by Roseanne Cole.(11) A more recent study was conducted by Stephen K. McNees of the Boston Federal Reserve.(12) Currently, Martin Fleming, Chairman of the Statistics Committee of the National Association of Business Economists, has completed a more recent review.(13)

As part of the study for the present report, data were developed by the Bureau of Economic Analysis for Haver Analytics of New York. That data base was used by Martin Fleming for his study. The data were also analyzed by George Feeney of Haver Analytics as an input to our book. The Feeney study used the third year revision as an estimate of truth for the individual GDP components, while Fleming used the second year revision as the benchmark for the "true" estimate. In the Feeney study the early quarterly estimates (estimates 1, 2, and 3) and the 1st and 2nd July revisions were compared with the 3rd July revision (assumed truth) to see the pattern of variance. The analysis showed little variance reduction in the 45- and 75-day estimates.

In a forthcoming study(14) Allan E. Young notes, "...the incorporation of additional or more accurate source data in the preliminary and final current estimates of GDP does not result in substantial improvement." He notes that most observers are puzzled by this lack of improvement, and he identifies two factors that are overlooked. They are:

1. "the small role played by the second and third months of the quarter in determining the change from the previous quarter,(15) and

2. "certain sources of errors in the preliminary and final current estimates to which the advance estimates are immune."(16,17)