The U.S. Bureau of Economic Analysis finds that early estimates of real Gross Domestic Product are reliable, with evidence indicating only minor improvements to accuracy from advance to second to third quarterly estimates, as measured by mean absolute revisions or standard deviations (see Fixler, Greenaway-McGrevey, and Grimm, “The Revisions to GDP, GDI, and Their Major Components,” Survey of Current Business, August 2014). BEA estimates GDP quarterly in three monthly vintages, with the “advance” estimate 30 days following the reference quarter, the “second” estimate 60 days following and the “third” estimate 90 days following. Annual revisions occur each year for a three-year period, with comprehensive revisions across the entire time series occurring in years ending in three and eight. BEA regularly tests the reliability of each successive vintage of GDP estimates.
BEA’s Alyssa Holdren (see Holdren, “Gross Domestic Product and Gross Domestic Income Revisions and Source Data,” Survey of Current Business, June 2014) illustrates that successive vintage estimates of real Gross Domestic Product contain increasing amounts of quality source data covering the components of GDP. More than three-fifths of the initial, or advance, estimate of GDP is based on trend projections or indirect indicators, and only two-fifths are based on direct indicators. With each successive quarterly vintage the portion of quality data increases, with slightly more than one-half of the second estimates and two-thirds of the third estimates based on data rather than proxies.
With respect to the annual revisions, the three current annual vintages contain still greater amounts of increasingly high quality annual-frequency source data. Reflecting the better data, the correlations between the successive early vintage estimates and the latest estimates increase from 0.78 for the advance estimates to 0.96 for the third annual estimates, with reasonably smooth increments between successive vintages. This pattern indicates a gradually tightening relationship, with no apparent jumps with later vintages.
The relationship between the advance and the latest estimates may also be visualized in a scatter diagram.
The scatter portrays a reasonably tight relationship. This suggests that the early estimates, which contain much more judgmental inputs, are not particularly less reliable than the latest estimates, which contain all available source data and thus relatively little judgment. As such, the early estimates should be viewed as generally accurate and not substantively less reliable than later vintage estimates.
The various estimates reflect a balance between timeliness of early estimates to inform policy and business investment decisions, and the accuracy accrued with the incorporation of increasingly quality source data. The relationships between the earliest estimates and the latest suggest an appropriate balance, although continuing to improve the reliability of the earliest vintages remains a priority for BEA.