Focus
Quarterly GDP data underpin business cycle analysis and estimates of economic slack used in Phillips curve models. How much do reconstruction choices change our account of prewar business cycles? Reporting of official US quarterly GDP begins in 1947. Earlier quarterly histories combine annual accounts with indicators such as industrial production. Annual GDP constrains the yearly average but does not determine recession dates or depths. These choices matter for accounts of the Great Depression, the recovery and economic risk.
Contribution
This paper systematically compares 58 specifications for reconstructing quarterly GDP in 1919–46, holding annual GDP fixed. A common prediction test temporarily conceals observed quarterly GDP in four periods between 1947 and 1963. It ranks specifications by their errors in predicting the concealed quarterly growth rates. Separate exercises vary estimated coefficients and change estimation assumptions to test which historical conclusions survive. This distinguishes disagreement across reconstructions from uncertainty within one specification.
Findings
Reconstruction uncertainty is substantial but differs across recessions. The interquartile range of estimated contraction depths is narrow for the Great Depression but much wider for the 1937–38 recession, especially relative to the size of the contraction. Full ranges are wider still. These ranges describe disagreement across specifications, not confidence intervals. Production regression predicts concealed quarters well. Denton allocation gives a modest further gain and the lowest prediction error. Richer indicator panels, expenditure detail and more flexible relationships do not outperform that combination. Its 1937–38 decline is about 7.1%, compared with 10.5–10.6% in two widely used published series at common annual GDP. This milder decline survives coefficient uncertainty and alternative estimation assumptions, even when the best-performing method changes. Precise recession dates remain less secure.
Abstract
Official quarterly US GDP estimates begin in 1947; earlier quarterly histories must be reconstructed using annual benchmarks and historical indicators. To investigate uncertainty in these reconstructions, this paper systematically compares specifications drawn from the literature for reconstructing quarterly GDP during 1919–1946. The comparison evaluates predictions on concealed postwar quarters and separates variation across specifications from coefficient uncertainty within a model. Reconstruction choices matter, but their effects differ across episodes. Across 58 reconstruction specifications, the interquartile range of estimated peak-to-trough declines is 30.9–31.6% for the Great Depression, compared with 6.6–7.8% for 1937–1938. A reconstruction that augments the production-regression approach of Balke and Gordon (1986) with the Denton allocation used by Ramey and Zubairy (2018) has the lowest prediction error. Industrial production supplies information about quarterly movements, while annual accounts constrain the resulting path. Richer indicator panels, additional expenditure detail and more flexible relation ships do not improve on this reconstruction’s predictive accuracy. Its estimated 7.07% decline in 1937–1938 is milder than the approximately 10.5–10.6% declines in those published reconstructions. The reconstruction matches all six NBER recessions, but dates the first contraction two quarters earlier, from 1919Q3 to 1921Q1 rather than 1920Q1 to 1921Q3; its peak is the earliest allowed by the dating window.
