The Bank for International Settlements (BIS) has published a working paper finding that reconstruction choices affect estimates of US business cycles before official quarterly gross domestic product (GDP) data began in 1947, with sensitivity differing across recessions. The method with the lowest prediction error estimates a 7.07% contraction in 1937–38, compared with approximately 10.5–10.6% in two widely used published series adjusted to the same annual GDP. The comparison holds annual GDP fixed across 58 specifications for reconstructing quarterly output in 1919–46. Methods are ranked by their ability to recover observed quarterly growth temporarily concealed in four periods between 1947 and 1963. A regression using industrial production combined with Denton allocation, which reconciles quarterly estimates with annual accounts, records the lowest prediction error. The allocation step provides a modest improvement over production regression alone. Broader indicator panels, additional expenditure detail and more flexible relationships do not outperform this combination in the comparison. Across specifications, the interquartile range of estimated contraction depths is 30.9–31.6% for the Great Depression and 6.6–7.8% for 1937–38. These ranges measure disagreement between reconstructions, not confidence intervals, and the full ranges are substantially wider. The milder 1937–38 decline survives coefficient uncertainty and alternative estimation assumptions, even when a different method records the lowest prediction error. Precise recession dates remain less secure, and narrow coefficient intervals do not capture all uncertainty about missing quarterly GDP.