The article by Shingo Watanabe (Reference Watanabe2016) analyses whether technology shocks—the prime driver of business cycles in Real Business Cycles (RBC) models—could have set off the Great Depression, adding to a recent discussion in this Journal. In line with our own work (Inklaar, de Jong, and Gouma Reference Inklaar, de Jong and Gouma2011), he concludes that an appropriately constructed technology series does not have the positive relationship with factor inputs that is required for RBC models to hold true.

His article and ours are inspired by the work of Susanto Basu, John G. Fernald, and Miles S. Kimball (Reference Basu, Fernald and Kimball2006), who analyse the same question (“are technology shocks contractionary?”) and we each reach the same conclusions as Basu, Fernald, and Kimball (BFK) did for the post-WWII period. However, compared with the “gold standard” of BFK, analysing the pre-WWII period requires compromises, for instance, regarding the data. In our article, this led us to analyse biennial census data on gross output and inputs for 19 manufacturing industries for the 1919–1939 period, which precluded us from analysing annual fluctuations. Watanabe (Reference Watanabe2016) used annual value added data for the private non-farm sector for the period 1892–1996, which requires more restrictive assumptions regarding technology. Neither empirical setting matches BFK's annual industry-level data which covered the entire non-farm sector after WWII, but taken together the article of Watanabe (Reference Watanabe2016) and our own (Inklaar, de Jong, and Gouma Reference Inklaar, de Jong and Gouma2011) present a compelling case against RBC models of the Great Depression.

Although Watanabe (Reference Watanabe2016, p. 924) argues that our results “may be partly due to coincidence,” we would argue that the criticisms raised do not warrant such a drastic assertion and that, instead, the two articles should be seen as providing convincing and complementary evidence.

Comparing the empirical analyses, Watanabe argues that we should have run a weighted regression and excluded oil prices as an instrument. Yet both changes would have led to a fairly modest increase in the estimated return to scale (from 1.17 to 1.23) with no effect on our findings.Footnote ¹ Another criticism is that the change in his (non-farm sector) technology series is differently signed than in our (manufacturing) technology series in many periods. However, there are very few years in which the sign of non-farm technology change can be determined with (statistical) confidence (Watanabe Reference Watanabe2016, Figure 1) and no compelling reason why the signs should match in the two different sectors. The argument that we should have been able to recover an (effectively) one-year negative response of technology on hours worked with biennial data is even less convincing.Footnote ²

A larger discussion is whether our specification lets us adequately capture unmeasured changes in input utilization. It is true that by not finding a significant effect of changes in average hours paid per worker, we cannot follow the same identification strategy as BFK and Watanabe, ending up, rather, with a different specification than the theoretical ideal. At the same time, in his analysis, Watanabe has to resort to stringent assumptions about production technology; specifically that intermediate inputs and factor inputs are used in strict proportion. Such an assumption has been repeatedly rejected in the literature, including in our work when we demonstrate the bias in coefficients when estimating our parameters in a value added rather than a gross output framework (Inklaar, de Jong, and Gouma Reference Inklaar, de Jong and Gouma2011, p. 847 and Online Appendix Table 8).

It is not hard to come away from this discussion with a sense of disappointment about the quality of the data available to hold this debate. However, we would offer a more optimistic conclusion, namely that whatever the shortcomings in the analyses, they are apparently not critical to the central argument of these two articles. And the fact that two studies, each with its own limitations, comes to the conclusion that technology shocks did not drive the Great Depression should certainly be considered scholarly progress.