Is there really a gap between aggregate productivity and technology?

Abstract

The important contribution by Basu and Fernald (2002) shows that, in practice, there is a statistically significant gap between aggregate productivity and technology that can be attributed to inefficient product and labour markets. This is important, as it implies that the Solow residual is an imperfect index for aggregate technology change. We take a related approach and find that when we control for capacity utilization, time varying markup and account for externalities between industries, by employing a superior system estimator, the gap between the aggregate productivity and technology is shown to narrow considerably.

Acknowledgements

We are grateful to John Fernald for providing data and software codes. We also thank the editor and two anonymous referees for helpful comments. The usual disclaimer applies.

Notes

¹Indeed, BF write in the abstract (p. 963) that ‘[e]mpirically,…gaps are important, even though we abstract from variations in factor utilization and estimate only small average sectoral markups.’

²Implicitly, our specification has the obvious advantage of not assuming constant returns to scale and perfect competition in product markets when estimating technical change, unlike Mullen (2001), for example, and studies therein.

³Based on Jorgenson et al. (1987), downloaded from Dale Jorgenson's webpage and combined with the data used in BF.

⁴Our replication looks different from the one in BF Fig. 2, p. 988. In their article, they mistakenly used the standard Solow residuals instead of the revenue weighted residuals. The corrected plot looks similar to our Fig. 1 even for their 34 industries.

⁵Another alternative is GMM system estimation. However, heteroskedasticity and autocorrelation tests at the sectoral level do not reveal significant problems, thus confirming the appropriateness of 3SLS.

⁶Two specification tests (LR and Breusch–Pagan) reject at the 1% significance level the null of cross-sectoral uncorrelated disturbances, thus favouring the system specification against the equation-by-equation estimation.