Why are initial estimates of productivity growth so unreliable?

https://doi.org/10.1016/j.jmacro.2015.11.004Get rights and content

Highlights

  • This paper analyzes revisions in annual and long-run US productivity growth.

  • Revisions are surprisingly important, and persist.

  • Revisions are often not news.

  • Revision errors contribute significantly to forecast errors.

Abstract

This paper argues that initial estimates of productivity growth will tend to be much less reliable than those of most other macroeconomic aggregates, such as output or employment growth. Two distinct factors complicate productivity measurement. (1) When production increases, factor inputs typically increase as well. Productivity growth is therefore typically less variable than output growth, meaning that measurement errors will tend to be relatively more important. (2) Revisions to published estimates of production and factor inputs tend to be less highly correlated than the published estimates themselves. This further increases the impact of data revisions on published productivity estimates.

To assess the extent of these problems in practice, we detail the importance of historical revisions to the most commonly-used measures of US aggregate productivity growth, expanding on previous empirical work by Aruoba (2008) and Anderson and Kliesen (2006). We find that such revisions have contributed substantially to policymakers’ forecast errors for US productivity growth.

Section snippets

Alan Greenspan on the measurement of US productivity growth

  • One would certainly assume that we would see this in the productivity data, but it is difficult to find it there. In my judgment there are several reasons, the most important of which is that the data are lousy.

    Transcript: Meeting of the Federal Open Market Committee, December 19, 1995, p. 37.

  • The one thing we know about the official data on productivity is that they are wrong.

    Transcript: Meeting of the Federal Open Market Committee, February 4-5, 1997, p. 101

  • The productivity numbers are very

On the reliability of productivity growth estimates

To better understand the reliability of productivity growth estimates and the scope for their improvement, we trace the sources of their revisions. Productivity growth can be decomposed in several ways. Corrado and Slifman (1999), for example, decompose aggregate productivity growth by sector. Here we decompose the noise/signal ratios of productivity measures to understand their relationship to the reliability of the underlying series used to calculate productivity.

We begin by defining the

Measures of productivity growth

No single measure of productivity is best for all purposes and care needs to be taken in matching the appropriate productivity measure to the problem at hand. Aggregate labor productivity, rather than aggregate or sectoral total factor productivity, is the relevant concept for many of the problems we mentioned at the outset. For consumption/savings decisions, individuals are concerned about the productivity of their labor, whether this is due to variations in total factor productivity or

Greenbook projections

In this section we compare the size of data revisions to the size of productivity growth rate forecast errors to examine the relative importance of data revisions from a policy perspective. The forecasts that we analyse here are those prepared by the staff of the US Federal Reserve Board for each meeting of the Federal Open Market Committee (FOMC) as part of their regular Greenbook projections.8

Conclusion

This paper investigated the statistical reliability of aggregate productivity estimates for the US. We explained why “recent” productivity growth estimates are much less reliable than those of other series (particularly output and employment.) The importance and robust nature of revisions reflect fundamental problems in productivity measurement rather than an unusual failure of government statisticians. The residual nature of macroeconomic productivity measurement causes productivity to be less

Acknowledgments

This paper was written during visits of the first author to CIRANO, of the second author to the research school SOM of the University of Groningen, and of both authors to the Centre of Applied Macroeconomic Analysis (CAMA), Australian National University, and the University of Tasmania (UTAS). The hospitality and support of these institutions, as well as that of CIREQ, is gratefully acknowledged. We would like to thank Robert Inklaar and Marcel Timmer of the Groningen Growth and Development

References (31)

  • Aruoba, S.B., 2004. Data uncertainty in general equilibrium. Available at SSRN: http://ssrn.com/abstract=639361 or...
  • AruobaS.B.

    Data revisions are not well behaved

    J. Money Credit Bank.

    (2008)
  • BenatiL.

    Drifts and breaks in labour productivity

    J. Econ. Dyn. Control

    (2007)
  • CroushoreD.

    Frontiers of real-time analysis

    J. Econ. Lit.

    (2011)
  • EdgeR.M. et al.

    Learning and shifts in long-run productivity growth

    J. Monet. Econ.

    (2007)
  • Cited by (7)

    • Reducing revisions in hedonic house price indices by the use of nowcasts

      2022, International Journal of Forecasting
      Citation Excerpt :

      Other nowcasting methods use models that exploit past revisions for deriving the best forecasts of the present. Authors like Garratt, Lee, Mise, and Shields (2008), Matheson, Mitchell, and Silverstone (2010), Jacobs and van Norden (2016), Kishor and Koenig (2012), and Clements and Galvão (2017) argue that revisions contain useful information for prediction. For example, Matheson et al. (2010) use qualitative responses from a panel business survey to predict revisions of GDP changes, based on nowcasts over 52 quarters (25 quarters for manufacturing output).

    • Testing for news and noise in non-stationary time series subject to multiple historical revisions

      2019, Journal of Macroeconomics
      Citation Excerpt :

      Consequently there exists a growing interest for investigating this type of data (see inter alia Croushore and Stark (2001), Orphanides and van Norden (2002), and Croushore (2011a, 2011b)). Several aspects of real-time data can be investigated: (i) structural or trend breaks (see Jacobs and van Norden (2016)) for a summary of the reliability of productivity growth rate trends); (ii) forecastability, i.e., whether revisions reduce noise or are news (the literature is briefly reviewed in Section 3.1); (iii) historical revisions, which affect the whole vintage of time series due to redefinitions, methodological innovations, etc., make testing difficult. The standard approach to dealing with historical revisions is either to employ growth rates to mitigate the effects of historical revisions, or to ‘clean’ the series in an attempt to get rid of the effects of historical revisions.

    View all citing articles on Scopus
    View full text