Explaining a productive decade

Brookings Papers on Economic Activity

Volumn , Issue 1, 2007, Pages 81-137

  Oliner, Stephen D ^a   Sichel, Daniel E ^a   Stiroh, Kevin J ^b  

^a FEDERAL RESERVE BOARD   (United States)

^b FEDERAL RESERVE BANK OF NEW YORK   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 34848907447     PISSN: 00072303     EISSN: None     Source Type: Journal    
DOI: None     Document Type: Conference Paper

References (100)

1. See Economic Report of the President 2001, Basu, Fernald, and Shapiro (2001), Brynjolfsson and Hitt (2000, 2003), Jorgenson and Stiroh (2000), Jorgenson, Ho, and Stiroh (2002, 2005), and Oliner and Sichel (2000, 2002). In these papers IT refers to computer hardware, software, and communications equipment. This category often also is referred to as information and communications technology, or ICT. For industry-level evidence supporting the role of IT in the productivity resurgence,
2. see Stiroh (2002b). For an interpretation of the industry evidence that puts less emphasis on IT,
3. see Bosworth and Triplett (2007) and McKinsey Global Institute (2001).
4. See Corrado, Hulten, and Sichel (2005, 2006), Brynjolfsson and Hitt (2003), Bresnahan, Brynjolfsson, and Hitt (2002), Basu and others (2004), Black and Lynch (2001, 2004), and Nakamura (1999, 2001,2003). The National Income and Product Accounts (NIPAs) exclude virtually all intangibles other than software, although the Bureau of Economic Analysis, which produces the NIPA data, recently released a satellite account for scientific research and development;
5. see Okubo and others (2006).
6. Bresnahan and Trajtenberg (1995) were the first to write about IT as a general-purpose technology. See also Organization for Economic Cooperation and Development (2000), Schreyer (2000), van Ark (2000), Basu and others (2004), and Basu and Fernald (2007).
7. Jorgenson, Ho, and Stiroh (2007); Bosworth and Triplett (2007).
8. See Gordon (2003), Baily (2003), Schweitzer (2004), and Stiroh (forthcoming).
9. For references to the business press, see Gordon (2003) and Stiroh (forthcoming).
10. Several other researchers have examined industry data, including Baily and Lawrence (2001), Stiroh (2002b), Nordhaus (2002b), Corrado and others (2007), and Bosworth and Triplett (2007).
11. For references to the literature on industry-level data in Europe, see van Ark and Inklaar (2005).
12. See Economic Report of the President 2007, Congressional Budget Office (2007a, 2007b), and the latest available transcripts of the meetings of the Federal Open Market Committee (FOMC). The 2001 FOMC transcripts show that staff presentations on the economic outlook featured growth accounting in the discussion of productivity trends. Private sector analysts also rely on growth accounting;
13. see, for example, Global Insight, U.S. Executive Summary, March 2007,
14. and Macroeconomic Advisers, Macro Focus, March 22, 2007.
15. Much has been written about the link between management expertise and productivity, including Bloom and Van Reenen (2006), McKinsey Global Institute (2001), and Farrell, Baily, and Remes (2005). Gordon (2003) and Sichel (2003) provide reasons why offshoring and hours mismeasurement may have had a relatively limited effect on labor productivity growth, whereas Houseman (2007) argues that these factors could have had a significant effect in the U.S. manufacturing sector.
16. For a discussion of measurement issues related to the pace of technical progress in the semiconductor industry, see Aizcorbe, Oliner, and Sichel (2006).
17. For further discussion of issues related to critiques of the neoclassical framework, see Congressional Budget Office (2007b).
18. Gordon (2003).
19. Bosworth and Triplett (2007).
20. Roberts (2001).
21. Basu and others (2004).
22. Oliner and Sichel (2000, 2002).
23. Basu, Fernald, and Shapiro (2001; hereafter BFS)
24. Basu and others (2004; hereafter BFOS).
25. Although BFS also include adjustment costs for labor in their model, they zero out these costs in their empirical work. We simply omit labor adjustment costs from the start. For additional discussion of capital adjustment costs and productivity growth,
26. see Kiley (2001).
27. The results in BFS and in Basu, Fernald, and Kimball (2006) strongly support the assumption of constant returns for the economy as a whole. We invoke perfect competition as a convenience in a model that already has many moving parts.
28. Oliner, Sichel, and Stiroh (2007).
29. Kf proxies for this output elasticity. However, in the presence of adjustment costs, the first-order condition for the optimal choice of capital yields the more general result shown in equation 1. In effect, the income share captures both the direct contribution of capital to production and the benefit of having an extra unit of capital to absorb adjustment costs. The weight in equation 1 nets out the portion of the income share that relates to adjustment costs, as this effect is embedded in the MFP term discussed below.
30. Domar (1961).
31. Oliner and Sichel (2002).
32. In contrast to the expression for aggregate MFP growth in BFS, equation 2 contains no terms to account for reallocations of output, labor, or capital across sectors. The particularly clean form of equation 2 arises, in large part, from our assumption of constant returns to scale and the absence of adjustment costs for labor (which implies that competitive forces equate the marginal product of labor in all sectors). In addition, we have assumed that any wedge between the shadow value of capital and its user cost owing to adjustment costs is the same in all sectors. Given this assumption, reallocations of capital across sectors do not affect aggregate output.
33. For details on data sources, see the data appendix to Oliner and Sichel (2002).
34. The weight on the capital deepening term in equation 1 for type-j capital equals its income share minus its adjustment cost elasticity. As discussed below, empirical estimates of these asset-specific elasticities are not available, which forces us to approximate the theoretically correct weights. Note that the weights on the capital deepening terms in equation 1 sum to one minus the labor share under constant returns to scale. We replace the theoretically correct weights with standard income-share weights that also sum to one minus the labor share. This approximation attaches the correct weight to aggregate capital deepening but may result in some misallocation of the weights across asset types.
35. Oliner and Sichel (2000, 2002);
36. Jorgenson, Ho, and Stiroh (2002, 2007).
37. Year-by-year share weighting embeds the implicit assumption that firms satisfy the static first-order condition that equates the marginal product of capital with its user cost. Strictly speaking, this assumption is not valid in the presence of adjustment costs, as noted by BFS and by Groth, Nunez, and Srinivasan (2006). Both of those studies replace the yearby-year share weights with the average shares over periods of five years or more, in an effort to approximate a steady-state relationship that might be expected to hold on average over longer periods. We found, however, that our results were little changed by replacing yearby-year shares with period-average shares. Accordingly, we adhere to the usual share weighting practice in the literature.
38. Jorgenson and Stiroh (2000), Jorgenson, Ho, and Stiroh (2002, 2007), Oliner and Sichel (2000, 2002), and Triplett (1996), among others.
39. Oliner and Sichel (2002) give a nontechnical description of the way in which we implement the dual method, and the appendix in the working paper version of this paper (Oliner, Sichel, and Stiroh, 2007) provides the algebraic details.
40. BFS used a larger value for φ, 0.05, but subsequently corrected some errors that had affected that figure. These corrections caused the value of φ to be revised to 0.035.
41. Shapiro (1986).
42. Hall (2004).
43. Groth(2005).
44. Jorgenson (2001) argues that the steeper declines in semiconductor prices reflected a shift from three-year to two-year technology cycles starting in the mid-1990s. Aizcorbe, Oliner, and Sichel (2006) report that shorter technology cycles drove semiconductor prices down more rapidly after 1995, but they also estimated that price-cost markups for semiconductor producers narrowed from 1995 to 2001. Accordingly, the faster price declines in the late 1990s-and the associated pickup in MFP growth-partly reflected true improvements in technology and partly changes in markups. These results suggest some caution in interpreting price-based swings in MFP growth as a proxy for corresponding swings in the pace of technological advance.
45. The combined effect of adjustment costs and factor utilization remained essentially zero after 2000. Although the deceleration in investment spending after 2000 eliminated the negative effect of adjustment costs, the net decline in the workweek pushed the utilization effect into negative territory.
46. Of course, MFP growth is a residual, so this result speaks only to the proximate sources of growth and does not shed light on the more fundamental forces driving MFP growth.
47. Jorgenson, Ho, and Stiroh (2007).
48. Corrado, Hulten, and Sichel (2005, 2006).
49. See Brynjolfsson and Hitt (2000), Brynjolfsson, Hitt, and Yang (2002), and McKinsey Global Institute (2002) for interesting case studies regarding the creation of organizational capital.
50. The BFOS model focuses on intangibles that are related to information technology. This is a narrower purview than in Corrado, Hulten, and Sichel (2005, 2006), who develop estimates for a full range of intangible assets, regardless of their connection to IT. Although we do not provide a comprehensive accounting for intangibles, we highlight the intangible assets that are central to an assessment of the contribution of information technology to economic growth.
51. Hall and Jorgenson (1967).
52. Specifically, for the income share of intangible capital, we use the income share series for "New CHS intangibles," that is, those intangibles over and above those included in the NIPAs. We then adjust this series downward to account for the fact that some CHS intangibles are not related to IT and thus do not fit in the BFOS framework. As a crude adjustment, we remove the income share associated with brand equity and one-third of the income share for other components of "New CHS intangibles."
53. We use 2003 as the final year for this calculation because that is the last year of data in CHS.
54. See Yang and Brynjolfsson (2001) for an alternative approach to incorporating intangibles into a standard growth accounting framework. Their approach relies on financial market valuations to infer the amount of unmeasured intangible investment and shows that, through 1999, the inclusion of intangibles had potentially sizable effects on the measured growth of MFP.
55. Brynjolfsson and Hitt (2000), Brynjolfsson, Hitt, and Yang (2002), and McKinsey Global Institute (2002).
56. Aral, Brynjolfsson, and Wu (2006, p. 2). Some interpret the econometric results in Brynjolfsson and Hitt (2003) as support for a lag between the installation of IT capital and the accumulation of complementary capital. We believe this interpretation is incorrect. Brynjolfsson and Hitt show that the firm-level effect of computerization on MFP growth is much stronger when evaluated over multiyear periods than when evaluated on a year-by-year basis. Importantly, however, the variables in their regression are all measured contemporaneously, whether over single-year or multiyear periods. Accordingly, their results suggest that the correlation between the growth of IT capital and intangible capital may be low on a year-by-year basis, but that a stronger contemporaneous correlation holds over longer periods, boosting the measured effect on MFP growth.
57. For background on industry productivity analysis, see Jorgenson, Gollop, and Fraumeni (1987), Basu and Fernald (1995, 1997, 2001, 2007), Nordhaus (2002b), Stiroh (2002a, 2002b), Triplett and Bosworth (2004), and Bosworth and Triplett (2007).
58. Bruno (1978); Norsworthy and Malmquist (1983); Jorgenson, Gollop, and Fraumeni (1987).
59. Basu and Fernald (1995, 1997).
60. Basu and Fernald (2001).
61. Howells, Barefoot, and Lindberg (2006).
62. The underlying sources of these data are the BLS Current Employment Survey (for wage and salary jobs and average weekly hours), the Current Population Survey (for self-employed and unpaid workers, agricultural workers, and within-household employment), and unemployment insurance tax records.
63. Jorgenson, Ho, and Stiroh (2007).
64. As a comparison, Stiroh (2001, 2002b) reported an acceleration of ALP after 1995 for six of ten broad sectors, which accounted for the majority of output using earlier vintages of SIC data.
65. Corrado and others (2007); Bosworth and Triplett (2007).
66. BEA uses the "double deflation" method to estimate real value added for all industries as the difference between real gross output and real intermediate inputs (Howells, Barefoot, and Lindberg, 2006). Basu and Fernald (2001) show that this can be approximated, as in equation 16, by defining gross output growth as a weighted average of value added and intermediate input growth.
67. As in Stiroh (2002b).
68. This value-added approach is similar to the decomposition in Nordhaus (2002b).
69. We also aggregated the industry output data using a Fisher (rather than the Tornqvist) index and still found a small difference for the period 1995-2000. We do not have an explanation for this.
70. Jorgenson and others (forthcoming) show an increase in both the intermediate input and hours reallocation terms, although both are slightly negative through 2004. The results in Bosworth and Triplett (2007) are similar to ours in some respects (rising direct contribution of gross output productivity through 2000 followed by a substantial fall, and an intermediate reallocation term that switches from negative to positive after 2000), but their hours reallocation term remains negative through 2005. This divergence reflects differences in the estimation of the hours series. Bosworth and Triplett (2007) use the BEA series on fulltime/part-time employees, which they scale by total hours per employee from BLS for 1987 to 2004. They hold hours per full-time/part-time employee constant from 2004 to 2005.
71. This is analogous to the analysis of the sources of productivity growth within the U.S. retail trade sector by Foster, Haltiwanger, and Krizan (2002), who report that the majority of productivity gains reflect entry and exit, with a very small contribution from productivity gains within continuing establishments.
72. Stiroh (forthcoming).
73. Caballero and Hammour (1994).
74. See, for example, Smith and Lum (2005) and Howells, Barefoot, and Lindberg (2006).
75. Stiroh (2002b).
76. Appendix table A-1 shows this classification scheme for the sixty detailed industries based on both 1995 and 2000 IT capital income shares and reports the 2005 share. Baily and Lawrence (2001), Stiroh (2001), and Jorgenson, Ho, and Stiroh (2005) also use relative shares of IT capital in total capital to identify IT-intensive industries in the United States, and Daveri and Mascotto (2002), Inklaar, O'Mahony, and Timmer (2005), O'Mahony and van Ark (2003), and van Ark, Inklaar, and McGuckin (2003) do so in international studies.
77. This specification is identical to a difference-in-difference-style regression with a post-1995 or post-2000 dummy variable, an IT intensity dummy, and the interaction estimated with annual data for the full period.
78. Becker and others (2005).
79. We also estimated (but do not report) weighted least squares estimates, which are appropriate if the somewhat arbitrary nature of the industry classification system makes measurement error more severe in the relatively small industries. See Kahn and Lim (1998) for a more detailed discussion of weights in industry regressions. These weighted estimates are similar to those reported in table 7.
80. Stiroh and Botsch (2007) report similar results.
81. Stiroh (2006). These results could be consistent with an IT-based explanation if the pervasiveness of IT makes it difficult to identify a link econometrically. That is, if IT is integral for all industries, then measures of IT intensity may not be useful for classification purposes. This view, however, is inherently untestable.
82. Baily (2004) discusses the case study evidence of the impact of competitive intensity on firms' need to innovate and increase productivity and argues that competitive pressure gradually increased during the 1970s and 1980s.
83. Gordon (2003, p. 274). See Nordhaus (2002a) for details on profit trends over this period.
84. Schweitzer (2004).
85. Groshen and Potter (2003).
86. This has been documented by Schreit and Singh (2003) and by Aaronson, Rissman, and Sullivan (2004).
87. The significance of the cross-sectional correlation is robust to dropping the two major outliers-computers and electronics, and information and data systems-on the far left of figures 2 and 3.
88. r-tests for differences in the mean growth rates between the two groups of industries reject the hypothesis that the two had equal growth rates for hours and productivity, but fail to reject the hypothesis that the two had equal output growth rates.
89. As a robustness check, we estimated difference-in-difference regressions and found that industries with a below-median change in the profit share from 1997 to 2001 had a bigger decline in the growth of hours and a bigger increase in the growth of gross output labor productivity than did other industries. No significant difference emerged for valueadded labor productivity growth. We also ran regressions with more detailed measures of intermediate inputs, including energy, materials, and purchased service inputs, as the dependent variable, but those results were uniformly insignificant and are not reported. As a second robustness check, we compared hours, productivity, and output growth for 1992 with the change in the profit share from 1989 to 1991 and found largely insignificant results, suggesting that the latest cyclical episode was different from the previous one.
90. The figures in table 9 are calculated from BLS's quarterly Productivity and Costs data. The definition of nonfarm business in these data includes government enterprises. In contrast, the definition of nonfarm business in BLS's MFP data, the data we use to calculate the labor productivity growth rates in table 1, excludes government enterprises. This slight difference in sectoral coverage explains why labor productivity growth for 2000-06 differs by 0.1 percentage point across the two tables. The same explanation accounts for the slight difference in the average growth rate for 1973-95 between table 1 and the column for nonfarm business in table 10 below.
91. Jorgenson, Ho, and Stiroh (2007) show that such revisions are not unusual; for example, there was a steady stream of upward revisions to productivity growth in the mid-1990s.
92. There are a number of alternative historical series for labor productivity. Although they yield different results in some periods, the patterns of growth and long-run averages are qualitatively similar to the BLS data presented here. For example, see Gordon (2006).
93. In Roberts (2001) the Kalman filter is used to obtain time-varying estimates of trend growth in both potential output and labor productivity. Our implementation first uses a Hodrick-Prescott filter to estimate the trend in hours and then feeds this exogenous trend to the model. Hence we need to estimate a trend only for labor productivity.
94. For other estimates of trend productivity using Kalman filter techniques, see Brainard and Perry (2000) and Gordon (2003).
95. See the appendix to the working paper version of this paper (Oliner, Sichel, and Stiroh, 2007) for details.
96. i) appear in the weights on the capital deepening terms in equation 23, just as they did in the growth accounting equation that applies outside the steady state (equation 1 ). As in that case, we lack the information to specify these asset-specific elasticities. We proceed as we did before, by replacing the theoretically correct weights with standard income-share weights that sum to the same value (one minus the labor share).
97. These are listed in the appendix to Oliner, Sichel, and Stiroh (2007).
98. For the IT-producing sectors, the rate of advance in technology is determined endogenously from the assumed rates of change in prices for IT capital and a variety of other parameters.
99. Jorgenson, Ho, and Stiroh (2007).
100. The horizon in Kahn and Rich (2006) is five years, that in Economic Report of the President 2007 is six years, and that in the March 2007 Macroeconomic Advisers report is eight years.