Bias in Aggregate Productivity Measures

1
Bias in Aggregate Productivity Measures

• Mike Harper

2
I will be talking about three papers

• A paper published in 1999
• Todays main paper, which was published this past
  Spring
• A new research paper trying to explain a new
  issue identified in todays paper

3
The Old Paper
4
Monthly Labor Review February 1999
5
Many experts felt measured aggregate productivity
trends were too low paper investigates

• Uses a model to break out productivity by
  industry
• Uses the same data as is used in the aggregates
• Some industry trends are negative and implausible
• Possible explanation measurement error

6
Evsey Domars Productivity Model
Final outputs
YMD
YSD
Manufacturing (M)
Nonmanufacturing (S)
YMI
YSI
KM
LM
KS
LS
Primary inputs
7
1999 Papers Findings

• All of the MFP growth from 1979-1996 came from
  manufacturing
• Some detailed non-manufacturing industries must
  have had, and indeed did have negative MFP trends
  (1977-1992)
• Some of these negatives were too large and too
  negative to be plausible
• Measurement error was likely. Real output
  measures (not nominal ones) were singled out as
  the most likely source

8
Todays Paper
9
Monthly Labor Review March 2002
10

• Table 3. Multifactor Productivity (MFP) Growth in
  the U.S. Private Business (PB) Sector and the
  Contributions of Labor Composition Effects (LCE),
  Manufacturing (Mfg) MFP Growth, and
  Nonmanufacturing (Non-mfg) MFP Growth

PB PB Unadj. Mfg Contributions MFP
LCE PB-MFP MFP Mfg Non-mfg
1949-1999 1.4 0.2 1.6 1.2 0.6 1.0 1949-1973
2.1 0.2 2.3 1.5 0.8 1.5 1973-1979 0.6
0.0 0.6 -0.6 -0.3 0.9 1979-1990
0.5 0.3 0.8 1.1 0.5 0.3 1990-1995
0.6 0.4 1.0 1.2 0.5 0.5 1995-1999 1.3
0.3 1.6 2.5 1.0 0.6
11
Industry Results in Bias Revisited

• Paper contains estimates of long term MFP trends
  for a comprehensive set of 25 non-manufacturing
  industries in the United States business sector
  for 1963-1977 and 1977-1997
• Some of the MFP trends are still negative
• BLS is skeptical that these are plausible
  indicators of productivity

12

• Table 6. Effects of adjusting output,
  sufficiently to produce zero industry MFP growth,
  on private business MFP

Total effects 0.34 Construction 0.12 Insurance
carriers 0.08 Health services 0.05 Credit
agencies, etc. 0.03 Auto repair, etc. 0.03
13

• Are U.S. productivity trends still understated?
• Perhaps, but less likely with the new data
• Is there still a problem with service sector real
  output trends?
• Most definitely for several
• What, besides service output, could account for
  high MFP in manufacturing and low MFP elsewhere?
• An imbalance between mfg and services
• An overestimate of capital growth
• Candidate an overestimate of high tech
  quality

14
How much impact are high tech goods having on the
U.S. productivity story?

• Oliner and Sichel identify two contributions of
  computers to aggregate MFP
• The effects of the use of computers
• (Inputs of computers, as evaluated with the Solow
  productivity equation)
• The effects of producing higher quality computers
• (Outputs of computers, as evaluated with the
  Domar model)

15
News ReleaseMultifactor productivity trends,
1948-2000

• Internet address
• http//www.bls.gov/mfp
• USDL 02-128
• Tuesday, March 12, 2002

16

• Table B. Compound average annual rates of growth
  in output per hour of all persons and the
  contributions of capital intensity, labor
  composition, and multifactor productivity,
  private business, 1948 to 2000 (percent per year)

1948-73 1973-79 1979-90 1990-95 1995-2000
Output per hour 3.3 1.3 1.6 1.5 2.7
Contribution of capital intensity 0.9 0.7 0.8
0.5 1.1 Contribution ofinformation processing
equipment and software 0.1 0.3 0.5 0.4 0.9
Contribution of all other capital services 0.8
0.5 0.3 0.1 0.2 Contribution of labor
composition 0.2 0.0 0.3 0.4 0.3 Multifactor
productivity 2.1 0.6 0.5 0.6 1.4
17

• Table 7. Contributions of Manufacturing
  Industries to Private Business Multifactor
  Productivity

SIC Industry 49-73 73-79 79-90 90-95
95-99 20 Food 0.09 0.01 0.03 0.05 -0.02
21 Tobacco 0.00 -0.01 -0.03 0.02 -0.05 22
Textiles 0.06 0.06 0.03 0.02 0.02 23
Apparel 0.02 0.04 0.01 0.01 0.02 26
Paper 0.03 -0.03 0.00 0.00 0.03 27
Printing 0.01 -0.02 -0.03 -0.04 -0.02 28
Chemicals 0.11 -0.13 0.04 -0.01 0.04 29
Petroleum 0.03 -0.03 -0.01 0.01 0.02 30 Rubber 0
.02 -0.04 0.03 0.03 0.03 31 Leather 0.00 0.00 0.0
0 0.00 0.00 24 Lumber 0.03 0.01 0.04 -0.02 -0.01
25 Furniture 0.01 0.00 0.01 0.01 0.01 32 Stone,
Clay and Glass 0.02 -0.03 0.02 0.01 0.01 33 Prima
ry Metals 0.02 -0.11 0.01 0.02 0.04 34 Fabricated
Metals 0.02 -0.05 0.02 0.03 0.00 35 Industrial
Commercial Machinery 0.04 0.01 0.20 0.16 0.35 3
6 Electrical Machinery 0.08 0.05 0.13 0.23 0.34
37 Transportation Equipment 0.12 -0.05 0.01 0.03 0
.09 38 Instruments 0.03 0.03 0.04 0.00 0.02 39 M
iscellaneous Manufacturing 0.02 -0.01 0.01 0.00 0.
01 Total Manufacturing Contribution 0.77 -0.32 0
.57 0.55 0.93 Private Business
Sector Multifactor Productivity
2.10 0.60 0.50 0.60 1.30
18

• We find that 2/3 of the speedup in labor
  productivity between the early and late 1990s
  comes from high tech goods
• Calculations during 2002 by Ho, Jorgenson, and
  Stiroh indicate investments in high-tech account
  for 7/8 of the speedup in recent years
• Calculations by Oliner and Sichel in early 2002
  over-explained the speedup with high-tech
• The estimates of both the input and output
  contributions of computers rest on the large
  quality adjustments being made by U.S.
  statistical agencies
• I had been criticized on a draft of todays
  paper, a year ago, for suggesting that the
  quality adjustments could be problematic

19
The New Paper
20
Technology and the Theory of Vintage Aggregation

• Michael J. Harper
• Bureau of Labor Statistics
• July 3, 2002
• This paper has been prepared for discussion at
  the NBER/CRIW Summer Institute session, July
  29-31, 2002

21
Capital Input Measurement

• Obtain nominal data on investment
• Create a quality adjusted price index
• Deflate the investment with the price index
• Aggregate past investments using the perpetual
  inventory method weight past real investments
  with relative marginal products

22
Fig 2 A Family of Machine Functions
Output
f
g
h
Labor Hours
23
Fig 7 A Family of Homogenous Machine Functions
Output
A
B
Labor Hours
24

• Quality adjustment of investment prices assumes
  market prices of new capital goods of various
  models reflect their marginal products
• But marginal products of assets will decline due
  to obsolescence
• Older models will be affected proportionally more
  than newer ones
• In neoclassical theory, market prices reflect
  future returns, not just initial marginal
  products
• Quality adjustments are too large!

25
Fig 8 Tracing Vintage Marginal Products and
Prices
Price
Marginal Product
5 4 3 2 1
3 2 1
3 2 1
Year
Year
26
Capital Measurement Units

• Currently we define the capital quantity unit in
  terms of goods of given characteristics
• This treats capital like apples
• This rationalizes accumulating quality changes
  without removing any obsolescence
• Better quality capital is more capital, at any
  point in time
• But we also need to remove the temporal effects
  of obsolescence

27
Implications for Measuring the Outputs of Capital
Goods

• Capital investment gets a special status in being
  counted in GDP. (A status not granted
  intermediate inputs)
• Rationale for special status To credit GDP in
  the current period for activities that will
  enhance future consumption possibilities
• Present measurement methods try to identify and
  measure the capital goods
• Problem with defining the measurement unit in
  terms of the goods We cannot eat capital

28
Implications for Measuring the Outputs of Capital
Goods

• We need to count real savings and real investment
  in terms of either
• The marginal consumption foregone or
• The marginal increment to future output
• Consistent with accounting for obsolescence in
  capital inputs, we should recognize that, over
  time, obsolescence reduces the amount of real
  investment required to obtain specific durable
  goods

29
The Vintage Paper Supports the Conclusions of
Todays Paper

• U.S. statistics may still underestimate the real
  output trends in some service industries
• U.S. statistics may overestimate the quality
  adjusted real output trends in high tech
  industries
• Consistent with this, the U.S. output trends
  could be close to correct, but the underlying
  data may assign too much credit to the high tech
  sector and too little credit to some service
  activities