Empirical analysis of the effects of R&D on measurement?

advertisement
OECD workshop on productivity measurement and analysis
Bern, Switzerland
16-18 October 2006
Empirical analysis of the effects of R&D on
productivity: Implications for productivity
measurement?
Dean Parham
Motivation

Empirical uncertainty about magnitude of R&D’s effect on
productivity
 Shanks & Zheng (2006), Econometric Modelling of R&D and
Australia’s Productivity, Productivity Commission Staff Working
Paper
 Not just this study. Widespread through other studies/countries

Certainty about magnitude of effects will be implicit in national
accounts if proposals to capitalise R&D are implemented
 Canberra II group recommendations
 R&D capital would be incorporated into productivity estimates

Is there a problem here?
2
Outline

Formation of R&D capital stocks

The Shanks & Zheng study

Why the empirical uncertainty?

Capitalisation of R&D in the national accounts

Concluding remarks
3
1.
Formation of R&D capital stocks

R&D outputs are largely unobservable

Knowledge assets measured by use of R&D inputs

 Implicit assumption of constant relationship between
R&D inputs and R&D outputs  ie constant
productivity of R&D
Accumulated via the perpetual inventory method
(PIM)
K Rt 1  (1  )K Rt  R t
4
Business R&D capital stocks: levels
Index
120
15 per cent
5 per cent
100
10 per cent
80
60
40
20
0
1968
1976
1984
1992
2000
5
Business R&D capital stocks: annual growth
15%
15 per cent
5 per cent
10%
10 per cent
5%
0%
-5%
1968
1976
1984
1992
2000
6
Domestic and foreign business R&D stocks
120
15%
100
Australia
10%
80
60
5%
Foreign
40
Foreign
Australia
0%
20
0
-5%
1968
1976
1984
1992
2000
1968
1976
1984
1992
2000
7
Characteristics

Generally smooth

Timing and extent of growth in domestic v. foreign
stocks

 R&D tax concession
Change in structure of R&D
  business
 shift to services
 firm entry
8
2.
The Shanks & Zheng study

Conventional framework

Cobb-Douglas specification
ni
lnY  lnA  llnK o  2lnK R  lnL  t   WlnZ
i
i
i 1

‘Two step’ transformation
9
Estimation of standard models




Models with limited controls mis-specified
Models with extended controls OK
 returns to R&D  point estimates of 60%, but imprecise (include
zero)
 negative coefficient on either domestic or foreign stock commonly
found
 other explanators more robust  human capital, ERAs,
communications infrastructure, ICT, decentralised wage bargaining
Dynamics and lags
 little improvement
Sensitivity testing on PIM depreciation rate
 Variation in implied returns, but no improvement in precision
10
Further exploration

Specification in growth form
 elements of endogenous growth
 continuation of mixed results

Two equation specification
 separate specifications for determinants of domestic
R&D and for determinants of productivity
 showed more promise
 indications that foreign R&D had positive effect via
domestic R&D as well as directly
11
Summary


Effect of R&D on productivity hard to pin down
 Mis-specification in standard models
 Imprecise estimates
 Sensitive to reasonable changes in model and variable
specification
Some reasonable models and robust explanation from
other factors
12
3.
Why the empirical uncertainty?

Generic

 limited degrees of freedom
 multi-collinearity
 measurement problems
Country and period specific
 shocks to R&D and to productivity
 policy changes and ‘phantom’ effects of the R&D tax
concession
13
Measurement: Use of constructed variable to
proxy R&D knowledge asset

Smoothness of change. Contributed by two principal
assumptions

Constant productivity of R&D
 across projects  single price deflator on R&D inputs
subsumes differences in value of R&D outputs
 across time  same real input use generates same
increment to stock in all periods.

Constant (or at least steady change in) depreciation
rates
14
Criticisms

R&D outputs highly heterogeneous. Not same
price/value

Productivity of R&D affected inter-temporally by:
 technological opportunities
 organisation of R&D
 policy changes in Australia
Depreciation of knowledge
 diversity in depreciation rates
 changes in R&D composition affect average depreciation
 interactions lead to increasing returns and discontinuities

15
4. Capitalisation of R&D in the national accounts

Same essentials  use of the PIM

Open to similar criticisms

 concerns about accuracy of measurement of R&D-based knowledge
stocks
Flow-on effects
 R&D capital enters capital input measure in derivation of
productivity estimates
 deterministic effect on productivity
 smooth effect on productivity growth  smooth change in R&D
stocks
 small effect?  relative size of R&D capital and conventional
productive capital stock, relatively high rental price weight
16
Criticisms

Doubtful accuracy

 ‘Conservative’ but not accurate
R&D not the only form of knowledge accumulation

Different views on how knowledge relates to
productivity
 not just like a physical asset
17
Doesn’t look good, but ….

Problems in current procedures
 R&D expensed
 underestimate value added
 particular relationship between R&D and productivity is imposed by
default

Choose between the ‘lesser of two evils’
 current: incorrect MFP, errors related to size of current R&D
expenditure and to its expensing in the accounts
 proposed: inaccurate but ‘smoothed’ effect on MFP, errors related to
mismeasurement of knowledge and rental prices and to limitations of
specification of relationship between knowledge and productivity
18
5. Concluding remarks

Capitalisation may be lesser of the two evils

But that does not make it right

Transparency to assist users
 limitations
 assumptions
 choice?

Communication to improve broader understanding
19
Download