I wrote this a few years ago privately, and now that I am no longer employed by the Australian Government, can post it up here. I’m doing so both because it’s interesting in itself, and because I may want to refer to it when discussing ‘additionality’ of all kinds of grants, taxes and subsidies in other posts. I should reiterate that this does not represent the views of any past, present or future employers, and given its age, it may not represent even my view any more.
TL;DR is: Usually yes, but it’s hard to think about let alone measure by how much.
Do Innovation Programs Actually Increase Innovation?
“Productivity isn’t everything, but in the long run it is almost everything. A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker.”
Paul Krugman’s famous quote highlights the centrality of productivity growth to a nation’s material welfare. Having accepted its underlying truth the only remaining question is, ‘from where will this productivity come?’ It is only in recent decades that economists have developed ‘endogenous growth theory’ to explain the innovation which underlies most productivity growth (Romer, 1990). The primary market failure that the resulting ‘innovation’ programs attempt to address is that innovation has ‘spill-over benefits’ that an innovator cannot capture. Wgeb they fail, the original innovator privately bears all the costs; when they are successful they will attract numerous imitators who benefit from their entrepreneurship while reducing their profits. The spill-overs from major advances can be very large and mostly accrue to consumers (Nordhaus, 2004). This will lead to an undersupply of innovation relative to the social optimum. Nonetheless private rewards do exist and a substantial amount of innovation will occur in the absence of any government intervention. This raises the question: how can a government induce additional innovation, without paying for ‘infra-marginal’ research that would have occurred in its absence? This essay will briefly investigate the theory and empirical evidence regarding the ‘additional effect’ (additionality) of three popular mechanisms for promoting research and development (R&D)
- subsidising private sector R&D through subsidies, in particular tax concessions (e.g. in Australia, the R&D Tax Credit)
- directly performing research through universities and public research organisations (e.g. CSIRO)
- awarding funding for specific projects based on selection criteria (e.g. the Green Car Innovation Fund).
Their true impact is challenging to answer and yet crucial to assessing whether innovation programs provide value for money.
Crowding out, crowding in and cost-benefit analysis
Public R&D spending could reduce, or ‘crowd out’, private R&D spending through a variety of mechanisms, most notably,
- public research organisations (PROs) pursuing the most fruitful lines of research, encouraging private firms to imitate them rather than innovate on their own
- PROs and subsidies chasing an inelastic supply of researchers and other research inputs, driving up wages without expanding output
- subsidies inducing private firms to spend less of their own money where the price elasticity of demand for R&D spending by a given firm is low
- the government awarding money to ‘infra-marginal’ research projects that recipients would have conducted anyway, while ‘marginal’ projects go unfunded.
There are also ways public R&D could induce, or ‘crowd in’, private spending so long as it targets marginal rather than ‘infra-marginal’ spending, the elasticity of R&D demand is high, public R&D complements private R&D, or R&D has ‘economies of agglomeration’. The net effect of these will be revealed in the ‘additionality ratio’ (AR), or the amount of extra R&D induced for each dollar of public expenditure.1
For a program to provide any net benefit to the economy, equation 1 below would need to be positive. In Australia the average cost of public funds is estimated to be $1.20 and $1.30 for each dollar raised (Robson, 2005). The marginal cost is probably higher, but estimates differ widely depending on the tax that is increased. Estimates of the spill-overs from R&D also span a wide range and cannot be addressed here (PC Appendix E, 2007). Nonetheless, it is obvious from the equation that the AR will have a large impact on the admissibility of many programs under a cost benefit analysis.
R&D tax concessions aim to increase innovation by lowering the price of R&D investments for firms. Such concessions exist in 20 of 34 OECD countries (OECD, 2010). To illustrate the effect, imagine a firm with the following demand for R&D,
R&D Demand = 100 – 5 * Price of R&D,
facing a constant R&D price of $10, of which $1 is tax (Figure 1). Initially $450 is spent on R&D, net of tax. Now suppose a new government tax concession for R&D expenditure lowers the price to $9, inducing the firm to lift R&D spending by $45, to $495. Ignoring other substitution effects, this costs the government $50 in tax revenue, resulting in an AR of 0.9. The same calculations reveal that a demand curve of Demand = 100 – 7 * Price leads to an AR of 2.1 while the equation Demand = 100 – 3 * Price generates an AR of 0.4. Under perfectly elastic supply, the AR is dependent on the 1) how much R&D the firm would engage in without any concession (less is better) and 2) the price elasticity of the firm’s demand for R&D (higher is better).
Note that if the government had perfect information about the firm’s demand for R&D it could have offered to pay a mere $2.5 to induce the same expansion, resulting in an AR of 18. This highlights an important principle in public policy design: ideally the government would pay firms just enough to make them willing to undertake the desired activity. Anything more is an unintended and costly transfer from taxpayers to recipients. Unfortunately, the government does not know how much to offer, so businesses can exploit this to earn ‘information rents’, a point that will be discussed in more detail later.
Econometric estimates of tax concession ARs use panel data and two kinds of exogenous shock 1) changes to concession rates or eligibility criteria over time 2) estimates of R&D elasticity of demand based on any research input price changes. They are complicated by the fact that elasticities grow the longer firms have to adjust. A review of 23 estimates across the US, Canada, Australia, France and Italy found a mean AR of 0.77 and median of 0.67 (PC Appendix M, 2007), indicating moderate levels of crowding out. Reinforcing the value of subsidising only marginal research, Bloom et al. (2001) found that programs which lower the cost of all business R&D have an average ratio of 0.83 while those which only subsidise expansions above typical or historical levels can be as high as 2.9.
Public Research Organisations (PROs)
Rather than lower R&D costs for private firms, government can employ researchers directly through PROs. Public research into promising new inventions might discourage private firms from working on the same problems, who might instead imitate PRO research. However government control of PROs allows them to focus on research the private sector is least likely to undertake. In light of the market failure described earlier, that would be research that cannot be patented and is easily imitated, with high research costs and little immediate commercial potential. So called ‘basic research’ is much more likely to be additional, and where it reveals new opportunities for commercial research could even be a complement rather than substitute for private sector R&D.
It has been suggested that PROs use up the limited supply of researchers and crowd out private R&D by driving its cost (Goolsby, 1998). However, higher wages will encourage more people to train as researchers, so this is probably only a problem in the short run. In a small country like Australia, this crowding out effect will also be reduced by the pool of foreign researchers who would relocate here as wages rose.
Guellec and Van Pottelsberghe (2001) performed an econometric analysis of different sources of R&D spending over time in a panel of 17 OECD countries, applying a range of controls. It is the largest evaluation of its kind. Across the sample they found that a 10% increase in government R&D funding reduced private R&D expenditure by 3%. The AR for non-university R&D was 0.62, while the AR for universities was 1, implying that university R&D was more distinct from, or complementary to, private R&D. When defence related R&D is excluded, non-university research also has an AR of approximately 1. Guellec and Van Pottelsberghe explain this by observing that defence R&D employs more engineers than scientists and the supply curve for engineers is less elastic. Another explanation could be that defence R&D has fewer knowledge or agglomeration spillovers for civilian R&D. This analysis reveals that local labour supply elasticities and the nature of research undertaken in PROs are key to determining their AR.
Research grants and prizes
The final way to fund R&D would be to offer grants to private firms. A serious concern in this case is a form of adverse selection: firms will most aggressively pursue grants for the large infra-marginal projects that they expect to be most profitable. This may be a successful strategy both because public servants are unable to assess how much the firm would have spent in the absence of the grant, and because the public sector would like to be seen to fund commercially successful projects (Klette et al., 2000). Finally, private organisations may be more willing to pay for the best negotiators, to assist them in seeking rents from the government. If a grant does not allow the recipient to escape some form of borrowing constraint, the firm may substitute public funding for their own dollar for dollar, resulting in an AR well below 1. Governments have used many approaches to avoid these problems, for instance setting criteria to target less profitable projects, requiring large co-investment and structuring grants as loans or equity investment to attract financially constrained firms.
Empirical estimates of the AR of grants, using a range of techniques, have varied widely, but in many cases produced estimates above 1 (PC Appendix M, 2007). One recent program evaluation in Australia (CIE, 2003) took two approaches, 1) surveys asking firms about additionality and commercial returns, 2) comparison with a control group of rejected applicants. Firm survey responses claimed a high AR but this was inconsistent with their reports that projects were highly profitability. Comparison with the control group indicated the grants had no impact, though the selection of the control group was heavily disputed during peer review. This is typical of the methodological difficulties in assessing grant programs and the conflicting indicators of their effect.
For a variety of reasons public R&D expenditure may increase total R&D by much more or much less than public outlays would suggest. As demonstrated above different approaches can have very varied impacts, and efficient use of public money requires that serious thought is given to the likely additionality of programs.
1 A figure of 1 would mean on net neither crowding out nor crowding in, while a figure of 0 would mean full crowding out with no additional R&D.
Bloom, N., Griffith, R. and Klemm, A. (2001) Issues in the Design and Implementation of an R&D Tax Credit for UK Firms, The Institute of Fiscal Studies
Centre for International Economics (2003) Review of the R&D Tax Concession Program, Prepared for Department of Industry, Tourism and Resources.
Commonwealth Productivity Commission (2007) Public Support for Science and Innovation Research Report Appendix M. Accessed online: http://www.pc.gov.au/__data/assets/pdf_file/0016/37123/science.pdf
Goolsbee A. (1998) Does Government R&D Policy Mainly Benefit Scientists and Engineers? American Economic Review 88(2)
Guellec, D. and Van Pottelsberghe de la Potterie, B. (2001) R&D and Productivity Growth: Panel Data Analysis of 16 OECD Countries. Directorate for Science, Technology and Industry, STI Working Papers 2001/3, OECD.
Klette, T., Moen, J. and Griliches, Z. (2000) Do Subsidies to Commercial R&D Reduce Market Failures? Microeconometric Evaluation Studies, Research Policy, vol. 29
Nordhaus, W., (2004) Schumpeterian Profits in the American Economy: Theory and Measurement. NBER Working Paper No. 10433
OECD Industry and Entrepreneurship (2010) R&D tax incentives: rationale, design, evaluation. Research note. Accessed online: http://www.oecd.org/dataoecd/61/13/46352862.pdf
Robson, A. (2005) The Costs of Taxation, Perspectives on Tax Reform, CIS Policy
Monograph 68. Accessed online: http://www.cis.org.au/images/stories/policy-monographs/pm-68.pdf
Romer, R., (1990) Endogenous Technological Change. The Journal of Political Economy. Vol. 98, No. 5.