Productivity and Propensity: The Two Faces of the R&D Patent Relationship

Transcription

1 Université libre de Bruxelles (ULB) Solvay Brussels School of Economics and Management (SBS EM) European Center for Advanced Research in Economics and Statistics (ECARES) Productivity and Propensity: The Two Faces of the R&D Patent Relationship G. de Rassenfosse Patent workshop IPTS, May 21

2 «One of the longest lasting debates in the history of economic measurement has been whether the noise and the biases in patent count measures can be made small enough to make patents counts useful measures of innovative output in economic studies» Lanjouw et al. (1998) G. de Rassenfosse 21 2

3 Who dare to interpret patent statistics? Patents per mio R&D, Notes: Average number of priority filings per R&D expenditure (1988 to 1992). Source: Own computation, OECD MSTI Is Germany more than twice as productive as Belgium? G. de Rassenfosse 21 3

4 Patent statistics are blurred by the propensity to patent R&D Inventions Patents Productivity Propensity Objective of this study: Estimate the propensity component in order to clean patent statistics We propose a methodology to identify the productivity and the propensity effects G. de Rassenfosse 21 4

5 The objective is to decompose the patent to R&D ratio into its components of productivity and propensity Productivity component (hypothethical) Patents per mio R&D, Propensity component (hypothetical) G. de Rassenfosse

6 Existing solutions are of two types Tricky, because the number of inventions is unobserved. No study has explicitly solved the identification problem Two commonly adopted approaches for meaningful patent statistics: [A] Count only high value patents (e.g., triadic patents, see Dernis et al., 21; Grupp and Schmoch, 1999) [B] Patent counts weighted by some value indicator (e.g. renewal and family size, see Schankerman and Pakes, 1986) G. de Rassenfosse 21 6

7 The count of triadic patents is appealing Triadic patents per mio R&D, Notes: Average number of triadic patents per R&D expenditure (1988 to 1992). Source: OECD MSTI de Rassenfosse and van Pottelsberghe (29) have shown that triadic patents are more reflective of a productivity effect Patent to R&D ratio is still difficult to interpret Arbitrary G. de Rassenfosse 21 7

8 The solution proposed in this paper use differences in the use distribution of patent value to solve the identification problem Under two general assumptions, one can show that the distribution of patent value can be used to identify the effects A1. The distribution of invention value is universal A2. The probability to patent a high value invention is similar across units Methods [A] and [B] can be seen as a special case of the model Let us have a look at a (simplistic but) intuitive example G. de Rassenfosse 21 8

9 Consider two types of invention value R&D expenditures lead to inventions, which lead to patents: λ = productivity parameter δ = propensity parameter Two types of invention value: I L and I H Proportion of low (high) value inventions: σ L (σ H ) The probability to patent depends on the type: π L and π H Explicitly, number of high value inventions: Number of high value patents: G. de Rassenfosse 21 9

10 Identification problem since σ and π are unobserved The propensity to patent: The distribution of invention value (σ s) and the probabilities to patent (π s) are unknown, so δ cannot be estimated. We need identifying assumptions A1. The distribution of invention value is universal (σ) A2. The probability to patent a high value invention is similar across units (π H ) Still no explicit solution for δ (and λ) G. de Rassenfosse 21 1

11 However, it is possible to recover relative rates Let us begin with the productivity rates (λ) We have two countries (1 & 2) Relative productivity levels: Under A1: Under A2: Simplifying Relative ratio of high value patents G. de Rassenfosse 21 11

12 However, it is possible to recover relative rates Now with the propensity rates (δ) Remember, we have: So the relative rates: Hence we can write: Simplifying Relative share of high value patents G. de Rassenfosse 21 12

13 The paper explains the methodology in greater details Easy to see why the count of triadic patents is a special case of the model Remember the productivity parameter: Comparing the triadic to R&D ratio across countries is a measure of the relative productivities (i.e P H = Σ triadic patents) Now, we understand the ins and outs of this approach In the paper: Continuous (i.e. infinitely many value classes) Derive a criterion to identify the value threshold (i.e. P H ) Empirical application for six industries and four countries over the period 1988 to 199 Patent value is captured by the family size G. de Rassenfosse 21 13

14 A detailed empirical application at the industry level Input: GERD in million 2 USD PPP OECD/MSTI Output: Count of priority filings (fractional allocation by inventor country) de Rassenfosse, Dernis, Guellec, Picci and van Pottelsberghe, forthcoming soon (hopefully) Level of analysis: cross country comparison at the industry level (BE CHEM against DE CHEM) This presentation: rough estimate at the country level for ease of exposition G. de Rassenfosse 21 14

15 A (quick and dirty) empirical application at the country level Remember our primary objective? Let s do it! Productivity component (hypothethical) Patents per mio R&D, Propensity component (hypothetical) G. de Rassenfosse

16 A (quick and dirty) empirical application at the country level Now with real data Patents per mio R&D, Productivity component (real data) Propensity component (real data) Germany has a value of 1 by construction (reference country) G. de Rassenfosse

17 Wrap up The paper proposes a methodology to decompose the R&D patent relationship into its components of productivity and propensity Results should be appealing to policymakers (monitor) and scholars (improved patent statistics) What if the assumptions are violated? Mildly violated: still a better indicator than straight patent count Strongly violated: then in any case the use of patent statistics can be seriously questioned! Next step: more countries, more recent data, more proxies for patent value, more validity checks G. de Rassenfosse 21 17

18 Thank you G. de Rassenfosse 21 18