So we would say this is not a robust piece of evidence. In a Third World country if you replace a horse with a truck , you increase your productivity Jaffe, A.
Khan, M. Skip to content. Human capital. How Robust is the R D productivity Relationship. Public Policy and Resource Allocation. Yet, their empirical results raise doubts about the validity of the model for pharmaceuticals. For consistency and parsimony, these new estimates derive from a methodology that relies on published data, economic theory and panel data econometric techniques.
The paper utilises the new estimates to model research productivity and to re-examine 3 For more detail on the measurement of PPPs, see Schreyer and Koechlin It extends the literature with the construction of two complementary measures of patent quality by utilizing USPTO patents data. Furthermore, international inventor collaboration is a key indicator of patent quality.
This paper is organised in four sections. The empirical assessment of the Lanjouw and Schankerman hypothesis follows in section three. The last section presents a summary and concludes. Such a database is not currently available but its development has become a key objective of official statistical agencies such as the OECD4 and the National Science Foundation in the USA.
However, the accomplishment of this objective would require substantial resources when considering the need for national industry-specific surveys. O'Mahony and Van Ark. Given the lack of industry-specific value added deflators, they employ the GDP deflator as a proxy for the non-labor cost index. Dougherty et al. The Dougherty et al. The series is an input price index for the NIH budget that is dominated by labour compensation of academic and Federal employees.
This is the GDP Deflator approach. Following Dougherty et al. This crude measure is adjusted on the basis of relative value-added growth in pharmaceuticals to arrive at a second measure, RDP2. Third, we draw on economic foundations to subject the above price deflators to econometric scrutiny.
Bloom et al. The authors adopt the GDP Deflator approach and are able to confirm the prediction at the national level. We depart from Bloom et al. We choose to work with the model in first differences and focus on short-term dynamics for three reasons. First, it is unlikely that the three alternative price deflators are independent when considering, for example, that the labour costs index is by construction a major component of RDP1 RDP2 and RDP3.
Hence, the three series could share a common trend that would make it difficult to distinguish between the three. Second, literature concerns about spurious regression cannot easily subside with the use of panel unit root and residual-based cointegration tests given the deficiencies in existing tests.
Hence, all panel series in this study are in first differences. One option is the feasible GLS estimator. Another is the dynamic panel data estimator of Arellano and Bover Third, given that the standard assumption of cross-sectional independence is often rejected, we employ spatial econometrics to obtain consistent estimates of standard errors. We employ the Conley and Driscoll and Kraay GMM estimators that nonparametrically correct for cross-sectional 7 There are two major issues with standard panel unit root tests.
First, they assume cross-sectional independence; this seems implausible as countries often share common shocks Moon and Perron Second, they impose homogeneity since the null hypothesis is that, on average, all cross-section units contain a unit root Strauss and Yigit and Taylor and Sarno The former uses a weighting matrix based on economic distance while the latter is robust to general forms of cross-sectional dependence.
The Appendix provides a detailed account of data sources and variable definitions. Note that we use per capita values for expenditures and output data.
We begin with feasible generalized least squares FGLS estimation to correct for AR 1 autocorrelation within panels, and contemporaneous cross-sectional correlation and heteroscedasticity across panels.
Surprisingly, the industry-wide unit labour cost index also results in a coefficient estimate for price that has the wrong sign. The adjusted labour cost index, on the other hand, leads to a negative and statistically significant coefficient as expected.
Further, a similar result is obtained when the Jaffe-Griliches approach is employed. Consistent with evidence in Dougherty et al. On the first issue, it is important to test for the empirical validity of the cross-sectional independence assumption, given the growing literature emphasis on the geography of 12 Of course, mismeasurement of the non-labor price deflators cannot be excluded as a source of this.
The null hypothesis is rejected in the last two regressions. This is in line with Moon and Perron who question the realism of the cross-sectional independence assumption. Thus, the results in table 1 are suspect, for standard errors are inconsistent in the presence of spatial correlation Driscoll and Kraay In order to correct for spatial correlation, we first employ the Conley OLS estimator. This approach relies on prior knowledge of the structure of temporal dependence but applies a nonparametric correction that accounts for economic distance when measured with error.
The spatial OLS estimation results are summarised in part B of table 1 and are similar to those obtained in part A , though the standard errors are now larger than the standard OLS estimates. Next, we consider the complete model in 2 but ignore, for the time being, spatial dependence. We adopt the dynamic panel data estimation DPD approach pioneered by Arellano and Bond and fully developed by Arellano and Bover The procedure, known as the System GMM panel estimator, exploits information on all series to obtain separate instruments for each lag and each time period, and then uses GMM to weight them.
Table 2 reports robust two- step system GMM estimates of 2. We account for this with a finite-sample correction as in Windmeijer who shows that the correction makes the twostep robust GMM estimator more efficient than the onestep estimator. Also, the AB test results do not question the validity of the model specification. Given the assumption of cross-sectional dependence is violated in table 1, we press on with GMM estimation that is robust to spatial correlation.
This is in order to avoid a bias that arises when the number of instruments approaches the number of observations Bond Helfat discuss the sources of persistence. The results are almost identical to those in tables and are available upon request. Figure 1 shows that the two deflators have followed different paths in the USA.
Lanjouw et al. The Conley GMM estimates s. The pattern is similar in France and Germany but the recovery of the ratio is less pronounced. According to 3 , patent quality would increase and productivity would fall. Various measures of patent quality have been proposed including citations, claims and renewals.
Yet, there are drawbacks with these measures. For example, patent citations and claims may actually be associated with increased competition and litigation and, thus, imply reduced market value Bosworth et al. Alternatively, they may simply relate to growth in the practice of patent citation Hall et al. However, the multi- dimensional index of Lanjouw and Schankerman remains a single all- encompassing index. In this study, we utilise USPTO data on applications and patents granted to arrive at two distinct but complementary indicators of patent quality.
The first measure, q1, builds on the concept of patent grant intensity i. Given that the patent grant lag is about 2 years Hall et al. However, this assumption has been severely criticised in recent literature. Sanyal and Jaffe and Lerner have argued that the US Patent Office has founded increasingly difficult to distinguish between genuine innovation and imitation.
In order to account for this possibility, we adapt the approach taken by Hall et al. On the assumption that the processing of patent applications by the USPTO is free of a country bias, the above criticism does not affect our first measure of patent quality.
The second, q2, is the number of patents granted to inventors residing in different OECD countries as a share of total patents granted to a specific country. This measure is based on the literature of innovation that suggests that research collaboration and formal 19 Visual inspection of the q1 series supports this view since there is no apparent downward trend in the grant intensity ratio at the end of the sample period.
In this study, we minimize the truncation problem with a four-year lag between USPTO data collection and our last observation. Note, the Davidson and MacKinnon exogeneity test statistic DM advises against 20 Perhaps, it may not be apparent how the benefits of collaboration reflect patent quality. This can be accommodated with a broad definition of patent quality that incorporates both technical innovation and market value, as in Lanjouw and Schankerman This finding is consistent with the evidence in Lanjouw and Schankerman We proceed to investigate whether this is due to spatial correlation since the BP statistic clearly rejects the assumption of cross-sectional independence.
This is confirmed when we account for spatial dependence in part B of table 4. We obtain similar results when both measures of patent quality are utilised in table 5. When the model is expressed in first differences, the test translates into the null of the constant being greater or equal to zero. Due to space limitations, the estimation results are not reported here but they are available upon request.
They show that the null hypothesis cannot be rejected. The results are available upon request. The evidence presented in this study of pharmaceuticals is as follows. Third, international inventor collaboration is an important element of patent quality. References Achilladelis, B. Adams, J.
0コメント