- Work in Progress Inventor mobility and regions' innovation potential Riccardo Cappelli, U Insubria Dirk Czarnitzki, K.U.Leuven and ZEW Mannheim Thorsten Doherr, ZEW Mannheim Fabio Montobbio, U Insubria and Bocconi Introduction • In knowledge-based economies, human capital and innovation are usually seen as key driver of wealth and growth – „new growth theory“, see e.g. Aghion and co-authors • How to measure „knowledge“ that is present in an economy or region? • To what extent does knowledge contribute to growth? „Technology gap models“ Technology gap models attempt to explain growth (or „catching-up) in income per capita in economies or regions by • changes in knowledge stocks or innovation (see e.g. Fagerberg, 1994 in JEL for an overview) • and other common controls, e.g. – Lagged income per capita – investment into physical assets (change in stock of physical assets) – Size of the region or economy (usually measured by population) Technology gap models How to measure knowledge or innovation? • Scholars have used R&D expenditure to proxy the change in knowledge stocks of regions – e.g. Verspagen and Fagerberg, 2002, Research Policy • Later substituted or augmented by patent applications – Patents measure inventions but not innovations – Patents could generate a premium as they approximate „successful R&D“ or „valuable knowledge“ to a certain extent • As the value distribution of patents is very skewed, scholars have also used forward citations as proxy for patent value – Trajtenberg 1990, Hall et al., 2005 Measuring knowledge continued • Knowledge spillovers at both macro and micro level are important to explain the relative growth performance – Grossman and Helpman, 1991; Griliches, 1992 • Knowledge Spillovers are geographically localized – Jaffe et al., 1993; Bottazzi and Peri, 2003; Maruseth and Verspagen, 2002; Peri, 2005 • There are some factors that can explain the geographically localized diffusion of knowledge: – importance of face-to-face contacts to spread tacit knowledge – labor market (Almeida and Kogut, 1999) – inventor mobility and co-invention networks (Breschi and Lissoni, 2009) Measuring knowledge spillovers • Frequently, scholars have tried to control for knowledge spillovers“ using patent citations • Justified in US studies, as USPTO applies „duty of candor“ – Patentees have to cite all relevant prior art in the patent applications • At EPO, however, most citations are added by examiners – Citations as measure of knowledge flows and thus value of knowledge are questionable – Patentee might not have been aware of existing knowledge during the inventive process Our approach • Knowledge is embedeed in people • Thus, inventor mobility is a more direct measure of knowledge flows • Challenge: how to measure inventor mobility (see e.g. Trajtenberg‘s NBER WP „The name game“) – Name homonyms – Spelling variations and so forth Our approach: inventor mobility index that has just been presented by Thorsten. Data • 20 Italian regions from 1995 to 2007 • Dependent variable: %-growth of GDP per capita • Variables based on the inventor mobility index: – Intra-regional: inventor that „change jobs“ (switch applicants) within the same region. – Inter-regional inflow: incomnig inventors that change jobs and move to region i from a different region. – Inter-regional outflow: inventors formerly employed in region i that now move to a new job in a different region. – Inter-regional net inflow: difference between inflow and outflow. all mobility figures enter regions as ratio: mobility relative to stock of inventors in t-1 (derived by the perpetual inventory method with 15% of obsolescence rate) • (Stock is corrected for double counting of inventors) Data Controls: • GDP/Capita in previous period • Total R&D expenditure (public and private) per capita change in „knowledge stock“ • Patent applications per capita as proxy for „successful R&D“ change in stock of successful R&D • Investment into physical capital per capita in previous period (change in asset stock) both variables measured in million EUR in real terms (GDP deflator) Descriptive Statistics Tab. 1 Descriptive Statistics Variable gdp per capita population Capital/POP Patent applications/ Total R&D exp. Total R&D/ POP Obs Mean Std. Dev Min Max 240 0.0202 0.0051 0.0116 0.0283 240 2874971 2278932 117063 9545441 240 0.0044 0.0012 0.0022 0.0081 240 0.2541 0.1998 0.0151 1.2774 240 0.0002 0.0001 0.0000 0.0005 Variable Gdp per capita growth log(gdp/pop) t-1 log(pop) t-1 log (Capital/POP) t-1 (Patent applications/ Total R&D) t-1 log(Total R&D/POP) t-1 Intra regional t-1 Inter regional Inflow t-1 Inter regional Outflow t-1 Inter regional Net Inflow t-1 Obs Mean Std. Dev Min 240 0.011 0.016 -0.031 240 -3.949 0.271 -4.481 240 14.448 1.058 11.667 240 -5.490 0.282 -6.158 240 0.249 0.205 0.006 240 -8.870 0.680 -10.856 240 0.003 0.012 0 240 0.003 0.011 0 240 0.003 0.008 0 240 0.000 0.011 -0.080 P.S.: the values are expressed in millions of euro. Max 0.057 -3.544 16.064 -4.818 1.277 -7.655 0.145 0.127 0.080 0.079 Regression Results Tab. 2 Estimation results (OLS, Cluster standard error) Variables Model 1 log(GDP/POP) t-1 -0.039 log(POP) t-1 log (Capital/POP) t-1 (Patent applications/ Total R&D exp.) t-1 Model 2 ** ** -0.038 (0.146) (0.013) (0.013) -0.000 -0.000) -0.000 (0.001) (0.001) (0.001) 0.011 0.011 0.011 (0.008) (0.009) (0.009) 0.015 *** (0.005) log(Total R&D/POP) t-1 -0.038 Model 3 0.006 (0.003) 0.013 *** (0.004) * 0.006 0.013 ** *** (0.004) ** 0.005 (0.003) (0.003) 0.088 0.080 (0.096) (0.089) * Mobility Intra regional t-1 Inter regional Inflow t-1 0.072 (0.062) Inter regional Outflow t-1 -0.138 ** (0.064) Inter regional Net Inflow t-1 0.097 (0.044) Time year dummies Yes Yes Yes Number of observations 240 240 240 0,499 0,507 0,507 R-squared Notes: Year 1996-2007, 20 Italian regions; * < 0.1, ** <0.05, *** <0.01 ** Very preliminary conclusions…. • Inventor mobility appears to explain a change in GDP growth among Italian regions • To-Do: – Employ a revised version of the inventor mobility index • According to the new version of the algorithm there is more mobility among regions – Try to collect more data to enable controlling for region fixedeffects – Generate patent forward citations to control for heterogeneity in value of patents • More recent patent data required – Try to handle potential endogeneity of measures such as R&D, patenting and inventor mobility.