( 9 ) d i s t i j = d i s t j i
Methodology and Data
The model is estimated as a dynamic continuous time panel through the ESCONAPANEL program developed by Cliff Wymer (2002). Continuous time estimates have the appealing proprieties of skipping the problem of nonstationarity of the series for the treatment of residuals in continuous time and for the correspondence of stochastic systems of differential equations with the ones stated in discrete time, as shown in Phillips (1991), Gandolfo (1981) and Wymer manual (2002). Also the problem of autocorrelation of disturbances is correctly treated in continuous time especially with systems of mixed stock and flow variables such as in our case (Gandolfo (1981) and Maggi et al. (2002)).
We consider nine European countries, the United States, and Japan. We use a panel data for 1988-1998 period. Due to limitations in data availability on services we consider the following countries in Europe: Austria, Germany, Denmark, Finland, France, United Kingdom, Italy, The Netherlands, and Sweden. We consider United States and Japan as representative of the “rest of the world.” Data9 on output (GDP), services and human capital are taken from various OECD databases. Data on physical capital and labor are taken from the Penn World Tables (Summers and Heston, 1991). Data on ICT expenditures refer to gross fixed capital formation in Information and Communication Technologies and are taken from EUROSTAT. Data on the bilateral technology flows (Patij) are taken from the U.S. patent office and are represented by the citations in the patents between countries. The use of patents as a measure of innovation is now standard in the literature (see Eaton and Kortum, 1996 for a review of the international patent system). The use of patent citations rather than patent applications or patents granted has the advantage to have a bilateral dimension that allows capturing technology transfers. As mentioned above, citations received from country a by country b indicates a transfer of technology from the latter to the former. Citations internal to one country are not treated as technology transfers. Citations may be backward or forward if referred respectively to inventions discovered in the past or, from the point of view of the cited (source) country, in the future. This is not irrelevant if one wants to evaluate the transfers of technology with a limited time series given the risk to neglect potential citations in the initial and final part of the series. To cope with this problem we follow the method indicated by Hall, Jaffe, and Trajtenberg (2001) where it is suggested to divide each citation by the average number of citations received by other patents in the same cohort (fixed approach)10. Data on regulation are from Nicoletti, Scarpetta, and Boylaud (2000) and refer to product market regulation. Data for the structure
A more detailed description of data sources is reported at the end of the paper.
10 Indeed other methods, named structural, are suggested. They refer to a specific function to be estimated that should fit with different distorting effects to be eliminated (such as pure time effect, field effect etc). This method, while more formally appealing in its specification, embeds some strong hypothesis in the definition of the function to be used. For this reason we adopt the fixed approach.