; 46040_1_Knowledge-creation-model-II.docx - Transtutors
Learning Center
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

46040_1_Knowledge-creation-model-II.docx - Transtutors


  • pg 1
									Knowledge creation model II:

The model accounts for sector specific characteristics such as the technological opportunity
and the overall dispersion among European regions as well as region specific characteristics
such as the size and the endowment of knowledge sectors etc, which assumed to positively
influence the patenting activity of a region.

In sum, externalities arising from industrial diversity as well as externalities resulting
from industrial specialization positively influence the region’s knowledge creation
capacity. However, regional innovation externalities may differ from industry to
industry (within the high tech sector). The following high-tech industries are considered
in the analysis: Computer and electronics; Micro-organism and genetic engineering;
Aviation; Communications technology; Semiconductor; Lasers;

Specificities of the model:

The model is tested on an extended sample of 186 European regions mainly composed of
NUTS II regions that covers the entire European Union. For high-tech sector in tota and for
each of the 6 high-tech industries.

Since the dependent variable of the model has a discrete nature with an important proportion
of zeros, the use of conventional linear regression models may by inappropriate. Generally,
in order to deal with the discrete and non-negative nature of the patent dependent variable, the
simple Poisson regression model is considered as in the study of Feldman and Audretsch
(1999). However, an important shortcoming of the Poisson model is its implicit assumption of
equality between the first two conditional moments.

Given the evidence of overdispersion and the rejection of the Poisson restriction, the model is
estimated by allowing for mean-variance inequality. The adopted approach consists in
applying the Count Models for panel data: Poisson quasi-maximum likelihood estimator
(QMLE). I attach the presentation in PPP by my Professor1

It is worth noting that the interpretation of estimation results of non linear models such as the

Poisson, the Tobit or the negative binominal model is somewhat different from those
obtained by linear regression models. However, when the explanatory variables enter
logarithmically, then the coefficients can be interpreted as elasticities giving the percentage
change of the expected value of the dependent variable for a 1 % change of an explanatory
variable. For this reason the independent variables of equation have been expressed in
logarithmic form.

1 (This robustness of the Poisson QMLE is a special case of the more explicit analysis of
quasi-maximum likelihood estimation in the linear exponential family by Gourieroux,
Monfort and Trognon (1984).
Pij=a1 + a2HTMSijt-3 + a3 HTKSijt-3 + a8 NHCi + a9HMTEi t-3 + a10POPi + a4OKj + a5TDPj +
a6GDKi t-1 + a7GDpi t-1 + t-3 + + εij

Where ..

i– region; j- sector;

Pij     stands for the average number of patent applications over the period 1998-2007 of
region i and sector j to the EPO and proxies innovation output.


HTMSij regional production specialization in high-tech manufacturing sector 1998-2007

HTKSij regional production specialization in high-tech knowledge intensive services 1998-


GDK i regional measures of high-tech production diversity in 1998-2007

NHCi t-3 number of companies in high-tech in 1998-2007 of region i


HTMEi average ratio of employment in high-tech (HTEi) and medium-high (MHTEi) sectors
over the period 1998–2008 over total employment3;

POPi stands for the average over the period 1998-2008 of total population and is introduced
as a size control variable;


        technological opportunity/intensity indicator in 1998-2007

GDPi innovation diversity indicator (in specific high-tech patent groups) in 1998-2007

2 Determined time lag between specialization/diversity (S) and P in high-tech sectors is about 1-3 years
3 Assuming that any adjustment in employment due to a technology change is gradual, the technology
variable is lagged by three time periods.
TDPj spatial dispersion of high-tech industrial activity in 1998-2007 in EU

εij is a random error term.

The model should be estimated for the entire high-tech sector as well as for its six
industries separately: Aviation, Computer and electronics, Micro-organism and genetic
engineering, Communications technology, Semiconductor, Lasers.

I would also want you to first estimate the base models and then consider some time
lags, that I have introduced in my equation above.

The model results should be explained and interpreted in the in regards to correlation
between Pij and independent variables: HTMSij , HTKSij , GDPi , NHCi , HTMEi , POPi, OKj
and TDPj
for high-tech alltogether and each of the industry seperately.

To top