Co-evolutionary patterns in regional knowledge bases and

Co-evolutionary patterns in regional knowledge bases
and economic structure: evidence from European
Francesco Quatraro
To cite this version:
Francesco Quatraro. Co-evolutionary patterns in regional knowledge bases and economic structure: evidence from European Regions. 2013. <hal-00992080>
HAL Id: hal-00992080
Submitted on 16 May 2014
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destin´ee au d´epˆot et `a la diffusion de documents
scientifiques de niveau recherche, publi´es ou non,
´emanant des ´etablissements d’enseignement et de
recherche fran¸cais ou ´etrangers, des laboratoires
publics ou priv´es.
Co-evolutionary patterns in regional knowledge bases and economic
structure: evidence from European Regions
Francesco QUATRARO
University of Nice Sophia Antipolis, GREDEG-CNRS
250, rue Albert Einstein
06560, Valbonne (France)
Tel: +33 4 93954373
Email address: [email protected]
BRICK (Bureau of Research on Innovation, Complexity and Knowledge), Collegio Carlo Alberto
ABSTRACT. This paper presents an analysis of the co-evolutionary patterns of structural
change in knowledge and economics. The former is made operational through an analysis of
co-occurrences of technological classes in patent documents in order to derive indicators of
coherence, variety and cognitive distance. The latter, on the other hand, is made operational in
a synthetic way by implementing shift share analysis which decomposes labour productivity
growth into effects caused by changes in the allocation of employment, those ascribed to
intra-sector productivity growth and those caused by interaction of these two components.
The results of the analysis conducted on a sample of 227 European regions show that
increasing variety is associated with the reallocation of workforce across sectors whereas
within sector productivity is associated with high levels of both coherence and cognitive
distance of the regional knowledge base.
JEL Classification Codes: O33, R11
Keywords: Recombinant Knowledge, Coherence, Variety, Regional Structural Change, Shift
Share Analysis
1 Introduction1
The relationships between technological change and structural change have not
received due attention in empirical analyses. Indeed, the two subjects have been mostly
analysed independently or in relation to the mechanisms of economic growth, but not much
consideration has been given to their close connection.
Within the domain of regional economics, an early effort to integrate analysis of
technological and structural change was made by François Perroux (1955) who integrated the
role of technological change in his “growth pole” theory. Regional economic systems are
characterized by rounds of growth, i.e. periods in which firms in the propulsive industry grow
at faster rates, propagating positive effects across firms directly and indirectly related to the
propulsive industry. The main driving factor in this expansion is technical efficiency achieved
through innovation efforts.
More recently, the evolutionary approach to economic geography has articulated a
framework for analysis of linkages between technology and structural change at the local
level that is able to account for the inherent complexity of the dynamics at stake (Boschma
and Frenken, 2006 and 2011). In this strand of analysis, the competences accumulated at the
local level are important in shaping the process of industrial diversification so that a change in
the allocation of the workforce across different sectors is influenced by the degree of
technological relatedness between the involved activities.
This paper aims to contribute to the debate by bringing knowledge into an empirical
setting by analysing its relationship with cross-regional differential structural change
dynamics. In doing so, we will draw on the recombinant knowledge approach to propose the
concept of regional knowledge structure (Weitzman, 1998; Fleming, 2001; Fleming and
Sorenson, 2001; Quatraro, 2012). We will then couple three different methodological
approaches. The former elaborates a set of indicators that are able to provide a synthetic
account of the architecture of knowledge structure. In particular, we will draw upon cooccurrence matrixes to calculate coherence, cognitive distance and variety indicators.
A preliminary version of this paper has been presented at the annual conference of the Italian branch of the
Regional Science Association (AISRE) held in Rome in September 2012. The funding of European Commission
through the FP7 research project ‘Policy Incentives for the Creation of Knowledge: Methods and Evidence’
(PICK-ME), Grant number 266959, is also gratefully acknowledged.
Secondly, we will provide a synthetic account of the change in economic structure by
implementing “shift-share analysis” in order to disentangle the contribution to (labour)
productivity growth of within-sector productivity dynamics and reallocation of the labour
force across the different sectors. Finally, the relationships between these two sets of
indicators will be investigated by using a vector autoregression (VAR) model which we
estimate via ‘reduced form’ by applying the least absolute deviation (LAD) estimator due to
the distributional properties of the variables.
The analysis is carried out on European NUTS II regions and provides an interesting
insight into the dynamic feedbacks between economic and knowledge structure. Innovation
patterns dominated by established capabilities and organized search are indeed related to
productivity gains in fast growing sectors whereas innovation patterns typical of the early
stages of technological trajectories and dominated by random screening are mostly related to
changes in the cross-sectoral allocation of employment. The rest of the paper is organized as
follows. The next section outlines the theoretical framework, Section 3 elaborates a model,
introduces ‘shift-share’ analysis and proposes the working hypotheses and Section 4 provides
a description of the data used and outlines the econometric strategy. Section 5 presents the
results of the estimations and a general discussion and Section 6 offers a final conclusion.
2 Economic Structure and Knowledge Structure: the missing link
A wide body of literature in the domain of economics of innovation deals with the
importance that knowledge plays in the economic development process. On the contrary,
scarce attention has been devoted to the relationship between technological knowledge and
the dynamics of structural change 2 . This is particularly true for what concerns empirical
analyses grounded on econometric modeling. Indeed, most of the studies in this field revert to
case studies. For example, Lever (2002) provides an assessment of the link between
knowledge base and economic growth in 19 European cities. An important attempt to link
regional innovation systems to local knowledge bases can be also found in Asheim and
Coenen (2005), wherein the authors show the overlapping of features of the knowledge base
and regional industrial specialization by focusing on Nordic clusters.
The limited attention to the relationship between technological knowledge and
structural change is all the more surprising for at least two reasons. First, the process of
economic development is punctuated by changes in the relative weight of sectors (Kuznets,
1930; Burns, 1934; Fisher, 1939). Second, the close relationship between structural change
and technological change was already clear to influential scholars such as Kuznets (1930) and
Schumpeter (1939 and 1942), who, however, did not elaborate much on this from an
analytical viewpoint 3 . The interplay between Schumpeterian dynamics and the retardation
theory provides a fertile ground for improving our understanding of regional differences in the
transition dynamics typical of structural change processes (Quatraro, 2009a).
The evolutionary economic geography approach (Boschma and Frenken, 2006 and
2011) proposes a far reaching integration of these issues in a framework explicitly combining
industrial dynamics with economic geography. The main argument is that regional growth
emerges out of a process of industrial diversification. The emergence of new industries at the
local level, i.e. the shift of employment away from one sector to another, is influenced by
technological relatedness between sectors. Proximity matters not only in the geographical but
also in the technical and technological space so that the introduction of new varieties is
constrained by the competencies accumulated at the local level (Boschma, 2005; Usai, 2011;
Antonelli and Quatraro, 2013).
While the extant literature has focused on the role of technological relatedness of
sectors in the process of regional branching, less attention has been devoted to the importance
of the emergence of new sectors of knowledge accumulated at the local level. In this
direction, the grafting of the recombinant knowledge approach onto the ongoing debate may
enhance understanding of the relationship between patterns of knowledge generation and
structural change.
The recombinant knowledge approach provides a far reaching framework for
representing the internal structure of regional knowledge bases as well as enquiring into the
effects of its evolution. If knowledge stems from a combination of different technologies, the
François Perroux (1955) provided the former efforts to build a systematic account of the relationships between
structural and technological change in local contexts. He proposed a view according to which the development of
local economies is shaped by centripetal and centrifugal forces. Some sectors are likely to be stronger in some
areas and weaker in others so that the economic development of a specific area is influenced by the structural ties
of the propulsive sector with the rest of the local economic activities. Vertical and horizontal linkages can
therefore enhance the positive effects of outperforming sectors. Of the main sources of competitive advantage in
this framework, innovation plays a key role in the development of technical efficiency.
frequency with which two technologies are combined provides useful information on the basis
of which the internal structure of the knowledge base can be characterized according to the
average degree of complementarity of the technologies which knowledge bases are made of as
well as the variety of the observed pairs of technologies (Fleming, 2001; Fleming and
Sorenson, 2001).
The dynamics of regional knowledge bases can be seen as the patterns of change in
their own internal structure, i.e. the patterns of recombination across the elements in the
knowledge space (Quatraro, 2010). This qualifies both the cumulative character of knowledge
creation and the key role played by the properties describing knowledge structure, as well as
linking them to the relative stage of development of a technological trajectory (Dosi, 1982;
Saviotti, 2004 and 2007; Krafft, Quatraro, Saviotti, 2014). The ability to engage in a search
process within cognitive spaces that are close to the core competencies residing in the region
is indeed the outcome of a learning process that displays its effects once the technological
trajectory gets well established. On the contrary, in the wake of a new technological
trajectory, search processes are more likely to occur in areas that are distant from existing
competencies in the cognitive space (Nightingale, 1998; Sorenson et al., 2006).
The intertwining of regional innovation capabilities and recombinant knowledge is of
great importance in understanding the relationships between knowledge and structural change.
Innovation capabilities are indeed unevenly distributed across regions and technological
trajectories so that the emergence of new sectors is more likely to be associated with periods
of random screening in the knowledge space, typical of the early phases of new trajectories in
which capabilities are not yet sufficiently established (Lawson and Lorenz, 1999; Romijn and
Albu, 2002; Todling and Trippl, 2005; Antonelli, 2008; Quatraro, 2009b). As the trajectory
gets more familiar, innovating agents learn to move across the knowledge space and are more
likely to undertake organized search directed towards the combination of technologies that are
close to one another. The transition to organized search is typical of phases in which the
recombination activity occurs out of a sharply defined region of the knowledge space. The
likelihood of successful innovations is greater in this stage and they are more likely to foster
productivity growth at the regional level rather than engender changes in the economic
structure (Krafft, Quatraro and Saviotti, 2011 and 2014).
In view of the arguments outlined so far, we are now able to spell out our working
hypotheses. Technological change and structural change are closely intertwined. Indeed, the
accumulation of technological knowledge is likely to shape the processes by which new
industries, either in relative or absolute terms, emerge in regional contexts. Technological
knowledge, however, ought to be regarded as an heterogeneous asset stemming from the
combination of a variety of pieces of knowledge. The way these pieces are combined together
by regional innovating actors gives the structure of the local knowledge base a peculiar shape.
The economic structure can therefore be related to the structure of technological knowledge.
The latter may have different configurations according to the search patterns underlying its
production which are in turn related to the relative maturity of the technological trajectory
they impinge upon. When a technology trajectory is sufficiently established and innovation
capabilities diffused at the regional level, search patterns are likely to be dominated by
organized efforts directed towards well defined areas of the knowledge space. Innovating
agents are more able to identify pieces of knowledge that can be successfully combined
together and feature high complementarity and similarity. Knowledge at this stage is likely to
have a boosting effect on the productivity dynamics of established sectors. In the early stages
of a technological trajectory, innovation capabilities in the new domain still have to be fully
developed and diffused so that innovating agents still move in a highly uncertain
environment. Search patterns are dominated by random screening in the knowledge space
which is likely to lead to the combination of pieces of knowledge loosely related to one
another and hence with low levels of coherence and similarity. The switch to a new
technological trajectory is likely to be associated with the reallocation of employment across
sectors rather than with productivity gains of established sectors.
3 The implementation of key variables
3.1 Shift share decomposition
Shift share analysis provides an interesting methodology that allows
labour productivity to be decomposed in order to identify the differential contribution
provided by changes in the reallocation of employment across sectors, i.e. the most traditional
utilization of the concept of structural change in economics.
As noted by Houston (1967), the origins of shift-share analysis can be dated back to
seminal work by Daniel Creamer (1942), although it did not reach great success at least until
1960, when Perloff, Dunn, Lampard and Mutt employed it as an analytical tool in their work,
Regions, Resources and Economic Growth. It has been mostly used to investigate and
disentangle the compositional mix and competitive position of regions in the face of observed
changes in some relevant variables (Esteban, 1972 and 2000). In this paper, we will follow the
approach developed by Fagerberg (2000) who decomposed labour productivity into three
major components, i.e. the allocative and the productivity differential and interaction between
the two. We start by rearranging labour productivity as follows (region subscripts are omitted
for the sake of clarity):
∑ [
Labour productivity at the system level can therefore be decomposed into the
contribution provided by labour productivity of each sector j as well as by share of sector j in
total employment.
If we set:
The variation in labour productivity can therefore be expressed as follows:
Equation (5) can therefore be expressed in growth rates by dividing it by (Y/L):
∑ [
The first term between parentheses is the contribution to productivity growth from
changes in the allocation of labour between industries. It will be positive if the share of high
productivity industries in total employment increases at the expense of industries with low
productivity. The second term measures the interaction between changes in productivity in
individual industries and changes in the allocation of labour across industries. It will be
positive if fast growing sectors in terms of productivity also increase their share in total
employment. The third term is the contribution from productivity growth within
For the sake of clarity, let us assign a symbol and a name to each of the identified
Reallocation term =
Cross-term =
Within-sector productivity =
3.2 The Knowledge Indicators
The implementation of regional knowledge indicators rests on the recombinant
knowledge approach and on the model described in Section 2. In order to provide an
operational translation of such variables, we need to identify both a proxy for the pieces of
knowledge and a proxy for the elements that form their structure. For example, we could take
scientific publications as a proxy for knowledge and look either at keywords or at scientific
classification (the JEL code for economists, for example) as a proxy for the constituting
elements of the knowledge structure. Alternatively, we could consider patents as a proxy for
knowledge and then look at technological classes to which patents are assigned as the
constituting elements of its structure, i.e. the nodes of the network representation of
recombinant knowledge. In this paper we will follow the latter path4. Each technological class
j is linked to another class m when the same patent is assigned to both of them. The higher the
number of patents jointly assigned to classes j and m, the stronger this link is. Since
technological classes attributed to patents are reported in the patent document, we will refer to
The limits of patent statistics as indicators of technological activities are well known. The main drawbacks can
be summarized in their sector-specificity, the existence of non-patentable innovations and the fact that they are
not the only protecting tool. Moreover, the propensity to patent tends to vary over time as a function of the cost
of patenting and is more likely to feature large firms (Pavitt, 1985; Griliches, 1990). Nevertheless, previous
studies highlighted the usefulness of patents as measures of production of new knowledge, above all in the
context of analyses of innovation performances at the regional level. Such studies show that patents represent
very reliable proxies for knowledge and innovation, as compared with analyses drawing upon surveys directly
investigating the dynamics of process and product innovation (Acs et al., 2002). Besides the debate about patents
as an output rather than an input of innovation activities, empirical analyses showed that patents and R&D are
dominated by a contemporaneous relationship, providing further support to the use of patents as a good proxy of
technological activities (Hall et al., 1986). Moreover, it is worth stressing that our analysis focuses on the
dynamics of manufacturing sectors.
the link between j and m as the co-occurrence of both of them within the same patent
The four properties of the knowledge base (KB) which we will use in our analysis are
the knowledge capital stock, its variety, related or unrelated, its coherence and the cognitive
distance (see the Appendix for details).
The traditional regional knowledge stock (KCAP) is computed by applying the
permanent inventory method to patent applications. We calculated it as the cumulated stock of
Ei ,t  hi ,t  (1   ) Ei ,t 1 , where h i ,t is the flow of regional patent applications and δ is the rate
of obsolescence.
The variety (KV) of a knowledge base measures the extent of its diversification, with
related variety measuring it at a lower level of aggregation and unrelated variety at a higher
level of aggregation (Frenken et al, 2007). Technological variety can be measured by using
the information entropy index. It was introduced by Shannon (1948) to measure the
information content of messages and can be used as a distribution function in a number of
circumstances (Theil, 1967, Frenken 2004).
Systems characterized by high entropy will also be characterized by a high degree of
uncertainty (Saviotti, 1988). Unlike common measures of variety and concentration,
information entropy has some interesting properties (Frenken and Nuvolari, 2004). An
important feature of the entropy measure is its multidimensional extension. Consider a pair of
events (Xl, Yj), and the probability of co-occurrence of both of them plj. A two dimensional
(total) knowledge variety (KV) measure can be expressed as follows:
 1 
TV  H ( X , Y )   plj log 2 
 lj 
If we consider plj to be the probability that two technological classes l and j co-occur
within the same patent, then the measure of multidimensional entropy focuses on the variety
of co-occurrences of technological classes within regional patents applications.
Moreover, the total index can be decomposed into a “within” and a “between” part
anytime the events to be investigated can be aggregated into a smaller numbers of subsets.
It must be stressed that to compensate for intrinsic volatility of patenting behaviour, each patent application is
made to last five years.
Within-entropy measures the average degree of disorder or variety within the subsets, while
between-entropy focuses on the subsets measuring the variety across them. Frenken et al.
(2007) refer to between- and within-group entropy respectively as unrelated and related
It can be easily shown that the decomposition theorem also holds for the
multidimensional case. Hence, if we allow lSg and jSz (g = 1,…,G; z = 1,…, Z), we can
rewrite H(X,Y) as follows:
 1 
KV  H ( X , Y )   plj log 2  
p 
 lj 
where the first term of the right-hand-side is the between-entropy and the second term
is the (weighted) within-entropy. In particular:
UKV  H Q   Pgz log 2
g 1 z 1
RKV   Pgz H gz
g 1 z 1
Sg j
 1
lj gz
g z gz 
We can therefore refer to between- and within-entropy respectively as unrelated
knowledge variety (UKV) and related knowledge variety (RKV), while total information
entropy is referred to as general technological variety.
The coherence (COH) of a KB measures the extent to which different types of
knowledge can be combined. This is of fundamental importance since the types of knowledge
required by innovating agents to create new products or services are not necessarily found
within a discipline, but need to be combined to produce the desired output. The ability to
combine these different types of knowledge is not constant but can be expected to vary
systematically during particular phases of the evolution of the lifecycle. For example, we can
expect the ability to combine different types of knowledge to fall as a completely new type of
knowledge emerges as a discontinuity and to rise again as the new type of knowledge starts
To yield the knowledge coherence index, a number of steps are required. The
following calculations show how to obtain the index at whatever level of analysis i. First of
all, we need to calculate the parameter , i.e. technological relatedness, by deriving the
relatedness matrix as follows (Nesta, 2008). Let the technological universe consist of k patent
applications. Let Pjk = 1 if the patent k is assigned the technology j [j = 1, …, n], and 0
otherwise. The total number of patents assigned to technology j is O j  k Pjk . Similarly, the
total number of patents assigned to technology m is Om  k Pmk . Since two technologies
may occur within the same patent, O j  Om  , the number of observed co-occurrences of
technologies j and m is J jm  k Pjk Pmk . By applying this relationship to all possible pairs,
we yield a square matrix  (n  n) whose generic cell is the observed number of cooccurrences:
J j1
J n1 
 J 11
  
 
J jm
J nm 
   J 1m
  
 
 J 1n  J jn  J nn 
We can assume that the number xjm of patents assigned to both technologies j and m is
a hypergeometric random variable of mean and variance:
 jm  E ( X jm  x ) 
O j Om
 K  O j  K  Om 
 K  K  1 
 2jm   jm 
If the observed number of co-occurrences Jjm is larger than the expected number of
random co-occurrences jm, then the two technologies are closely related: the fact the two
technologies occur together in the number of patents xjm is not casual. The measure of
relatedness hence is given by the difference between the observed number and the expected
number of co-occurrences, weighted by their standard deviation:
 jm 
J jm   jm
 jm
 jm   ; . Moreover, the index shows a distribution similar to a t-student, so that if
It is worth noting that this measure of relatedness has lower and upper bounds:
 jm   1.96;1.96 , one can safely accept the null hypothesis of non-relatedness of the two
technologies j and m. The technological relatedness matrix ’ may hence be thought of as a
weighting scheme to evaluate the technological portfolio of regions.
Once the parameter τ has been calculated for each pair of technologies, we can
proceed to derive the weighted average relatedness WARj of technology j with respect to all
other technologies present within the relevant aggregate (say sector, firm or region).
Following Teece et al. (1994), WARj is defined as the degree to which technology j is related
to all other technologies j≠m in the aggregate, weighted by patent count Pmt:
WAR jit 
m j
 jm Pmit
m  j mit
Finally, the coherence of knowledge base within the aggregate i (be it a firm, a sector
or a region) is defined as the weighted average of the WARlt measure:
Rit  WAR jit 
j jit
It is worth stressing that the index implemented by analysing co-occurrences of
technological classes within patent applications measures the degree to which the services
rendered by the co-occurring technologies are complementary to one another (see Nesta and
Saviotti, 2005, 2006 and Krafft, Quatraro and Saviotti, 2014). The relatedness measure  jm
indicates that the utilization of technology j implies that of technology m in order to perform
specific functions that are not reducible to their independent use. This makes the coherence
index appropriate for the purposes of this study.
In addition to coherence, we also investigate the relationship between the terms of
shift-share and cognitive distance (CD) (Nooteboom, 2000), which expresses the
dissimilarities between different types of knowledge. This measure is of fundamental
importance when distinguishing the effect of the emergence of a discontinuity from that of the
subsequent period of normal or incremental development.
A useful index of distance can be derived from the measure of technological
proximity. It was originally proposed by Jaffe (1986 and 1989), who investigated the
proximity of firms’ technological portfolios. Subsequently, Breschi et al. (2003) adapted the
index in order to measure the proximity, or relatedness, between two technologies. The idea is
that each firm is characterized by a vector V of the k technologies that occur in its patents.
Knowledge similarity can first be calculated for a pair of technologies l and j as the angular
separation or un-centred correlation of the vectors Vlk and Vjk. The similarity of technologies l
and j can then be defined as follows:
S lj 
k 1
k 1
VlkV jk
k 1
The idea underlying the calculation of this index is that two technologies j and l are
similar to the extent that they co-occur with a third technology k. The cognitive distance
between j and l is the complement of their index of the similarity:
dlj 1Slj
Once the index is calculated for all possible pairs, it needs to be aggregated at the
industry level to obtain a synthetic index of technological distance. This can be done in two
steps. First of all, we can compute the weighted average distance of technology l, i.e. the
average distance of l from all other technologies.
l lj jit
l jit
where Pj is the number of patents in which the technology j is observed. Now the
average cognitive distance at time t is obtained as follows:
 lit
l lit
Table 1 provides the synthetic definition of the variables which we have described so
Methodology and Data
4.1 Methodology
The main focus of this paper is on the observation of the co-evolutionary dynamics
between knowledge and economic structure. In the previous section we proposed a synthetic
representation of change in economic structure by introducing shift share analysis.
In view of the complex and endogenous nature of the relationships between the
properties of knowledge and those of economic structure, we apply a VAR model.
The regression of interest is the following:
where wit is an m1 vector of random variables for region i at time t,  is an m[mz]
matrix of slope coefficients that are to be estimated. In our particular case m=9
corresponds to the vector [μ(i,t), π(i,t), α(i,t), growth of knowledge capital (i,t), coherence
growth (i,t), growth of cognitive distance (i,t), variety growth (i,t), related variety growth (i,t),
unrelated variety growth (i,t)].  is an m1 vector of disturbances.
In line with previous studies, the measure of growth rates is based on the difference in
the logarithms of the respective variables. Let Xi(t) represent the absolute value of the
variable in region i at time t. Define the normalized (log) value of the variable as:
Where N is the number of regions. In what follows, growth rates are defined as the
first difference of normalized (log) values according to:
In such a way, common macroeconomic shocks are already controlled because the
growth rate distribution was normalized to zero for each variable in each region in each year.
Following a growing body of literature (Coad, 2010; Buerger, Broekel and Coad,
2012; Colombelli, Krafft and Quatraro, 2014a), Equation (18) is estimated via ‘reduced form’
VARs which do not impose any a priori causal structure on the relationships between the
variables and are therefore suitable for the purposes of this analysis. These reduced-form
VARs effectively correspond to a series of m individual ordinary least squares (OLS).
However, previous studies have emphasized how the empirical distribution of the
growth rates is closer to a Laplacian than to a Gaussian distribution (Bottazzi et al. 2007;
Bottazzi and Secchi 2003; Castaldi and Dosi 2009). Such evidence suggests that standard
regression estimators, such as ordinary least squares (OLS), assuming Gaussian residuals may
perform poorly if applied to these empirical frameworks. To cope with this, a viable and
increasingly used alternative consists of implementing least absolute deviation (LAD)
techniques that are based on the minimization of the absolute deviation from the median
rather than the squares of the deviation from the mean.
It must be noted that we do not include any individual dummies in the analysis. Even
though unobserved heterogeneity can have important effects on the estimation results, the
inclusion of individual dummies along with lagged variables may engender some biases for
fixed-effect estimation of dynamic panel-data models, a problem known as Nickell-bias.
Some alternative approaches relate to the use of instrumental variable (IV) or GMM
estimators (Blundell and Bond, 1998). The main problem with this lies in the difficulty in
finding good instruments which is particularly hard when dealing with growth rates. When
instruments are weak, IV estimation of panel VAR thus leads to imprecise estimates. Binder
et al. (2005), on the other hand, propose a panel VAR model including firm-specific effects
which is, however, based on the assumption of normally distributed errors which is not the
case for the growth rates of the variables used in our regressions.
Since we are dealing with growth rates rather than of levels, we can assume that any
region-specific component has been largely removed. Moreover, we follow the wide body of
literature on analysis of firms’ growth rates which states that the non-Gaussian nature of
growth rate residuals is a far more important econometric problem deserving careful attention
even in regional level analyses (Buerger, Broekel and Coad, 2012).
4.2 The Data
In order to implement the methodology outlined in the previous section, we gather
together two datasets. The shift-share analysis has been conducted by using the branch
accounts of NUTS II European regions6 provided by the Eurostat within the European System
of Integrated Economic Accounts. As is well known, these data have only been available
We acknowledge that the use of administrative regions to investigate the effects of knowledge creation
represents only an approximation of the local dynamics underpinning such a process. Indeed, administrative
borders are arbitrary and therefore might not be representative of the spontaneous emergence of local
interactions. It would be much better to investigate these dynamics by focusing on local systems of innovation.
However, it is impossible to find out data at such a level of aggregation. Moreover, the identification of local
systems involves the choice of indicators and threshold values according to which one can decide whether to
unbundle local institutions or not. This choice is in turn arbitrary and therefore would not solve the problem, but
would only reproduce the issue at a different level. Thus, we think that despite the unavoidable approximation,
our analysis may provide useful information on the dynamics under scrutiny.
since 1995, the year in which the Eurostat implemented a standardized procedure to collect
data from European countries in order to build a coherent and homogeneous dataset. As a
result, we were able to calculate the μ, the π, and the  components for a subset of European
regions on a time span ranging from 1995 to 2007. The properties of knowledge structure, i.e.
coherence, cognitive distance and variety (based on the information entropy index) have
instead been calculated by using patent information contained in the OECD REGPAT
database which covers patent data that have been linked to regions utilizing the addresses of
the applicants and inventors. Analysis was conducted by adopting inventor-based
regionalization7, and by using 4-digit technology codes.
We merged the two sets of indicators on the basis of the NUTS II regional code and
the year. We ended up with an unbalanced panel of 227 regions observed on average over 8
years. The descriptive statistics for the whole sample are reported in Table 2 whereas Figure 1
shows the distributional properties of the variables under scrutiny, providing empirical
support for their non-Gaussian distribution. In particular, all the variables appear to follow a
Laplace-like distribution which makes our empirical strategy outlined in the previous section
the best approach to the analysis.
>>> INSERT Figure 1 AND Table 2 ABOUT HERE <<<
The elaboration of a regional breakdown of descriptive statistics turns out to be very
complicated when dealing with a sample of 227 regions. For this reason, we decided to show
the cross-regional distribution of average values by implementing a map for each of the
variables under consideration. The maps reported in Figure 3 and Figure 4 are based on
absolute (log) values of the properties of the knowledge structure.
In Figure 2, we report the cross-regional distribution of the three components
contributing to labour productivity growth. Let us recall that μ is the contribution of the
changing mix of regional industries and is positive if regions tend to specialize in highproductivity activities, π is the interaction between productivity growth and the change in the
industry mix and is positive to the extent that regions specialize in fast growing sectors
The assignment of patents to regions on the basis of inventors’ addresses is the most widespread practice in the
literature (see, for example, Maurseth and Verspagen, 2002; Henderson et al., 2005; Breschi and Lissoni, 2009,
Paci and Usai, 2009, to quote a few). A viable alternative may rest on the use of applicants’ addresses, above all
when the assessment of knowledge impact on growth is at stake (see Antonelli, Krafft and Quatraro, 2010).
However, when analysis is conducted at a local level of aggregation, and the geography of collective processes
of knowledge creation is emphasized, the choice of inventors’ addresses remains the best one.
whereas α is the contribution of within-sector productivity growth weighted by the sector
share on total employment.
>>> INSERT Figure 2 ABOUT HERE <<<
It is interesting to note that for most of the sampled regions, the effect of change in the
industry mix is positive, suggesting that structural change plays an important role in the
process of economic growth. Most European regions therefore tend to specialize in highproductivity sectors, with the only exception being some Greek regions and the British
Midlands. The process is more pronounced in Italy and in central-eastern Europe than in
Spain and France. The second diagram shows that the interaction term is positive again in
most Italian regions, Spain, France and Germany, while evidence is more mixed in other
regions. Italy, France, Spain and Germany in the observed period are subject to changes
favouring the increasing share of fast-growing sectors. Finally, within-sector productivity
growth seems to matter the most for Northern regions, such as Finland, Sweden and Denmark,
and to a somewhat lesser extent for some Eastern and Greek regions.
In Figure 3, we report the cross regional distribution of knowledge capital, coherence
and cognitive distance (log values). The top diagram reports the figures concerning the
knowledge capital. We can see how knowledge capital is higher in central European regions
and in northern regions whereas it is lower in the periphery of the continent. A look at the
coherence index reveals that on average search behaviours are more like organized search
than random screening, while cognitive distance is on average very low in most of the
European regions, suggesting that exploration is conducted across the safe boundaries of
established knowledge competences. We observe both low values of coherence and cognitive
distance only for a few scattered regions in France, Spain and Finland suggesting search
strategies characterized by exploration behaviours conducted within well defined boundaries
of the knowledge space.
>>> INSERT Figure 3 ABOUT HERE <<<
Figure 4 shows cross distribution of the variety index, articulated in unrelated and
related knowledge variety. The top diagram indicates that on average European regions are
characterized by a high degree of variety, with the only exception being some peripheral
regions in Portugal and Greece. When we look at the distinction between related and
unrelated variety, we can see that the distribution looks very similar to that of total variety. By
also observing the ranges assigned to each classes, we can also state that on average related
knowledge variety is higher than unrelated variety.
>>> INSERT Figure 4 ABOUT HERE <<<
Table 2 shows the correlation across the variables used in the empirical analysis. It can
be noted that only in a few cases correlation coefficients appear to be significantly correlated,
but in any case the coefficients are pretty low.
Since the relationships under scrutiny are investigated at the regional level, the spatial
dependence of the relevant variables can engender some biases in the estimation results. In
order to check the extent to which spatial dependence can be a problem, we calculated the
Moran’s I index for all variables by relying on a row standardized contiguity matrix (Moran,
1950). As is well known, the Moran’s I index is used to calculate the spatial autocorrelation
across OLS residuals. Average growth rates were used because they reflect the fundamental
relations between the regions in contrast to fluctuating yearly growth rates. The results are
shown in Table 4, which suggests that only in two cases the statistics are significant and hence
the null hypothesis of no spatial correlation ought to be rejected. However, the correlation
coefficients are fairly low and this should not bias the subsequent estimations in any serious
way. This is even more so considering that LAD estimation techniques are preferred over
standard OLS.
In the following section we will implement the estimation of equation (18), which is
based on the normalized growth rates of such variables.
5 Econometric results
The results of the econometric estimations are reported in Tables 5 to 8. Tables 5 and
6 present the results of exploratory regressions conducted by implementing the OLS
estimator. Table 5 shows the estimations in which the shift share components are dependent
variables and Table 6 shows the estimations in which the properties of the regional knowledge
base are dependent variables.
With regard to the observed autocorrelation, it is impressive to note that none of the
variables under scrutiny shows any degree of persistence. On the contrary, coefficients are
negative and significant across all three lags considered, suggesting erratic growth dynamics
for all the variables. Such results on knowledge-related variables are consistent with the
findings of Buerger, Broekl and Coad (2012) who ascribe this kind of evidence to intrinsic
uncertainty and volatility characterizing innovation.
>>> INSERT Table 5 ABOUT HERE <<<
We will now analyze in more detail the lead-lag relationship between the change in
knowledge and economic structure. Let us start with Table 5. As far as the first lag is
concerned, knowledge coherence (COH) and knowledge capital (KCAP) show a positive and
significant coefficient on α which is consistent with the literature linking knowledge structure
and productivity growth (Nesta, 2008; Quatraro, 2010). The α component stands for the
contribution stemming from within-sector productivity growth which is positively affected by
the growth of knowledge coherence and that of knowledge capital. Knowledge coherence is
indeed likely to increase as a technological trajectory gets established and innovation
capabilities begin to spread to regional innovating agents. Technological learning enhances
the capacity to identify and retain profitable combinations of technologies in the generation of
new knowledge. At the aggregate level, the knowledge base appears to be therefore
characterized by a high degree of integration as a result of a cumulative process. Agents are
now able to focus on a well defined area of the knowledge space and to engage in successful
innovation efforts that are more likely to exert relevant effects on economic performances.
The knowledge variety (KV, RKV and UKV) indexes as well as CD do not seem to affect the
economic structure significantly.
When we move to the second lag, we see that knowledge coherence (KOH) affects the
cross-term π positively and significantly, i.e. faster growth of coherence is associated with a
faster increase in faster growing sectors. Once again, the ability to screen the knowledge
space in an organized and systematic way allows for improved economic performance.
Knowledge variety, on the other hand, negatively affects the cross-term. These negative
coefficients can be interpreted in the light of the mixed nature of the cross-term π. We would
expect increasing technological variety to affect the change in the industry mix positively. On
the contrary, higher growth rates of technological variety are typical of instability phases of
the technological trajectory which are more likely to be associated with productivity
Finally, the third lag presents an interesting positive and significant coefficient of KV
and UKV on the reallocation term (µ). This is consistent with the idea that increasing
technological variety is likely to be associated with changes in the industry mix. The
displacement of the workforce from mature activities to new emerging activities is indeed
related to an increase in the scope of technological activities, focusing in particular on the
combination of pieces of knowledge that belong to different technological domains.8.
The effect of KV, RKV and UKV on the cross-term π is again negative, signalling the
prevalence of the negative effects on productivity dynamics. Finally, it is also worth noting
the positive and significant coefficient of knowledge coherence (COH) and cognitive distance
(CD) as far as effects on within-sector productivity (α) are concerned. The former is in line
with the evidence observed when looking at the first lag. Coherence is likely to increase as an
effect of organized search activities typical of exploitation dynamics. These in turn are likely
to engender significant productivity gains. The coupling of positive effects of coherence and
cognitive distance suggest that productivity gains are associated with a combination of
complementary but not similar pieces of knowledge. This is in line with the literature in the
evolutionary economic geography field which suggests that too much similarity is detrimental
to regional economic development (Boschma and Wenting, 2007; Colombelli and Quatraro,
Table 6 presents the results of the OLS estimation concerning the determinants of
growth rates of the properties of regional knowledge bases.
The displacement effect on employment of course is unlikely to manifest itself in the shortest run. This may be
the reason why this effect is grasped only when lookings at the third lag. However, one would expect this
evidence to become more robust in the subsequent lags. Unfortunately, the available data do not allow us to go
further back in time.
See Colombelli, Krafft and Quatraro (2014a) and Colombelli and Quatraro (2014) for similar evidence at the
firm level.
The evidence regarding the first lag shows that knowledge coherence (COH) appears
to be positively affected by π, suggesting that the increasing share of fast growing sectors is
associated with increasing integration of regional technological activities. It is also interesting
to note that within-sector productivity (α) negatively affects the growth of cognitive distance.
This evidence is in line with previous work (Colombelli, Krafft and Quatraro, 2014a)
according to which higher performances are likely to create the economic conditions to
stimulate patterns of knowledge creation based on the search across domains that are not so
familiar and yet not too far from the established technological competences. The positive and
significant effect of α on KV and RKV provides further support for such an interpretation.
The coefficients for the second lag of the explanatory variables reveal that the
reallocation term (μ) does not yield any significant effect on the knowledge characteristics.
The same applies to the cross-term (π). In line with the evidence for the first lag, within-sector
productivity (α) has a negative effect on cognitive distance whereas it positively affects KV.
As far as the third lag is concerned, we can see that no significant effects of the shiftshare terms on the properties of the regional knowledge base can be devised. The only
exception is the negative effect of the reallocation term (µ) on knowledge variety. It would
therefore seem that increasing variety can foster the changing allocation of labour across
sectors, but that this in turn is likely to be followed by a reduction in variety.
The evidence discussed so far suggests that the effects of the properties of the regional
knowledge base on the terms of shift-share decomposition are broader than the effects that the
latter yield on the former. However, as already discussed in Section 4, OLS estimations are
likely to be biased due to the distributional properties of the variables at stake. For this reason,
we estimated the reduced form VAR by applying the LAD estimator. The results are reported
in Tables 7 and 8.
Table 7 presents the evidence concerning the determinants of the shift share
components. In columns (1a)-(1c), the dependent variable is the reallocation term (µ). Over
the three lags, the autoregressive coefficient is negative and significant suggesting erratic
movement of this term. With regard to the effects of the properties of the regional knowledge
base, only KV shows a positive and significant coefficient at the third lag, which was already
present in the previous OLS estimation. This is in line with expectations. The enlargement of
the scope for recombination activities signals the opening up of new trajectories for
exploitation which attract economic agents. In the aggregate this is like to be associated with
displacement of the workforce from old and mature sectors to new ones. Columns (2a)-(2c),
on the other hand, show the results for the cross-term (π). Here too, the autoregressive terms
over the three lags are negative and significant. As far as the knowledge-related variables are
concerned, none of the coefficients is significant in the first lag. When we move to the
following lags, KV turns out to be negative and significant both at the second and the third lag
whereas RKV and UKV show negative and significant coefficients only on the third lag. As
in the previous estimations, these results can be interpreted as an outcome of the dominance
of the productivity gain component of the cross-term. Finally, in columns (3a)-(3c), the
dependent variable is within-sector productivity (α). Here too, the autoregressive terms are
negative and significant over the three implemented lags. Out of the knowledge-related
variables, only CD shows a significant (and positive) coefficient on the third lag. This is in
line with the previous estimation and suggests that in order to observe productivity gains, the
regional knowledge base ought to be the outcome of the recombination of dissimilar pieces of
In Table 8, we report the results of the reduced-form VAR estimations in which the
properties of the knowledge base are dependent variables. In columns (4a)-(4b), the
dependent variable is COH. In the first lag, no significant effects of the knowledge structure
are observed. In the second lag, the cross-term (π) shows a positive and significant coefficient.
This can be interpreted as an outcome of the dominance of the effect of the reallocation
component. After displacement of the workforce from one sector to another, as an outcome of
increasing variety which signals exploration strategies, technological activities are oriented
towards increasing integration of the knowledge base in the new fields. This interpretation is
also supported by the negative and significant coefficient of within-sector productivity (α). A
technology-lifecycle interpretation is in order. When increasing, growth rates are indeed the
outcome of exploitation dynamics that allow for the selection of profitable trajectories. After
some time, high growth rates are followed by decreasing coherence due to the need to identify
new and unexploited technological opportunities which allows for further productivity gains
in the future.
In columns (5a)-(5c) the dependent variable is KCAP. In this case, the only shift-share
variable showing a statistically significant coefficient is the cross-term and this applies to all
of the three investigated lags. The sign of the coefficients is always negative. This could be
interpreted as an outcome of the dominance of the reallocation component of the cross-term.
A change in the sector allocation of the workforce in one region is indeed a sign of maturity
of established activities and a search for new avenues of development. In this phase, the
technological competences associated with old sectors may have exhausted the technological
opportunities leading to a reduction of knowledge production which is not yet
counterbalanced by identification of new profitable trajectories. The evidence on CD reported
in columns (6a)-(6c), is consistent with the regional technology lifecycle interpretation
articulated so far. Indeed, we can observe a negative and significant coefficient of withinsector productivity (α) in the first lag and a negative and significant coefficient of the crossterm (π) in the third lag. According to the evolutionary economic geography approach, high
levels of CD are likely to positively affect regional competitiveness. When productivity gains
are achieved, most technological opportunities have been exploited and the scope for
recombination in the knowledge space gets smaller and smaller. As an outcome, the average
degree of similarity amongst combinable technologies increases. This could also explain the
negative sign of the cross-term which combines both the reallocation and the productivity
components. Insofar as increasing reallocation is associated with the exhaustion of
technological opportunities in incumbent sectors, the narrowing of the set of profitable
combinations leads to the increasing likelihood that technologies with a high degree of
similarity will be combined. Columns (6), (7) and (8) report the estimations for KV, RKV and
UKV respectively. As far as KV is concerned, we can observe that the within-productivity
term (α) shows a positive and significant coefficient in the second lag, while the cross-term
(π) shows a negative and significant coefficient in the third lag. With regard to RKV, the only
significant shift-share variable is the reallocation term which is featured by a negative and
significant coefficient in the second lag. The estimation on UKV shows that only the crossterm is characterized by a significant and positive coefficient in the first lag. The evidence
concerning the variety variables provides further support for our interpretation. Productivity
growth is likely to be followed by an increase in knowledge variety due to the need to look for
new profitable trajectories. Consistently, displacement of the workforce from incumbent to
new sectors is typical of exploration phases. When this happens, we will probably observe a
reduction in knowledge variety in the following periods as a result of the gradual selection of
successful combinations.
6 Conclusions
In this paper we have conducted an exploratory analysis of the co-evolutionary
patterns of knowledge and economic structure. The concept of knowledge structure has been
in particular elaborated so as to qualify the regional knowledge base according to its average
degree of similarity, complementarity and variety. Drawing upon a theoretical framework that
stresses the dynamic nature of the interactions between these two components as well as the
endogenous character of the change process, we decided to implement an empirical
framework based on the recombinant knowledge approach in order to characterize the
knowledge structure. We coupled this methodological approach with the shift-share technique
which allows the effects of the change in economic structure to be grasped in a synthetic way
and in particular, we focused on the changing allocation of the labour force across sectors.
The empirical analysis, given the dynamic effects feeding back from economic and
knowledge structure and vice versa, was conducted by implementing a set of ‘reduced-form’
VAR estimations which allowed us to investigate the lead-lag relationships between the two
systems without imposing any aprioristic causal structure. The results of the analysis are
encouraging and call for further research in this direction, showing a clear interactive pattern
between the two structures. Changes in knowledge structure that signal the patterns of
knowledge creation based on organized search are likely to engender increasing within-sector
productivity growth, while knowledge generation activities characterised by random
screening across the knowledge space are likely to be followed by changes in the allocation of
the labour force across sectors. Moreover, the increasing share of faster growing sectors
appears to stimulate the establishment of knowledge creation patterns dominated by organized
search within the comfortable fences of established competences. Moreover, econometric
analysis also allowed us to appreciate some interesting dynamics such as the one relating
variety and μ, according to which increasing variety fosters the changing allocation of labour
across sectors, but which in turn is likely to be followed by a reduction in variety10.
These results represent an interesting starting point for investigating and shedding new
light on issues traditionally addressed in regional economics and economic geography such as
the importance of the effects of the surrounding environment on firms’ economic
performances or the possible effects of the structure of regional (or local) knowledge bases on
firms’ location choices, which can be enriched by fully appreciating the role of the properties
of local knowledge bases. These issues are also important to the policy realm, too often
focused on the design of measures targeted to restoring regional competitiveness by providing
funding to incumbent activities. Although measures targeted to restore productivity in
incumbent sectors are less complicated, these are more likely to yield temporary outcomes.
On the contrary, the results of our analyses suggest that policymakers should promote the
entry of new activities grounded on new and profitable technological trajectories. Measures
aiming at reshaping the structure of economic and technological activities at the local level
are likely to yield enduring effects since they are based on structural interventions.
In this direction, the scope for demand-driven policy instruments at the local level is
enriched. Indeed, new activities cannot emerge out of the blue. The incentives to the local
creation of new technology based activities should be therefore grounded on the accurate
analysis of both the comparative advantages developed over time in a specific area and of the
relative position of such technologies in the technological landscape (Colombelli, Krafft and
Quatraro, 2014b; Asheim and Coenen, 2005). Stimulating local agents to jump to new
activities far away from their cumulated competencies can be inefficient and unsuccessful.
Strategic demand pull emerges here as an ingredient to the strategic management of
places, the goal of which should be the promotion of knowledge-based entrepreneurship as a
vehicle for the employment growth and global competitiveness at the local level (Audretsch,
2003). Demand-driven innovation policies should therefore be particularly cautious in the
identification of the key sectors to be promoted, above all when implemented at the European
level, as this latter task cannot be performed without a careful screening of the patterns of
local technological specialization in the different areas upon which the policy instrument
An important limitation of this exercise concerns the lag structure used in the analysis. Although most of the
existing studies using the reduced VAR approach do not go beyond the third lag (Coad and Rao, 2010; Buerger,
Broekl and Coad, 2012), it would be useful to investigate longer lag structures. The available data on regional
economic accounts unfortunately do not allow analysis to be extended in this direction.
impinges. These policy measures should be rather customized, so as to ensure effective
implementation and the reduction of duplication of efforts and waste of resources.
7 References
Acs, Z.J., Anselin, L. and Varga, A., 2002. Patents and innovation counts as measures of
regional production of new knowledge. Research Policy, 31, 1069-1085.
Antonelli, C., 2008. Localized Technological Change: Towards the Economics of Complexity.
London: Routledge.
Antonelli, C., Krafft, J. and Quatraro, F., 2010. Recombinant knowledge and growth: The
case of ICTs. Structural Change and Economic Dynamics 21, 50-69.
Antonelli, C. and Quatraro, F., 2013. Localized Technological Change and Efficiency Wages:
the Evidence across European Regions. Regional Studies 47, 1469-1483.
Asheim, B. T., & Coenen, L., 2005. Knowledge bases and regional innovation systems:
Comparing Nordic clusters. Research policy 34, 1173-1190.
Audretsch, D. B., 2003. Entrepreneurship policy and the strategic management of places. In
D. M. Hart (Ed.) The emergence of entrepreneurship policy: Governance, start-ups,
and growth in the US knowledge economy, 20-38. New York: Cambridge University
Binder, M., C. Hsiao and C. H. Pesaran, 2005. Estimation and inference in short panel vector
autoregressions with unit roots and cointegration. Econometric Theory 21, 795–837.
Blundell R.W and Bond S.R., 1998. Initial conditions and moment restrictions in dynamic
panel data models. Journal of Econometrics 87, 115-143.
Boschma, R. 2005. Proximity and innovation: A critical assessment. Regional Studies 39: 6174.
Boschma, R. and Frenken, K., 2006. Why is economic geography not an evolutionary
science? Towards an evolutionary economic geography. Journal of Economic
Geography 6, 273-302.
Boschma, R. and Frenken, K., 2011. Technological relatedness and regional branching. In:
Bathelt, H., Feldman M.P. and Kogler, D.F. (ed.) Dynamic geographies of knowledge
creation and innovation. London and New York, Routledge.
Boschma, R. and Wenting, R., 2007. The spatial evolution of the British automobile industry:
Does location matter? Industrial and Corporate Change, 16, 213-238.
Bottazzi, G., Cefis, E., Dosi, G. and Secchi, A., 2007. Invariances and Diversities in the
Patterns of Industrial Evolution: Some Evidence from Italian Manufacturing
Industries. Small Business Economics 29, 137-159.
Bottazzi, G., Secchi, A., 2003. Common properties and sectoral specificities in the dynamics
of U.S. manufacturing companies. Review of Industrial Organization 23, 217–232.
Breschi, S. and Lissoni, F., 2009. Mobility of skilled workers and co-invention networks: an
anatomy of localized knowledge flows. Journal of Economic Geography 94, 439-468.
Breschi, S., Lissoni, F., and Malerba, F., 2003. Knowledge relatedness in firm technological
diversification, Research Policy 32: 69-97.
Buerger, M., Broekel, T. and Coad, A., 2012. Regional dynamics of innovation: Investigating
the co-evolution of patents, research and development (R&D), and employment.
Regional Studies 46, 566-582.
Burns A. F., 1934. Production trends in the United States since 1870. NBER, Boston.
Castaldi, C. and Dosi, G., 2009. The patterns of output growth of firms and countries: Scale
invariances and scale specificities. Empirical Economics 37, 475-495.
Coad, A., 2010. Exploring the processes of firm growth: evidence from a vector autoregression. Industrial and Corporate Change 19, 1677-1703.
Coad, A. and Rao, R., 2010. Firm growth and R&D expenditure. Economics of Innovation
and New Technology 19, 127-145.
Colombelli, A., Krafft, J. and Quatraro, F., 2014a. High Growth Firms and Technological
Knowledge: Do gazelles follow exploration or exploitation strategies?, Industrial and
Corporate Change 23, 261-291.
Colombelli, A., Krafft, J. and Quatraro, F., 2014b. The emergence of new technology-based
sectors at the regional level: A proximity-based analysis of nanotechnologies.
Research Policy, forthcoming.
Colombelli, A. and Quatraro, F. 2013. New Firm Formation and the properties of local
knowledge bases: Evidence from Italian NUTS 3 regions, Working Papers hal00858989, HAL.
Colombelli, A. and Quatraro, F. 2014. The persistence of firms' knowledge base: A quantile
approach to Italian data, Economics of Innovation and New Technology, forthcoming.
Creamer, D., 1942. Shifts of Manufacturing Industries, in Industrial Location and National
Resources. Washington, D. C.: U. S. National Resources Planning Board.
Dosi, G., 1982. Technological paradigms and technological trajectories: A suggested
interpretation of the determinants and directions of technical change. Research Policy
11, 147–162
Esteban, J.M., 1972. Shift-share analysis revisited. Regional and Urban Economics 2, 249-61.
Esteban, J.M., 2000. Regional convergence in Europe and the industry mix: a shift-share
analysis. Regional Science and Urban Economics 30, 353-364.
Fagerberg, J., 2000. Technological Progress, Structural Change and Productivity Growth: A
Comparative Study. Structural Change and Economic Dynamics 11, 393-411.
Fisher, A.G.B, 1939. Production, primary, secondary and tertiary. The Economic Record 15,
Fleming, L., 2001. Recombinant uncertainty in technological search. Management Science 47,
Fleming, L., Sorenson, O. 2001. Technology as a complex adaptive system: Evidence from
patent data. Research Policy 30: 1019-1039.
Frenken, K. 2004. Entropy and information theory. In: Hanusch, H, and Pyka, A. (ed.), The
Elgar Companion to Neo-Schumpeterian Economics. Cheltenham, Edward Elgar, pp.
Frenken, K., Nuvolari, A., 2004. Entropy Statistics as a Framework to Analyse Technological
Evolution. In: J. Foster and W. Hölzl (ed.) Applied evolutionary economics and
complex systems. Cheltenham, U.K. and Northampton, Mass.: Edward Elgar.
Frenken, K., von Oort, F., Verburg, T., 2007. Related Variety, Unrelated Variety and
Regional Economic Growth. Regional Studies 41: 685-97.
Griliches, Z., 1990. Patent statistics as economic indicators: A survey. Journal of Economic
Literature 28, 1661-1707.
Hall, B.H., Griliches Z. and Hausman J.A., 1986. Patents and R and D: Is there a lag?.
International Economic Review 27, 265-283.
Henderson, R. M., Jaffe, A. and Trajtenberg. M., 2005. Patent citations and the geography
of knowledge spillovers: A reassessment: Comment. American Economic Review
95, 416-464.
Houston, G.B., 1967. The Shift and Share Analysis of Regional Growth: A Critique. Southern
Economic Journal 33, 577-581.
Jaffe, A.B. 1986. Technological opportunity and spillovers of R&D: Evidence from firms’
patents, profits, and market value. American Economic Review 76, 984–1001.
Jaffe, A., 1989. Real Effects of Academic Research. American Economic Review 79: 957-70.
Krafft, J., Quatraro, F. and Saviotti, P.P., 2011. The knowledge-base evolution in
biotechnology: a social network analysis. Economics of Innovation and New
Technology 20, 445-475.
Krafft, J., Quatraro, F. and Saviotti, P.P., 2014. Evolution of the knowledge base in
knowledge intensive sectors. Industry and Innovation, forthcoming.
Kuznets S., 1930. Secular Movements in Production and Prices. Houghton Mifflin, Boston.
Lawson C. and Lorenz, E., 1999. Collective learning, tacit knowledge and regional innovative
capacity. Regional Studies 33, 305-317.
Lever, W. F., 1999. Competitive cities in Europe. Urban studies 36, 1029-1044.
Maurseth, P. B. and Verspagen, B., 2002. Knowledge spillovers in Europe: A patent citations
analysis. Scandinavian Journal of Economics 104, 531-45.
Moran, P.A.P., 1950. Notes on Continuous Stochastic Phenomena. Biometrika, 37, 17-23.
Nelson, R.R. and Winter, S., 1982, An Evolutionary Theory of Economic Change, Cambridge,
Harvard University Press.
Nesta, L., 2008. Knowledge and productivity in the world’s largest manufacturing
corporations. Journal of Economic Behavior and Organization 67: 886-902.
Nesta, L., and Saviotti, P.P., 2005. Coherence of the Knowledge Base and the Firm's
Innovative Performance: Evidence from the U.S. Pharmaceutical Industry, Journal of
Industrial Economics 53: 123-42.
Nesta, L., and Saviotti, P.P., 2006. Firm Knowledge and Market Value in Biotechnology.
Industrial and Corporate Change 15: 625-52.
Nightingale, P., 1998. A cognitive model of innovation. Research Policy 27, 689-709.
Nooteboom, B., 2000. Learning and innovation in organizations and economies. Oxford:
Oxford University Press.
Paci, R. and Usai, S., 2009. Knowledge flows across European regions. The Annals of
Regional Science 43, 669-690.
Pavitt, K., 1985. Patent statistics as indicators of innovative activities: Possibilities and
problems. Scientometrics 7, 77-99.
Perloff, H. S., Dunn, E. S., Lampard, E. E. And Muth, R. F., 1960. Regions, Resources and
Economic Growth. Baltimore, Maryland: Johns Hopkins Press.
Perroux F., 1955. Note sur la notion de ‘pole de croissance’. Èconomie Appliquèe 7, 307-320.
Quatraro, F.. 2009a. Innovation, structural change and productivity growth. Evidence from
Italian regions, 1980-2003. Cambridge Journal of Economics 33, 1001-1022.
Quatraro, F., 2009b. The diffusion of regional innovation capabilities: Evidence from Italian
patent data. Regional Studies, 43, 1333-1348.
Quatraro, F., 2010. Knowledge Coherence, Variety and Productivity Growth: Manufacturing
Evidence from Italian Regions, Research Policy, 39, 1289-1302.
Quatraro, F., 2012. The economics of structural change in knowledge. London and New York,
Romijn H. and Albu, M., 2002. Innovation, networking and proximity: Lessons from small
high technology firms in the UK. Regional Studies 36, 81-86.
Saviotti, P. P., 1988. Information, variety and entropy in technoeconomic development.
Research Policy 17: 89-103.
Saviotti, P.P., 2004. Considerations about the production and utilization of knowledge,
Journal of Institutional and Theoretical Economics 160: 100-121.
Saviotti, P.P., 2007. On the dynamics of generation and utilisation of knowledge: The local
character of knowledge, Structural Change and Economic Dynamics 18: 387-408.
Schumpeter, J. A., 1939. Business Cycles. A Theoretical, Historical and Statistical Analysis of
the Capitalist Process. New York and London, McGraw Hill.
Schumpeter. 1942. Capitalism, Socialism and Democracy, London, Unwin.
Shannon, C.E., 1948. A mathematical theory of communication, Bell System Technical
Journal 27: 379-423.
Sorenson, O., Rivkin, J.W. and Fleming, L., 2006. Complexity, network and knowledge
flows. Research Policy 35, 994-1017.
Teece, D.J., Rumelt, R., Dosi G. and Winter, S., 1994. Understanding Corporate Coherence:
Theory and Evidence. Journal of Economic Behavior and Organisation 23: 1-30.
Theil, H., 1967. Economics and Information Theory. Amsterdam: North Holland.
Tödtling, F., & Trippl, M., 2005. One size fits all?: Towards a differentiated regional
innovation policy approach. Research policy 34, 1203-1219.
Usai, S., 2011. The Geography of Inventive Activity in OECD Regions. Regional Studies, 45,
Weitzman, M. L., 1998. Recombinant growth. Quarterly Journal of Economics 113, 331-360.