Regional Income Growth in Europe - Insights from a Spatial Econometric Perspective

Austrian Science Fund (FWF) - Project No. P19025-G11

Duration: Aug 2006 - Jul 2010

Principal investigator: Manfred M.Fischer

Project assistants:

Lukas Leuprecht (Aug 2006 - Jun 2007)

Monika Bartkowska (Aug 2007 - Jul 2010)

Student assistants:

Peter Stumpner Sept 2006 - Jul 2007)

Andrea Kunnert (Oct 2007 - Sept 2008)

Christoph Peyerl (Sept 2009 - present)

University assistants:

Sascha Sardadvar (Aug 2006 - Mar 2009)

Aleksandra Riedl (Dec 2008 - Sept 2009)

Research collaborators:

Bernard Fingleton (University of Strathclyde)

James LeSage (Texas State University - San Marcos)

Jan Mutl (IHS Vienna)

1.1 Information on the development of the research work

Almost a quarter of a century has elapsed since William Baumol started the empirical debate on economic convergence. Since then, dozens of researchers have taken up his lead on this and related topics, generating a vast literature on cross-country and cross-regional studies of economic growth and its determinants. Research has developed in different directions, but empirical research has focused predominantly on investigating beta-convergence, namely running what are known as cross-sectional growth regressions. The vast majority of these studies have relied on the unrealistic assumption that economic growth of each region (country) is independent of that of neighbouring regions (countries), and consequently ignored the issue of spatial dependence in economic growth. Ignoring this has the implication that all estimates and inferences are potentially biased and inconsistent.

Our research activities focused around two major task areas, as described in the research proposal. The first involved the development of spatial statistical/econometric modelling approaches along with relevant methodologies and algorithms for implementing these, and the other related to their application in a context of European regions, along with substantive conclusions regarding the nature of regional growth.

Change of direction between the start and the end of the project

We changed the balance of research activities within the project. Partly, this reflected natural progression as the research has evolved, partly it was a response to what research results told us are important areas for further research, and partly it reflects the changing context within which the project had to operate. In particular, we extended our research activities in two directions: first, by a – complementary –distribution perspective that views the catching-up question as a question about the evolution of the entire cross-economy income distribution, and second, by focusing some attention on the role of knowledge capital in aggregate economic growth, with a prominent role for knowledge spillovers.

1.2 Research output, most important results and brief description of their significance

Written research output

We have published nine papers in peer-reviewed journals, including Geographical Analysis, Annals of Regional Science, Journal of Geographical Systems, Spatial Economic Analysis and Letters in Spatial and Resource Sciences. Two papers are under review for publication in Economic Modelling and International Regional Science Review, and we hope to publish a further two (currently available in form of working papers) in the immediate future. Two of the published papers have been selected for reprint in the Handbook of Applied Spatial Analysis (2010, Springer). See Section 4.a for more details.

Our research has attracted a great deal of attention in the respective scientific communities, as evidenced by 104 citations (measured in terms of Hartzing’s Publish or Perish 3).

In addition to written outputs, we have undertaken major efforts to disseminate new research findings through conferences, symposia and workshops across the world. In fact, we have taken part in 26 conferences/symposia/workshops. See Section 4.b for more details.

Main research findings

The project has generated important results, in terms of both new spatial statistical/econometric approaches with relevant methodologies and substantive conclusions regarding the nature of regional growth.

A Spatial Mankiw-Romer-Weil Model

We have developed a theoretical growth model that extends the well-known Mankiw-Romer-Weil (MRW) model by taking into account both within and cross-region physical and human capital externalities as well as technological interdependencies among the regional economies. By making use of new developments in spatial econometrics, we provide an econometric implementation of the reduced-form theoretical model, based on a spatial Durbin model specification along with the relevant methodology for estimating and correctly interpreting the model.

This extension is important because practitioners in growth empirics benefit by having at hand theoretical foundations (linked to the MRW model) to justify the use of a spatial Durbin model specification to explore spatial externalities within and across regions.

A system of 198 regions across 22 European countries over the period from 1995 to 2004 has been used to empirically test the model. Testing was performed by assessing the importance of cross-region technological interdependence, and measuring direct and indirect (spillover) effects of the MRW determinants. The results indicate that the model is consistent with the empirical evidence on technological interdependencies and these work through physical rather than human capital externalities.

A Cross-section Growth Regression Framework

Growth theories are not sufficiently explicit about which specific factors underlie the data-generating process for growth regressions, so researchers are faced with a dilemma regarding the large number of potential regressors. There is a trade-off between arbitrary selection of a small subset of variables which may give rise to omitted variables bias, and introduction of a large set of variables that will tend to increase the dispersion of the estimated coefficients, making it difficult to identify important factors. An additional complication is spatial dependence that has for the most part been ignored in this literature, which complicates the task of finding appropriate measures of factors that influence economic growth.

In contrast to non-spatial growth regressions, long-run steady-state regional income depends on (i) own-region as well as (ii) neighbouring region characteristics, (iii) the spatial connectivity structure of the regions, and (iv) the strength of spatial dependence. Given this, the search for regional characteristics that exert important influences on income levels or growth rates should take place using spatial econometric methods that account for spatial dependence as well as own and neighbouring region characteristics, and the type of spatial regression model specification and weight matrix. In co-operation with James LeSage we developed a unifying framework for dealing with these issues. We used a Markov Chain Monte Carlo Model Comparison method (labelled MC in the literature) to strategically sample from the high dimensional model space.

Spatial growth regression models produce estimates and inferences that are conditional on both the particular spatial weight matrix used to specify which regions are linked and the set of explanatory variables employed. Selection of an appropriate spatial weight matrix and explanatory variables are central to the analysis of growth empirics and substantive interpretation of the research. Competing specifications are usually non-nested alternatives so that conventional statistical procedures such as likelihood ratio tests are inappropriate.

Our empirical findings indicate that indirect effects or spatial spillovers are perhaps more important than the direct effects of regional characteristics that have been the focus of non-spatial growth regressions. For example, when appropriately measuring the direct as well as indirect impact of changes in explanatory variables such as human capital on income levels we find that spatial spillovers may negate the direct positive impact on income levels (or equivalently income growth). While the direct (own region) impact on income of this variable is positive as we would expect, the spatial spillover impact on neighbouring regions is negative, producing an overall insignificant impact.

Bayesian Spatial Methods for Modelling and Interpreting Club Convergence

In co-operation with James LeSage we developed a two-step approach to analyze regional club convergence. The first step involves identifying the number and composition of clubs using a cross-sectional spatial growth regression in conjunction with Bayesian model comparison methods. A second step uses a Bayesian dynamic space-time panel data model to assess how changes in the initial endowments of variables (that explain growth) impact regional levels of income over time. These dynamic trajectories of regional income changes over time allow us to draw inferences regarding the timing and magnitude of regional income responses to changes in the initial conditions for the clubs that have been identified in the first step. This is in contrast to conventional practice that uses a cross-sectional growth regression for club classification purposes and for estimation and inference.

The first step uses a formal Bayesian model comparison methodology to classify European regions into convergence clubs. Each region must be classified into one of m clubs. The classification takes place conditional on a cross-sectional spatial growth regression model relationship. Since observations are regions in our spatial regression model, the model comparison problem is one of comparing models based on different assignments of each observation/region to one of the m club categories.

The second step uses a Bayesian dynamic space-time panel data model to estimate the parameters for each club suggested by the first step Bayesian classification scheme. We derive analytical expressions for the partial derivative impacts of changes in the initial endowments on regional levels of income over time. These dynamic trajectories of regional income changes over time are informative regarding issues of regional convergence.

Income Distribution Dynamics and Cross-region Convergence in Europe

In our research, growth regression analysis has been complemented by distribution analysis that (i) diverts attention from the individual regional economy to the entire distribution of income across the economies, (ii) puts emphasis on the change in its external shape as well as intra-distribution dynamics, and the law of motion describing transition dynamics, and (iii) views the catching-up question as a question about the evolution of the entire cross-economy income distribution.

We developed a continuous distribution dynamics model along with relevant methodology that utilizes (continuous) kernel estimation and more powerful graphical devices [highest density regions (HDR) boxplots] for revealing persistence, mobility and polarization features, and spatial filtering to account for the role of spatial effects. HDR boxplots and spatial filtering are important extensions to a distribution perspective to analyze income dynamics, since overlooking spatial dependence in income growth may result in misguided inferences.

Application of this new empirical strategy has furnished some interesting facts, such as bimodality and polarization features. The bimodal nature of the ergodic distribution provides indication for two types of processes at work over time: a gradual and slow catching-up of the poorest regions which turned out to be regions in Central and Eastern Europe [CEE], and simultaneously a tendency towards polarization – a small group of richer regions (including Inner London, Brussels, Luxembourg, Ile-de-France) separating from the rest of the cross-section. The results obtained, moreover, clearly indicate that there is no development trap in the long-run into which poorer CEE regions will be permanently condemned.

1.3 Effects of the project outside the scientific field

Our project involved two branches of science, one involving development of spatial statistical/ econometric modelling methodologies as well as algorithms for implementing those, and the other related to substantive conclusions regarding the nature of regional growth processes in Europe.

The first aspect regarding methodological developments and software for implementing these as well as reliable sample data added to the infrastructure of the spatial modelling branch of science. The second aspect of the project relating to substantive inferences regarding factors that promote or retard regional economic growth, the spatial extent of (human capital, physical capital and knowledge) spillovers, and specific issues relating to the geography of growth among European regions is useful to policy makers, and of practical significance for socio-political as well as economic development.

Broader Implications

A broader implication of the project is that scholars undertaking empirical investigations have to take spatial dependence among sample data observations into account in their modelling efforts. Ignoring this has the implication that all estimates and inferences are potentially biased and inconsistent. For this reason, it is no surprise that many of the substantive conclusions from our project differ from accepted knowledge regarding regional economic growth. Conventional modelling methods employ the unrealistic assumption that economic growth of each region is independent of that from neighbouring regions. 

Relevance for Developments in Teaching

The project along with the research results has motivated the principal investigator to develop a novel course on Spatial Economics, a field that has emerged at the interface between economics and geography. This course is integral part of the WU Master Program in Economics that started in the study year 2009/2010. The main objective of the course is to expose students to the state of art in the field of spatial economics with special emphasis laid on growth theory, spatial econometric methods, growth empirics, and the geography of knowledge spillovers.

Research findings also had an important impact on the design and development of WU’s PhD Program in Applied Economics that – in cooperation with our partner universities in Austria – will be will be implemented in near future.

4 Related PhD theses/diploma theses

PhD theses

Sardadvar, S.: Economic growth in the Regions of Europe. Theory and empirical evidence from a spatial growth model. Doctoral thesis, completed on April 16, 2009. Vienna University of Economics and Business

Bartkowska, M.: Determinants of regional economic growth in Europe, Doctoral thesis. Vienna University of Economics and Business, work in progress

Diploma theses

Stumpner, P.: Regional income convergence in the enlarged European Union – Evidence from the distribution approach. Diploma thesis, completed on June 29, 2007. Vienna University of Economics and Business

Kunnert, A.: Empirical growth models and spatial dependence with special emphasis on model specification issues. Diploma thesis, completed on Feb 22, 2009. Vienna University of Economics and Business