|
POSITION STATEMENT
There has been a valuable reinvigoration and exciting new impetus produced by developments in new economic geography theory (Fujita, Krugman and Venables, The Spatial Economy : Cities, Regions and International Trade, MIT Press 1999), although I think it is important to now give emphasis to models designed for empirical as well as theoretical consistency, in order to develop a truly scientific approach in which theory is confronted by data and new, more realistic, and ultimately more useful, theory is the outcome. With this as a motivating force, I model productivity and GDP per capita growth, mainly for regions of the European Union, using my models to monitor across-time changes in the strength of spillover and increasing returns, and to simulate regional dynamics and equilibria. Recent publications illustrating this are Fingleton B (2000) 'Spatial econometrics, economic geography, dynamics and equilibrium : a third way?' Environment & Planning A, 1481-1498 and Fingleton B (2001) 'Equilibrium and economic growth : spatial econometric models and simulations' forthcoming in Journal of Regional Science, Feb (a recent *.pdf version is available at http://www.landecon.cam.ac.uk/) and Fingleton B (2001) 'Theoretical economic geography and spatial econometrics : dynamic perspectives' forthcoming in Journal of Economic Geography, Feb.
This empirical work can be shown to be consistent with Chamberlinian monopolistic competition theory developed by Spence (1976) and Dixit and Stiglitz (1977), involving explicit micro-level market structure assumptions typical of new economic geography theory. Assuming a constant elasticity of substitution production function for intermediate immobile non-traded services, we see increasing returns and greater intermediate variety in denser areas. The (increasing returns) intermediate services combines with labor in a Cobb-Douglas production function (degree one) as the final manufactured good technology. The preferred underlying theory also assumes diseconomies due to congestion effects, following Ciccone and Hall, 'Productivity and the density of Economic Activity', America Economic Review, 1996, although reduced form parameter estimates indicate that the net outcome is increasing returns.
The foregoing theory represents an advance on our modeling of pecuniary externalities since it precisely and explicitly shows the origins of agglomeration economies, but it has nothing to say about technological externalities (previously described by Marshall). One way forward, allowing a more comprehensive suite of (spatial) externalities, is via spatially varying technical progress rates. In this vein, I model the rate of technical progress as a function of across-region knowledge spillovers plus the local rate of innovation creation and adoption determined by schooling/human capital and the initial technology level. Low initial technology regions benefit more from technology diffusion and thus achieves faster technical progress (the existence of a significant positive 'catch-up' effect has the important consequence that dynamics converge to a steady state). Fast productivity growth in 'neighboring' regions boosts technical progress locally as knowledge spills across region boundaries, and gives an explicit spatial econometric orientation to my empirical analysis. Spatial spillover implies that productivity growth is an endogenous variable, with productivity growth in an area partly determined by, and partly determining, productivity growth in 'neighboring' areas.
Consequently, a reduced form I and others have used to try to capture spatial externalities/spillovers is the space lag model familiar to spatial econometricians, involving an endogenous dependent variable (in my case productivity growth) requiring maximum likelihood or instrumental variables estimation. Various single equation specifications with these general characteristics have been estimated using SPACESTAT, with diagnostics indicating the necessity of the lag. Faute de mieux, I write my own programs in the GENSTAT programming language (occasionally going down to FORTRAN) to support the SPACESTAT results and to take the analysis further. Generally, I find GENSTAT useful for data manipulation, simulation and graphics, and other types of modeling (eg generalized linear modeling) not available in SPACESTAT. For the purpose of monitoring across time the varying effects of externalities and increasing returns, I fit spatial seemingly unrelated regressions, as set out in Luc Anselin's 'Spatial Econometrics', 1988. Estimation for these multi-equation models is usually via LIMDEP and PcFIML. I also use ARCVIEW to visualize and organize my data, and to link to SPACESTAT. This overall combination is messy and time consuming, and it would be an excellent development if a more completely integrated platform for spatial modeling combining these various elements could be developed under the aegis of CSISS.
This raises issues regarding the data and tools available facilitating complementary and alternative approaches to modeling spatial externalities. I feel that more emphasis should be given to generalized linear modeling approaches to spatial interaction analysis to model for example knowledge externalities created by labor migration between firms.
There are numerous other questions generated by our consideration of spatial externalities, for instance what is the empirical basis of our assumptions about the spatial reach of externalities and how can this be enhanced? Can progress be made modeling knowledge spillovers due to the cooperative and competitive interaction of firms? How do we introduce into our modeling overarching exogenous developments in communications technology and economic integration, which change the scale and scope of spatial interaction? What is the relationship between sectoral structure and spatial externality effects? Are regions with more small firms more likely to produce more externalities and what are the consequences for regional development?
Also, dynamics are very important. We need to keep in mind the paths implied by cross-sectional models, whether they lead to deepening core-periphery patterns and stronger clustering or the ultimate dominance of centrifugal forces, and the role played by spatial externalities in this. My preferred approach is probably close to what Fujita, Krugman and Venables call 'quantification', meaning theory consistent models whose parameters are 'based on some mix of data and assumptions, so that realistic simulation exercises can be carried out'. Thus my simulations are driven by estimated parameters and assumptions about the values of variables in empirical spatial econometric models. This focus on empirical and theoretical consistency differs from simulation exercises currently typifying new economic geography, which purposefully abstract from the real world to illustrate particular theoretical outcomes, but is seen as 'the way forward'.
|