1、D S EWorking Paper Urban Growth Drivers and Spatial Inequalities: Europe - a case with geographically sticky peoplePaul CheshireStefano Magrini Dipartimento Scienze EconomicheDepartment of EconomicsCa Foscari University ofVenice ISSN: 1827/336XNo. 32/WP/2008Working Papers Department of Economics Ca
2、Foscari University of Venice No. 32/WP/2008 ISSN 1827-3580 The Working Paper Series is availble only on line (www.dse.unive.it/pubblicazioni) For editorial correspondence, please contact: wp.dseunive.it Department of Economics Ca Foscari University of Venice Cannaregio 873, Fondamenta San Giobbe 301
3、21 Venice Italy Fax: +39 041 2349210 Urban Growth Drivers and Spatial Inequalities: Europe - a case with geographically sticky people Paul Cheshire London School of Economics Stefano Magrini Ca Foscari University of Venice First Draft: October 2008 Abstract We try to combine theory with empirical an
4、alysis to investigate the drivers of spatial growth processes, welfare and disparities in a context in which people are markedly immobile. Drawing on two of our recent papers (Cheshire and Magrini, 2006 and 2008), we review the evidence on the drivers of differential urban growth in the EU both in t
5、erms of population and output growth. The main conclusion from our findings is that one cannot reasonably maintain the assumption of full spatial equilibrium in a European context. This has a number of wider implications. It suggests that i. differences in real incomes in Europe - and more generally
6、 where populations are relatively immobile - are likely to persist and indicate real differences in welfare; ii. there is no evidence of a unified European urban system but rather of a set of national systems; iii. there are significant but theoretically consistent, differences in the drivers of pop
7、ulation compared to economic growth. Keywords Growth, urban system, spatial equilibrium JEL Codes O18, R11; R13 Address for correspondence: Stefano Magrini Department of Economics Ca Foscari University of Venice Cannaregio 873, Fondamenta S.Giobbe 30121 Venezia - Italy Phone: (+39) 041 2349194 Fax:
8、(+39) 041 2349176 e-mail: email This Working Paper is published under the auspices of the Department of Economics of the Ca Foscari University of Venice. Opinions expressed herein are those of the authors and not those of the Department. The Working Paper series is designed to divulge preliminary or
9、 incomplete work, circulated to favour discussion and comments. Citation of this paper should consider its provisional character. 21. Introduction In this chapter we try to combine theory with empirical analysis to investigate the drivers of spatial growth processes, welfare and disparities in a con
10、text in which people are markedly immobile. Much work has been done on regional growth processes in the USA (for example, Rey and Montouri, 1999; Glaeser et al, 1995), but the assumption has been either explicitly or implicitly, that a reasonable underlying assumption is of full spatial equilibrium.
11、 This is explicitly the case with (Glaeser et al, 1995) who argue that given full spatial equilibrium, since people are unable to improve their welfare by moving from one place to another, flows of people indicate changes in the distribution of spatial welfare - as people move to places offering sup
12、erior opportunities or lifestyles - more directly than do changes in income levels or rates of growth of income. Research in Europe, however, shows that people tend to be quite immobile. Net migration between geographically similarly sized regions in the USA is 15 times greater than in Europe (Chesh
13、ire and Magrini, 2006) despite differences in real incomes and employment opportunities being very substantially greater and geographic distances being smaller. Even mobility within countries is restricted compared to the US but national boundaries - as we illustrate here - offer particular barriers
14、 to spatial adjustment. Thus it is unreasonable to assume full inter regional or inter urban equilibrium in a European context and differences in per capita incomes are persistent and are likely to signal real spatial welfare differences. Furthermore, it implies that the drivers of what population m
15、ovement there is, may differ from the drivers of spatial differences in productivity or output growth. In this chapter, drawing on two of our recent papers (Cheshire and Magrini, 2006 and 2008), we review the evidence on the drivers of differential urban growth in the EU both in terms of population
16、and output growth. The conclusion is that while environmental goods, in the form of climate differences have significant influences on urban population growth, there is no apparent process of the European population moving to the sun. Climate differences are only significant as they vary from nation
17、al values: not as they systematically vary from European values. Moreover while there do appear to be some Europe-wide economic drivers of population movement, their influence is restricted compared to the case of economic growth differences; and spatial econometric evidence also reveals substantial
18、 national boundary barriers to both population and economic adjustment. Together these findings suggest one cannot reasonably maintain the assumption of full spatial equilibrium in a European context. Apart from increasing our understanding of the drivers of spatial growth and adjustment processes t
19、his has a number of wider implications. It suggests differences in real incomes in Europe - and more generally where populations are relatively immobile - are likely to persist and indicate real differences in welfare. Although it does not tell us how significant they are 3compared to other sources
20、of welfare differences between individuals, it does imply that people of similar personal characteristics may have different life chances because they are born in one region rather than another. It suggests, contrary to some recent assertions (for example Kresl, 2007) that there is no evidence of a
21、unified European urban system but rather of a set of national systems, with weak responses to variations in local economic opportunities when national boundaries intervene. It also shows there are significant but theoretically consistent, differences in the drivers of population compared to economic
22、 growth, Agglomeration economies, concentrations of R and below that the Kreise (NUTS Level 3) - 439 of them in 2003. Britain has 12 NUTS 1 regions corresponding in mean size to the Lnder but only one of them - Scotland - has any real administrative or fiscal independence. In Britain there are only
23、133 of the smaller units supposedly equivalent to the Kreise. Bavaria, despite including major cities such as Munich, had a population density of only 174 people per square km compared to 4,539 in the NUTS Level 1 region of London or 2,279 in Hamburg (CEC, 2004). More significant than their heteroge
24、neity in size and administrative powers is the fact the official NUTS regions are economically heterogeneous, in some cases containing very different local economies within the same statistical unit (for example, Glasgow and Edinburgh in Scotland or Lille and Valenciennes in Nord-Pas-de-Calais) and
25、in others dividing a single city-region between as many as three separate units. The functional reality of Hamburg, for example, is divided between three different Lnder, Hamburg, Schleswig-Holstein and Niedersachsen. There are thus many NUTS regions with large scale and systematic cross border comm
26、uting and some contain mainly dormitory suburban areas of large cities. Others (for example, Brussels, London, Bremen or Hamburg) are effectively urban cores or only small parts of urban cores. This means that residential segregation influences the value of variables such as unemployment, health or
27、skills if measured on the basis of the boundaries of NUTS; and measures of Gross Domestic Product, Value Added or productivity can be grotesquely distorted since output is measured at workplaces and people are counted where they live. Even measured growth in GDP pc can be seriously distorted since o
28、ver time residential (de)centralisation may occur at different rates to job (de)centralisation. The reported growth in GDP pc for the NUTS region of Bremen during the 1980s, for example, was 40 percent higher than for the Bremen functionally defined region. These problems are concentrated in the 2No
29、menclature des Units Territoriales Statistiques (N.U.T.S.) regions. This is a nesting set of regions based on national territorial divisions. The largest are Level 1 regions; the smallest for which a reasonable range of data is available are Level 3. Historically these corresponded to Counties in th
30、e UK, Dpartements in France; Provincies in Italy or Kreise in Germany. 5larger cities, because these tend to spill over their administrative boundaries, and in the richer regions. This last facet of the distortions to official regional statistics results not only because richer regions tend to inclu
31、de larger cities but because a significant proportion of larger cities extend functionally beyond their administrative boundaries, so their recorded GDP pc is overstated. These are obvious points, causing serious reservations in relation to the many published analyses of regional growth rates in Eur
32、ope using the official Eurostat data for NUTS regions. They mean official measures of so-called regional disparities - showing, for example, that in 2001 the region of Inner London was 2.5 times as rich in per capita GDP as the mean for the EU of 15 and 3.2 times as rich as the UKs poorest region, a
33、re complete nonsense. It is for these reasons that we rely on our own data for FURs. There is one additional advantage of this choice in the present context which is that FURs are as economically independent divisions of national territories as it is possible to construct. They represent concentrati
34、ons of jobs and all those people who depend on those jobs - the economic spheres of influence of major cities. So the benefits of additional employment or output are as confined to those who live within them as is possible for any sub national regions. We have data for the 121 largest FURs in the ol
35、d EU of 12 member states, excluding Berlin and the cities of the former Eastern Germany. We are comparing the drivers of population growth, analysed as quasi-net migration between 1980 and 2000, with those of growth in GDP per capita measured at PPS between 1979 and 1994. Two idiosyncrasies of our m
36、odels should be noted. In our analysis of growth in GDP pc we do not include the initial level of GDP pc. So this is not a contribution to the regional growth regression literature stemming from the work of Barro and Sala-i-Martin (Barro, 1991; Barro and Sala-i-Martin, 1991; 1992 or 1995). Not only
37、do we find this literature theoretically suspect (see the discussion in Cheshire and Malecki, 2004, for reasons) but we find it empirically suspect, too. In our better specified models, including the initial level of GDP pc clearly introduces multicollinearity and leads to very unstable parameter es
38、timates for the variable - even signs flip. In essence, it is possible to generate either apparent -convergence or -divergence in equally respectable looking models but in all the better specified ones the effect of initial GDP pc on subsequent growth performance is statistically entirely insignific
39、ant. The second idiosyncrasy is our approach to, and interpretation of, issues of spatial dependence. We interpret a finding of spatial dependence as not surprising since systematic patterns in growth should be expected; but as an indication of omitted variable(s) and so failing to account for those
40、 patterns. In our models instead of reporting results with spatial lags (which cure any problems of spatial dependence we encounter) we find additional variables directly reflecting spatial processes. In testing for spatial dependence and formulating our variables to reflect spatial economic process
41、es we also find 6that results critically depend on how the spatial weights matrix is formulated. Following standard procedures to specify the spatial weights matrix, using contiguity, geographic or even time-distance test statistics reveal no apparent problems of spatial dependence with the theoreti
42、cally more satisfying models. Problems of spatial dependence are only indicated when an additional time-distance penalty for national borders is introduced. This is consistent with other findings - for example that climatic differences only influence population mobility if expressed relative to a co
43、untrys mean indicating the continuing barrier national borders present in Europe to processes of spatial adjustment. We start by summarising the results on the drivers of population growth reported in detail in Cheshire and Magrini (2006) and then summarise the results of a more recent analysis of t
44、he drivers of growth in FUR GDP pc (Cheshire and Magrini, 2008). Because we find strong indications of population immobility and sluggish migration response across national borders, we expect the drivers of productivity/GDP growth to be significantly independent and also that over any period we can
45、actually observe we will not be observing a full spatial equilibrium. So differences in economic growth rates across space in Europe are likely to represent real differences in welfare. In analysing the drivers we pay particular attention to the role of highly skilled human capital, concentrations o
46、f R Rappaport, 2004) have shown that migration is - other things equal - sensitive to better weather. Likewise in the Quality of Life literature (for example, Blomquist et al, 1988; Gyourko and Tracey, 1991) climate is an important driver of quality of life. Data does not allow us to estimate full Q
47、uality of Life models in Europe. However we can include measures of weather and the results of including a selection of these are shown in the fourth to seventh columns of Table 2. We can see that these are statistically highly significant and, if anything, perform rather better than the geographic
48、position of a FUR. The functional form that is most appropriate seems to be quadratic, although the relationship is quite close to linear. These results confirm that it is only the climate of a FUR relative to the mean for its country that is significant. Again expressing climatic differences relati
49、ve to the mean for the EU proves entirely non-significant. Table 3 shows the results for some better performing models and shows that the best result are achieved if measures of both dryness and warmth relative to national means are included. Table 3 about here Diagnostic test results are reported in Cheshire and Magrini, 2006. These suggest that there are no problems of either heteroskedasticity or non-normality of errors. The value of the multicollinearity condition number is relatively high in most of the models in which climate variables are included
