TWO SPEED REGIONAL CONVERGENCE IN PORTUGAL AND THE IMPORTANCE OF STRUCTURAL FUNDS ON GROWTH

TWO SPEED REGIONAL CONVERGENCE IN PORTUGAL AND THE IMPORTANCE OF STRUCTURAL FUNDS ON GROWTH Micaela Antunes and Elias Soukiazis Faculty of Economics, University of Coimbra, Portugal Abstract The aim of this paper is twofold: in first place we want to ascertain if there is any difference in the convergence process between the Littoral (coastal) and the Interior (inland) areas in Portugal in terms of per capita income. In second place, we examine the relevance of Structural Funds (European Regional Development Funds) as conditioning factors influencing the convergence process in Portugal and to what extent these funds contributed to the growth of regional per capita income. In doing so, we apply a panel data approach to the convergence in per capita income among the 30 NUTS III regions in Portugal, and a separate analysis is given for the regions constituting the Littoral and Interior zones. The evidence shows that the distinction between the Littoral and Interior areas is important in the convergence process in Portugal and that convergence is slightly faster in the regions of the Interior. On the other hand, structural funds have a significant positive effect only in the Littoral area helping these regions to grow faster. Keywords: per capita income, Littoral/Interior division, European Regional Development Fund, absolute and conditional convergence, panel regressions. Correspondence Address: Faculdade de Economia Universidade de Coimbra, Av.Dias da silva, 165, 3004-512 Coimbra, Portugal. E-mail: elias@fe.uc.pt; micaela.antunes@ci.uc.pt. - - 1

1. Introduction The purpose of this paper is mainly to find whether there is any difference in the convergence process of per capita income among the Portuguese regions which constitute two basic areas, the Littoral (coastal) and the Interior (inland) zones at the NUTS III 1 level. The study covers the 1991 2000 period, where data is available. The neoclassical approach to convergence is initially used, derived from the Solow s (1956) neoclassical model of the production function with diminishing marginal returns to capital properties and exogenous technical progress. This approach predicts that poorer economies tend to grow faster than richer ones in earlier stages (due to the lower capital stock they possess) and then in the long run they all grow at similar rates. Convergence is unconditional (or absolute) to a common steady-state for all economies and divergence is a transitory short term phenomenon reflecting adjustments towards a long run equilibrium level of per capita income. Absolute convergence is found when the inverse relationship between the growth of per capita income and its initial level is confirmed and this result is more likely to occur for a set of economies with similar economic and institutional characteristics. The higher the distance from the steady-state the higher the speed of convergence is expected to be found. The second approach used, is derived from the new theory of endogenous growth (Barro, 1991, Sala-i-Martin, 1994). Convergence is conditional to some structural factors with increasing returns to scale properties, such as, human and capital accumulation, technological progress, innovation, among others. Economies converge to different steady-states because of differences in economic structures. Convergence is not the rule, but rather the exception, occurring when the economies are able to develop activities with increasing returns to scale characteristics. Convergence is found after differences in the steady states across economies are controlled for. In this study, Portugal is divided into two main areas, the Littoral constituted from 16 coastal regions and the Interior constituted from 14 inland regions. Both approaches of absolute and conditional convergence are employed to find differences in the speed of convergence between these two areas. The ERDF (European Regional 1 NUTS stands for the Nomenclature of Territorial Units. It is a regional territorial division defined by Eurostat that enables the elaboration of credible regional statistics at the European level. - - 2

Development Fund) is used as a conditioning factor to check its relevance in explaining the growth of per capita income in the Portuguese regions. The convergence equations, both absolute and conditional, are estimated by using a Panel data approach. The advantage of this methodology is that it takes into account the individual specific effects of the aggregate production functions across regions. This approach is preferable to the single cross-sectional analysis 2, allowing to control the omitted variable bias and to introduce dynamics into the estimated convergence equations. The remainder of the paper is organized as follows: in section 2, statistical information is given on the per capita income position of the Littoral and Interior areas relatively to the national average. In section 3 we provide evidence of σ- convergence in per capita income for all the 30 NUTS III regions and the subsets of the Littoral and the Interior areas. Section 4, makes a short theoretical development of the convergence equation derived from the Solow s neoclassical growth model and discusses the relevance of panel data estimation of the convergence equations. Section 5, shows the importance of the Littoral/Interior division on the estimation of the convergence equations. Section 6, initially gives information about the distribution of structural funds (ERDF) among the Portuguese regions and later provides evidence, through an econometric analysis, of how these funds affected regional growth and the convergence process among all regions and in particular among the regions that constitute the Littoral and Interior areas. The final conclusions are summarized in the last section. 2. Per capita income of the Portuguese regions and the dichotomy between the Littoral and the Interior zones. Portugal is divided into 30 regions according to the NUTS III classification (see the map presented in the Appendix). It is well known that coastal regions are more developed than inland regions and that there is a lower population density in the inland area comparatively to the coastal zone. For this reason it makes more sense to divide Portugal, geographically, into two main areas, the Littoral (coastal) and the Interior (inland area) than between the North and the South as usually is done in regional studies (i.e. Italy, Spain). According to this division based on geographical location 2 For this methodology see Soukiazis(2003) - - 3

criteria the Littoral is constituted by 16 regions: Minho-Lima, Cávado, Ave, Grande Porto, Entre Douro e Vouga, Baixo Vouga, Baixo Mondego, Pinhal Litoral, Oeste, Grande Lisboa, Península de Setúbal, Lezíria do Tejo, Alentejo Litoral, Algarve, Açores (islands) and Madeira(islands) and the Interior by 14 regions: Tâmega, Douro, Alto Trás-os-Montes, Pinhal Interior Norte, Dão-Lafões, Pinhal Interior Sul, Serra da Estrela, Beira Interior Norte, Beira Interior Sul, Cova da Beira, Médio Tejo, Alto Alentejo, Alentejo Central and Baixo Alentejo. Table 1 presents the per capita income 3 relative position of the Littoral and the Interior areas in relation to the national average, for the period 1991-2000 where data is available. The data shows that per capita income in the Littoral area is about 15% higher than the national average in contrast to the Interior where per capita income is almost 18% lower that the country s average. It is also observed that the relative gap between the two areas and the national average persists during the 90 s. Table 1. Per capita income of the Littoral and the Interior areas relatively to the national average, 1991-2000. 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Littoral 115.38 116.03 115.32 113.93 115.05 114.83 116.09 116.26 116.20 115.14 Interior 82.42 81.67 82.49 84.09 82.80 83.05 81.62 81.42 81.49 82.69 Data source: National Institute of statistics (INE), Regional Accounts, 1995, 1995-1999 and 2000. 3. Dispersion of per capita income among the Portuguese regions. (σ-convergence). The σ-convergence concept is a measure used to determine the dispersion of per capita income over time for a sample of different economies. The coefficient of variation is normally used to measure σ-convergence given by the standard deviation over the mean of the sample. Figure 1 illustrates the evolution of regional disparities in per capita income among the 30 NUTS III regions and separately along with the 16 Littoral and the 14 3 The ratio of GDP to total population (product per head) is used as a proxy for per capita income. GDP is at current prices since there are not regional price deflators to express the variable in real terms. However, considering the fact that overall inflation in this period has been kept to low levels (around 2%), the differences in price levels is not very substantial between regions, to the extent to bias the results. - - 4

Interior regions. As concern with the whole sample the dispersion of per capita income remained almost unchanged in the whole decade of the 90 s with a slight reduction only in the period 1993-1996. However, the behaviour of the Littoral and the Interior sets of regions is different. The disparities in per capita income between the regions of the Littoral area are slightly declining thought time, showing evidence of a moderate σ-convergence. The behaviour of the regions belonging to the Interior is more unstable. The differences in per capita income remained constant from 1991 to 1994, but after that there is a substantial increase in the dispersion of per capita income. For the whole period the Interior area shows evidence of divergence among its regions. Figure 1. σ-convergence in per capita income among the Portuguese regions, 1991-2000. 0,5 Coefficient of Variation 0,4 0,3 0,2 0,1 0 1991 1993 1995 1997 1999 Total NUTSIII Litoral Interior Years 4. The neoclassical approach to convergence and the relevance of panel data regressions. The idea of absolute convergence emerged from the Solow s growth model based on the Cobb-Douglas production function incorporating a labour-augmenting technological progress of the type: 4 α 1 α () t = K() t [ A() t L() t ] Y, 0<α<1, (1) 4 The theoretical development follows closely Islam (1995), with the necessary adaptations. - - 5

where Y is output, K and L are capital and labour, respectively, A is technology and α the elasticity of output with respect to capital. In this model L and A are assumed to grow exogenously at rates n and g, () ( ) nt gt respectively, so that: L t = L 0 e and A() t = A( 0) e. The model also assumes that s is the constant fraction of output that is saved and invested (s=s/y) and defines output and capital stock per unit of effective labour as y = Y AL and k = K AL, respectively. From the definition of output per unit of effective labour y = Y AL and the expression of steady state output, it is possible to define per capita income at the steady state, as: 5 Y () t α α ln ( ) ( + + δ () = ln A 0 + gt + ln( s) ln n g ) (2) L t 1 α 1 α In equation (2) gt is a constant (technological progress is assumed to be the same for all economies and t is fixed), A(0) reflects not only the technological level but also resource endowments, legal system and institutions, among others, and so it may be different across countries or regions (Mankiw et al., 1992). Therefore, the term ln A( 0) = a + ε can be decomposed into two parts: one is constant (a) and the other is random (ε), representing a stochastic shock or a country (region)-specific change. After some rearrangement of the above expression and also by considering the equation describing per capita income out of the steady state behaviour, we arrive to the dynamic panel data model 6, given by: ln y βt βt ( t ) ln y( t ) = ( 1 e ) ln() s ( 1 e ) ln( n + g + δ ) 2 1 (1 e where ( 1 e βt ) ln A( 0) βt ) ln y α 1 α α 1 α βt βt ( t1 ) + ( 1 e ) ln A( 0) + g( t 2 e t1 ) + v it is the time-invariant individual effect term reflecting country (region) specific effects and v it is the error term that varies across countries and time periods. Estimating equation (3) by using panel data techniques is the way to control for (3) 5 Analytically: Y ln L ln () t () t Y () t L () t ln A α 1 α () t = ln( s ) ln ( n + g + δ ) α 1 α α 1 α α 1 α [ ln A ( 0 ) + gt ] = ln () s ln( n + g + δ ) 6 For this development see Soukiazis and Antunes (2004) - - 6

the individual country (region) effects. Another advantage is that in the single crosssection regression, s and n are assumed to be constant for the entire period studied. Such hypothesis is more realistic in panel data estimations that consider shorter periods of time, say, annual data. The main problem with the cross-sectional regressions is that the individual specific effects of the aggregate production function are ignored. These effects can be correlated with the explanatory variables included in the convergence equation, creating estimation bias due to the omission of relevant variables. Hence, an apparent difficulty in cross-sectional regressions (especially in conditional convergence) lays on the fact that only differences in preferences and technology can be accounted for as can be properly observed and measured. Yet, differences in these parameters have dimensions not easily detectable and measured and so they are not considered in cross-section analysis. The use of panel data allows to take care of the variable omission problem and to test for convergence in a more consistent way. The fact that, in empirical studies, conditional convergence has been found to be higher in the panel regressions reinforces the idea for a higher policy activism. In order to increase the steady state level of per capita income, authorities must not only care about the rates of saving and labour force growth but also with every tangible and intangible factors that may be related to individual effects. These structural factors have direct positive effects on the long-run income level and also indirect ones, through their influence on saving and population growth rates. 5. Estimation of the convergence equation and the importance of the Littoral / Interior division. We shall now adopt a more systematic mode in finding evidence of convergence in per capita income among the Portuguese regions through a panel data estimation approach. Since our interest is to find if there is any difference in the convergence process between the Littoral and the Interior areas we introduce initially a dummy variable, labeled L, to indicate that the region belongs to the Littoral area 7. The statistical significance of this variable will show that it makes difference in the convergence process if the region belongs to the Littoral or to the Interior area. The simplified convergence equation to be estimated assumes the following specification: 7 The Dummy variable L takes the value 1 if the region belongs to the Littoral area and 0 otherwise. - - 7

L ln y = a + b ln y + γ i, t i, t 1 i + u (4) i, t where α is the steady state level and b=(1-e -βt ) the convergence coefficient. The rate of convergence is given by the relation β=-ln(1-b)/t. In this equation the growth of per capita income of the regions is related to their initial level (lagged income) and the fact that the region belongs or not to the Littoral zone. We initially use pooled data to estimate equation (4) and we split the whole period into two subsets (1991-1994) and (1995-2000) to be consistent with the σ-convergence finding that suggested a lower dispersion of per capita income in the first sub-period than in the second. The estimation results are shown in Table 2. Table 2. Convergence in per capita income among the Portuguese Regions Estimated equation: lny i,t = a + b lny i,t-1 + γ L i 0.1476 0.04786 87 2.35-0.0120 Method Period Constant lny i,t-1 β L i R 2 SEE D.F. DW 1991-2000 0.1582 (9.4406) -0.0489 (-4.9881) -0.0477 0.0168 (2.6268) 0.0853 0.04564 267 2.24 Pooled 1991-1994 0.2248-0.0895-0.0857 0.0199 OLS (6.5367) (-3.7865) (1.5759) (n) 1995-2000 0.0881-0.0121 0.0033 0.0094 0.02807 147 1.88 (4.6499) (-1.1735) (n) (0.5797) (n) Absolute 1991-2000 0.1438-0.0361-0.0355 0.062 0.0461 268 268 convergence (8.9828) (-4.1981) 1991-1994 0.20-0.0673-0.0651 0.123 0.0483 88 2.35 (6.4859) (-3.5179) 1995-2000 0.0827 (5.0162) -0.0085 (-1.0348) (n) -0.00846 0.010 0.028 148 1.87 Notes: Figures in parenthesis are t-ratio. (n) Indicates that the estimated coefficient is not statistical significant at the 5% significance level. L- Littoral dummy. Table 2 shows that the annual convergence rate in per capita income among the Portuguese regions, in the whole period, runs at a higher rate (4.77%) than that observed in the case of absolute convergence (3.55%), where no distinction is made between the Littoral and the Interior areas. The dummy variable is statistically significant (only in the whole period) and has a positive effect suggesting that it makes difference in the convergence process if a region belongs or not to the Littoral area. Regions of the Littoral have the advantage to grow faster and contribute significantly to the speed of converge across the Portuguese regions. These results seem to confirm the known - - 8

convergence club hypothesis introduced by Chatterji (1992). The higher convergence speed in the first period 1991-1994 confirms the early finding of lower dispersion in per capita income in this period, through the σ-convergence pattern (Figure 1). 6. The importance of structural funds on regional growth. In this section we want to examine how Structural Funds (ERDF) 8 affected regional growth and the process of convergence of the Portuguese regions, during the period 1991-1999 9. In Table 3 we explain the distribution of structural funds, ERDF per inhabitant 10, along the Portuguese regions, the period average and total values of ERDF received by each region over the whole period. In the same table we provide two ways of ranking the regions according to the amount of structural aid received: first, we ranked the regions according to the average ERDF per capita received, and second, according to the total amount received in the whole period. The two ranking measures give a different position of the region, with respect to the structural aid received. For instance, Grande Lisboa and Grande Porto, are in the 1 st and 3 rd places when the total amount of funds is used but in the 17 th and 21 st positions when per capita figures are used. Another illustrative case is Alto Alentejo situated in the 22 nd position when the total amount of funds is taken into account but in the 9 th position when per capita amount is considered. To our opinion the ERDF per capita measure is more coherent since it takes into account the population dimension of each region. Using per capita figures, it is interesting to see that in average the Littoral area (the more developed area) received 137.79 euros per inhabitant whereas the Interior (the less developed area) received a lower amount, 124.78 euros per capita. This kind of distribution contradicts the basic principle of the structural funds orientation policy which aims to assist more the less developed regions. The same picture we get if we consider the total amount of structural funds the regions received in the period 1991-8 ERDF - European Regional Development Fund. Created in 1975, it finances regional development programs oriented to less developed regions, with a co participation in the projects that ranges from 20% to 85%. The aim is to contribute for the reduction of socio-economic disparities among regions of the European Union, financing programmes which contribute to infrastructure improvement. The ERDF funds are mostly absorbed by regions with per capita GDP less than 75% of the Community s average. 9 Data on ERDF at NUTS III level is available for the 1991-1999 period only. 10 ERDF per capita is obtained by the ratio ERDF/Resident Population for each region and it is expressed in euros per inhabitant. - - 9

1999. The regions of the Littoral area appear at the top of the ranking positions whereas the regions of the Interior occupy the last positions of the scale. Table 3. Distribution of structural funds among the Portuguese regions. ERDF per capita, 1991-1999 (euros per inhabitant). Regions Years Period Average Total of ERDF (thousands of euros) 1991 1992 1993 1994 1995 1996 1997 1998 1999 Norte 1097.7 661.0 271.4 896.7 1.223.7 827.8 709.5 730.3 961.8 820.0 3125841 Minho-Lima L 308.5 235.0 67.62 94.9 175.9 118.1 94.4 112.1 217.3 5º 158.2 9º 356145 Cávado L 120.6 70.1 36.6 76.1 130.5 103.1 77.9 85.3 94.0 23º 88.2 12º 294429 Ave L 92.4 45.3 47.2 130.6 128.8 102.0 54.3 70.7 100.6 26º 85.8 8º 375882 Grande Porto L 132.1 54.7 12.4 131.3 133.7 91.8 124.5 120.4 104.8 21º 100.6 3º 1090385 Tâmega I 56.6 32.9 6.1 53.7 89.9 140.3 64.5 60.1 77.9 30º 64.7 11º 308632 Entre Douro e Vouga L 58.8 98.0 35.1 224.2 133.5 117.7 106.2 97.8 113.6 15º 109.4 16º 259426 Douro I 116.1 48.0 30.9 88.1 164.8 61.5 73.2 71.0 135.9 24º 87.7 21º 181920 Alto Trás-os-Montes I 212.8 77.1 35.6 97.8 266.5 93.5 114.4 113.0 117.6 13º 125.4 17º 259022 Centro 1196.9 872.6 357.7 1148.0 1875.8 1844.2 732.5 761.6 1163.4 1105.8 1687445 Baixo Vouga L 112.5 43.1 20.4 84.9 169.3 169.4 73.8 109.1 164.4 18º 105.2 26º 347025 Baixo Mondego L 86.2 73.3 28.7 113.6 228.4 136.2 35.1 38.9 104.1 22º 93.8 13º 281700 Pinhal Litoral L 117.1 56.8 7.9 169.1 152.6 115.2 64.0 69.1 179.2 19º 103.5 19º 217958 Pinhal Interior Norte I 95.2 106.4 11.3 88.8 128.2 68.5 52.3 53.0 118.0 28º 80.2 28º 99830 Dão-Lafões I 115.7 77.4 38.9 238.0 221.3 109.7 51.5 54.3 67.9 16º 108.3 15º 276660 Pinhal Interior Sul I 275.2 78.2 2.8 67.7 169.6 67.6 76.3 71.5 112.0 20º 102.3 29º 44699 Serra da Estrela I 28.8 55.9 12.6 106.8 135.8 109.5 56.4 47.7 46.2 29º 66.6 30º 31385 Beira Interior Norte I 24.1 129.8 99.9 78.1 320.4 332.5 158.1 159.3 106.0 6º 156.5 23º 163879 Beira Interior Sul I 205.7 139.8 82.2 108.4 230.4 124.7 120.0 127.7 204.4 7º 149.3 27º 106849 Cova da Beira I 136.5 111.9 53.0 92.5 120.0 610.7 45.2 31.1 61.0 10º 140,2 26º 117460 Lisboa e Vale do Tejo 1388,8 356.0 126.8 481.6 919.3 634.8 413.0 426.1 342.8 565.5 3532113 Oeste L 143.7 69.6 8.2 44.7 143.1 135.4 69.9 73.5 56.8 27º 82.8 14º 278923 Grande Lisboa L 129.8 63.4 33.1 73.9 180.1 216.2 84.2 80.4 113.6 17º 108.3 1º 1818054 Península de Setúbal L 673.9 98.2 77.4 155.2 280.4 75.7 53.7 60.6 46.3 4º 169.0 2º 1010050 Médio Tejo I 228.8 91.6 2.6 121.4 195.4 112.9 112.4 116.9 79.4 14º 117.9 18º 238408 Lezíria do Tejo L 212.7 33.2 5.4 86.5 120.3 94.6 92.9 94.7 46.8 25º 87.5 20º 186679 Alentejo 797.4 399.9 207.0 602.6 601.7 684.3 963.3 1056.7 859.3 685.8 864038 Alentejo Litoral L 192.7 104.2 27.7 162.0 195.3 128.2 147.8 166.3 116.5 11º 137.9 25º 119862 Alto Alentejo I 130.4 90.0 65.0 204.6 150.3 161.3 139.3 187.2 156.8 9º 142.8 22º 164640 Alentejo Central I 215.1 120.5 70.9 154.9 152.7 221.3 611.2 636.1 322.3 2º 278.3 7º 423946 Baixo Alentejo I 259.2 85.3 43.4 81.1 103.4 173.6 65.0 67.1 263.6 12º 126.9 24º 155591 Algarve L 355.5 144.3 45.5 111.6 92.9 239.5 105.5 95.4 102.7 8º 143.6 6º 458493 R. A. Açores L 402.5 214.0 82.7 864.7 596.5 183.3 343.3 341.4 216.0 1º 360.5 4º 774344 R. A. Madeira L 302.4 136.7 187.3 614.7 878.7 90.7 54.9 56.1 112.3 3º 270.4 5º 606879 Notes: L-Littoral (16 regions); I-Interior (14 regions) 1º-30º_ranking of regions by increasing order of the ERDF received. Dta source: DGDR (Direcção-Geral de Desenvolvimento Regional) - - 10

Having in mind the distribution of structural funds among the Portuguese regions, we next try to test the impact that these funds had on regional growth and how the funds affected the convergence process. To do so, we estimate the conditional convergence equation assuming that structural funds play an important role in explaining regional growth and regional convergence. For comparison, we also report the estimation results obtained from the absolute convergence equation. Three methods of estimation are used: the pooled data OLS estimation, the fixed effect method by using individual dummies to capture specific regional characteristics, and the random effect GLS estimation method where we assume that specific effects are captured in the stochastic error. The estimated results are reported in Table 4 and give some interesting insights. Table 4. The importance of structural funds on regional convergence. Panel data estimations, 1991-1999. Pooling OLS Fixed Effects LSDV Random Effects GLS Estimated equation: lny i,t = a + b lny i,t-1 (absolute convergence) Constante lny i,t-1 β R 2 SEE D.F. DW 0.1434-0.0358-0.0352 0.054 0.0476 238 2.32 (8.0680) (-3.6711) Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constante lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.137-0.0374-0.0367 0.00007 0.086 0.0469 237 2.31 (7.7619) (-3.8886) (2.901) Estimated equation: lny i,t = a i + b lny i,t-1 (absolute convergence) Constante lny i,t-1 β R 2 SEE D.F. DW -0.0982-0.0936 0.226 0.0459 209 2.51 (*) (-6.3325) Estimated equation: lny i,t = a i + b lny i,t-1 + c ERDF i,t (conditional convergence) Constante lny i,t-1 β ERDF i,t R 2 SEE D.F. DW -0.101-0.0962 0.00008 0.252 0.0453 208 2.47 (*) (-6.592) (2.6704) Estimated equation: lny i.t = a + b lny i,t-1 (absolute convergence) Constante lny i.t-1 β R 2 SEE D.F. DW 0.1422-0.0352-0.0346 0.053 0.0477 238 2.32 (8.0622) (-3.6334) Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constante lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.1333-0.0354-0.0348 0.00007 0.083 0.0471 237 2.3 (7.7334) (-3.7664) (2.9263) Notes: ERDF is the ratio ERDF/Resident Population, for the 30 NUTS III regions. Figures in parenthesis are t-ratio. (*)- All dummies are statistically significant. - - 11

First of all, and independently of the method of estimation used, the annual convergence rate and the statistical significance of the convergence coefficient b (and R 2 ) are higher when we take into account the ERDF per inhabitant variable, as a conditioning factor in the convergence equation. Second, the impact of structural funds on regional growth is positive and statistically significant in all methods of estimation, a clear evidence that structural funds contributed positively to the growth of per capita income in the Portuguese regions. However, the marginal impact of structural funds on regional growth is not very substantial: the evidence shows that each additional euro per inhabitant received from the ERDF contributed to only 0.00007% increase in per capita income. Next question to address is whether structural funds received by regions belonging to the Littoral area contributed more significantly to the regional growth. To test this hypothesis we introduce into the conditional convergence equation a multiplicative dummy variable, ERDF*L, where ERDF are structural funds per capita received by each region (each year from 1991 to 199) and L a dummy variable with the value of 1 when the region belongs to the Littoral area and 0 when it belongs to the Interior area. The specified conditional convergence equation relates, in this way, the annual growth of per capita income of each region to the convergence factor (the lagged per capita income), the per inhabitant structural funds received and the funds received separately by the Littoral regions. Table 5 illustrates the estimation results of this equation. Table 5. The importance of the Littoral area on regional growth, through the structural funds impact, 1991-1999. Pooling OLS Fixed Effects LSDV Random Effects GLS Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t + γ ERDF*L i Constant lny i,t-1 β ERDF i,t ERDF*L R 2 SEE D.F. DW 0.1514-0.0447-0.0437-0.000003 0.0001 0.128 0.0459 236 2.32 (8.5065) (-4.6272) (-0.0740) (n) (3.3623) -0.0956-0.0913 0.000009 0.0001 0.264 0.0450 207 2.48 (*) (-6.1589) (0.1772) (n) (1.8611) + 0.1504-0.0441-0.0432-0.000002 0.0001 0.127 0.0460 236 2.32 (8.5024) (-4.5963) (-0.0599) (n) (3.3627) Notes: ERDF is the ratio ERDF/Resident Population, for the 30 NUTS III regions. ERDF*L is the multiplicative dummy for the Littoral regions. * - All dummies are statistically significant. Figures in parenthesis are t-ratio. (n) - Indicates that the estimated coefficient is not statistically significant at 5% significance level. + - The coefficient shows statistical significance at 10% level. - - 12

By observing the results we can infer the following: First, the speed of convergence increases (relative to Table 4) when the multiplicative dummy variable for the Littoral is introduced into the convergence equation. Second, the coefficient of the multiplicative dummy variable is statistically significant (in the fixed effect estimation at 10% only) and has a positive sign. Third, the marginal impact of this variable is higher to that found in Table 4, where the distinction between the Littoral and the Interior was not made. The evidence shows that regional per capita income increases by 0.0001% for each additional euro per capita received from the ERDF by the Littoral regions. In other words, the fact that a region belongs to the Littoral zone is an a priori advantage for the region it self to grow faster. This advantage in terms of regional growth can be explained by the early observation that regions belonging to the Littoral zone benefit more from the ERDF distribution. Another explanation might be that structural funds received by the Littoral regions are invested in a more efficient way. Two last questions to address are whether the convergence process in per capita income is different between the Littoral and the Interior areas and whether structural funds affected differently regional growth in these two areas. To answer to these questions we provide in Table 6 two separate estimations of the convergence equations for the Littoral and the Interior sets of regions. Two main conclusions can be derived from the estimated results, independently the method of estimation used: First, regional convergence among the Interior regions is slightly higher than the convergence in the Littoral area. This means that regions in the Interior become more homogeneous with the passage of time and that they converge to a different steady state comparing to the Littoral regions. Second, and more importantly, the effect of structural funds (through ERDF) on the regional growth of capita income is statistically significant only in the case of the Littoral area and this is consistent with the prior finding of Table 5. Additionally, the degree of explanation of the regressions increases substantially in the case of the Littoral area when structural funds are included as conditioning factors to control for differences in the steady state. Once more, the evidence shows that structural funds contributed more substantially in the improvement of living standards in the regions of the Littoral area. - - 13

Table 6. Convergence in per capita income in the Littoral and Interior areas and the impact of structural funds on regional growth, 1991-1999. Pooling OLS Fixed Effects LSDV Random Effects GLS Pooling OLS Fixed Effects LSDV LITTORAL Estimated equation: lny i,t = a + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW 0.1810-0.0513-0.0500 0.118 0.0415 126 2.13 (7.3739) (-4.1145) Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.1539-0.0445-0.0435 0.0001 0.215 0.0393 125 2.09 (6.3437) (-3.7239) (3.9222) Estimated equation: lny i,t = a i + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW -0.0838-0.0805 0.195 0.0517 111 2.35 * (-4.6292) Estimated equation: lny i,t = a i + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW -0.0780-0.0751 0.0001 0.364 0.0377 110 2.26 * (-4.5686) (3.9726) Estimated equation: lny i,t = a + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW 0.1900-0.0559-0.0544 0.124 0.0406 126 2.18 (7.2735) (-4.2217) Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.1644-0.0500-0.0488 0.0001 0.223 0.0383 125 2.14 (6.3595) (-3.9217) (3.9767) INTERIOR Estimated equation: lny i,t = a + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW 0.1676-0.0559-0.0544 0.064 0.0524 110 2.45 (5.0004) (-2.7383) Estimated equation: lny i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.1721-0.0609-0.0591 0.00003 0.067 0.0526 109 2.44 (4.9942) (-3.7239) (0.5914) (n) Estimated equation: lny i,t = a i + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW -0.1167-0.1104 0.195 0.0518 97 2.65 * (-4.4237) Estimated equation: lny i,t = a i + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW -0.1199-0.1132 0.00002 0.196 0.052 96 2.65 * (-4.335) (0.4019) (n) Estimated equation: lny i,t = a + b lny i,t-1 (absolute convergence) Constant lny i,t-1 β R 2 SEE D.F. DW 0,1503-0,0452-0.0442 0.047 0.0532 110 2,43 Random (4.7324) (-2.3346) Effects Estimated equation: lny GLS i,t = a + b lny i,t-1 + c ERDF i,t (conditional convergence) Constant lny i,t-1 β ERDF i,t R 2 SEE D.F. DW 0.1547-0.0503-0.0491 0.00003 0.050 0.0533 109 2.41 (4.7233) (-2.3688) (0.6094) (n) Notes: * - All dummies are statistically significant. Figures in parenthesis are t-ratio. (n) - Indicates that the estimated coefficient is not statistically significant at 5% significance level. - - 14

7. Summary and Main Conclusions In this study, an attempt has been made to explain the convergence process among the Portuguese regions at the NUTS III level and more specifically among the regions of the Littoral and the Interior areas. For this reason, the convergence equations in per capita income of the neoclassical type, both, from the absolute and conditional perspective have been estimated by using a Panel Data approach. The role of structural funds in improving living standards and affecting the speed of convergence has been examined from the perspective of the conditional convergence using structural funds as a conditioning factor. The empirical analysis provides some interesting remarks which can be summarized as follows: Statistical evidence shows that there is a substantial gap in per capita income between the regions of the Littoral and the Interior zones. Per capita income in the Littoral area is 15% higher than the national average, while in the Interior area is 18% lower from the national average and this gap is sustained during the decade of 90`s. In the convergence process it makes difference if a region belongs to the Littoral area. Evidence from the estimated convergence equations shows that regions belonging to the Littoral area grow faster in terms of capita income. The distribution of structural funds is irrational in Portugal, benefiting more the more developed regions of the Littoral zone than the less developed regions of the Interior zone. The former receives higher structural aid than the latter both in terms of per capita figures or in terms of the total amount received by the regions, as Table 3 shows. Empirical evidence reveal that structural funds increase the speed of convergence in per capita income as concern the whole set of regions. Structural funds contributed positively to the growth of per capita income but the marginal impact is not so substantial. Regional per capita income increases by only 0.00007% as a result of each additional per capita euro received from the structural funds. This shows that structural funds are used in a less productive manner, mostly orientated for improvements in infrastructure networks or used as direct income support. The fact that a region belongs to the Littoral area is an advantage for the region to grow faster. This is shown from the dummy variable used to distinguish if a region - - 15

belongs to the Littoral area having statistical significance and positive effect on the growth of per capita income. Regional convergence in per capita income is slightly higher between the regions of the Interior area revealing that these regions become more homogeneous over time and they converge to a different steady-state than the Littoral area. Finally, our results clearly show that structural funds contributed more efficiently for improving the living standards of the regions of the Littoral area. This is a rather pessimistic result from the point of view of the regional policy followed in Portugal showing a wrong orientation which can create even higher regional asymmetries. - - 16

APPENDIX Map of the 7 NUTS II and 30 NUTS III Portuguese regions. Source: National Institute of Statistics (INE), Regional Accounts, 1995 - - 17

References Barro, R. J. (1991) Economic Growth in a Cross-Section of Countries, The Quarterly Journal of Economics, 407-443. Barro, R. J.; Sala-i-Martin, X. (1991) Convergence across States and Regions, Brooking Papers on Economic Activity, 1, 107-182. Barro, R. J.; Sala-i-Martin, X. (1992) Convergence, Journal of Political Economy, 100, 2, 223-251. Barro, R. J.; Sala-i-Martin, X. (1995) Economic Growth, McGraw Hill, New York. Dobson, Stephen; Ramlogan, Carlyn (2002) Convergence and Divergence in Latin America, 1970-1998, Applied Economics, 465-470. Ferreira, A. (2000) Convergence in Brazil: Recent Trends and Long Run Prospects, Applied Economics, 32, 79-90. Islam, Nazrul (1995) Growth Empirics: A Panel Data Approach, The Quarterly Journal of Economics, 1127-1171. Kangasharju, A. (1998) β Convergence in Finland: Regional Differences in Speed of Convergence, Applied Economics, 30, 679-687. Mankiw, G.; Romer, D.; Weil, D. N. (1992) A Contribution to the Empirics of Economic Growth, The Quarterly Journal of Economics, 407-437. Marques, A.; Soukiazis, E. (2000) Per Capita Income Convergence across Countries and across Regions in the European Union. Some New Evidence, Discussion Paper nº1, CENEUROP, FEUC. Pekkala, S. (2000) Aggregate Economic Fluctuations and Regional Convergence: The Finish Case 1988-1995, Applied Economics, 32, 211-219. Pigliaru, F.; Adams, J. (1999) Economic Growth and Change: National and Regional Patterns of Convergence and Divergence, Cheltenham, UK. Sala-i-Martin, X. (1994) Economic Growth, Cross-Sectional Regressions and the Empirics of Economic Growth, European Economic Review, 38, 739-747. Sala-i-Martin, X. (1996) Regional Cohesion: Evidence and Theories of regional Growth and Convergence, European Economic Review, 40, 1325-1352. - - 18

Solow, R.M. (1956) A Contribution to the Theory of Economic Growth, Quarterly Journal of Economics, LXX, 65-94. Soukiazis, E. (2000 a) Are Living Standards Converging in EU? Empirical Evidence from Time-Series Analysis, Discussion Paper nº3, CENEUROP, FEUC. Soukiazis, E. (2001) The Cumulative Growth Model as an Alternative Approach to the Convergence Process: Some Theoretical and Empirical Considerations, Notas Económicas, 15, FEUC, 19-39. Soukiazis, E. (2003) Regional Convergence in Portugal, Discussion Paper nº14, CENEUROP, FEUC. Soukiazis, E.; Antunes, M. (2004) The evolution of real disparities in Portugal among the NUTS III regions. An empirical analysis based on the convergence approach., Estudos Regionais, Nº6, p.165-179. - - 19