Territorial Convergence in Ecuador: the Role of Economic Sectors and Spatial Spillovers

Territorial Convergence in Ecuador: The Role of Economic Sectors and Spatial Spillovers

Rodrigo Mendieta Muñoz1, Nicola Pontarollo2

October, 2015

Abstract The paper analyses the subnational convergence process of Ecuador during the period 2007-2013 through a spatial panel econometric technique. The advantage of this technique is to provide a reliable estimation because it takes into account the spatial interaction in the territory. Ecuador is characterised by severe cantonal disparities, reflected in a heterogeneous economic and social geography that can undermine a balanced development and a positive spatial multiplier effect within the country. In this extent we measure the sectoral effects on economic growth proving that, despite the change of productive matrix pushed by the government, this process if far to be completed. In particular the country is too much focussed into low productive sectors which depress economic growth and the manufacture sector is too much concentrated in few areas, preventing its possible positive effect into the whole economy.

Keywords: Subnational convergence, Panel Spatial Econometrics, Economic Sectors

1 Facultad de Ciencias Económicas y Administrativas and Grupo de Investigación en Economía Regional, University of Cuenca, Ecuador: [email protected] 2 Department of Economics, University of Verona, Italy, and Grupo de Investigación en Economía Regional, University of Cuenca, Ecuador. Email: [email protected] 1

1. Introduction

Ecuador has been characterized by persisting severe cantonal disparities, reflected in a heterogeneous economic and social geography, which accounts for cantons with asymmetric characteristics in terms of productivity and competitiveness, as well as in terms of differentiated population and social dynamics (Mendieta, 2015a; Ramón-Mendieta et al., 2013; Alvarado, 2011). These asymmetries between subnational areas can inhibit the growth of domestic production and contribute to its instability (CEPAL, 2010), becoming a problem of circular causation that can undermine the future development of the whole country. This process of unbalanced growth justifies the implementation of compensatory territorial policies such as incentives for private investment, tax breaks, and provision of infrastructure in lagging provinces (Espina, 1994). This kind of interventions, that started in the 90s together with policies and reforms whose aim was to increase the decentralization and the autonomy in of the institutions that manage development, obtained limited benefits in terms of reduction of asymmetries (Barrera, 2007). From 2008, with the new constitution, the process of territorial compensation in Ecuador made another push, with the creation of the National Secretariat of Planning and Development (SENPLADES), which coordinates the processes of autonomy, promotes governance decentralization, and seeks to expand local development capacities. In this context, the Central Government has started the project called “Changing Productive Matrix” which wants to achieve “productive diversification based on adding value; promotion of the exports and their expansion in terms of products and destinations: substitution of imports, including the different actors; deconcentration of production from the existing poles to the territories, and the continuous improvement of productivity and competitiveness across all sectors of the economy” (Plan Nacional del Buen Vivir, PNBV, 2013 - 2017: 73). More explicitly, as a policy guideline, it aims at closing economic and social gaps at territorial level. In this extent, in the last two decades, considering that the prevailing growth and development theories could no longer explain empirical growth patterns, we assisted to a rethink of how economic development takes place and of its relation with economic geography. Globalization has made localities and their interaction more important for economic growth and prosperity (Rodríguez-Pose, 2011). The territory became an active part in the economic structure and the importance of aspects such as human capital and innovation (endogenous growth theory), agglomeration and distance (new economic geography), and institutions (institutional economics) was taken to the fore (Barca et al., 2012). This study focuses on the role of economic sectors into economic growth of the 221 Ecuadorian cantons using spatial econometric tools. The classical growth regression, thanks to the recent

2 development of high-quality statistics at subnational level, has been augmented including sectoral weights into the analysis. Ecuador is characterised by a relatively strong share of non-financial services and agriculture, while it is widely differentiated in terms of manufactory, with some cantons and provinces in which it is very concentrated. Furthermore, Ecuador has still low infrastructure level in some areas of the country that represents a problem in allowing connectives and, finally, spatial diffusion of economic phenomena. In principle, widespread differences among neighbour locations could prevent and/or make relatively complicated the application of policies because their effects may be confined to a very limited spatial dimension. In this extent, to test for spatial spillovers and for the role of economic sectors in context of “Changing Productive Matrix” objective, the adopted methodology is a spatial panel data estimation (Elhorst, 2009). The paper is organised as follows. In the second section a brief overview of the economic structure of Ecuador is given. In the third section it is describe the empirical model and the estimation technique, while in fourth we illustrate the results of our analysis. In last part, finally, we discuss the conclusions of this research.

2. Approach to subnational economic disparities in Ecuador

The Republic of Ecuador, located northwest of South America, between Colombia (north) and Peru (south), is divided into 24 provinces, 221 munipalities or cantons and 1,228 parishes, in an area of 283,500 squared kilometrs, with around 16 millions inhabitants. During the eighties and nineties, like many Latin American countries, Ecuador's economy was characterized by severe economic downturns. These were accompanied by political, social and institutional instability. At subnational level this performance is reflected in a sharp economic and social disparities (Mendieta, 2015a). In the last years, according to the PNBV, in order to smooth the territorial gaps, many strategies has been implemented such as an unprecedented level of public investment deployed throughout the country, especially on roads, hydroelectric projects and in various areas among which health and education, which was made possible from the significant government revenues derived mainly from high oil prices and a more efficient tax collection3. Also, through the Code of Land Management, Autonomy and Decentralization (COOTAD) in force since 2010, several institutional mechanisms that promote decentralization of governance, and seeks to expand the capacity for autonomy and local development have been implemented.

3 Since the seventies, the oil extraction is the most important activity for Ecuadorian economy. In 1974, oil represented 42.51% of public sector revenues, 62.01% of exports and 13.15% of national value added. By 2014, these proportions were 18.47%, 51.70% and 10.41% respectively (Central Bank of Ecuador, 2015). 3

Figure 1: Provinces of Ecuador ID Province Area (km2) 1 Azuay 8 639 2 Bolívar 3 254 3 Cañar 3 908 4 Carchi 3 699 5 Chimborazo 6 479 6 Cotopaxi 6 569 7 El Oro 5 988 8 Esmeraldas 14 893 9 Galápagos 8 010 10 Guayas 17 139 11 Imbabura 4 599 12 Loja 11 027 13 Los Ríos 6 254 14 Manabí 18 400 15 Morona 25 690 16 Napo 13 271 17 Orellana 20 773 18 Pastaza 29 520 19 Pichincha 9 494 20 Santa Elena 3 763 21 Santo Domingo 4 180 22 Sucumbíos 18 612 23 Tungurahua 3 334 24 Zamora 10 556 Source: authors’ elaboration on the basis of INEC.

These actions and strategies begin to show their effects in terms of economic growth Martin (2012) and poverty reduction (Mideros, 2012). World Bank data, on the other hands, highlight that, from 2006 to 2011, the rate of extreme poverty was reduced from 16.9% to 7% and the Gini index decreased from 54% to 48.7%. In this extent, the economic growth can be considered inclusive, consistent with the Boston Consulting Group report according to which, from 2006 to 2012, Ecuador was the oil country that better transformed its oil wealth on well-being (Beal et al., 2012). But were these apparent positive results distributed equally within the country? Is it possible to speak of balanced results? Are these performances accompanied by a process of territorial convergence?. These questions implicitly imply to evaluate how national and local productive matrix has evolved in order to determine if the process of improvement in well-being is sustainable over time. This is shown in table 1 and 2 where we have the average sectoral weight by province in 2007 and

2013, respectively4. In addition, in the last two columns provincial Gross Value Added per capita (GVA/pop) and population per squared kilometer (Density) is reported. The data are provided by the

4 Following the indications of the Central Bank of Ecuador and some other authors, we excluded the gross value added related to oil production because it is does not create wealth in the cantons where it is produced (Mendieta, 2015a; Ramón- Mendieta et al., 2013). Provinces of Santa Elena and Santo Domingo were created after 2007 from the provinces of Guayas and Pichincha respectively and then they were included only in 2013. 4

Central Bank of Ecuador and the GVA in USD is in constant prices with base year 20075. Excluding Galápagos, Pichincha, Guayas and Azuay which have the highest level of Gross Value Added per capita in 2007 and 2013, a significant performance is shown by Tungurahua and El Oro in the last considered year. On the contrary, Morona Santiago, Napo, Zamora Chinchipe, Bolivar, Chimborazo and Orellana have the lowest level of development in both years. Besides persisting gap between rich and poor provinces, the interesting point is that the more developed ones have the highest population density, just concentrated in provincial capital cities. In contrast, less developed provinces have greater dispersal of population. With regard to the provincial production structure, minimal changes are observed between 2007 and 2013 (first 10 columns of Tables 1 and 2). The subnational structure is predominantly based on non- financial services, that include trade and accommodation and food services, transportation, information and communications, and real estate professionals services. The manufacturing sector maintained a weight of around 16% of domestic Value Added. In 2013, in addition to Pichincha, Guayas and Azuay provinces, also Esmeraldas, Santa Elena, Manabi, Tungurahua, Sucumbíos, Santo Domingo and Imbabura reach an industrial weight equal to 10% of Value Added. These provinces are also characterised by the highest rates of population density. It is worth nothing that the manufacturing sector is very concentrated in few cantons. In particular Guayaquil and Quito create around 60% of the manifacturing Value Added and, if we consider only the provincial capitals, they produce the 80% of total manifacturing Value Added. The weight of the agricultural sector is important in some provinces with low levels of development like Los Rios, Esmeraldas, Cotopaxi, Carchi y Bolivar. In contrast El Oro, which has a important banana production, shows higher GVA per person. According to the actual development model, the public administration sector, plus the education and health services, are important for creation of economic value especially in poor provinces with low levels of population density like Morona Santiago, Napo, Bolivar, Pastaza, Zamora Chinchipe and Orellana. In connection with this, the construction sector shows an increase between 2007 and 2013 in all provinces. These results are associated with public investment in infrastructure and housing.

5 Central Bank of Ecuador does not produce annual cantonal data on Gross Domestic Product. Anyway, GVA per head is one of the headline indicators used, for example, in UK regional policy (Dunnell, 2009). According to BIS (2010: 3), in fact, “Gross Value Added per head is typically used for considering performance levels within a country. Although there are some criticisms of this metric it has the advantage that it provides a full picture of performance implicitly including both productivity and employment effects”. In addition, GVA, which measures the contribution to the economy of each individual producer, industry or sector is used in the estimation of Gross Domestic Product (GDP) when using the production or income approaches. In this extent, GVA can be used as a proxy of GDP. 5

Table 1: Percentage of contribution by sector to total provincial Value Added 2007 Province Agricult. Mines Manufac. Water Construct. Basic serv. Fin. serv Pub. adm. Teaching Health GVA/pop Density Azuay 0.0550 0.0090 0.1480 0.1170 0.1040 0.3460 0.0480 0.0700 0.0550 0.0350 3635.236 81.2036 Bolivar 0.3400 0.0000 0.0170 0.0000 0.1030 0.2150 0.0140 0.1390 0.1220 0.0340 1568.134 45.2529 Cañar 0.2070 0.0030 0.0850 0.0000 0.1560 0.3170 0.0290 0.0710 0.0900 0.0340 2232.576 69.4800 Carchi 0.2100 0.0010 0.0510 0.0050 0.1110 0.3550 0.0170 0.1370 0.0790 0.0280 2070.465 42.4252 Cotopaxi 0.2690 0.0000 0.0700 0.0090 0.1340 0.2990 0.0150 0.0710 0.0890 0.0320 2130.334 63.4738 Chimborazo 0.1340 0.0010 0.0820 0.0070 0.1590 0.3370 0.0240 0.1070 0.1000 0.0380 1852.564 67.9295 El Oro 0.2540 0.0230 0.0430 0.0020 0.1100 0.3420 0.0170 0.0750 0.0740 0.0340 2662.382 99.8291 Esmeraldas 0.2030 0.0000 0.3920 0.0000 0.0760 0.1710 0.0050 0.0620 0.0650 0.0180 3236.061 28.4787 Guayas 0.0780 0.0040 0.1930 0.0080 0.0850 0.4560 0.0230 0.0440 0.0560 0.0270 3528.432 194.7800 Imbabura 0.0960 0.0010 0.0720 0.0020 0.1660 0.4170 0.0310 0.0870 0.0910 0.0260 2218.044 82.5790 Loja 0.1600 0.0010 0.0400 0.0010 0.1670 0.3350 0.0380 0.1420 0.0670 0.0450 2051.096 39.1593 Los Rios 0.3860 0.0000 0.0360 0.0030 0.0720 0.2680 0.0090 0.0660 0.0950 0.0520 2113.852 101.646 Manabi 0.2110 0.0010 0.1560 0.0010 0.1130 0.2970 0.0140 0.0780 0.0900 0.0300 2056.027 68.7892 Morona Santiago 0.1770 0.0000 0.0170 0.0590 0.1130 0.2380 0.0170 0.2170 0.1090 0.0400 1401.639 5.68900 Napo 0.1380 0.0000 0.0150 0.0140 0.1610 0.2780 0.0090 0.2180 0.1030 0.0530 1558.224 7.52789 Pastaza 0.0750 0.0000 0.0480 0.0000 0.1440 0.3780 0.0200 0.1980 0.0840 0.0440 2124.35 2.55119 Pichincha 0.0510 0.0020 0.1870 0.0040 0.0770 0.4550 0.0490 0.0560 0.0440 0.0280 4585.378 211.580 Tungurahua 0.0680 0.0010 0.1180 0.0790 0.1260 0.3850 0.0330 0.0630 0.0660 0.0480 2821.255 142.247 Zamora Chinchipe 0.1550 0.0340 0.0170 0.0070 0.1220 0.2410 0.0120 0.2410 0.1190 0.0480 1558.292 8.22337 Galapagos 0.1790 0.0000 0.0080 0.0000 0.0930 0.5020 0.0090 0.1550 0.0150 0.0090 7115.918 2.8239 Sucumbios 0.1250 0.0000 0.3340 0.0000 0.0800 0.2330 0.0090 0.1030 0.0840 0.0220 2385.675 8.75076 Orellana 0.1930 0.0000 0.1880 0.0000 0.0540 0.2230 0.0110 0.1950 0.1010 0.0240 1881.13 5.81498 Santo Domingo Santa Elena Total 0.11181 0.00371 0.16411 0.01372 0.09411 0.39811 0.03053 0.06523 0.06161 0.03052 3138.64 53.1945 In bold the main sector

Table 2: Percentage of contribution by sector to total provincial Value Added 2013 Province Agricult. Mines Manufac. Water Construct. Basic serv. Fin. serv Pub. adm. Teaching Health GVA/pop Density Azuay 0.0349 0.0099 0.1844 0.0362 0.1787 0.3425 0.0497 0.0535 0.0537 0.0429 3827.7411 94.0983 Bolivar 0.2083 0.0000 0.0204 0.0100 0.1311 0.2645 0.0277 0.1446 0.1252 0.0496 1669.8398 50.1126 Cañar 0.1101 0.0024 0.0767 0.0104 0.2155 0.3275 0.0361 0.0857 0.0776 0.0509 2621.9414 79.2407 Carchi 0.2423 0.0005 0.0325 0.0104 0.1089 0.3509 0.0223 0.1026 0.0792 0.0400 2346.3485 46.7308 Cotopaxi 0.2408 0.0006 0.0516 0.0107 0.1268 0.3334 0.0203 0.0804 0.0850 0.0378 2390.9688 72.7545 Chimborazo 0.1279 0.0005 0.0880 0.0145 0.1670 0.3036 0.0292 0.0991 0.0993 0.0580 2035.5050 75.6534 El Oro 0.2514 0.0378 0.0480 0.0111 0.1140 0.3273 0.0190 0.0652 0.0616 0.0382 3326.9562 113.3054 Esmeraldas 0.3376 0.0002 0.1876 0.0101 0.0809 0.1905 0.0044 0.0638 0.0872 0.0273 2806.5159 36.0169 Guayas 0.0843 0.0033 0.2108 0.0129 0.1253 0.3835 0.0271 0.0403 0.0526 0.0362 4087.5734 256.8649 Imbabura 0.0727 0.0010 0.1193 0.0172 0.1844 0.3785 0.0250 0.0668 0.0762 0.0469 2883.4970 94.2963 Loja 0.1041 0.0004 0.0308 0.0106 0.1713 0.3940 0.0353 0.1090 0.0812 0.0561 2496.1167 43.7998 Los Rios 0.4123 0.0000 0.0350 0.0085 0.0883 0.2607 0.0077 0.0623 0.0831 0.0309 2503.6945 116.8272 Manabi 0.1261 0.0010 0.1656 0.0104 0.1595 0.3091 0.0138 0.0728 0.0817 0.0480 2511.0319 77.4618 Morona Santiago 0.0724 0.0001 0.0287 0.0212 0.1209 0.3182 0.0224 0.1595 0.1541 0.0905 1638.0233 6.9158 Napo 0.1072 0.0000 0.0162 0.0121 0.1154 0.3282 0.0123 0.1755 0.1398 0.0795 1927.7632 9.1479 Pastaza 0.0994 0.0000 0.0518 0.0137 0.1328 0.3242 0.0316 0.1491 0.1179 0.0662 2315.2185 3.1802 Pichincha 0.0406 0.0033 0.1691 0.0087 0.1118 0.4021 0.0486 0.1130 0.0356 0.0274 5730.6372 297.3308 Tungurahua 0.0532 0.0004 0.1541 0.0207 0.1032 0.4446 0.0474 0.0526 0.0625 0.0465 3035.9004 160.6691 Zamora Chinchipe 0.0664 0.0232 0.0168 0.0151 0.1413 0.3216 0.0089 0.1962 0.1413 0.0642 1696.1812 9.7016 Galapagos 0.0898 0.0000 0.0109 0.0073 0.0938 0.5715 0.0072 0.1273 0.0395 0.0214 4940.1039 3.4939 Sucumbios 0.1118 0.0000 0.1477 0.0064 0.1361 0.3583 0.0094 0.0884 0.0947 0.0290 2405.7446 10.8324 Orellana 0.1793 0.0000 0.0256 0.0260 0.0733 0.3246 0.0147 0.1516 0.1416 0.0434 1766.9825 6.7289 Santo Domingo 0.0999 0.0002 0.1235 0.0121 0.1195 0.3855 0.0178 0.0912 0.0884 0.0514 2535.3713 116.9389 Santa Elena 0.0668 0.1599 0.1783 0.0161 0.2438 0.3136 0.0064 0.0560 0.0949 0.0125 2282.6649 92.7796 Total 0.1033 0.0037 0.1576 0.0127 0.1283 0.3657 0.0317 0.0767 0.0593 0.0365 3563.7570 61.5688 In bold the main sector

Summarising the reported results, the production structure has only little changed, both at national and subnational levels. The main critical points are the persistent importance of agricultural and non- financial sectors characterised by low productivity and labor skills and the excessive concentration of manufacturing sector. Furthermore, in accordance with the increase of central government spending, sectors like public administration, education and health have increased in recent years. Their magnitude, anyway, according to the tables below, seems to be not enough to create a multiplier effect in order to push the change to a more productive structure. On the contrary, in provinces with the lowest population density and GVA per hear they produce a large part of Gross Value Added both in 2007 and 2013. If, from a side, the effects of public sector and of connected activities can be hampered by the little population concentration (i.e. absence of economy of scale), on the other hand, their potential positive impact would have been connected to the capacity of local policy makers to tailor policies related to the specificity of each territory (Barca et al., 2012). This eventuality, in consideration of the low performances of sparsely populated provinces and of the unchanged economic structure, has not been accomplished, with the risk to undermine their long-run development perspectives. These disparities, according with CEPAL (2009) and Silva (2005) would form different typologies of territories, which would amplify inequalities at a higher level of geographical breakdown. To check this regularly in the Ecuadorian case, figure 2 plots the average rate of economic growth between 2007 and 2013 at cantonal level on the vertical axis against the GVA per capita on the horizontal axis (in logarithms). The names in the figure correspond to the provincial Capital cantons. In this period a cantonal average growth rate of -1.7% is recorded, with 58% of cantons characterized by negative growth, and a similar proportion with a GVA per person in 2013 less than cantonal average in 2007. The subnational economic gap is evident, especially between provincial capitals and other cantons. Most of these capitals have both a higher average level of Gross Value Added per capita and a higher rate of growth. According to the aforementioned, Quito, Guayaquil and Cuenca, capital of Pichincha, Guayas and Azuay respectively, which jointly account for over 45% of produced GVA, present the best performance. Similarly, the poorest capitals correspond to the most backward provinces. Thus, behind the negative relation between the initial cantonal GVA per person and growth, which means that convergence is present, there is a strong problem related to the lack of growth performances of most advanced cantons. In addition, as shown by Mendieta (2015b) this process does not correspond to a decrease of disparity levels.

Figure 2: Relation between GVA per person in 2007 and average growth 2007-2013.

The cantonal low or negative growth is often related to the too strong dependence form row materials (essentially agricultural products and oil) which is subject to the fluctuations of international prices. In addition, these cantons and the surrounding ones have often grown for the services and productions provided to the workers that migrate from other cantons. This created a “bubble” that involved in particular agricultural, non-financial services and construction sectors, which create employment but that are not the basis for a lasting growth. As a consequence, the row materials prices fluctuations affect the economy in two ways: the first regards the various directly and indirectly connected sectors, and the second regards to the multiplier effect on neighbour cantons. On this bases, in the next point we delve over the Ecuadorian subnational convergence and the roles of sectoral structure.

3. Convergence and econometric model

The specification of the empirical model for panel growth regressions is:

푔푟푖,푡 = 훼 + 훽푙표푔(푦푖,푡−1) + 훿sectors푖,푡−1 + 휇푖 + 휀푖푡 (1)

Where the dependent variable gri,t represents the cantonal annual growth rate of per capita Gross

Value Added; 훼 is a constant term; 휇푖 are dummies specific to canton i which control for unvarying

6 factors determining differences in the steady states across cantons ; 푦푖,푡−1 is the per capita GVA in canton i, of which there are 221, in the period 2007-20137; 훽, if negative, is the coefficient related to the annual rate at which an economy converges to the long-run steady state. The additional variables sectors푡−1 represent the relative weight of the GVA produced by the different economic sectors, which assume an important role in light of the “Changing of Productive Matrix” plan and that, if jointly significant, means that each economy converge to its own steady state because each one has different structural characteristics and production function. In this last case we speak of conditional convergence. On the contrary, i.e. if the additional variables are not significant, there is absolute convergence: all economies converge to the same steady state because they are not structurally different. When dealing with data at regional or subnational level, standard growth regression models can suffer of misspecification problems (McMillen, 2003; Fingleton and Lopez-Bazo, 2006). A number of factors either unobservable or not properly measured related, for example, to culture, social capital and institutional characteristics can affect regional performances. These factors, often spatially correlated, i.e. with similar values in space, and can lead to spatially correlated error in regression models, that, if ignored, may lead to biased results and hence misleading conclusions. In order to deal with this issue, Fingleton and Lopez-Bazo (2006) show that scholars tend to use growth models that incorporate spatial dependence as either autoregressive term or spatial error. Another possibility, according to LeSage and Fischer (2008), is, in order tackle the issue of the presence of spatial dependence in the disturbances and of omitted variables that exhibit correlation with included variables is to use a spatial Durbin model. As shown by the results of the analysis in the followin section, the ad-hoc tests to check which is the model to use in this context confirm that the framework proposed by LeSage and Fischer (2008) is the correct one. The regression model, then, becomes:

푔푟푖,푡 = 휌퐖푔푟푖,푡 + 훼 + 훽푙표푔(푦푖,푡−1) +

휑퐖푙표푔(푦푖,푡−1) + 훿sectors푡푖,−1 + 휂퐖sectors푖,푡−1 + 휇푖 + 휀푖푡 (2) Where W represents the spatial weights matrix that, as customary in literature, it is a queen standardized by row and the convergence speed λ is determined from the following relation: (훽 + 휑)/(1 – 휌) = (1 − e−휆×푇).

6 As observed by Elhorst (2009), often find weak evidence in favor of spatial interaction effects is found when time period fixed effects are also accounted for. This is much more valid when dealing with a country in which most variables tend to increase and decrease together in different spatial units because follow the national evolution over time. 7 In literature Henley (2005) uses GVA per head to measure convergence of UK regions. 10

According with Elhorst (2009), the spatial econometric model for a panel of N observations over T time periods can be estimated along the same lines as the cross-sectional model, i.e. using Maximum Likelihood (ML), provided that all notations are adjusted from one cross-section to T cross-sections of N observations. The partial derivative of Spatial Durbin Model, for the presence of the spatial autoregressive parameter, is not straightforward. Differently from linear regression the partial derivative of the dependent variable with respect to the explanatory variable, the marginal effects are described as follow: 휕푔푟 푖,푡 = (퐈 − 휌퐖)−1(퐈훽 − 퐖휑) (3) 휕log (푦푖,푡−1) Estimated effects do not depend only from β, but also from the sign and magnitude of ρ and φ. At this regard LeSage and Pace (2009, 2009a) suggest the following scalar summary measures: a. the average direct effect constructed as an average of the diagonal elements of (퐈 − 휌퐖)−1(퐈훽 − 퐖휑); b. the average indirect effect constructed as an average of the off-diagonal elements of (퐈 − 휌퐖)−1(퐈훽 − 퐖휑), where the off-diagonal row elements are summed up first, and then an average of these sums is taken; c. the average total effect is the sum of the direct and indirect impacts. The key variable trough which spatial spillovers effects are transmitted is ρ: it expresses the relation between dependent variable and its spatial lag. According to Fischer et al. (2009), the indirect effect estimates, that correspond to spatial spillovers, can be interpreted in two ways. One interpretation reflects how a change in the level of a variable for the study canton impacts the GVA per head growth of other cantons which in turn negatively influences our typical canton’s GVA per head growth due to the presence of negative and significant spatial dependence (ρ) on neighboring cantons’ GVA per capita growth. According to the second interpretation, spatial spillovers measure the cumulative impact of a change in the initial level of a variable in a canton averaged over all other canons. The impact of a changing in a single canton’s variable level on each of the other cantons’ GVA per head growth is small, but cumulatively often exceeds the direct effect. The interpretation of the estimate of the indirect impact is closely related to the concepts of average total impact from and to an observation (LeSage and Pace, 2008). In the first case a change in initial variable level impacts other cantons’ GVA per head growth, which, though spatial autocorrelation coefficient ρ, influences study canton’s GVA per head growth. Essentially this is given by the sum of the j-th column of (퐈 − 휌퐖)−1(퐈훽 − 퐖휑). In the latter case a change in the initial level of a variable

11 of all other cantons affects GVA per head growth of a typical canton. In this case the effect is given by the sum of the j-th row of (퐈 − 휌퐖)−1(퐈훽 − 퐖휑).

4. Estimation results The estimation results are in table 3. The estimation has been performed with both standard and spatial panel technique and for both growth and level regression. To test the hypothesis whether the spatial Durbin model can be simplified to the spatial error model one may perform a Wald or LR test. The results reported for the two regressions confirms the choice of spatial Durbin both with Wald test and LR test. Similarly, the hypothesis that the spatial Durbin model can be simplified to the spatial lag model must be rejected for both Wald test and LR test. This implies that spatial error and spatial lag models must be rejected in favor of the spatial Durbin model. Regarding the choice of fixed effect, the results of Hausman's specification test indicate that the random effects model must be rejected in all cases in favor of fixed effect. As highlighted in the previous paragraph, table 3 does not allow to read the partial effects of the variables of Spatial Durbin Model8, but it reports at least other two important outcomes: the first is the joint statistical significance of regressors related to sectoral structure in growth regression (F-test = 4.52, with p-value < 0.01), which implies that there is conditional convergence, and the second is the statistical significance and magnitude of the spatial lag parameter, which gives information regarding spatial spillovers. The spatial autoregressive term ρ is equal to around 0.27 in both cases. The multiplier effect, thus, is around 1.40, which means that 40% of growth (GVA per head) is already reflected in neighborhood growth (GVA per head), through indirect reaction effects from neighbors that depends from the interaction among cantons in the country. As shown in the table, failure to account for this redundancy would lead the OLS estimates of the direct (marginal) impacts biased upwards in magnitude, because the model is misspecified by omission of spatial spillover effects. The spatial Durbin model accounts for the redundancy induced by spatial autocorrelation in explanatory variables as well as spillover effects (interactions) among cantons that lead to interdependence in cantons’ behaviour.

8 In Appendix A the two models are estimates using two alternative spatial weights matrix specifications: queen matrix of degree 2 and knearneigh of order 5 and the results are confirmed. The spatial weights matrix chosen for the model presented in the text guarantees the lowest AIC. 12

Table 3: Estimation results Growth regression Level regression Fixed effect Sp. Durbin Fixed Effects Fixed effect Sp. Durbin Fixed Effects GVA/pop -0.558646 *** -0.484414 *** (-19.0563) (-14.89264) Agricult. -0.660982 *** -0.623907 *** -0.323446 *** -0.422615 *** (-4.76412) (-4.267607) (-2.639105) (-3.320153) Mines -0.18782 -0.116192 -0.301372 * -0.317334 * (-0.89808) (-0.542392) (-1.714591) (-1.767734) Manufac. 0.004634 0.032708 0.02471 0.012859 (0.058346) (0.389477) (0.329596) (0.163173) Wather -0.638562 *** -0.505776 ** 0.150382 0.143614 (-2.62580) (-2.017313) (0.721993) (0.672185) Construct. -0.345284 ** -0.32589 * -0.773991 *** -0.996743 *** (-2.18238) (-1.888022) (-5.52865) (-6.696764) Basic serv. -0.399461 *** -0.311218 ** -0.479902 *** -0.56558 *** (-2.99249) (-2.235447) (-4.042562) (-4.605736) Fin. serv -2.76394 *** -0.919766 -3.702141 *** -3.240624 *** (-3.32684) (-1.037408) (-5.679752) (-4.688952) Pub. adm. -0.891434 *** -0.596144 *** -1.987847 *** -2.170926 *** (-4.43791) (-2.798952) (-11.73778) (-12.350536) Teaching -0.721497 *** -0.155097 -3.209762 *** -3.493777 *** (-3.26248) (-0.633182) (-17.75647) (-18.240437) Health -0.841962 ** -0.45467 -2.094742 *** -2.150659 *** (-2.35733) (-1.22645) (-7.068351) (-7.04409) W×GVA/pop -0.088618 (-1.616013) W×Agricult. -0.063956 0.68959 *** (-0.253164) (3.153001) W×Mines -0.093373 0.016743 (-0.521087) (0.109976) W×Manuf. -0.140098 -0.043021 (-1.020512) (-0.333831) W×Wather -0.089434 0.006115 (-0.197299) (0.0158) W×Construct. -0.094116 1.091239 *** (-0.336522) (4.49605) W×Basic serv. -0.307921 0.670966 *** (-1.257106) (3.113458) W×Fin. serv -5.56791 *** -0.434561 (-3.593327) (-0.353163) W×Pub. adm. -0.14004 1.392347 *** (-0.369571) (4.278844) W×Teaching -1.284715 *** 2.522441 *** (-3.071721) (7.30082) W×Health -0.615525 0.412515 (-0.925208) (0.740366) ρ 0.263962 *** 0.278986 *** (7.929891) (9.125238) R-squared 0.2645 0.4433 0.4141 0.9467 corr-squared 0.2589 0.3115 0.4106 0.4357 sigma^2 0.0175 0.0182 0.0164 0.017 N° obs 1326 1547 log-likelihood 869.78 1049.37 F-test cond. regr. 4.52 (p-val. < 0.01) Wald sp. lag 48.87 (p-val. < 0.01) 82.55 (p-val. < 0.01) LR sp. lag 55.44 (p-val. < 0.01) 93.64 (p-val. < 0.01) Wald sp. error 52.25 (p-val. < 0.01) 38.30 (p-val. < 0.01) LR sp. error 62.35 (p-val. < 0.01) 44.19 (p-val. < 0.01) Hausman test 383.39 (p-val. < 0.01) 630.05 (p-val. < 0.01) *Significant at 1%, ** significant at 5%, *** significant at 10%. t-stat in brackets. Source: authors’ elaboration. 13

Table 4 reports the estimated direct, indirect and total effects of the growth regression. The results show that convergence is taking place and the average rate is 9.59%, with a half-life of only 7.22 years9. As the reported values concern the conditional convergence, this means that Ecuadorian cantons are very close to their own steady state, which signifies that, in order to guarantee a lasting growth, it is extremely urgent to intervene in order to increase technology levels to make more efficient cantonal productive structure. In addition, in the extent in which we have conditional convergence, this makes that cantons will reach very differentiated steady state because each single economy will converge to its own steady state. The result of this process is to increase country inequality, which is exactly what PNBV is trying to avoid. The signs related to the significant sectors are negative. Agriculture, which has a prevailing role in the creation of gross value added in Ecuador, is also characterized by a low productivity and this is reflected into its contribution to growth (Hausmann et al., 2010). As expected, it has a global effects, which means that it involves the entire country through the multiplier effect. Water procurement, as well as construction have only a strictly direct effect, which is probably related to the fact that these sectors are well developed only in some cantons, which are located mainly in the province of Zamora and in cantons where mining and big public infrastructures, as hidroelectric, are been build. Basic services, as well as agriculture, is a very low productive sector, which doesn’t only characterize less developed cantons. This is shown by the significant indirect and total effect, which proves that the negative effects spreads over the whole economy exceeding cantonal borders. Financial sectors have a negative indirect and total impact which can be related to the lack of financial services in many cantons and the concentration in only some of them. This causes that less endowed ones suffer the competition of neighbor endowed cantons which exploit the resources of the weaker. The negative impact of public administration and teaching is related to the high weight of these sectors into low productive areas, where they are much more concentrated. At this regard, here we can measure the inefficiency of these sectors into promoting growth in two extents: the first is in failure into fostering the change in the productive matrix shown also in tables 1 and 2, and the second in being productive by themselves.

9 The implied speed of convergence is calculate as 휆 = −ln (1 − (훽 + 휑)⁄(1 – 휌))⁄푇 . The time τ it takes to move half way to the balanced growth path is calculated as: 휏 = −ln (0.5)/휆 .

Table 4: Estimated effects growth regression Direct Indirect Total GVA/pop -0.500575 *** -0.282579 *** -0.783155 *** (-15.45896) (-4.620095) (-11.72415) Agricult. -0.641355 *** -0.292135 -0.93349 ** (-4.285698) (-0.910377) (-2.514647) Mines -0.129515 -0.177567 -0.307082 (-0.58713) (-0.702314) (-0.771559) Manufac. 0.020931 -0.16579 -0.14486 (0.255783) (-0.967374) (-0.753326) Water -0.521766 ** -0.292607 -0.814373 (-2.032121) (-0.502003) (-1.205938) Construct. -0.337087 ** -0.241795 -0.578882 (-1.98465) (-0.69636) (-1.494991) Basic serv. -0.342233 ** -0.508159 * -0.850392 ** (-2.413002) (-1.660662) (-2.383499) Fin. serv -1.32613 -7.485566 *** -8.811696 *** (-1.518299) (-3.870734) (-4.102852) Pub. adm. -0.61922 *** -0.401273 -1.020493 * (-2.831179) (-0.828133) (-1.822438) Teaching -0.245075 -1.718108 *** -1.963183 *** (-1.019649) (-3.487147) (-3.738077) Health -0.50816 -0.957237 -1.465397 (-1.376327) (-1.133507) (-1.445231) *Significant at 1%, ** significant at 5%, *** significant at 10%. t-stat in brackets.

Table 5 shows results in line with the previous table. Agriculture has direct negative effect on local production, but positive indirect effects. This confirms that agricultural activity does not directly benefit rural areas in which it is concentrated, but a change in canton’s i level of agriculture averaged over all other cantons is positive. The interpretation is related to the multiplier effect due to the use of agricultural products by transforming industries, distributors and hotel and restaurants services. Anyway the global effect is not significant because the two effects tend to cancel out reciprocally. A similar interpretation can be applied to construction and basic services that have a positive indirect effect that is not enough strong to widespread over space. For both public administration and teaching, which have a negative direct and total effect, the spatial spillovers are positive and significant. Own-canton sectoral magnitude increases will restrain GVA per head in their own canton, but it has a positive effect in neighbouring cantons. This finding may be associated with more concentrated development in big cities, reflected in higher GVA per capita, that has left other cantons less-developed, which in turn has led to increase GVA per capita in those cantons. The negative indirect effect, for the limited spatial multiplier effect, does not cancel out the negative direct effect, resulting in a significant negative total effect. Financial services and health show a direct and total negative effect while indirect effect is not significant. Finally the manufacturing sector still does not have a benefit to the cantonal production, like mines and water. These findings warns about the poor results in changing production model in terms of generating higher subnational value added. 15

Table 5: Estimated effects level regression Direct Indirect Total Agricult. -0.379671 *** 0.763705 *** 0.384035 (-2.899162) (2.677163) (1.149186) Mines -0.323055 * -0.105577 -0.428632 (-1.755977) (-0.491694) (-1.279971) Manufac. 0.012132 -0.044125 -0.031992 (0.15821) (-0.263025) (-0.172227) Water 0.149023 0.062401 0.211424 (0.665215) (0.11911) (0.347261) Construct. -0.933389 *** 1.083902 *** 0.150513 (-6.546167) (3.396886) (0.410045) Basic serv. -0.527864 *** 0.684126 ** 0.156262 (-4.217578) (2.385036) (0.467082) Fin. serv -3.361519 *** -1.707566 -5.069085 *** (-4.845014) (-1.076643) (-2.91838) Pub. adm. -2.113009 *** 1.042635 ** -1.070374 ** (-11.93012) (2.553989) (-2.277204) Teaching -3.384312 *** 2.044559 *** -1.339753 *** (-17.69287) (4.720775) (-2.749389) Health -2.151766 *** -0.222133 -2.373898 *** (-6.845171) (-0.299453) (-2.648202) *Significant at 1%, ** significant at 5%, *** significant at 10%. t-stat in brackets.

The results for the production function estimates are rather close to the estimates for the growth regression, with the exception of indirect effects. These have negative sign in growth regression and positive sign in level estimation. The discrepancy in level regression are related to the heterogeneous economic structure in which some cantons take (indirect) benefit from the weakness of the others. This determines that few areas (the main urban centres) “absorb” too much economic resources mainly in terms of skilled workers, productive sectors and technology, making extremely difficult to create a virtuous multiplier effect for the whole economy. The result of this unbalanced and heterogeneous production structure is, as shown in section 2, a negative average GVA per capita growth in which some sectors amplify the territorial imbalances.

5. Conclusions The paper examines cantonal convergence using a spatial panel econometric approach that allows to account for issues related to spatial dependence, which is a typical issue when regional or subregional data are used. In addition, the role of sectoral economic structure in order to produce growth and spatial spillovers is checked. The results support the importance of considering spatial relationships in analyzing subnational development in Ecuador, which appears asymmetrically distributed in space, with the most developed areas that absorb potentially more productive sectors like manufactury. In period 2007-2013 the cantonal conditional convergence rate is 9.59% annually. This implies that each canton converges to its own steady state, which is marked by stark differences that meet specific local production 16 structures. The high convergence rate, furthermore, means a half-life of only 7.22 years, showing that Ecuatorian cantons are very close to their steady state making crucial to improve efficiently modifing the productive structure of the economy. Despite the Central Government’s project to change productive matrix, the weight of non-financial and agricultural sector is still too strong and accounts for almost 40% of Gross Value Added, reaching, in some provinces, the 41% of production. The recent construction boom, involving unskilled labour work, may further explain the losing GVA per capita of the country. In this context, public sector is not able to generate a multiplier effect, both for their probable lack of efficiency and for the unfavourable context. In this extent the too strong focus into low productive sectors which depress GVA per capita growth extend their negative effects throughout the whole country via spatial multiplier. The multiplier, if compared with other contexts like the European, is very low and the magnitude of indirect effects, as a consequence, is lower than the direct effects. Considering that the indirect effects correspond to the average contribution of the neighbour (and neighbour of the neighbour) cantons, this indicates that the spatial spillover effects are very low, and their effects are strongly clustered in space. This is due to a heterogeneous growth, which, for the structural limitations due, among other causes, to the heterogeneous production structure, do not allow exploiting endogenous cantonal potentials and economy of scale. This has some important policy implications and opens various problems for the objectives of the PNBV. The low value of spatial multiplier means that a policy intervention in a determined canton tends to spread out and reinforce its effects generating a spillover effect only in a limited extent, exhausting after a certain distance. This makes localities and their interaction much more important for economic growth and prosperity implying that policies tailored on the specificity of each territory are fundamental (Barca et al., 2012). The priority aims of these policies have to be multiple. The first one is decentralizing manufactory sector and/or creating incentives related to the creation of collateral services and productions in order to exploit the maximum from the existing manufactory sectors. This requires an in-deep analysis of the actual situation with the involvement of institutional actors and territorial stakeholders. The second point is to reinforce the local networks investing in both ‘harder’ (i.e. routes) and ‘softer’ infrastructure (human capital and research capacity). The third point is to add an explicit spatial dimension to the policy objectives. In addition to the reduction of existing disparities, the aim has to be avoiding territorial imbalances making both sectoral policies which have a spatial impact and subnational policy more coherent through an improved territorial integration and cooperation. The Ecuadorian case proves that economic development of lagging development areas is not an automatic process but involves a wide variety of aspects related to the peculiarities of each territory that cannot be ignored.

References

Anselin, L., (1988), Spatial econometrics. Methods and models, Dordrecht, Boston, London: Kluwer Academic Publisher.

Alvarado, R., (2011), Measuring the competitiveness of the provinces of Ecuador, Universidad Católica del Norte. MPRA Paper No. 34244, posted 22.

Barca, F., McCann, P., Rodríguez-Pose, A., (2012), The case for regional development intervention: Place-based versus place-neutral approaches, Journal of Regional Science, 52: 134-152

Barrera Guarderas, A. (2007), Agotamiento de la descentralización y oportunidades de cambio en el Ecuador. En La descentralización en el Ecuador: opciones comparadas. Fernando Carrión M., (eds.). FLACSO sede Ecuador, Quito: 175-206.

Beal D., Rueda-Sabater E., and Espírito Santo T., (2012), From wealth to well-being. Introducing the BCG sustainable economic development assessment. November 2012, The Boston Consulting Group.

BIS (2010) Understanding Local Growth, BIS economics paper no. 7, Department for Business, Innovation and Skills and the Department for Communities and Local Government. October 2010.

CEPAL (2010), La hora de la igualdad brechas por cerrar, caminos por abrir. Trigésimo tercer período de sesiones de la CEPAL. Brasilia, 30 may - 1 june 2010.

Dunnell, K. (2009), National statistician’s article: Measuring regional economic performance. Economic & Labour Market Review. 3: 18-30

Elhorst, J.P., (2009), Spatial panel data models. In Fischer M.M., Getis A (eds.) Handbook of applied spatial analysis. Springer, Berlin, Heidelberg and New York: 377-407.

Espina, A., (1994) La Estrategia de Hirschman Revisitada, Claves, 39: 51-55.

Fingleton, B. and López-Bazo, E. (2006), Empirical growth models with spatial effects. Papers in Regional Science, 85: 177-198.

Fischer, M.M., Bartkowska, M, Riedl, A., Sardadvar, S. and Kunnert, A., (2009), The impact of human capital on regional labour productivity in Europe. Letters in Spatial and Resource Sciences, 2: 97-108.

Hausmann, R. and Klinger, B., (2010), Structural transformation in Ecuador. Inter American Development Bank. Country Departament Andean Group (CAN). Policy Brief No. IDB-PB- 112.

Henley, A., (2005), On regional growth convergence in Great Britain, Regional Studies, 39: 1245- 1260.

LeSage, J.P. and Fischer M.M., (2008), Spatial growth regressions: model specification, estimation and interpretation. Spatial Economic Analysis, 3: 275-304.

LeSage J.P., Pace, R.K., (2009), Introduction to spatial econometrics, Taylor & Francis CRC Press, Boca Raton.

LeSage, J.P. and Pace, R.K., (2009a), Spatial econometrics models, in: Fischer M.M., Getis A. (eds.) Handbook of applied spatial analysis. Springer, Berlin, Heidelberg and New York: 355-376.

Martín, M.F., (2012), El desempeño de la economía Ecuatoriana durante el gobierno del economista Rafael Correa, in Balance de la Revolución Ciudadana, Mantilla S. and Mejía S. (eds.). Centro Latinoamericano de Estudios Políticos, Editorial Planeta del Ecuador. Quito: 237-266.

McMillen, D.P., (2003), Spatial autocorrelation or model misspecification? International Regional Science Review, 26: 208-217.

Mideros, M.A., (2012). “Ecuador: Definición y medición multidimensional de la pobreza, 2006- 2010”. Revista CEPAL 108: 51-70.

Mendieta Muñoz, R., (2015a), Remesas y disparidades económicas territoriales: El caso ecuatoriano. 1a ed. Miguel Ángel Porrúa y Universidad de Cuenca. México.

Mendieta Muñoz, R., (2015b), La hipótesis de la convergencia condicional en Ecuador. Un análisis a nivel cantonal. Revista Retos, 9: 13-25.

Plan Nacional del Buen Vivir 2013-2017 (PNBV 2013-2017) (2013), Secretaría Nacional de Planificación y Desarrollo SENPLADES. Quito.

Ramón-Mendieta, M.G., Ochoa-Moreno, W.S., Ochoa-Jiménez, D.A., (2013), Growth, clusters, and convergence in Ecuador: 1993-2011, in Regional Problems and Policies in Latin America, Cuadrado-Roura J.R. and Aroca P. (eds.). Springer-Verlag. 2013.

Rodríguez-Pose, A., (2011) Economists as geographers and geographers as something else: On the changing conception of distance in geography and economics, Journal of Economic Geography, 11, 347-356.

Appendix A: robustness check

Table A1: Spatial Durbin estimation results with alternative Spatial Weights Matrices Growth regression Level regression Queen deg. 2 K = 5 Queen deg. 2 K = 5 GVA/pop -0.464737 *** -0.46239 *** (-14.152282) (-14.370022) Agricult. -0.584853 *** -0.651377 *** -0.450953 *** -0.422615 *** (-3.959386) (-4.508056) (-3.56179) (-3.320153) Mines -0.087951 -0.166631 -0.262695 -0.317334 * (-0.410789) (-0.795009) (-1.483652) (-1.767734) Manufac. 0.014609 0.003645 -0.026979 0.012859 (0.175536) (0.044094) (-0.34924) (0.163173) Wather -0.497293 ** -0.580454 ** 0.151837 0.143614 (-1.983264) (-2.343626) (0.71902) (0.672185) Construct. -0.305497 * -0.395657 ** -0.996354 *** -0.996743 *** (-1.781397) (-2.348089) (-6.777835) (-6.696764) Basic serv. -0.333317 ** -0.35631 *** -0.589438 *** -0.56558 *** (-2.376166) (-2.59047) (-4.832514) (-4.605736) Fin. serv -0.501772 -0.678434 -3.654764 *** -3.240624 *** (-0.568467) (-0.779205) (-5.393371) (-4.688952) Pub. adm. -0.549062 *** -0.566672 *** -2.201225 *** -2.170926 *** (-2.581979) (-2.727135) (-12.859507) (-12.350536) Teaching -0.112852 -0.211785 -3.591825 *** -3.493777 *** (-0.46135) (-0.874261) (-18.924875) (-18.240437) Health -0.388279 -0.54308 -2.184121 *** -2.150659 *** (-1.052125) (-1.498179) (-7.291658) (-7.04409) W×GVA/pop -0.139235 * -0.08539 (-1.85582) (-1.34479) W×Agricult. -0.354089 0.048475 0.940496 *** 0.68959 *** (-0.860461) (0.16447) (2.676025) (3.153001) W×Mines 0.353218 0.431663 0.107074 0.016743 (0.885391) (0.753383) (0.320602) (0.109976) W×Manuf. -0.225165 0.025805 0.086026 -0.043021 (-0.998072) (0.148229) (0.412261) (-0.333831) W×Wather -1.26266 0.211253 -0.864707 0.006115 (-1.582343) (0.376196) (-1.282727) (0.0158) W×Construct. -0.515446 0.063391 1.577881 *** 1.091239 *** (-1.133258) (0.19328) (4.056908) (4.49605) W×Basic serv. -0.492242 -0.29499 0.876097 *** 0.670966 *** (-1.253622) (-1.007835) (2.577091) (3.113458) W×Fin. serv -9.202847 *** -6.732083 *** 1.967005 -0.434561 (-4.270335) (-3.687686) (1.191924) (-0.353163) W×Pub. adm. -0.057171 -0.386153 2.414625 *** 1.392347 *** (-0.098601) (-0.867066) (4.913865) (4.278844) W×Teaching -1.661652 *** -0.699409 3.283791 *** 2.522441 *** (-2.941126) (-1.566162) (6.952903) (7.30082) W×Health 0.09706 -0.880269 1.059557 0.412515 (0.100015) (-1.026984) (1.316279) (0.740366) ρ 0.315965 *** 0.380913 *** 0.393958 *** 0.278986 *** (6.916456) (10.23494) (10.072037) (9.125238)

R-squared 0.4456 0.4679 0.9481 0.9467 corr-squared 0.3287 0.3257 0.4495 0.4357 sigma^2 0.0181 0.0174 0.0165 0.017 N° obs 1326 1547 log-likelihood 877.86 896.01 1072.06 1049.37 F-test cond. regr. 3.82 (p-val. < 0.01) 3.62 (p-val. < 0.01) *Significant at 1%, ** significant at 5%, *** significant at 10%. t-stat in brackets.