Regional inequality in France : Structural change dynamics

Regional inequality in France 1860‐2010: Structural change dynamics Autores y e-mail de la persona de contacto: M.Teresa Sanchis (Universidad de Valen...
Author: Derek Fisher
0 downloads 3 Views 421KB Size
Regional inequality in France 1860‐2010: Structural change dynamics Autores y e-mail de la persona de contacto: M.Teresa Sanchis (Universidad de Valencia), [email protected] Joan Ramón Rosés (London School of Economics) Alfonso Díez (Universidad de Valencia) Departamento: Análisis Económico Universidad: Universidad de Valencia Área Temática: Crecimiento Económico Regional Resumen:

Este trabajo analiza la evolución de la desigualdad regional en Francia entre 1860 y 2010. Para ello se ha construido una base de datos original de PIB por sectores (agricultura, industria y servicios) para las 22 regiones francesas de la clasificación NUTS‐2 en los años censales 1860, 1896, 1911, 1921, 1931, 1954, 1962, 1982, 1990, 1999 and 2010. La base se compone de las estimaciones de Combes et al (2011) para 1860, 1896 y 1930, los datos de EUROSTAT para 1982 y las estadísticas del INSEE para 1990‐1999 y 2010 y de estimaciones originales para el resto de cortes transversales (1911, 1921, 1954 y 1962). Se ha utilizado la desarrollada por Geary y Stark (2002). Los resultados revelan que la desigualdad en Francia ha seguido la forma de U‐invertida (Williamson, 1965), con una aumento de la desigualdad en las primeras etapas de la industrialización y un claro descenso entre 1896 y 1982. El análisis de shift‐share nos permite descomponer la desigualdad regional en tres componentes: (a) convergencia de la productifividad del trabajo dentro de las industrias de distintas regiones, (b) reasignación del trabajo entre sectores, y (c) convergencia en la productividad entre distintos sectores. Se toma la Región de París como referencia para hacer el análisis.

Palabras Clave: Convergencia regional; Francia; 1850-2000; Historia Económica Clasificación JEL: R11, N93, N94

1. Introduction In 2010 the Paris Region with only 2.2 % of the France’s surface, sheltered 18.79 % of its population and produced 30.95 % of total GDP. At the opposite side was Limousin, a French region that expands over 3.11 % of continental France but only contributed to national GDP in 0.87 % and to total population in 1.17%. In average, a person settled down in the Paris region had 2.21 times the income of someone living in Limousin and 1.66 times that of an average French citizen. Things were not so different 150 years before. In 1860, according to Combes, Lafourcade, Thisse and Toutain (2011), the Paris Region was by far the richest region of the country with a per capita income that more than doubled the national average and was 3.13 times per capita income of Britany, the poorest region of France in those years. Although regional disparities in per capita income distribution have been reduced since the beginning of the industrialization process, the distance between the leader region and the rest of the country is still huge1 and official statistics reveal that divergence between regions has increased at least since 19902. The aim of this research is to investigate the historical forces that stand behind the evolution of regional disparities in income per capita among the French regions throughout the process of modern economic growth. To achieve this objective we will focus on the theoretical literature that explains the uneven regional development in terms of the process of structural change (Caselli and Coleman, 2001) and factor reallocations between sectors and regions (Williamson, 1965). According to this view, as a country develops and becomes more economically integrated and industrialized, the distribution of economic activity becomes more unequal. Regions within nations do not typically possess equal capacity for growth and when development begins in those more fortunate, regional inequality will clearly increase. The manufacturing and the high value activities tend to concentrate in a few more advanced regions and the rest of the country experiences a process of retardation, concentrating its activity in lower productivity activities such as agriculture or more traditional manufacturing. As productivity in the modern industries progress more quickly fueled by technological change and economies of scale, income per capita will 1

According to the coefficient of variation (CV) calculated at the NUTS-2 level, France reached a maximum level of 0.319 in the period 1860-2010 which is twice the maximum level reached by United Kingdom (Geary and Stark, 2015) for the same period and similar to the maximum level of other European countries such as Sweeden (Enflo et al, 2010) and Belgium (Buyst, 2011). 2 The official regional GDP series published for the INSEE for 1990-2010 bring an unweighted CV of that runs from 0.172 in 1990 to 0.214 and a weighted CV runs from 0.265 to 0.325.

tend to grow faster in the industrializing regions than in the traditional ones. Hence, in the early stages of modern economic growth regional income differentials will increase. Additionally, labour and capital will flow from the laggard regions to the industrializing ones. Migrants used to be the most educated people at their productive age. This kind of selective migration increases the participation rates and the human capital in the recipient region and accentuates the tendency towards regional income divergence. By the same way, capital will flow to the richest regions looking for more attractive opportunities to invest (Williamson, 1965).However, as it has been observed in several countries, somewhere during the course of development, the disequilibrating forces will diminish, causing the convergence of the backward regions with the industrialized ones. Consequently, the statistical representation of regional income inequality will trace out an inverted U-shaped curve as suggested by Williamson (1965). Caselli and Coleman (2001) found that in the mature stages of economic growth most of the regional convergence is attributable to the structural transformation of the less developed regions. Over time, declining education/training costs increase the proportion of the labor force that abandon agriculture looking for higher wages in the nonagricultural sector. The decline in the agricultural labor force increases relative agricultural wages and productivity. Both effects benefit the less developed regions and lead to convergence in labor productivity and per capita income. Additionally, migration becomes less selective as wage differentials between skilled and unskilled tend to converge among both kinds of regions. The less developed regions increase their probability to retain educated and skilled workers, while loose the excess of unskilled. By the same way, net capital flows will arrive to the less developed regions. The reallocation of labour from the less developed regions to the industrial ones and the subsequent change in the economic structure of the laggard regions, moving from agriculture towards industry and services will drive convergence in income per capita. The inverted U-shaped curve was originally observed for the United States by Williamson (1965) and more recent studies for a myriad of European countries reveal a wide variability in the timing of the U-shaped curve.3 In France, historical debate has 3

In Spain the reversal took place around the first decade of the 1900 (Martínez-Galarraga, Rosés and Tirado, 2010 and 2015), in Italy in the 1950’s (Felice, 2009), in Great Britain the new estimations of Geary and Stark (2015b) date it back to 1901 meanwhile Crafts (2005) dated it around 1911 and in Portugal around 1970’s (Badiá-Miró, Guilera and Lains (2012). Buyst (2010) observed a decrease in regional inequality in Belgium since the beginning of the twentieth century and for Sweeden Enflo and Rosés (2015) observe that inequality dropped from 1860 to 1980 and did not show the U-shaped pattern observed in other European countries.

traditionally focused on the aggregate performance of the economy in the nineteenth and early-twentieth century, with little research into regional development. However, in the last years some studies analyze regional inequality in France with special attention to the ninetieth century (Caruana, 2013; Bazot, 2014) or making a quick look into the twentieth century (Combes et al, 2011). In general, all of them find that the start of the industrialization process in the first decade of the ninetieth century brought an increase in income inequality between departments that stopped with the turn of the century. Combes et al. (2011) offer a long term view using data for French departments for only three benchmark years (1860, 1930 and 2000) and find that the spatial distribution of economic activity (Manufacturing; Services) followed an inverted U-shaped relationship, with a decreasing trend since 1930. But, as these authors recognize, one of the main fragilities of their analysis is to check the validity of the bell-shaped curve with data for only three benchmark years. Later published studies increase the number of benchmark estimates for the ninetieth century. Caruana (2013) presents new estimations of GVA per capita in constant terms for 1860, 1872, 1886, 1901 and 1911. According to his calculations, on average, inequality between departments in France was growing until 1911. Bazot (2014) using the patente, a tax on non-agricultural value-added, presents even backward estimates of Gross Domestic Product (GDP) by department, starting in 1840, and dates the reversal in the bell-shaped curve around 1890. Apart from the above mentioned papers and a pioneering indirect estimation of regional evolution by Toutain (1981), there is a lack of studies that carry out a comprehensive statistical analysis of the French regional inequality following a long run perspective and covering with more detail the twentieth century. For this purpose in this paper we present new estimations of regional value added by sectors (agriculture, industry and services) to give a picture of the evolution of regional GDP per capita over the period 1860 to 2010. More precisely we fill in the gaps for 1896, 1911, 1921, 1954 and 1962 following the Geary and Stark (2002) method and these estimates are then linked to a EUROSTAT estimate for 1982 and to the INSEE official estimates from 1990 to 2010. The preliminary results indicate that regional income inequality in France followed an upward trend during the very early stages of industrialisation, 1860-1896. In contrast, regional GDP per worker converged mainly in the early 1900s and during 1954-1982. These findings go in line with the traditional view of an inverted U-shaped relationship. Nevertheless, the late decades of the twentieth century have witnessed a considerable

upswing in regional inequality that has recorded in 2010 levels close to 1930. In the last section of this study we realize a shift-share analysis to determine role of the process of structural change in the shaping of France’s economic cohesion between 1860 and 2010. The paper is organized as follow. In section 2 we present the new database of French regional per capita GDP. In section 3 income inequality and polarization are measured by applying a vast battery of inequality indicators. Finally section 4 presents a shift share analysis in order to show the role of the structural change process in explaining the relative performance of the French regions.

2. A new database on French regional per capita GDPs In order to analyze the long term evolution of regional inequality in France throughout the period 1860-2010 we have estimated new regional GDP figures for the years 1911, 1921, 1954, 1962 and 1982 following the standard NUTS-2 classification.4 One advantage in the analysis of the France regions is that the administrative organization by departments has remained quite stable since 1790.5 For the remaining years the data have been collected from different well known sources. We use the Combes et al (2011)’s cross-section which covers employment and value added in current francs for each French department in 1860 and 1930 in manufacturing and services.6 For 1896 4

Administratively, France is divided in 22 regions which corresponds to the European Union NUTS-2 level of classification: Îlle de France, Champagne-Ardenne, Picardie, Haute-Normandie, Centre, BasseNormandie, Bourgogne, Nord-Pas-de-Calais, Lorraine, Alsace, Franche-Comté, Pays de la Loire, Bretagne, Poitou-Charentes, Aquitaine, Midi-Pyrénées, Limousin, Rhône-Alpes, Auvergne, LanguedocRoussillon, Provence-Alpes-Côte d’Azur, Corse. The offshores regions (Guadeloupe, Martinique and Guyane) are not taken into account for the analysis that is centered in the European territories. The NUTS-2 classification is an aggrupation of the “Departments” division that was adopted in 1790 during the French Revolution.. 5 The administrative classification has remained quite stable across time. This classification initially accounted for 82 departments and since 1968 refers to 94 continental departments The main changes correspond to the consequences of 1871 French-German Treaty with the loss of Alsace (Haut Rhin and Bass Rhin) and part of Lorraine (Moselle) and the integration of the Belfort Territory into the FrancheComte region. These territories were recovered after WWI. We have adopted the convention of maintaining these departments into France estimates to give coherence to the whole period. The estimates corresponding to 1896 and 1911 for Alsace and Lorraine with Moselle have been facilitated by Niko Wolf who uses data provided by the censuses and wage surveys of Germany in those years. Another change refers to the 1968 post period. In that year the three departments of Ile de France were divided into eight, but this change does not affect to our classifications because it does not change the territory of the region. Something similar occurred in Corsica that was divided in two departments: North and South Corsica, but maintaining the boundaries of the region. 6 These estimates are based on Toutain’s work which provides data for each French department on population, employment and output for 1860 and 1930. The estimates are available at Combes et al (2010)’s data Appendix (tables A.1, A.2 and A.3). Agriculture estimates are based on the Agriculture Surveys for 1861 and 1929 (Toutain, 1992 and 1993). For industry the estimates for 1860 comes from Desaigues (2012)and are based on the Manufacturing Survey for 1860 (for Paris) and 1861-1865 8rest of

these authors also provide estimations for employment and value added, but not for value added in the service sector. Then we have estimated the service sector by region in 1896 taking into account the employment data published by Combes et al (2011) and the average labour productivity of the sector between 1860 and 1930 by departments. For more recent decades (1990, 1999 and 2010) we have taken the regional GDP at current values published by the INSEE at NUTS-2 level and for 1982 the estimates published by EUROSTAT7. For those years with no regional data available, the estimation of French per capita regional GDP is mainly based on the method developed by Geary and Stark (2002). This method uses national GDP estimates, breaking them down according to the regional employment structure and the corresponding regional relative wages by sector. It departs from the basic principle that the national per capita GDP, represented by YFR, is equal to the sum of all regions per capita GDP, Yi . Algebraically the total GDP of the French economy is the sum of all regions GDPs: (1) However, given that Yi by regions and departments is unknown, it will be proxied according to the following equation:

(2)

Where yij is the average value added per worker in the i-region at the j-sector, and Lij is the number of workers in each i-region at the j-sector. As we do not have data for yij this value is proxied by taking the national output per worker (yj) for each sector (agriculture, industry and services) and assuming that regional labour productivity in each sector is reflected by its wage relative to the French average wage in this sector (ωij/ωj). Then, the regional GDP will be given by:

France). Combes et at (2010) use the Appendix to the 1931 census which provides estimates of valueadded for a number of manufacturing industries for estimating the value-added for industry. 7 These data have been kindly facilitated by Miren Lafourcade.

(3)

Where ωij is the wage paid in the region i in sector j, ωj is the French wage in each sector j and βj is a scalar which preserves the relative region differences but scales the absolute values. This coefficient makes that the new estimations of regional GDPs sum up the national values and hence the new series constructed yearly from a myriad of different sources present chronological continuity in current values. The national values are the Toutain (1987)’s historical national GDP estimates for agriculture, industry and services8 for 1860 to 1930 and the official national accounts for the remaining benchmark years. Table 1. Main data sources for new estimations of French regional GDP Year 1911

1921

1954

1962

Wages

Employment

Wages survey, sociales. SGF Interpolation: -

1910: Ministère du Travail et de la prévoyance Population census, 1911 Agriculture: daily male wages Industry: 34 male professional occupations Services: Average agriculture & industry NUTS3: departments Population census, 1921 Wages in 1910 Wages in 1929: Bulletin de la Statistique Générale de la France et du Service d‘Observation des Prix. SGF - 13 male professions - NUTS3: departments Interpolation: Population - Wages in 1937: 13 male professions; Bulletin de la census, 1954 Statistique Générale de la France et du Service d‘Observation des Prix. SGF - Wages in 1962 - NUTS2: regions Wages survey, 1962: Recencement Industriel de 1963. Resultats pour Population census, 1962 1962 - Agriculture: daily wages - Industry: 66 industrial branches - Services: Average agriculture & industry - NUTS2: 22 regions

Source: Own elaboration.

8

The French historical national GDP are available at the Groningen Growth and Development Centre, the Historical National Accounts Database. http://www.rug.nl/research/ggdc/data/historical-national-accounts

Hence, Geary and Stark (2002) lets, in the absence of regional output figures, to estimate the GDP by region at factor cost, in current values.9 The estimate requires three sets of data: the national output per worker by sector, relative wages comparing sector wages in each region with the national sector wages and labour by sector and region. This methodology allows us to provide new estimates not only of regional GDPs but also figures for the different industries (agriculture, industry and services). The historical series of national output by sector (agriculture, manufacturing and services) are taken from Toutain (1987) for the years 1860, 1896, 1911, 1921 and 1930 and from the National Accounts for post-1950. All these figures can be found at the Groningen Growth and Development Centre Historical National Accounts Database. The data on working population come from the national censuses of population which provide detailed information about active population by sectors at different territorial levels (departments, municipalities…) for every ten years, approximately10. Table 1 resumes the main sources used. We would like to find precise correspondence between the census years for occupied persons and the sources used for data wages and value added, but this is not always possible. In general, the census year (1911, 1921, 1954 and 1962) determines when to match wage data with active population figures. French statistics are rich and provide vast information on regional wages, but unfortunately it was not able to find regional or departmental wages for each of the population census years. Only in two moments, 1911 and 1962, it was possible to stablish a contemporaneous correspondence between active population and wages. In both cases, there were current wage surveys which contained detailed information on salaries for different municipalities at the department level (for example the chêf de lieux)11. The survey published in 1911, "Salaires et coût

9

The Geary and Stark (2002) method presents the advantage of bringing the researchers the opportunity to estimate historical regional GDP with data sources based exclusively on relative wages by regions and national labor productivity. However, its main critical point is not to take into accounts the capital incomes as suggested by Crafts (2005). In a recent publication, Geary and Stark (2015a) compare their method with the variation suggested by Crafts (2005). They test that their method generates accurate estimates of regional GDP and find that there are practical and theoretical problems to test h the Crafts’s extension. They find their method robust with regard to wage coverage limitation by comparing their results with modern official estimates. 10

The population censuses used are the following. 1911, 1921, 1954 and 1962. The division of France into “departments” was adopted in 1790 during the French Revolution. This administrative organization substituted the old provinces and aimed to create more uniform units to submit them into a common central administration. The criteria used to establish their size was that people from any point in the department could arrive to the capital city in no more than two days. At the beginning there were 83 departments which were gradually increased to 94 continental departments. In

11

de l'existence: à diverses époques, jusqu'en 1910", collects wage data for 34 professions at the department level in 191012 and the publication "Recensement Industriel de 1963. Resultats pour 1962"13 collects wage data for 60 economic activities in 196214. In those years with no available information on current regional wages, the relative wage of one region with regards to the national wage has been calculated by interpolation between relative wages in two consecutive years with available information. For example, the relative wages for 1921 have been calculated as a weighted average of relative wages in 1911 and the relative wages in 1929. The source for regional wages in 1929 is the "Bulletin de la Statistique Générale de la France et du service d'observation des prix” published in 1930 which recorded figures on salaries for different working categories by departments and big municipalities for 192915. The same interpolation exercise has been done for 1954 with original wages for 1937 obtained from the "Bulletin de la Statistique Générale de la France et du service d'observation des prix” and the salaries of the 1962 survey. Figures for 1975 have been estimated by the Geary and Stark method using data of employment and salaries from several publications of the INSEE belonging to the collection of regional studies16. Finally, as Geary and Stark (2002) we have distributed regional GDPs in three different sectors (agriculture, industry and services) because this is the maximum level of disaggregation at the French Historical National Accounts (Toutain, 1987). In agriculture, we have not followed the Geary and Stark (2002) method but we have applied the distribution of the Gross Value Added in agriculture proposed by Toutain (1992-1993) to the historical national GDP in agriculture (Toutain, 1987).

1956, the departments were grouped into 21 continental regions in order to design policies at a larger spatial scale. 12

For 1911 we have used the wages survey: "Salaires et coût de l'existence: à diverses époques, jusqu'en 1910" which contains daily nominal wages (for males) in old francs for 34 male professions. You can find the list of professions at the Appendix. 13 We have found a similar survey for 1962 ,"Recensement Industriel de 1963. Resultats pour 1962". It contains Wages by region (Nuts-2) in thousands of new francs for 60 activities. You can find de detail of activities and their correspondence with the sector at the Appendix. 14 . We take the relative wages and not the wages in levels because wages in 1911 are in old francs and wages in 1929 in new francs. Additionally wages in 1911 are daily wages and in 1929 are hour wages. 15

The source for wages in 1929 is the "Bulletin de la statistique générale de la France et du service d'observation des prix", tome XIX, fascicule II (Janvier-Mars, 1930). Wages in 1937 comes from "Bulletin de la statistique générale de la France et du service d'observation des prix", tome XIX, fascicule II (Janvier-Mars, 1938). 16 Muet, Bolton and Cozin (1970); Chanut and Trêca (1975); Chanut and Monfort (1978); Mary and Turpin (1981); Donnellier, Malverney and Montlouis (1987).

Usually the wage surveys do not provide information of the wages in the service sector. That is the case for 1911, 1929 and 1937. In these cases we assume that wages in the service sector are a simple average of wages in agriculture and industry.

3. Regional income inequality in France, 1860-2010 Having introduced the dataset, we describe the main patterns of regional (NUTS2) income inequality in France, 1860-2010. Figure 1 displays σ-convergence for our period of study. More specifically, it shows the Gini index of per-capita GDP with bootstrapped standard errors for each year.

.05

.1

.15

.2

Figure 1. Regional per-capita GDP inequality in France, 1860-2010 (Gini)

1860

1890

1920

1950

1980

2010

Sources: main text. Notes: Solid line illustrates the Gini index. Dashed lines indicate one bootstrapped standard error.

Overall, regional disparities in France have declined since 1860. The Gini index fell from 0.15 in 1860 to 0.07 in 1990, increasing thereafter. This convergence was more acute during the Golden Age or “Trente Glorieuses” (1945-1975), but came to a halt in the last decade of the twentieth century. To account for demographic-related aspects,

figure 2 illustrates a simple coefficient of variation (CV) and a population-weighted coefficient of variation (WCV)17.

.4

Figure 2. Regional per-capita GDP inequality in France, 1860-2010 (CV; WCV) CV

.15

.2

.25

.3

.35

WCV

1860

1890

1920

1950

1980

2010

Sources: main text. Notes: The solid line connects the coefficients of variation, while the dashed line connects the populationweighted coefficients of variation. Our regional GVA estimates correspond to 1911, 1921, 1954, 1962, and 1982. Regional GDPs for 1860, 1896, and 1930 were already published (Combes et al., 2008), whereas INSEE provided estimates for 1990, 2000, and 2010.

Regional population dynamics are accounted for in figure 2. On the whole, the patterns are rather similar, but the population-adjusted coefficient of variation displays some peculiarities. For example, regional inequality, once adjusted for population, started to decrease in the late 1800s. This goes in line with existing empirical evidence. Bazot (2014), using decadal data from 1840 to 1910, noted that per-capita GDP inequality across Départements declined after 189018. Furthermore, our estimates can be compared with other case studies. Table A.1 in the appendix shows the single coefficient of variation for a selection of European countries. France ranges in the middle, with Great Britain and Sweden displaying the lowest values in the late twentieth century, and the 17

Williamson (1965) used a population-weighted coefficient of variation to measure the dispersion “of the regional income per capita levels relative to the national average while each regional deviation is weighted by its share in national population” (Williamson, 1965: p. 11) 18 Bazot (forthcoming) uses the patente, a tax on non-agricultural value-added, to estimate Gross Domestic Product (GDP) by department for the period 1840-1911.

Mediterranean countries the highest ones. Interestingly, regional inequality in France doubles that of Great Britain throughout the twentieth century19. This can be seen in figure A.1 in the appendix. Regional (NUTS2) disparities in France are in between Spain (NUTS2) and Great Britain (SSR). Furthermore, population dynamics also played a relevant role as figure 2 shows above. From 1860 to 1896, the population-adjusted coefficient of variation increased. This is contrary to what we observed in the single coefficient. In the early stages of economic development, rural-urban migrations are common. France was no exception. Although annual population growth rates across regions (NUTS2) ranged between -1.0% and 2.0%, spatial and time disparities were notable. Map A.2 in the appendix illustrates the annual rate of population growth for five major periods: 1860-96; 1896-1930: 1930-54; 1954-82; 1982-2010. From these maps, two contrasting patterns emerged between 1860 and 1930. On the one hand, Île-de-France grew rapidly, while other industrial regions (i.e. Nord-Pas-de-Calais, Haute-Normandie, Provence-Alpes-Côte d’Azur, Lorraine, Rhône-Alpes, Languedoc-Rousillon) did moderately. Rural regions, on the other hand, stagnated and/or lost population20. This is reflected in figure 2, and also in figures A.2 and A.3 in the appendix. Île-de-France, and particularly Paris, gradually increased its size in terms of population and gross-value added. By 1860, Île-de-France represented 7% and 15% of France population and GDP, while it was 15% and 25% in 193021. As stated before, it can be easily observed that the twentieth century exhibited a downward trend in regional income inequality, somewhat reversed in the interwar years and the late decades. This can be reconciled with the long-standing view of an inverted U-shaped relationship between economic development and regional inequality (Williamson, 1965). In this line of thinking, Combes et al. (2011), using data for 1860, 1930 and 2000, argue that the spatial distribution of economic activity (Manufacturing; Services) followed an inverted U-shaped relationship, as economic geography would have predicted. Deeper market integration (i.e. falling transport costs) stimulated the concentration of certain economic activities. Nonetheless, the spatial concentration of economic activity and regional per-capita GDP inequality did not go hand-in-hand. Labour reallocation, rural-urban migrations, and labour productivity catch-up spurred regional convergence. 19

Geary and Stark (2015b) provide two coefficients of variation. CV1 excludes London, whereas CV2 includes London in the Southeast Region. Hence, CV1 is computed for 11 regions, while CV2 for 10. 20 Between 1896 and 1930, just Île-de-France, Nord-Pas-de-Calais and Provence-Alps-Côte d’Azur grew above 1%. 21 Jean-François Gravier noted this in his 1948 seminal work: “Paris et le désert français”.

.5

1

1.5

2

Figure 3. Regional (NUTS2) per-capita GDP inequality, France 1860-2010 (France=1)

1860

1896

1911

1921

1930

1954

1962

1982

1990

2000

2010

Source: main text

To better understand the long-run evolution of regional inequality in France, figures 3 and 4 show a box-plot and kernel densities, single and population-weighted, for each year. Regional per-capita GDP has been indexed with respect to the average of France (FRA=1)22. A box-plot is simply a graphical representation where a box contains 50 per cent of the probability mass distribution. The bigger the box, the more inequality in income distribution is, the smaller the box the lesser inequality. The horizontal line inside the box represents the median value, while the horizontal lines outside are known as adjacent values. The scattered points, represented with small circles in figure 4, are outliers. In our case, the upper-outliers are Île de France for the whole period of study, Haute Normandie in 1860 and 1911, and Provence-Alpes-Côte d’Azur in 1921; whereas the lower-outliers were Corse in 1911, 1921 and 1954, Bretagne in 1911, and PoitouCharentes in 1962. It is noticeable that regional per-capita GDP inequality has compressed over the period of study, while Île-de-France has remained outside the interquartile box. Likewise, the distance between Île-de-France and France (FRA=1) resembles the Gini index and WCV depicted above in figures 1 and 2. Figure 4. Regional (NUTS2) per-capita GDP inequality, France 1860-2010 (France=1)

22

Table 1 in section 2 shows per-capita GDP values for each region 1860-2010 normalised (FRA=1).

1896

1911

1921

1930

1954

1962

1982

0

2

4

6

0

2

4

6

1860

.5

2000

1.5

2

2010

4

6

1990

1

unweighted

0

2

weighted

.5

1

1.5

2

.5

1

1.5

2

.5

1

1.5

2

Source: main text

The corresponding kernel density functions can be drawn for each year, and are shown in figure 4. Again, regional per-capita GDP has been indexed with respect to France (FRA=1). This, in turn, provides an accurate description of the regional per-capita GDP distribution. To do so, we use non-parametric estimation techniques. In the x-axis, we plot the regional per-capita GDP normalised (FRA=1), while the associated density is presented in the y-axis. We use a Gaussian kernel function. As in figure 3, regional percapita GDP has compressed since 1860, but Île-de-France has stretched the upper tail of the distribution. Overall, the initial dispersion in per-capita GDP has converged into a club, thus creating one major pole: Île-de-France23. Moreover, figure 4 show both, single and population-weighted kernel densities. When we adjust for population size, regional inequality exhibits a similar pattern. Also, the estimated density functions, population-adjusted, suggests the advent of some degree of polarisation. Although the absolute distance seems to have been reduced, Île-de-France represents roughly 30% of GDP and 15% of population in 2010, doubling our estimates for 1860. Figure A.3 in the appendix shows the long run evolution of Île-de-France GDP and population as a share of France. The above descriptive evidence thus hints a compression of regional per-capita GDP, and the growing presence of Île-de-France. This has been a consistent leader during our 23

Figures A.2 and A.3 illustrates Île-de-France growth throughout the period of study.

period of study, but its demographic and, above all, economic weight has risen over time. In fact, it is noteworthy that the compressed pack of French regions has moved away from Île de France in recent decades, as figure 4 shows. This, in turn, points to a potential bimodality in the future, or polarisation. Table 1 in section 2 illustrates percapita GDP normalised (FRA=1) for each region (NUTS2) and year. Apart from Île-deFrance, few of the rich regions in late 1800s remain at the top of the regional income distribution in 2010. By 1900, Île-de-France, Haute-Normandie, Champagne-Ardenne, Nord-Pas-de-Calais, and Picardie were in the top-5, but not in 2010. In fact, Nord-Pasde-Calais (14#) and Picardie (21#) were in the lower tail of the income distribution. Rhône-Alps and Provence-Alpes-Côte d’Azur, Alsace, and Pays de la Loire, on the other hand, witnessed upward mobility, being in the top-5 in 2010. The spatial localization of French GDP presents significant changes from 1860 to 2010. The overwhelming advance of the Paris region, that doubled its share between 1860 (15.5 %) and 2010 (31%), has hidden the progresses made by the other regions. Only two regions have increased their presence in the national GDP, Rhone-Alpes from 8 to 9.8% and Provence-Alpes-Côte d’Azur from 4 to 7.3%. The five regions around the Parisian core have halved their participation in the French economy (Champagne, Picardie, Haute-Normandie, Centre and Basse-Normandie).

4. What accounted for regional income convergence in France, 1860-2010? Regional disparities in per-capita GDP can be reduced through changes in the size and structure of populations and labour productivity catch-up. Heterogeneities in the composition of the population (i.e. working-age population) and/or labour force participation might be relevant for our analysis. If a given region A has more people working relative to its total population than region B, and labour productivities do not differ, then the former would be richer in per-capita GDP. To further examine the contribution of demographic factors to regional income inequality we should decompose regional per-capita GDP inequality into two major components: labour force

and labour productivity

. Enflo and Roses (2015)

found that labour productivity explains more than 70 per cent of regional income

inequality in Sweden 1860-2000. In our case, a preliminary look at the contribution of both across regions pointed to the preponderance of labour productivity24. Regional convergence in labour productivity has been broadly studied and examined. Theoretically, we should expect that factor-scarce regions experience higher rates of return, becoming poles of attraction. If factors freely flow, technological gaps should be reduced, and hence regional disparities. But, the “immobility of the factors and the secular inertia of the structures” (Toutain, 1981: p. 303) could prevent this process from happening. Furthermore, greater market integration does not necessarily lead to convergence. In the early stages of development, production (i.e. manufacturing) tends to concentrate in specific locations, as Combes et al. (2011) already illustrates for France. Thus, it is relevant to understand the sources of labour productivity convergence or divergence. In our case, convergence in per-capita GDP has prevailed in the period of study. Following Enflo and Roses (2015), who adapted Caselli and Tenreyro (2006) approach, we decompose labour productivity

convergence into three major components: (a)

Within-industry; (b) Labour reallocation; (c) Between-industry. Enflo and Roses (2015) propose to use as benchmark for the comparisons the richest spatial unit and not the mean observation as did Caselli and Tenreyro (2006). Algebraically,

(4) While

captures labour productivity, measured as value-added per worker, in

industry (agriculture, industry and services)

and region ,

is the share of

employment. Both terms are indexed, i indexes the counties, j industries (agriculture, industry and services) and t time. In other words, overall labour productivity is a weighted sum of sector’s labour productivity with weights given by the share in total employment of each sector. Given that we use Ile the France as the numeraire in our convergence analysis, i= PR, convergence to the Paris region is equal to:

(5)

24

A first look into the break-down of the Theil index into labour force and labour productivity suggests that the latter one explained around 85-95% of regional income inequality in France, 1860-2010. In turn, the periods in which relative differences in labour force played a more relevant role were 1860-1930 and 1990-2010.

As Caselli and Tenreyro (2006) demonstrates this measure of convergence could be decomposed into three different sources of convergence: within industry convergence, labour reallocation and between industry convergence. Within-industry convergence’ captures labour-productivity catch-up of each sector with the corresponding sector in Île de France, weighted by the average labour share. Labour reallocation’ captures the convergence caused by labour flows across sectors. Each sector is weighted by its relative productivity. Between-industry convergence’ captures the convergence in labour productivity across sectors into a region, weighted by each region share with respect to Île de France. The final decomposition could be written such as:

We call “Total convergence” the quantity on the left-hand side of the equation; “withinindustry convergence” is the quantity in the first line of the right-hand side; “labour reallocation” is the quantity in the second line and “between-industry convergence” is the third line. Since Île de France is the leading region for the whole period of study, we use Île de France as benchmark. That is to say, our decomposition of regional per-worker GDP convergence can be understood as a catching up between the leader and the rest. To simplify matters, we have grouped our 22 regions (NUTS2) into 8 major regions (NUTS1)25. Table 2 shows the decomposition of labour productivity for the whole period.

25

Metropolitan France is divided into 8 major regions (NUTS1): Îlle de France; Bassin Parisien (Champagne-Ardenne, Picardie, Haute-Normandie, Centre, Basse-Normandie, Bourgogne); Nord-Pas-deCalais; Est (Lorraine, Alsace, Franche-Comté); Ouest (Pays de la Loire, Bretagne, Poitou-Charentes); Sud-Ouest (Aquitaine, Midi-Pyrénées, Limousin); Centre-Est (Rhône-Alpes, Auvergne); and Méditerranée (Languedoc-Roussillon, Provence-Alpes-Côte d’Azur, Corse).

Table 2. Decomposition of convergence in per-worker GDP, 1860-2010 Region (NUTS1) Bassin Parisien Nord-P.-Calais Est Ouest Sud-Ouest Centre-Est Méditerranée France1

Overall -0.0362 -0.1105 0.1368 0.1541 0.2351 0.2221 0.1439 0.1219

All -0.0438 -0.1121 0.0775 0.0339 0.0576 0.0149 0.0054 0.0125

Within-industry Agr. Ind. 0.0706 0.0166 0.0202 0.0042 0.0913 0.0926 0.0671 0.0206 0.0870 0.0585 0.0663 0.0374 0.0776 0.0312 0.0748 0.0350

Ser. -0.1311 -0.1365 -0.1064 -0.0538 -0.0878 -0.0888 -0.1034 -0.0973

Labour reallocation 0.0545 0.0487 0.0949 0.1604 0.2168 0.2527 0.1536 0.1486

Betweenindustry -0.0469 -0.0471 -0.0356 -0.0401 -0.0393 -0.0455 -0.0152 -0.0392

Source: main text. Notes: (1) France excluding Île-de-France. Île de France is our benchmark.

Overall, our results suggest a catch up in labour productivity between Île-de-France and the rest, except for Bassin Parisiene and Nord-Pas-Calais. This is noteworthy, because the latter were among the richest regions in the late 1800s. Table 2 also indicates that the principal force of labour productivity convergence throughout our period of study was labour reallocation, whereas within- and between-industry played a minor role26. These results provide further evidence in support of the structural interpretation of regional income inequality. In fact, if labour productivity convergence is decomposed in shorter periods (i.e. 1860-96; 1896-1930; 1954-82; 1982-2010), it can be shown that per-worker GDP convergence essentially occurred between 1896 and 1982. Moreover, this resulted from convergence in productivities across sectors or between-industry, and labour reallocation, as figure 5 illustrates.

26

Given our period of study, 1860-2010, within-industry played a minor role in comparison with labour reallocation. Between-industry, on the other hand, did not contribute to labour productivity convergence.

Figure 5. Decomposition of convergence in per-worker GDP by period, 1860-2010 .2

Within!industry Labour:realloca>on

!.2

!.1

0

.1

Between!industry

1860!96

1896!1930

1954!82

1982!2010

Source: main text. In the annex, tables 3-5 you can find the decomposition of the three effects by periods and for a NUTs-2 aggregation (22 regions). Notes: Île de France is our benchmark.

This confirms that little, if any, convergence in labour productivity occurred in the early stages of modern economic growth. Economic activity (i.e. Manufacturing) tends to concentrate in specific locations. This is reflected in figure 5 where between-industry convergence is counteracting within-industry and labour reallocation. Thenceforth, it is expected that “internal factor mobility should tend to eliminate interregional income per capita differentials, geographic dualism, or spatial polarization” (Williamson, 1965: p. 5). Figure 5 shows that convergence in labour productivity occurred between 1896 and 1982, with between-industry and labour reallocation as major forces. This can be easily reconciled with the historical experience. The primary sector in France employed a large share of the labour force in the early 1900s. The mechanisation of agriculture, and rural-urban migrations played a great role in the convergence process. Capital-scarce regions, mainly agrarian, where labour productivity was rather low, welcomed investment and pushed labour out. This structural transformation was at the core of regional income convergence. After World War II (WWII) economic planning aimed to reduce regional disparities implemented a set of policies that tried to promote investment out of the Paris region. Therefore, rapid structural change and economic

policy somewhat explain regional convergence in per-capita GDP. However, technological change (i.e. ICT revolution) has triggered further socioeconomic changes. But, this time is different, with regional income inequality at a minimum, Île de France has benefitted more than the rest, as figures 3 and 4 clearly indicate. Whether this is a different wave seems apparent. Our results indicate that the catch up between Île de France and the rest has concluded. In fact, the twenty first century presents new challenges, especially the growing muscle of specific regions or cities. This might alter the territorial cohesion of nation-states, bringing back old debates such as “Paris et le désert français”.

5. Conclusions

This study, using a new database of regional per-capita GDP, explores regional income inequality in France, 1860-2010. In line with the process of modern economic growth regional inequality increased in the early stages of industrialization. Then, convergence occurred for most of the twentieth century, coming to a halt in the last three decades. Our study provides further evidence in favour of an inverted U-shaped pattern between economic development and regional inequality (Williamson, 1965). Regional population dynamics are acknowledged, especially for the leading region: Îlede-France. It is noteworthy that Île-de-France doubled its share in France population from 15% in 1900 to a 30 % in 2010. However, our main focus lies with the role played by structural change. We find that convergence towards the leading region (Île-deFrance) in labour productivity occurred mainly between 1896 and 1982, with betweenindustry and labour reallocation as major forces. The only period with a clear catch-up was the Golden Age. This can be easily reconciled with the historical experience. The primary sector in France employed a large share of the labour force in the early 1900s. The mechanisation of agriculture, and rural-urban migrations played a great role in the convergence process. This was reinforced after World War II (WWII) by central planning aimed to reduce regional disparities. Once structural change was exhausted, other forces might drive regional inequality up, e.g. globalisation, tertiarisation.

References Badià-Miró, M., Guilera, J.,

Lains, P. (2012). Regional Incomes in Portugal:

Industrialization, Integration and Inequality, 1890-1980”. Revista de Historia Económica. Journal of Iberian and Latin-American Economic History, 30, pp. 225-244. Bazot. G. (2014). Interregional Inequalities, Convergence and Growth in France from 1840 to 1911. Annals of Economics and Statistics, N. 113-114, pp 309-346. Bolt. J. and J. L. van Zanden (2013). The First Update of the Maddison Project; ReEstimating Growth Before 1820. Maddison Project Working Paper 4. Broadberry, S.N., Federico, G., Klein, A. (2010). Sectoral developments, 1870-1914. In S.N. Broadberry and K.H. O’Rourke, ed. Cambridge Economic History of Modern Europe. Volume 2, 1870 to Present. Cambridge: Cambridge University Press, pp. 164-186. Buyst, E. (2010). “Reversal of fortune in a small, open economy: Regional GDP in Belgium, 1896-2000”, Rivista di Storia Economica, 26, pp. 75-92. Caselli, F., Colleman, W.J.(2001). “The U.S. structural transformation and regional convergence: A reinterpretation.” The Journal of Political Economy, N. 109 (3), pp. 584-616. Caselli, F., Tenreyro, S. (2006). Is Poland the Next Spain? In Clarida, R.H., Frankel, J., Giavazzi, F. and West, K.D., NBER, Intenational Seminar in Macroeconomics 2004. Cambridge MA, pp. 459-533. Chanut, J.M., Trêca, L.(1975). “Analyse régionale et indicateurs regionaux”. Les Collections de l’INSEE, C3, INSEE. Chanut. J.M.. Heffer. J. Mairesse. Postel-Vinay. G. (1995). “Les disparités des salaires en France au XIXe siècle”. Histoire and Mesure. vol. 10 (3-4). Chauet, A., Monfort, J. (1978). “Les comptes régionaux des branches industrielles en 1975”. Les Collections de l’INSEE, C3, INSEE. Combes. P-P. Lafourcade. M.. Thisse. J-F. Toutain. J-C. (2011). “The rise and fall of spatial inequalities in France: a long run perspective”. Explorations in Economic History. vol 48(2).

Crafts, N. (2005). “Regional GDP in Britain, 1871-1911: some estimates”. Scottish Journal of Political Economy, 52, pp. 54-64. Donnellier, J.C., Maliverney, J., Montlouis, E.(1987). “L’appareil productif regional de 1975 a 1984. Un analyse des valeurs ajoutées”. Les Collections de l’INSEE, C3, INSEE. Enflo, K., Henning, M., Schön, L. (2010). “Swedish regional GDP 1855-2000: Estimations and general trends in the Swedish regional system. Working Papers in Economic History, WP 10-03. Department of Economic History and Institutions, Universidad Carlos III, Madrid, Spain. Enflo, K., Rosés, J.R. (2015). “Coping with regional inequality in Sweeden: Structural change, migration and policy, 1860-2000”. The Economic History Review, 68(1), 191-217. Felice, E. (2011). “Regional Value Added in Italy, 1891-2001, and the Foundation of a Long-Term Picture”, Economic History Review, 64 (2011), pp. 929-50. Krugman. P. (1991). “Increasing returns and economic geography”. Journal of Political Economy. vo. 99(3). pp. 483-499. Krugman. P. and Venables. A.J. (1995). “Globalization and the inequality of nations”. The Quartely Journal of Economics. vol 110(4). pp. 857-880. Mary, S., Turpin, E. (1981). “Panorama économique des régions françaises”. Les Collections de l’INSEE, C3, INSEE. Martínez-Galarraga, J., Rosés, J.R., Tirado, D.A. (2010). “The upswing of regional income inequality in Spain (1860-1930)”. Explorations in Economic History, 47(2), 244-257. Martínez-Galarraga, J., Rosés, J.R., Tirado, D.A. (2015).“The long-term patterns of regional income inequality in Spain (1860-2000). Regional Studies, 49(4), 502517. Morrison. C., Snyder. W. (2000). The income inequality of France in historical perspective”. European Review of Economic History. vol 4(1). pp.59-83.

Muet, P.A., Bolton, P. and Cozin, F. (1970). “Études de démographie régionale. L’emploi de 1954 a 1968”. Les Collections de l’INSEE, C3, INSEE. O’Brien, P. and Keyder, C. (1978). Economic Growth in Britain and France, 1780-1914. Two paths to the Twentieth Century, Londres, Allen and Unwin. Recensement Industriel de 1963. Resultats pour 1962. Salaires et coût de l'existence: à diverses époques. jusqu'en 1910. Ministère du Travail et de la prévoyance sociale. Statistique générale de la France. Toutain.J.C. (1987). “Le produit interieur brut de la France de 1789 á 1982”. Économies et Société. Série AF 15. http://www.rug.nl/research/ggdc/data/historical-nationalaccounts. Toutain. J-C. (1992-1993). “La production agricole de la France de 1810 à 1990: départements et régions”. Économies et Sociétés. Série AF 17. 3 tomes. Toutain, J-C (1996). Comparison entre les différents évaluations du Produit Intérieur Brut de la France de 1851 á 1938 ou l’histoire économique quantitative a-t-elle sens? Revue Économique, N. 47, pp. 893-919. Toutain. J-C. (1997). “La croissace française. 1789-1990. nouvelles estimations”. Économie et Sociétés 31. Série HEQ 1. 5-136. Williamson, J. G. (1965). “Regional Inequality and the Process of National Development”. Economic Development and Cultural Change, 13, pp. 1-84.

APPENDIX Table A.1. Regional income inequality in Europe: A summary. Country Belgium Finland France Great Britain Great Britain Italy Portugal Spain Sweden

Coeff. Var. Regions Years Max. Min. Source

Period 1896-2000 1880-2010 1860-2010 1871-2001 1901-2001 1891-2001 1890-1980 1860-2000 1860-2000

9 12 21-22 10 10 16-19 18 17 24

5 14 11 10 10 7 8 16 15

0.28 0.42 0.32 0.16 0.18 0.39 0.45 0.44 0.33

0.17 0.17 0.17 0.08 0.08 0.22 0.20 0.21 0.07

Buyst (2011) Enflo (2014) (see text) Crafts (2005) Geary and Stark (2015b) Felice (2012) Badia-Miro, Guilera and Lains (2012) Martinez-Galarraga et al. (2015) Enflo et al. (2010)

.5

Figure A.1. Regional income inequality in Great Britain, Spain, and France 1860-2010 (WCV) GBR ESP

.1

.2

.3

.4

FRA

1860

1890

1920

1950

1980

2010

Sources: Great Britain: Geary and Stark (2015b); Spain: Martinez-Galarraga, Roses and Tirado (2015); and France: main text. Notes: There are 10 regions in Great Britain, and the coefficient of variation corresponds to CV2 in Geary and Stark (2015b). In Spain and France, on the other hand, there are 17 and 22 NUTS2 regions. In all three cases the coefficient of variation is population-weighted, as Williamson (1965) proposed.

Figure A.2. Île de France GDP and Population, 1860-2010 (as % of France)

Map A.2. Regional population growth in France by major period, 1860-2010 (%)

Source: main text. Notes: Annual population growth rates have been classified in 6 equal classes ranging from -1.0% to 2.0%.

Suggest Documents