2016
Aniello, Gianmarco
S2548402, University of Groningen,
Faculty of Economics and Business
Supervisor: Dr. Raquel Ortega Argiles,
University of Groningen
Co-assessor: Marco Maria Sorge,
University of Göttingen
DOES PUBLIC EMPLOYMENT
DISTORT LOCAL LABOUR
MARKETS IN DEPRESSED
AREAS?
The case of Italian Provinces
Abstract1
Table of Contents
1. Introduction
3
2. Debate and Literature Review
4
3. Data
11
4. The Model
16
5. Descriptive Statistics
19
6. The Hypothesis
27
7. Results
28
8. Robustness
37
9. Discussion, Limitations and Conclusion
39
2
Tables
1. Commuters
13
2. Total Employment by different Censuses
14
3. Description of the variables
19
4. Descriptive statistics (all provinces)
20
5. Descriptive statistics of not southern provinces
20
6. Descriptive statistics of southern provinces
21
7. (a/b) Correlations among variables
21
8. Wage Gaps by region
22
9. Industrial sector estimation results
29
10. Tertiary sector estimation results
30
11. Agricultural sector estimation results
32
12. Agricultural sector estimation results (WG)
34
13. Tertiary sector estimation results (WG)
35
14. Agricultural sector estimation results (WG)
36
15. Mundlak Test (total population )
38
16. Mundlak Test (working age population )
39
Graphs
1. Distribution of industrial entrepreneurs across provinces
23
2. Distribution of tertiary entrepreneurs across provinces
24
3. Distribution of agricultural entrepreneurs across provinces
25
4. Distribution of public employees across provinces
26
3
1.Introduction
Since its constitution Italy has been characterized by huge geographical disparities in terms of economic development. In a simple neoclassical framework per capita growth is just due to accumulation of capital (both physical and human), hence in a national context, with low barriers to movement of capital, labour and knowledge, some convergence between “rich” North and the “poor” should have taken place easily.After 150 years the North-South gap widened, despite some modest and not sustained catching up periods. The “ Questione
Meridionale” ( southern issue) has been often liquidated in the public opinion debate as just the
economic consequence of a lack of infrastructures and poor institutions , but while the direction of causality might be debatable, these factors typical could only explain the fact that total factor productivity (as showed in Di Giacinto et al. 2006) is lagging behind in southern regions. If labour is less productive, it should be cheaper enough to naturally attract investments from the more high-productive and expensive North, especially if capital per worker is lacking and labour force is overabundant. On the contrary investments in the industrial sector were massively carried out directly and incentivized by the State after the Second World war, and in spite of that, almost no catching up was experienced: manufacturing sector in South never took off, and in the last decades it is even retrenching (see Literature review). This elementary economic forecast, has the fallacy to assume labour cost to reflect lower labour productivity in the South, in facts when we consider the large manufacturing companies, the national wage bargaining system might have obviously hindered the competitiveness of South. Yet it is less obvious, in the case of smaller enterprises , where wage bargaining is depending more on local labour market condition, to individuate a similar bottleneck. The largest employer in Italy, namely the State , could play a role, because its wages are mostly set on a national level, with no regard towards local labour market conditions, price levels, productivity . This would determine a sort of twisted Balassa-Samuelson effect , where it is the non-tradable sector (public services) to drive up salaries and even prices in all the economy, with no relationship with labour productivity.
4
to productivity and global competitiveness. So without any competitiveness in the tradable sector growth stunts, and unemployment would be large because of too high formal salaries. An “exception” to low competiveness in tradable sector could be represented by the agriculture, where an high level of informality reduces labour cost, thanks to a loose control performed by institution on immigrant workers and local ones who prefer to register themselves as unemployed in order to get public subsidies. This black economy together with an high tax evasion, it might be needed to maintain this economy, and then demanded to local institution, which in turn become “poor” and feed a vicious circle of stagnant growth . At the same time palatable public jobs can be used to create political consensus, whereas efficiency of public services could be undermined further depressing private sector competitiveness.
In this paper, we are aiming at analysing empirically whether public employment affects local labour markets in economically depressed areas of the country. To do so we will use panel data on public and private employment (by macro-sectors), so that we can carry out a within estimation. Furthermore our data have a level of aggregation NUTS-3 (by provinces), while previous analysis mostly do not go beyond the NUTS-2 (regions).
The within estimation and provincial level data represent one of our major contribution about the Southern Italy issue. Furthermore we will estimate heterogeneous impacts of public employment according to public-private wage differentials, bringing together two strands of literature, about public-private wage gap and public employment effects on labour market .
2.Debate and Literature review
In 2014 Italy was the only “ big” economy of EU to have still a negative growth rate (-0.4%), yet this aggregate figure conceals wide regional differences. According to the 2015 Economic Report by SVIMEZ1 (Association for the Industrial Development of South Italy), GDP decreased by 1.3 % in the southern regions (by 0.2 in the Centre-North) , while between 2008 and 2014 the cumulative contraction of the GDP accounted for 13% in the South , but only by 7.4% in the rest of the country. Investment between 2001-2014 diminished by ca. 60% in the already modest industrial sector, and the same time the labor productivity diminished in absolute and relative term w.r.t. the Centre-North. Also the other two sector contracted in comparison to
5
North2. The figures about are even worse: employment rate in the male population between 20 and 64 is just 45.5 (67.7 for North); during the crisis 70% of the ca. 811 thousand jobs lost where from the South. Finally 38.7% of people between 15 and 34 are “NEETs” (Not in Education, Employment, or Training), compared to 20% in the Centre-North. To sum up the study shows how in the last 15 years no convergence was experienced between sub-developed and developed regions, on the contrary the gap is dramatically widening. Conversely it shows that in the EU some catch-up is taking place between poor and rich areas, while in another “dualistic” economy, Germany, the growth rates have moved together (West-East). For 2014 the per capita GDP of a southern Italian was 53.7% (55.5% in 2006) of a northern conational, but population drop and internal migration mitigated this figure. The authors individuate as main drivers of this: the drop in capital stock, and the retrenching of public sector employment and intergovernmental transfers. They argue that massive public investment is needed, to compensate for a “market failure” in capital accumulation, and at the same time incentives and fiscal exemptions must be provided to private enterprises to attract investments. The policy proposed is basically the one implemented in the 30-40 years after 1950 through the institution of the “ Cassa del Mezzogiorno” , a public institution whose purpose was to trigger a “big push” to industrialize the South. Scarlato (2011) provides a review of the policies implemented to favor the convergence between the “ two Italies” after the Second World War. The Agricultural Reform was the first pillar to relaunch the depressed southern economy, since the inheritance of the feudal system was still predominant . Then massive emigration played a huge role, around 4 million people (in 1946-1973) moved to North or abroad. Finally following the development theories of the time, the central State acted as an economic planner, since the critical threshold of capital accumulation determinant for the growth take-off could not be achieved through market forces . The idea was that market failed because spillover-effect and agglomeration , links in the upward and downward value chain and lack of infrastructure were externalities that needed a public coordination. So by law 40% of investments from public enterprises (very influential at the time) had to take place in the South public works, while private enterprises received eased credit and other incentives to invest. This resulted in a massive capital accumulation in the 60s-70s, but unfortunately most of the it was in heavy, chemical and naval industry, whose inputs were not locally produced and output were bought by other industries in
6
7
depress private entrepreneurship. In this perspective a low social capital would be just part of a vicious circle, difficult to break, if not faced in its entire complexity .
Since we focus on labor market distortions determined by national policies, public employment is likely its one main component (but also rules, pensions, subsidies plays a role), and probably the most persistent. First source of concern is its absolute and relative size. Santoro ( 2014) treats also about the “southernizzation” of public employees working in northern regions; although this does not affect directly the southern economy, it affects the educational choices of many young people aiming at safe public jobs. Alesina et al (1999), analyze the imbalance in the distribution of public employment across Italian region;, the simple descriptive analysis shows that for every 100 inhabitants there are 6.1 public employees (5.1 in the North), whereas they are 22.1 (12.4 in the North) for every 100 employed given the low employment rate (Centre shows some peculiarity because of central government employment). In addition by analyzing such figures, it must be taken into account that a poorer economy will demand less public administration because of Wagner’s law. For example with regard to the postal service, the per employee volumes of shipping are 10 times larger in North than South (Alesina et al. 1999). This is strictly connected with a second source of concern, namely inefficiency: on the one hand the number of employee is not justified by the demand of services, since the southern economies are smaller. On the other the quality of services it was estimated to be worse in the South of the country by a number of studies studies (Putnam, Bank of Italy, Censis ), again social capital plays a role. For example, in a recent paper, Giordano et al. (2015) study how local public sector inefficiency hinder labor productivity of enterprises across Italian provinces.
8
of 26% for Southern public employees against private employees with similar characteristic, while in the North the premium is reduced to just 12%.
A more refined estimation of the wage premium is carried out in Dell’Arringa et al. (2007), in order to analyze the impact of the labor market reforms implemented in the early 90s. The authors also provide a description of the institutional framework. As already partly mentioned above, in the 80s the rise in public jobs degenerated and often was used to sustain income and employment in the most depressed areas, on the other hand also highly centralized wage bargaining system and wage indexation to national inflation made labor market really rigid and not sensitive to the conditions of the economy in the local areas. To counteract this, apart from a massive privatization of public enterprises, the determination of salaries shifted away from legislation to national collective bargaining (increasing the importance of unions) that fixed national minimum standards, whereas local administrative areas were allowed to bargain an increase of their local wage level. This system should have better reflected geographical economic disparities in public employees’ wages. Some flexibility was also introduced by allowing private temporary work agency to work in the public sector , and public employment hiring was subjected to severe limitations. Nonetheless geographical wages dispersion did not take place among public employees3, in contrast local autonomy was often used to increase wages over national standards, and public manager compensation bill reached levels comparable with the private sector. Using the same database from the Bank of Italy survey (likewise Alesina et al.), the empirical analysis is implemented through different estimations technique: OLS with the typical wage equation controls, but also allowing for different regional coefficients, Oaxaca decomposition, quantile regression and Geographical Weighted Regressions. Yet the main interesting findings are that public wage premium for men is relevant just in the South, while for women is more widespread. In addition in South men wage premium declines in the upper quartile, while for northern women wage gap widens with income ( conversely from southern ones). But what are the consequence of this public wage premium? Pagani (2003) calculated through a bivariate probit model the simulated probability affecting job search decision, and found that southern Italians exhibit a lower probability to have carried out any job search activity (as result of discouragement and shadow employment), at the same time the probability of the job
9
seekers looking for job only in the public sector is almost double for southern Italians compared to northern. Again all this strand of literature assesses that private sector employment is highly unsecure in the South, while public jobs ensure life employment and vacations, so this combined with the wage premium, as stated by Alesina et al.(1999), generates a “culture of dependence” where occupational choices are public sector-driven -- and further exacerbates “wait” unemployment and low participation rates. The same authors also analyse the individuals choice of undertaking an entrepreneurial activity (using a logit model), and found that while low productivity encourages entrepreneurship, typical characteristic of the South (see Giacinto et al. 2006), this effect is offset by the share of public employment present in the region.
Public sector wage differentials have also been analyzed in other countries, Disney (2007) reviews the methodology used in the literature, and for the Italian case is interesting to mention how unaccounted shadow economy could underestimate wage gaps, since in the South informal labor represents a large share of employment (20% in some southern regions4).
Borjas (2002) analyse the trend of public-private wages for the USA, where sectorial divide was present just for women, but declined with the time. On the contrary public sector did not manage to catch-up with rising private wages of highly skilled worker.
Pedersen (1990) analyse the wage-twist in favour of private wages, that was beneficial to stimulate employment and alleviate tax burden.
Furthermore there is a strand of literature that analyse the impact of public sector on labour market and unemployment rate. Algan et al.(2002) carry out a cross-country analysis over a sample of OECD countries. Through a simultaneous equation model they study how the creation of extra public job affect private employment and unemployment. Firstly they found that unemployment is generally not reduced by public jobs increase. Secondly they sort the countries according to wage premium ( estimated by Blanchflower (1996)) and corruption indexes (by Lambsdorff (2000)), finding that for above median observations there is a significant crowding effect of private jobs (namely labor force and employment rate diminish). Both indicators as mentioned above would be high in the case of southern regions (low social capital, and high-wage gap).
10
Yet this works does not study regional differences within a country, where public wages are set equal across the territory. Faggio et al. (2014) examine the case of England, where public jobs where used to fight unemployment in depressed areas. They studied how the local labour markets react to public employment, by analyzing the change in private employment of different sectors. Analysing a short time span (2003-2007), Faggio et al.(2014) found that public employment have no multiplying effect on private employment; but it affects sectorial distribution: jobs in the non-tradable sectors were positively affected, while tradable ones were crowded out with same magnitude. Conversely, using a longer time span (1999-2007), they also found that tradable sector crowd-out remain persistent, whereas non-tradable beneficial effect tend to disappears with time. Still this last study misses to consider wage premium differences across local areas that could be crucial in affecting the labor markets.
Burdett (2012) proposes a theoretical model to incorporate public sector in the local labor markets. He divide local areas according to their public wage premium, so that in its model a public employment program would fail to reduce unemployment in High or Medium Wage Premium areas, while it will succeed in Low Wage Premium areas.
In conclusion we will include some measure of wage gap in our empirical analysis on local labor markets
Unfortunately this literature ignores the problem of different price levels between poor and rich local areas, or at least does not face it empirically, and we will not do it either. Alesina et al.(1999) wrote:
“[…] while public wages are very similar across regions, the price level instead is lower in the South, so that real wages are higher in the South.
11
3.Data
Almost all the data used in the analysis were collected by ISTAT ( the Italian National Institute for Statistics). The Institute carries out a survey on Labor Force with quarterly frequency, where a sample of households is inquired with a great level of detail about the working conditions of the components. Although the use of this panel microdata would have been idealistic in the framework of this study, the results of the survey are not fully available for the general public. A particular source of problem was the availability of data on provincial level, which is one the crucial contribution the this analyses is aiming at providing. Freely available Labor Force Survey microdata files conceal the provenience (and residence) of the surveyed individuals with regard to the provincial and municipal level, and no long time-span is freely accessible. Similarly the Survey on Household Income and Wealth (SHIW) from Bank of Italy, which offers a yearly panel dating back to 1977, provides openly only the region of residence of the sampled individuals. Finally, even the aggregated database derived from ISTAT Labor Force Survey, does not allow any disaggregation to provincial level. The purpose of the analysis was to estimate impact of public employment on local labor market across sectors, in order to proxy private employment sectorial dimensions other data than employment might have been used, e.g. enterprises datasets. On the other hand the same could not be done with regard to public workers. So the major constraint originated by the lack of public employment provincial data. The State General Accounting Department (Ragioneria Generale dello Stato - RGS) “is responsible for the
consistency and reliability of national accounts and for the assessment and the analysis of public expenditure trends”, this include the provision of detailed figures on public employees.
12
maintain a fixed geographical repartition across different census (and then years), since some provinces and municipalities were newly created/or dissolved during these decades. So we use the most recent repartition following the creation of new provinces in 2009, for two main reasons: a)it offers the largest amount of local areas at the NUTS-3 level, namely 110 provinces; b) data from 2011 Censuses are only available with this repartition5. This last census data were directly downloaded by the online data warehouse from ISTAT. The variable obtained is the number of employees working in all the local units of public institutions in a given province, this means that public employees could work in a province, but reside somewhere else. Indeed, in order to reduce the errors deriving from this commuting factor, we did not go beyond the provincial level of analysis, i.e. by using a municipal level of aggregation, some very small municipality appeared to have up to more than 1000 thousands public sector worker each thousands of inhabitants. On the other hand the commuting factor would not be either excluded using a regional level analysis, so that a provincial level seems a fair compromise given the increased width of the panel and at the same time a reasonable dispersion of the public employment per inhabitant. Furthermore commuting across provinces does not seem to us a particular concern in the case of South, since we think that the natural geographical isolation, the worse connections and a lower living cost dispersion ( e.g. big vs small city) discourage it.
Data on commuting are just available in the 2011 Census. Tab.1; more than 10% of employed people commute to other provinces within the same region in North-west, while just 4% do this in the South and Islands .
5
13
Tab.1- Commuters
Employed persons resident in private household moving to the usual place of work (absolute values)
Census year 2011
Location of place of work
different province of the same region of place of usual residence
provinces in other region abroad Territory Italy 1463557 279853 63938 North-west 616197 100593 54813 North-east 307683 73735 6297 Centre 264673 54356 2821 South 208098 49550 7 Islands 66907 1619 .. Source:Istat
Yet data on public sector provide data about employees operating in a given province, which is most important to us, since we want to estimate their impact on the local labor market, while their place of residence it has a marginal relevance for our purpose.
14
The Population Census is carried out directly on households and individuals on their place of residence, so that give us a more accurate picture on people having any sort of occupation.
Tab.2
Total Employment by different Censuses (2011)
Territory Industry and services Census
Population Census Difference(%)
Italy 19946950 23017840 0,13341347 North-West 6251064 6745925 0,07335703 North-Est 4643780 5073810 0,08475485 Centre 4188603 4729040 0,11428049 South 3316314 4396231 0,2456461 Islands 1547189 2072834 0,2535876 Source: Istat
15
in the economy was a major key factor . In particular the employment in industrial sector resulting from the massive intervention to industrialize the South (“Intervento Straordinario”) implement in the 60-70´s, which favored an industrialization in the heavy industry , chemical and petrochemical sector, could seriously prejudice one of the major objective of our analysis, namely the estimation of the impact of public sector on the industrial-tradable sector. On the other hand this also means that even the data on public employment acquired by the Census of Industry and Services could not fully represent of the public sector employment (e.g. transports are not included) . Furthermore other factors determined at a national and/or political level, which affect local labor dynamics are not taken into account, namely trade unions, national contracts system (for big firms), public subsidies for “strategic” companies, companies owned partially by public institutions etc. This means that the dependent workers of public institutions are just a proxy to measure the impact of public sector and central institutions on the local economies. Nonetheless, with regards to our dependent variables, in order to address this sort of problems deriving by dealing with general employment, we use data on entrepreneurs and self-employed individual, and, rather measure the impact of public sector on the entrepreneurship and self-employability of individuals. The Census of Population provides data about the position of individuals in their jobs, allowing a macro-sectorial distinction by Agriculture, Industry and “Others”, namely the tertiary sector .
So we just consider workers who declared to be either entrepreneurs and professionals, autonomous or co-helpers (“coadiuvanti”, namely individuals working with an autonomous or entrepreneur familiar without any formal employment contract). According to ISTAT, individuals tend to declare themselves entrepreneurs even though they should be classified as autonomous workers, for these reason, we aggregated all these 3 sub-categories.
16
that some activities in the “Industry” ( sector from B to F in the Ateco6, Istat classification) are not strictly tradable, e.g. construction, reparations and maintenance of machinery etc.
From the same Census, we also took provincial data on total employment, total population, individuals with tertiary education. Furthermore we aggregated different classes of age to obtain the population between 15 and 64 years.
An exception to our data source, are the estimates of wage gaps. In order to go beyond a simple and maybe too discretionary distinction between South against other regions, we divide our provincial observations according the degree of wage gap between public and private sector workers in their respective regions. We used the average of two wage gaps, estimated using two GWR by Dell´Arringa et al (2007). It is important to mention, that in this estimation some small regions (e.g. Abruzzo and Molise) were aggregated, so that we have less than 20 different regional values (see Descriptive Statistics).
4.The Model
With the Census data we obtain a balanced panel of 110 province for 4 non-consecutive years, i.e. 1981, 1991, 2001 and 2011. These gaps over time are not per-se source of big concern, since we assume that impact of public sector employment on private employment, is quite slow in taking effect, and also persistent. In addition public employment is very static on its own. Similarly Algan et al. (2001) used 5 years averaged observations, even though they had available a yearly panel.
Yet by exploiting this panel dimension, we can control for fixed effect of provinces. Our basic model equation (1) was first estimated by a pooled OLS, then to control for unobserved heterogeneity we used first LSDV, and finally a Fixed effect within estimator that provides different results than a LSDV.
Eq. (1)
17
The dependent variable (indXth in this case) is the number of entrepreneurs, autonomous workers and co-helper in the industrial sector, adjusted per each thousands of inhabitants (from now we will broadly refer to entrepreneurs). To control for possible structural difference in the population age across provinces and overtime, we also make use of an alternative rate, namely per each thousands of inhabitants in the working age (between 15-64 years of age). We think that the first rate could better represent a weight by population more related to the demand side (at least for “tradable” sectors), e.g. the number of public employees might be determined in function of the services demanded by the entire population, in particular the share of non-active population could have even an higher weight (Schools, Hospitals) . On the other hand, the number of working people in any sector directly depends on the size of the population that is in the working age . For these reasons we use a different specification for each rate.
Furthermore we substitute our dependent variable and estimate the impact on the tertiary and agriculture sector, while pubXth ( the rate of public employees) it is stably in the RHS of the equation. Then we make use of South to account the effect of being situated in a Southern region, and more importantly we use the cross-effect between South and pubXth to account for a special impact of public employment in the South. Finally we include as control variables Part_rate ( the share of labor force participants on working age population) and eduXth (the rate of university graduated).
The first control variable is meant to proxy for the general economic conditions, and at the same a propensity to enter the labor market. For this second reason we preferred the participation rate to the unemployment rate (highly correlated among them), that would be more related to short-term busyness cycle dynamics, and would deshort-termine an ambiguous effect on the entrepreneurial attitude of individuals; e.g. Alesina et al (1999) found that unemployment can foster entrepreneurship in economically depressed areas.
EduXth controls for change in education level that could somehow affect directly the number of
18
To conclude in all the specification we control for year fixed effect, since they explain a great part of variability in our panel7.
The eq.2. shows our alternative specification to estimate the heterogeneous impact
of public sectors on self-employment (from now we mean by this the aggregate value of entrepreneurs, autonomous and co-helpers).
(Eq.2)
Using the public-private wage gap estimates by Dell’Arringa et al. (2007), we constructed a dummy variable for each of the quartile of the regional wage gap distribution; then we constructed the respective cross-effect (with pubXth). We left out the lowest quartile, so that we can compare how public employment have a different impact in medium, medium-high and high wage gap regions with respect to low wage gap regions. As explained in the literature, public wage premium has a crucial role in determining a crowd out or a positive effect of public employment on private employment. Even though our model lacks of a time variant and provincial level variable, it permits to make some distinction among southern regions, and similarly among central and northern ones. In particular, our four regional categories, are determined accordingly to the factor of major interest in our research, namely the degree of labor market distortion. In this optic, the private-public wage gap is seen as a sort of economic adjustment, so that we use its magnitude as proxy for the magnitude of the distortion. In parallel, the other outcome of this distortion would be an high-labor cost that depress entrepreneurship and then private employment.
7
19
5.Descriptive Statistics
Since we introduced our model and its variables , Table 3 provides a detailed description of each variable. Tab.4 provides the descriptive statistics of the time variant variables, and Tab.5 and 6 compare southern against non-southern regions.
Table 3 – Description of the variables Variable Description
agrXth: number of entrepreneurs, autonomous and co-helper per thousand of inhabitants, operating in agricultural sector (sector A, Ateco)
agrX1564th number of entrepreneurs, autonomous and co-helper per thousand of inhabitants between 15 and 64 years , operating in agricultural sector (sector A, Ateco)
indXth number of entrepreneurs, autonomous and co-helper per thousand of inhabitants, operating in industrial sector (sectors B-F, Ateco)
indX1564 number of entrepreneurs, autonomous and co-helper per thousand of inhabitants between 15 and 64 years , operating in industrial sector (sectors B-F, Ateco)
terXth number of entrepreneurs, autonomous and co-helper per thousand of inhabitants, operating in tertiary sector (sectors G-U, Ateco)
terX1564th number of entrepreneurs, autonomous and co-helper per thousand of inhabitants between 15 and 64 years , operating in tertiary sector (sectors G-U, Ateco)
pubXth number of employed of public institutions per thousand of inhabitants pubX1564th number of employed of public institutions per thousand of inhabitants
between 15 and 64 years
eduXth number of individuals holding an university degree per thousand of inhabitants
eduX1564th eduXth= number of individuals holding an university degree per thousand of inhabitants between 15 and 64 years
unemployment unemployed individuals (including first job seekers) as share of individuals part_rate labor market participants ( employed and unemployed individuals) as
share of inhabitants between 15 and 64 years
South dummy variable for southern regions (including Islands) SouthXpubXth Cross-effect between South and pubXth
SouthXpubX1564th Cross-effect between South and pubXth
high_wg dummy for regions in the highest quartile of public-private wage gap distribution
20
Table 4 – Descriptive statistics (all provinces)
Variable Obs Mean Std. Dev. Min Max
agrXth 440 16.73097 13.33155 1.218371 83.25984 indXth 440 42.08305 35.34665 5.664919 143.3825 terXth 440 54.92161 12.91968 25.16727 104.2293 pubXth 440 50.54465 9.399528 31.50485 86.07637 eduXth 440 53.11423 31.16832 7.926277 153.4492 agrX1564th 440 25.4178 20.53943 1.821879 132.8045 indX1564th 440 63.37318 53.12944 9.001953 219.5554 terX1564th 440 83.01405 19.23533 40.68399 148.6189 pubX1564th 440 76.55651 14.41487 46.99514 133.0757 eduX1564th 440 80.77663 48.50928 12.66433 233.437 unemployment 440 0.139181 0.0889233 0.023244 0.4269824 part_rate 440 0.6295731 0.0602053 0.4835932 0.7775457
Tab.5 – Descriptive statistics of not southern provinces
medium_wg dummy for regions in the 2° quartile of public-private wage gap distribution
high_wgXpubXth Cross-effect between high_wg and pubXth high_wgXpubX1564th Cross-effect between high_wg and pubX1564th mhigh_wgXpubXth Cross-effect between mhigh_wg and pubXth mhigh_wgXpubX1564th Cross-effect between mhigh_wg and pubX1564th medium_wgXpubXth Cross-effect between medium_wg and pubX1564th medium_wgXpubX1564th Cross-effect between medium_wg and pubX1564th
Variable Obs Mean Std. Dev. Min Max
21
Tab.6 - Descriptive statistics of southern provinces
Variable Obs Mean Std. Dev. Min Max
indXth 164 29.88 26.95 5.66 103.31 terXth 164 43.37 9.14 25.17 68.87 agrXth 164 17.17 14.14 2.66 83.26 pubXth 164 52.49 7.94 34.97 69.06 eduXth 164 49.82 29.46 7.93 126.23 part_rate 164 0.5755 0.0370 0.4836 0.6729 indX1564th 164 45.08 40.30 9.00 158.63 terX1564th 164 65.83 12.90 40.68 104.13 agrX1564th 164 26.37 22.24 4.11 132.80 pubX1564th 164 79.93 11.94 51.97 105.75 eduX1564th 164 75.37 44.11 12.66 194.69
By splitting our descriptive statistics for North and South (tab. 5-6), we can notice how people in southern provinces have a lower attitude towards entrepreneurship/self-employment in all the sectors except the agriculture one, in addition they are slightly less educated and have a very significantly lower participation rate. On the contrary public employment has a comparable size across the country, and public employees are even slightly more numerous in the South.
Table 7a –Correlations among variables (total population)
22
Table 7b –Correlations among variables (working age population)
indX1564t h terX1564t h agrX1564t h pubX1564 th eduX1564 th part_rat e unemployme nt indX1564th 1 terX1564th 0.3907 1 agrX1564th -0.1159 -0.1018 1 pubX1564th -0.0015 -0.0141 0.0366 1 eduX1564th 0.1638 0.3937 -0.4619 0.1359 1 part_rate 0.2148 0.688 -0.0487 -0.1362 0.4686 1 unemployme nt -0.3959 -0.7189 0.0073 0.1562 -0.2806 -0.7588 1
Tables 7a and b show how public employment does not appear to be strongly correlated with our dependent variables, and this is line with the fact that public sector is not strongly affected by the local economies characteristic. In contrast participation rate is strongly and positively correlated with the entrepreneurship in the industrial and tertiary sector, whereas , interestingly, it is negatively correlated with the number of public employees . Education is positively correlated with all the sectors, except agriculture, suggesting some room for structural shift to the other sectors. Finally we can see how participation rate and unemployment are strongly intertwined, since high unemployment might discourage labor market participation..
Tab.8-Wage Gaps by region
REG Region WG_1 WG_2 Mean_WG Quart_WG South
23 14 Molise 0.123 0.079 0.101 3 yes 15 Campania 0.161 0.104 0.1325 4 yes 16 Puglia 0.186 0.134 0.16 4 yes 17 Basilicata 0.186 0.134 0.16 4 yes 18 Calabria 0.221 0.148 0.1845 4 yes 19 Sicilia 0.183 0.118 0.1505 4 yes 20 Sardegna 0.081 0.051 0.066 3 yes
Table 8 shows the details of regional time invariant variables, namely the wage gap extracted from Dell’Arringa et al. (2007), the average calculated and its quartiles. Then correlation tables are shown. We can notice how southern regions are all either in the third or fourth quartiles of the estimated wage gaps distribution.
Last but not least the graphs below show the geographical distribution across provinces of entrepreneurs (including autonomous workers and co-helpers) by sectors, public employees and participation rate.
24
Graph 2 - Distribution of tertiary entrepreneurs across provinces
25
Graph 3- Distribution of agricultural entrepreneurs across provinces
26
Graph 4- Distribution of public employees across provinces
27
Graph 5- Participation rate across provinces
Graph 5 shows again, through participation rate, the net North-South economical divide.
6.The Hypothesis
28
H1a: Public sector employment has a negative effect on industrial (tradable) sector employment,
this is effect will be larger in Southern provinces.
H1b:Public employment has a sharper negative impact on industrial (tradable) sector in provinces where public-private wage gap is higher
Conversely public employees might increase the demand for non-tradable services, and then positively affect the tertiary sector. Yet the effect on entrepreneurship could be ambiguous and even negative where high labor market distortions could push up prices and labor cost. So our second hypothesis
H2a: Public employment has generally a positive impact on the tertiary (non-tradable) sector. In Southern regions this impact may be offset and turn to be negative.
H2b:Public employment has a negative effect on tertiary sector in provinces where the public-private wage gap is high.
Finally, we consider agriculture as a tradable sector, hence we do not expect any demand side positive effect from higher public employment. Yet it is a sector where informal labor and illicit workers play a big role, especially in the South. So we expect public employment to have no influence, since on the one hand informality offset labor market distortions, whereas time invariant territorial characteristic play the most relevant role in determining the size of this sector.
H3a: We expect public employment to have no significant impact on the agricultural sector, independently of the northern or southern geographical location.
H3b: We expect public employment to have no significant impact on the agricultural sector , independently of the public-private wage gap magnitude
7.Results
29
Table 9- Industrial sector estimation results
30
hypothesis 1a seems to be partially confirmed. Participation rate has in all the estimations a positive and significant effect, while public education turns to be weakly significant and negative only in the FE, yet this negative coefficient does not surprise us, as mentioned in our model specification. Finally it is worthwhile to notice that in both LSDV estimations, South has a positive and significant coefficient. This could appear surprising, but since we are already controlling for negative fixed effect of southern provinces, the positive value might be due to a lower chance of a dependent employment in South, that might stimulate self-employability.
31
Table 10 shows the results of Eq.1, when we plug the tertiary sector as dependent variable. First we notice that South has a systematical lower number of tertiary entrepreneurs (after controlling for participation rate), but again when we control for provincial heterogeneity in both LSDV estimations, we have a significant positive effect . Secondly impact of public employment is positive and significant just in the first LSDV estimation (rate per total population),yet we can assess it has generally no negative impact. In contrast the cross effect between public employment and South has always a negative and significant coefficient in all the 6 estimations. On the other hand, as expected, the impact seems to be milder than the one estimated with regard to the industrial sector. So we can argue that H2a seems to be only partially confirmed, since we did not find a convincing proof that public employment has a general positive effect on tertiary sector entrepreneurship.
32
Table 11- Agricultaral sector estimation results
33
Table 12 shows how public sector tends to have in general a negative effect in the pooled OLS and, more importantly, in the LSDV estimations. Furthermore the crossed effect is negative and significant for the high wage gap regions (namely in the fourth quartile ) in the 2 LSDV estimations, whereas medium-high wage gap cross effect are not significant, and medium wage gap are even positive and significant. Apparently the heterogeneous impact of public employment follows the magnitude of the labor market distortions. The FE specification still confirms an heterogeneous impact of public employment, since the cross effect of high_wg and mhigh_wg are both negative and significant8, but the first one has a stronger magnitude: namely in regions where the wage gap is in the fourth quartile, the public employment has a more strongly negative impact with respect to regions in the third quartile (ca 1.6 vs 0.65). Conversely in the FE, medium_wg are not significant, so that positive impact is not confirmed, and similarly the general impact of pubXth is not significant.
So in our preferred specifications (and partially in the LSDV) , i.e. the FE estimations, H1b appears to be confirmed.
Table 13 shows the results with regard to the tertiary sector. Firstly the positive general impact of is public employment (pubXth and pubX1564th) is confirmed (as in regression table 2) just in the OLS and LSDV specifications. Secondly the negative impact of the cross effects increase again together with the wage gap. Yet in our preferred specifications (FE ) the cross coefficient is negative and significant only for the regions with high wage gaps. Hence the tertiary sector seems not to suffer from public employment size, except in region in the highest category of market distortion, and still by a lower degree than industrial sector: H2b is confirmed.
Finally Table 14 shows the results for the agriculture sector. Public sector coefficients are generally not significant, apart from some unexpected weakly significant cross coefficient for medium-high and medium wage gap. Nonetheless we cannot either reject our hypothesis 3.
34
35
36
37
8.Robustness
In all our estimations we use robust standard errors for provincial clusters, addressing so potential heteroscedasticity typical of panel observation. By doing so we also address the problem of not-normal distributions of the variables used, also because we have a relative large sample. Furthermore we weight our variable according two different kind of populations and still our result do not change significantly.
Yet a source of concern could rise by noticing the unusually high R-squares of our estimated results. These are due to the fact that potentially all our explanatory variables could serve as proxy for the general population structure, capable to explain a big portion of variability. Yet the estimated coefficient about public employment shows a lot of variability according to sectorial and wage gap (or North-South) diversity, that cannot be straightforwardly addressed to any spurious correlation.
We also estimated Random Effect model, and in order to choose which model to prefer we run an Hausman test, but we discarded its results, since the covariance matrix was not positive definite. Alternatively we implemented the Mundlak approach to check whether fixed effect would make RE inconsistent. Regression Table 7 and 8 reports all the different RE estimations results and the extended RE that include group means of time variant variables. The group means of the time invariant variables are, in general, jointly significant9, consequently we preferred FE results to RE to avoid the potential endogeneity stemming from time invariant characteristic not accounted in the RE.
Yet even with this alternative estimation technique, coefficient are roughly in line with the ones predicted in our hypothesis. The cross effect South with public employment has a negative and significant impact on industrial and tertiary sector, whereas the general coefficient of pubXth is negative for the industrial sector and positive for tertiary one.
38
39
Table 16- Mundlak Test (working age population)
9. Discussion, Limitations and Conclusions
The aim of this thesis was to isolate the impact of public sector employment on private employment across Italy; a country where the huge geographical economic disparities are not reflected in public wage geographical dispersion.
40
behavior of local institutions. Then, following Alesina et al. (1999), we focused on the role of public sector employment on curbing entrepreneurial choices. We aimed at contributing to the literature by carrying out a panel analysis on public employment impact for the case of Italy, at a provincial level of disaggregation. The existing literature on public employment impact on labor market, did not seem to go beyond cross-country panel analysis, where public wage are set to country level, and not equally set like in the Italian case. Furthermore no distinction across economic sectors of activity were carried out, whereas the focus is given at the general impact on private employment and unemployment (see Algan et al. 2002) . Yet Faggio et al (2014) carried out a subnational analysis for the case of UK local labor markets, splitting tradable from non-tradable sector, nonetheless they did not implement a within estimation, and more importantly, they did not allow for heterogeneous impacts depending on public-private wage gaps. Conversely we use the wage gap estimates from Dell’ Arringa et al.(2007) , and we were able to estimate how public sector has a different impact in depressed areas, where labor market presents the highest level of distortion.
We analyzed census data for 4 different decades of 110 provinces, using data on entrepreneurships and self-employment for the three main economic sectors as a proxy for private employment, and data on all employees working for public institutions. After we saw how pooled OLS estimated were missing to account for provincial time invariant characteristic , we checked whether those effect could have been more efficiently accounted with a Random effects model, yet after running a Mundlak test, we preferred our FE estimates, to avoid the problem of endogenous time invariant characteristic. Despite that, even RE estimates were in line with our predicted coefficient. In addition we implemented two ways to weigh our variables by population, namely using the total population and only the one in the working age, nonetheless the results were really similar.
41
employment was found to be negative , but of a smaller magnitude with respect to the industrial sector. Furthermore we found that provinces situated in regions were the public-private wage is higher, tend to experience a sharper negative impact of public sector on industry and services. Last but not least, we found no evidence of similar patterns in the agricultural sector, where we think that informality and illicit workers play a major role in mitigating the distortions, whereas at the same time territorial characteristic shape the size of the this sector.
In our preferred specifications, we exploited the panel dimension to control for provincial fixed effect that might otherwise bias our results. By doing so we tried to control for social capital, attitude towards work, territorial characteristic and infrastructure, that we assume to be static overtime or, at least, would show a common growth rate (e.g. infrastructure) given the national context. Yet these assumptions are quite strict , and represent a limitation of this analysis. In order to improve our control variables set, we might have included time variant data on firms productivity, criminality rate, public administration efficiency, though not easy to obtain and/or compute.
Other major limitation is the lack of a proper counterfactual or control group, namely we do not know what would have happened if public employment wage was set according to local conditions. Nonetheless, by exploiting the difference within time in the rate of public employment we managed to capture an impact where labor market distortion are high. As said e provided a sort of counterfactual, with the case of the agricultural sector, where high informality rate may offset labor market distortion. Nonetheless informal labor is a limitation of our data, that we addressed in part by using household censuses en lieu of enterprises censuses. Yet using this sort of data, we extracted a general aggregated figure of entrepreneurs, self-employed and familiar co-helpers, without any reference to the size of the enterprises and the total private employment created. A far more accurate analysis could be done with the use of Labor Force Survey, that was not available to us. By using micro level data, we would have been able to have more detailed information about the profession, education and working location of the individuals, and then even able to account for any sort of employment, and not overlook dependent employment in private sector.
42
43
10. References
Alesina, Alberto, Stephan Danninger, and Massimo Rostagno. Redistribution through public
employment: the case of Italy. No. w7387. National Bureau of Economic Research, 1999.
Algan, Yann, Pierre Cahuc, and André Zylberberg. "Public employment and labour market performance." Economic Policy 17.34 (2002): 7-66.
Borjas, George J. The wage structure and the sorting of workers into the public sector. No. w9313. National Bureau of Economic Research, 2002.
Burdett, Ken. "Towards a theory of the labor market with a public sector."Labour
Economics 19.1 (2012): 68-75.
Dell'Aringa, Carlo, Claudio Lucifora, and Federica Origo. "Public Sector Pay And Regional Competitiveness. A First Look At Regional Public–Private Wage Differentials In Italy*." The
Manchester School 75.4 (2007): 445-478.
De Gregorio, Carlo, and Giordano, Annelisa. “The heterogeneity of irregular employment in Italy:some evidence from the Labour force survey integrated with administrative data” Istat Working Paper (2015)
Disney, Richard. "Public-private sector wage differentials around the world: methods and evidence." University of Nottingham (2007).
Faggio, Giulia, and Henry Overman. "The effect of public sector employment on local labour markets." Journal of urban economics 79 (2014): 91-107.
Giacinto, Valter Di, and Giorgio Nuzzo. "Explaining labour productivity differentials across Italian regions: the role of socio‐economic structure and factor endowments*." Papers in
Regional Science 85.2 (2006): 299-320
Pagani, Laura. "Why do people from southern Italy seek jobs in the public sector?." Labour 17 (2003): 63-91.
Pederson, P. J., et al. "Wage differentials between the public and private sectors." Journal of Public Economics 41.1 (1990): 125-145.
Santoro Patrizia “Deboli ma forti. Il pubblico impiego in Italia tra fedeltà politica e ammortizzatore sociale” (2014)
Scarlato, Margherita. "Le politiche di sviluppo: effetti sulla convergenza o divergenza dei territori." Economia Italiana/Review of economic conditions in Italy 3 (2011).
