Government spending and its components in Italy, 1862-2009: drivers and policy implications

Using a new historical dataset over the time period 1862-2009, this paper tests the validity of Wagner’s Law of public spending (WL) in Italy. To this aim, cointegration and Granger causation are used to investigate the long run relationship between GDP and government expenditure. Moreover, DOLS method is applied to estimate consistent long run elasticity between these two variables. In contrast to previous studies, we evaluate WL for both total government expenditure and some specific items of spending. Our main findings are that WL does not hold in the long run for total government expenditure. However, we find strong support for WL in the shorter time span from 1862 to the end of the 19 th century. Here WL is confirmed as regards both total government expenditure and all the specific items of spending we have considered. Conversely, in the post-Second World War years, WL holds only for capital expenditure, compensation of employees, justice and national security, welfare and redistribution by the state. Thus, it seems that Italy invested a great deal and for a long period in infrastructures, justice, national security, and welfare, and less in items such as education and culture that play a paramount role in the formation of human capital.


Introduction *
In the last century, government expenditure underwent constant growth in Western countries. Today in Western Europe the public sector absorbs around fifty percent of the national product, and in the United States around thirty percent. This growth in government's involvement in market economies is one of the most debated topics in public economics (Atkinson and Stiglitz 1980;Lindert 1994Lindert , 1996Lindert , 2004aLindert , 2004bNorth andWallis 1982, Tanzi andSchuknecht 2000).
A number of explanations have been provided to identify the drivers of growth in public sector. Among these, Wagner's Law of public spending (henceforth WL) has drawn widespread attention. WL suggests three main reasons for increased government involvement in the long run (Wagner 1883(Wagner , 1893. Firstly, the increasing complexity of the economy generates increased demand for government intervention. The urbanization and greater division of labour accompanying industrialization required more government regulation and higher expenditure on contractual enforcement and law and order. Another reason is the supposed superior income-elasticity of publicly provided goods and services (such as cultural and welfare expenditures). Lastly, the growing need to finance large-scale investments with public good characteristics (i.e., infrastructures) also boosted public spending.
In this perspective, the growth of government expenditure is considered to be a consequence of the expansion of the state produced by a country's social and economic progress. 1 Wagner was the first to attempt an empirical verification of his hypothesis. He did so for some European countries 2 that underwent industrialization in the late 19 th century. Subsequent studies tried to verify whether this "empirical uniformity" also existed in more recent periods, and whether it has been maintained in other contexts: for example, the developing countries in the 20 th century.
Up until the early 1990s, WL received strong support with few exceptions. Since then, however, according to a recent review of over forty studies (Durevall and Henrekson 2011), the evidence is more controversial: around 35 percent of these studies fail to find support for WL, while around 30 percent provide indirect support by controlling for other variables or focusing on specific expenditure items, and around 35 percent provide direct support.
However, there are very few studies that analyse very long time spans and all of them reject WL. Henrekson (1993) and Bohl (1996) find no support for WL in Sweden from 1861 to 1990 and in the UK from 1870 to 1995. Ghate and Zak (2002) do not find any empirical evidence in the USA from 1929 to 2000. Durevall and Henrekson (2011) find that in both the UK and Sweden WL does not hold in the long run (i.e., in the years 1800-2006 for Sweden and 1830-2006 for the UK); there are, however, two shorter time spans during which WL holds: approximately 1860-1913 and 1920-1975 for both countries. Kuckuck (2014) finds that WL can be rejected in five European countries in the period 1850-2010; nevertheless, he finds evidence that WL has a higher validity during early stages of development, but becomes weaker at an advanced stage of development. * We would like to thank Maria Ambrosanio, Giovanna Messina, Aline Pennisi, and Paolo Silvestri for their useful help in the interpretation of the data provided by the Ragioneria Generale dello Stato. We also thank Federico Biancardi for his research assistance. The usual disclaimer applies. 1 The major alternative approaches are the displacement effect (Peacock and Wiseman 1961) and the ratchet effect (Bird 1971) which claim that in a downturn public expenditure declines more slowly than percapita GDP so that public expenditure as a share of GDP rises, whereas the reverse occurs in upturns. 2 Prussia, Bavaria, England and Switzerland. Some explanations have been provided to account for the stronger support WL receives in early stages of development and especially in the 40-50 years prior to the First World War. Economic growth in the late 19 th century made it possible for several countries to introduce a system of income tax collection more cost-efficient than a system based on custom and excise taxes, which allowed an expansion of fiscal capacity that was not possible before (Levi 1988). Moreover, Meltzer and Richard (1983) and Lindert (1994Lindert ( , 2004a argue that government spending grows when the franchise is extended to include more voters below the median income level. This led to a distribution of income less equal than the distribution of votes, so that voters with the lowest incomes pushed the policy-maker to implement redistributional programmes in their own favour. Some hypotheses have been put forward also to explain the fading of WL in recent years, especially since the mid-1970s. Technological and institutional innovation has increased the scope for marketbased production of social services, i.e., private agents can use insurance and financial markets to handle many desired redistributions across individuals and over the life cycle (Durevall and Henrekson 2011). This change has been spurred by the resurgence of neoliberalism that induced governments to cut government spending and carry out massive privatization programmes. Furthermore, since the early 1990s, globalization has increased the competition between governments to cut taxes in order to create a business-friendly environment, thereby undermining their fiscal capacity and scope for government spending (Sinn 1998;Bergh and Karlson 2010).
This paper analyses WL in Italy from 1862 to 2009. It differs from existing studies on this country because it relies on a new dataset recently provided by Italy's State General Accounting Department containing data on both total government expenditure and some specific items of spending over a long time span of about one and a half century. This allows to address the evolution of the composition of government expenditure in response to the changing demands and needs of the society. 3 Our dataset also makes use of the most recent series of Italy's national accounts that have been recently produced by the Bank of Italy.
To this aim, we use cointegration analysis to investigate the long run relationship between government expenditure and GDP also taking into account the evolution of population structure. Moreover, we test two conditions implicit in WL: the unidirectional Granger causation from GDP to government spending and the superior income-elasticity of government spending. This paper shows that WL holds for total government expenditure only from 1862 to the end of the nineteenth century. Here, WL is confirmed for both total government expenditure and all the specific items of spending we have considered. This suggests that the national income, and therefore the development process, appears to have played a fundamental role in the creation of a more evolved and modern central state.
Evidence with regard to WL changed for the years after the Second World War, where WL is no longer valid for total government expenditure. Now WL only holds for expenditure on economic affairs and environmental protection; and expenditure on health and social protection in the late 1950s-early 1990s.
There is strong evidence in favour of WL for expenditure on education in the period 1862-1939, when reducing illiteracy was on top of the agenda of state's policy, but afterward investment in education seems to fade away.
Furthermore, there three items of spendingcapital expenditure, compensation of employees, and justice, public order and safetyfor which WL is verified for the whole period 1862-2009. Overall, our results suggest that investment in infrastructures, justice, national security, and welfare (to which WL applies for large part of the period under consideration) seems to have permanently played a central role in the action of the Italian state. However, this was not the case for investment in education and culture, that was substantial only until 1939.
The paper is structured as follows. Section 2 presents the sources and data we have used in our analysis. Section 3 illustrates the evolution of Italy's government spending from 1862 to 2009. Section 4 provides a theoretical framework for the empirical analysis, whose results are displayed in Section 5. The subsequent Section 6 discusses the results of the empirical analysis, and Section 7 concludes.

Sources and data
We use annual time series data of Italy's central government spending that comprise the years 1862-2009. Data on total government expenditure and on the specific economic and functional items of spending are at current prices and are drawn from the series that have recently been provided by Italy's State General Accounting Department (Ragioneria Generale dello Stato 2011). 4 Spending refers to the total payments disbursed in the year and have been obtained from the final budget of the state. 5 Different items of government spending have been selected, according to their economic and functional nature, in order to reveal any empirical evidence in favour of WL, as shown in Table 1: Items have calculated by aggregating various chapters of Italy's state budget according to the procedure described in the Appendixes A, B, and C.
Our data differ from those of the two most recent papers that test WL for Italy. In fact, Magazzino (2012) relies on the Informative Public Base (IBP), a database developed by the Bank of Italy that covers the shorter 1960-2008 time span and refers to the expenditure of the Italian public administration as a whole, i.e., it includes not just the expenditure of Italy's central government but also the expenditure of local governments (regional, provincial and municipal administrations). Our series differ also from that of Kuckuck (2014) who uses data drawn from Mitchell (2007) for the years 1850-1995 and from Eurostat for the years 1996-2010. These data are also provided by Italy's State General Accounting Department, but, differently from ours, they refer to the expenses accrued and not to the actual payments in the fiscal year.
For GDP at current prices and real per capita GDP (2010 prices) we rely on the new series of Italy's national accounts that have been recently provided by the Bank of Italy (Baffigi 2011). 6 These new 4 Our series stops in 2009 because since then the General Accounting Department has not updated the database of the specific economic and functional items of spending. 5 From 1884 to 1964 Italy's fiscal year ran from July 1st to June 30th. Data have been attributed to the solar years by adding half of the expenditures disbursed in two consecutive fiscal years and assuming an equidistribution of the expenditure over each fiscal year. 6 Thus, our GDP data differ from those of Kuckuck (2014), who for the years 1850-1995 relies on the GDP series published by Mitchell (2007). The latter are drawn from Istat's (1957) first series of Italy's national accounts, that were only partially improved by a team of scholars led by the economist Giorgio Fuà (1969). These series are heavily flawed (Cohen and Federico 2001) so that on the occasion of the 150 th anniversary of Italy's political unification, the Bank of Italy launched the project for a thorough reconstruction of the nation's national accounts, on which we rely. series were the basis of the recently published Oxford Handbook of the Italian Economy since Unification (Toniolo 2013).
Population data are from the Ricostruzione della popolazione residente e del bilancio demografico database (Istat 2012).

The evolution of Italy's government spending
In the first fifty years after Unification, Italy's total government spending in real terms increased slowly. On average, total spending growth did not exceed that of national income. However, as it can be seen in Figure 1, there were drastic swings from one year to the next, with peaks in the mid-1860s and in the latter half of the 1880s. Between 1862 and the mid-1890s, the share of total government spending to GDP and the growth in real per-capita GDP show very similar trends. Subsequently, the two series diverge considerably: the per-capita GDP begins to grow at much faster rates compared to previous years, 7 while total spending as a share of GDP drops constantly until the outbreak of the First World War.
On average, between 1862 and 1913 total government spending was around 10 percent of GDP and was mostly destined for defence and interest payments on public debt. In the first decade after Unification, the share of government spending for public works was around 10 percent of the total, and increased to 20 percent after 1870, staying on those levels until 1904, when it began a new surge to above 30 percent in 1909. The majority of spending for public works (around threequarters) was destined for railway construction. 8 Spending for education and culture also increased constantly from Unification to the First World War, both as a percentage of total state spending (from 1.6 to 5 percent) and as a share of GDP.
Participation in the First World War led to a drastic increase of total government spending, from just over 10 percent to more than 35 percent of GDP. In the years after the conflict, it dropped again and in 1926 it settled to the pre-war values compared to the GDP. Spending returned to its pre-war composition: defence costs dropped, whereas investments in public works and other economic interventions resumed. Investment in the railways was particularly significant, with the construction of the express lines and the electrification of the main lines, as well as the first toll motorways (Felice 2015).
Between the end of the Second World War and 1963, total government spending increased drastically in real terms. Despite this, it remained well below 25 percent of GDP that was growing at its fastest rates in the whole Italian history ( Figure 2). The state budget began to suffer in the second half of the 1960s. However, in 1970, despite a relevant deficit (3.7 percent) the total public spending (33 percent) was, compared to the GDP, below the average of the EEC countries (Ciocca 2007).
From the 1950s a substantial change occurred in the composition of government spending. After the defeat of the Second World War, Italy was prohibited from reconstructing its own independent military power. This led to a drastic reduction in defence spending, both compared to GDP and as a share of overall government spending (De Cecco 2003). On the other hand, the social components of spending increased: infrastructures, welfare, and redistributing action. Most social spending increases concerned pensions, for seniority, old age and invalidity (Franco 1993).
From the mid-1970s state spending began to grow more rapidly than the GDP, reaching a 44 percent peak in 1986 ( Figure 2). In 1983 Italy reached the EEC average in the ratio of government spending to GDP, and from that moment on systematically exceeded it (Ciocca 2007).
In 1978, the creation of a universalistic welfare system and the redistributing action of public spendingdriven by the social conflicts of the 1960s and 1970scontributed to a rapid reduction in the inequality of income distribution and poverty in Italy. From 1968 to 1982 the percentage of poor people fell from 20 to 3.6 percent, while in absolute terms their number fell from 11 to 2 million (Felice 2015).
From 1993, with a view to Italy's signing up to the single European currency, the imbalance of the national accounts began to be tackled, and clear results were seen in 1995, making decisive progress in 1997, when the deficit dropped to 2.7 percent of GDP, below the 3 percent target laid down in the Maastricht Treaty. However, after Italy joined the Euro, public spending began to increase again in absolute terms, with frequent swings compared to the GDP (Ciocca 2007).

Theoretical framework and empirical analysis
As originally formulated by Musgrave (1969) and recently proposed by Shelton (2007) and Durevall and Henrekson (2011), we employ the following specification to model WL: where g is the log of total government expenditure in nominal terms as a share of nominal GDP, and y is the log of real per capita GDP. Eq. (1) is a long run relationship between government spending and national income and β is a long run coefficient that according to WL should be higher than 0. In this case, government expenditure increases faster than GDP, i.e. government expenditure is income elastic, or in other words it is a superior good. 9 As stressed by Durevall and Henrekson (2011), Eq. (1) is probably the most common specification for modelling WL, and most models in the literature are reformulations of it. Contrary to the alternative specification that considers real total government expenditure and real GDP, it has the advantage, by using real per capita GDP, of better considering a nation's prosperity and spending capacity of its citizens. In fact, the latter is the crucial variable to model the evolution of the demand for public services. Moreover, the use in Eq. (1) of total government expenditure as a share of GDP instead of real total government expenditure considers the possibility that differences in productivity growth between the government and the private sector production lead to an increase in government spending due to "Baumol's disease". 10 To test for the existence of a long run relationship between g and y, we reformulate Eq. (1) as follows Eq. (2) is a simple econometric model to estimate the long run parameter β. But, if g and y are nonstationary variables, i.e. they contain a unit root, this invalidates conventional statistical inferences and hypothesis tests on β. Hence, we need to analyse the integration properties of government expenditure (and all its items) and national income performing ADF test (OLS/GLS) and KPSS tests. From these tests, we are unable to reject the null hypothesis of non-stationarity for the levels and stationarity for the first differences of all the series: they are I(1). The ADF and KPSS results are not presented here for space consideration. 11 Although I(1) series tend to wander over time without returning to a constant mean, economic forces should not allow these variables to wander apart in a permanent way. In this case, these series share a common path of growth and at least one of the two variables has to adjust to deviations from this long run path. In other words, if there is a long run relationship between g and y, movements in the former must respond to a disequilibrium in the latter, or vice versa, and the disequilibrium from the long run path is transitory. Cointegration offers one method to check whether WL holds. If there is a positive long run path between g and y, WL would require at least that the estimated residual t u in Eq. (2) be stationary, i.e. I(0). In this case, g and y share a common stochastic and deterministic trend, and this is the so-called deterministic cointegration.
We also check for a less stringent concept of long run relationship implicit in the definition of stochastic cointegration. If stochastically cointegrated, g and y share a common stochastic trend but different deterministic trends. It is useful when government expenditure and national income are not deterministically cointegrated (Eq. 2). We test for stochastic cointegration by using the following specification: If g and y are not cointegrated as stated by Eq. (2), we additionally test for cointegration by augmenting Eq. (2) by the Dependency ratio, i.e. controlling for the age structure of population as in Durevall and Henrekson (2011) 12 where the Dependency ratio is 64 15 . That is, the DR is defined as those aged 0-14 and 65+, divided by those aged 15-64. This measure should capture spending connected to policy reforms that should have substantially strengthened the link between government expenditure and the age structure of population, such as expenditures on child care, schooling, pensions, and health care. requirement to compete for employees with jobs that experienced gains and hence can naturally pay higher salaries (Baumol 1967). 11 The results of these tests are reported in a preliminary version of this paper. See Pistoresi, Rinaldi and Salsano (2015). 12 Including the Dependency ratio increases the probability to find cointegration among the series. Moreover it is possible to imagine the Dependency ratio to work like the "particular trend" in the stochastic cointegration framework.
For all these cases (Eq. 2, 3, 4), we test cointegration by Engle and Granger (1987) procedure performing ADF tests on the estimated residual t u with the asymptotic p-values for cointegration proposed by MacKinnon (1996).
Note that WL firstly requires that a long run relationship exists, i.e. that g and y are cointegrated and then that g Granger-causes y in the long run and not vice versa (unidirectional long run causality). To this aim, we estimate the Error Correction Models in Eq. (5) and Eq. (6). We test that the error correction parameter δ in Eq. (5) be statistically significant (one side t-test), while ~ in Eq. (6) be not suggesting that only g adjusts to the long run equilibrium. In fact, WL suggests that government expenditure reacts to a change in national income in the long run and this is driven, for example, by a change in the demand for public goods consequent on increasing prosperity.
Hence, WL requires cointegration between g and y, unidirectional causality from y to g, and also that β be greater than zero. The statistical inference on the long run parameter β may be problematic if levels representation of g and y contains a unit root and even if g and y are cointegrated. In fact, the OLS estimator of the coefficient in the cointegration regression is consistent, but the OLS estimator has a non-normal distribution and the inference based on its t-statistics is misleading. To avoid this problem, we use DOLS (Dynamic Ordinary Least Square) estimator suggested by Stock and Watson (1993). The DOLS is based on a modified version of Eq.
(2) that includes past, present and future values of changes in y: If g and y are cointegrated, DOLS is efficient in large samples and statistical inference about β based on HAC standard errors has a standard normal distribution in large samples. 13 Hence, thanks to this efficient estimator, on Eq. (5)

Results
In this section we present the result of our empirical analysis with regard to both total government expenditure and to the selected items of economic and functional categories of spending as defined in Table 1.  Tables 2 and 3 present the cointegration results according to the long run regression stated by Eqs. (2), (3) and (4). Only for cointegrated series, Tables 4 and 5 outline the findings about Granger causation from the estimated ECM in Eq. (5) and (6).
We stress that the Granger causation result is always unidirectional from y to g in all the cases considered, as suggested by WL. For this reason, in the remainder of this section, we will not return to comment the Granger causality results again. Finally, Tables 4 and 5 also present the inference on the long run parameter β estimated by DOLS.
Note that in all the Tables, findings are presented for the whole period 1862-2009 and for the two shorter sub-periods 1862-1939 and 1948-2009. There are two main reasons for this periodization: 14 1) The Second World War represents the most significant watershed in post-unification Italy.
For a certain period (from September 1943 to April 1945), the unified Italian state disappeared, with the partition of the country between the Kingdom of the South (allied to the Anglo-Americans) and the Italian Social Republic (allied to Nazi Germany). The conflict also led to the sharper institutional break-up in the country's recent history: the abolition of the monarchy and the establishment of the republic in 1946. For the purposes of this study, therefore, there is justification for breaking the analysis down into the periods prior to  and following ) the Second World War.
2) Disaggregation of the analysis into shorter sub-periods is often present in the literature on other countries, and several authors have suggested it is an effective method of investigation. When Durevall and Henrekson (2011) were unable to obtain significant results from their analysis of co-integration over the long period 1800-2006 with regard to the UK and Sweden, they examined the two sub-periods 1800-1913 and 1920-2006, and subsequently also the four even shorter sub-periods 1800-1859, 1860-1913, 1920-1973, and 1974-2006. An approach of this kind was also adopted by Kuckuck (2014), who evaluated WL for five countries (the UK, Denmark, Sweden, Finland, and Italy) between 1850 and 2010 in relation to their various stages of development approximated by three classes of per capita GDP. The analysis in sub-periods (Tables 2 and 4) suggests that WL holds for total government expenditure between 1862 and 1897, with 90 . 1  dols  , while as regards the post-Second World War period it holds in the years 1980-2009 with a negative β = -0.46. The latter result implies an elasticity of total government expenditure to GDP that is less than unity. In other words, total government spending during that period grew less than proportionally to national income. 14 Apart from a historical viewpoint or that of the quality of the data discussed above, the sub-periods selected are consistent with an analysis of the breaks. We find that in 1946 there was a structural break in the historical series of GDP that already justifies this periodization. Moreover, the various historical series of expenditure show other significant breaks at times that are not always uniform but consistent with our periodization. The breaks we find are consistent with those that are found for Italy in Kuckuck (2014). Our results are available on request. These results are in line with those of other studies on other countries for the years from the mid-19 th century to the outbreak of the First World War, which also find strong support for WL (although in our case, evidence in favour of WL ends in 1897). 16 Conversely, Italy seems to diverge from some major European nations in the post-Second World War years. In fact, Durevall and Henrekson (2011) find evidence in favour of WL during the "Golden Age"i.e., the 30 to 35 years after the Second World Warwhich was not the case in Italy.
The absence of a relation à la Wagner with regard to total government expenditure over the entire period 1862-2009 is due to the behaviour of current expenditure, which only co-integrates with GDP until the beginning of the 20 th century, and to that of the expenditure on interests, which only co-integrates with the national income until 1939. However, Tables 2 and 3 show that WL is verified over the entire period 1862-2009 with regard to other significant items of government spending: capital expenditure (β = 0.59) and compensation of employees (β = 0.32) among the economic categories; and justice, public order and safety among functional categories (β = 0.17).
Detailed examination of items of spending by functional category (Tables 3 and 5) shows that WL holds over the entire sub-period 1862-1939 for all of them, with the exception of expenditure on defence, for which the relationship is valid only until 1913. The reason for this result is that the explosion of military expenditure during the First World War and its rapid reduction in the years immediately after the conflict gave to this item a dynamics significantly different from that of real per capita GDP in the years 1914-1939.
Expenditure on economic affairs and on environmental protection and expenditure on health and on social protection do not co-integrate with national income over the entire period 1862-2009. However, they moved with GDP from 1862 to 1939, and subsequently from the early 1960s to the early 1990s. In particular, WL holds for the expenditure on health and social protection with

Discussion
The overall results of our econometric analysis show a strong support for WL in the years from Italy's political unification in 1861 to the end of the 19 th century. Here WL holds for both total government expenditure and all the specific items of spending we have selected, whether economic or functional. One gains the impression that the increasing incidence of government expenditure on GDP was a consequence of the efforts made by Italian governments to construct the framework of the new state, endowing it with indispensable infrastructures both organisational (a functioning central and peripheral administration) and physical (roads, ports, a rail network and a national telegraph service) (Cohen and Federico 2001). A similar purpose was served by the power strategy precociously adopted by the Italian state in order to ensure its survival in the European geopolitics of the time (De Cecco 2003).
Also expenditure on education and culture at that time seems to have been part of a nation-building strategy. In fact, one of the greatest issues the Italian ruling class had to address when Italy became a nation-state was the creation of a unitary conscience among its citizens. In this respect, reduction of illiteracy was of a paramount importance and this task was largely assigned to government schools. The Coppino Law of 1877 17 was a turning-point that ensured that in the future closer attention and greater resources would be devoted to the educational and institutional problems of the younger generation. Growing expenditure on health and social protection was also instrumental to the construction of the infrastructure of the new state.
Thus, the case of Italy in the years from political unification to the outbreak of the First World War seems to be principally referable to the theory of independent action of the state, defined on the basis of the internal and international political objectives conceived by state policy-makers, and not as a response to a growing demand by its citizens for higher-category goods, as Wagner theorised (Mann 1984). With regard to this period, one finds unconvincing the explanations offered by the literature to account for the increase in government spending in the principal European countries in correspondence with the early stage of industrialisationbetween mid-19 th and early-20 th centuryand especially the extension of electoral suffrage to low-income voters. The latter induced median voters to prefer an increase in taxes, which would principally affect the segment of voters with incomes higher than theirs, in order to yield the resources to be used to finance new social programmes with strong redistributive content (Meltzer and Richard 1981;Persson and Tabellini 1990;Lindert 1994). However, suffrage in Italy during this period was extremely limited: 2 percent of the population in 1861, that was increased to 7 percent in 1880 and to 23 percent only in 1913, at which time the evidence in favour of WL had already weakened (Ballini 1988).
One interpretation that might be considered for this period is the one offered by Douglass North to explain the growth of government expenditure in the USA over the same period. He associates the demand for greater government expenditure with technological changes: "Technological changes have led to an enormous increase in specialisation and division of labour, and therefore a radical change in relative prices which fundamentally altered the traditional structure of the polity, the family, and economic organisation. The variety of interest groups that emerged from this expanded division of labour led to political pluralism. The demand for new institutional forms of organisation to replace functions previously undertaken by the family and traditional economic organisation could not be completely realized by voluntary organisations because of moral hazard, adverse selection, and the demand for public goods." (North 1985: 392).
The years 1914-1939 constitute a particularly crisis-ridden period with the First World War, the post-war downturn, and the Great Depression. Thus, it is not surprising that in these years we find not support for WL for total government expenditure, even if, as we have seen, it holds for some specific items of spending: capital expenditure; economic affairs and environmental protection; health and on social protection; justice and national security; education and culture. This was principally consequent on the effort in nation-building and social control by the Fascist regime.
Evidence with regard to WL changed in the years after the Second World War. In this period, WL is no longer valid for the entirety of government expenditure. However, WL holds for some specific items of spending. Amongst them, capital expenditure stands out. In fact, capital expenditure meant that the state intervened in the economy far more decisively than it had done with transfers. The latter, in that they consisted of the redistribution of monetary resources, did not affect the relative magnitudes of the public and private sectors in terms of production. The intervention of the Italian state began after the First World War and culminated in the 1933 with the creation of the big stateowned holding IRI (Istituto per la Ricostruzione Industriale) that took over Italy's three largest universal banks and their industrial securities. 18 State intervention further expanded after the Second World War with a massive provision of subsidized credit and grants to enterprises. In face of the weakness of the private sector, the state-owned enterprises (the energy giant ENI was added to IRI in 1953, and EFIM followed in 1961 in the mechanical engineering industry) played a crucial role in the development of capital-intensive sectorssteel, heavy industry, energy, motorways, and telephonyin which the investments required were too high and profitability too long deferred to be within the capacities of private investors. They therefore acted as a "substitution factor" à la Gerschenkron for the missing prerequisites for the country's industrialisation (Gerschenkron 1962;Amatori 2000;Toninelli 2004). As a result, the influence of state-owned enterprises in Italy grew constantly and only diminished with the privatisations of the 1990s. 19 In the post-Second World War years WL holds also for compensation on employees, and justice and national security. The former is someway a necessary consequence of the expansion of state administration. The latter is a fundamental service required of the state, whose expansion is driven by the growing demand for regulation and the enforceability of contracts that arises as a society becomes more complex and the division of labour increases (Kuznets 1967;North and Wallis 1982). Moreover, Italy's pro-American stance during the Cold War and the possibility that the Italian Communist Party (the largest Communist party in the Western world) might organise a revolution meant that a large proportion of government expenditure was allocated to national security. In the 1970s, this expenditure was further increased to tackle political terrorism.
As we have seen, from the late 1950s to the late 1980s there is evidence in support of WL also for the expenditure on economic affairs and on environmental protection and, from the early 1960s to the early 1990s, for the expenditure on health and on social protection. This seems to reflect the growth of the economic and social components of expenditure: infrastructure, welfare, and redistribution by the state. Under pressure from the expansion of suffrage (universal suffrage was 18 At the time when IRI was created, the companies comprising it possessed 83% of the share capital of joint-stock companies operating in Italy in the telephone sector, 56% in shipping, 39% in the banking sector, 38% in steel works, 29% in the electricity sector, and 21% in the machinery industry (Cianci 1977, pp. 278-79). 19 The percentage of the total of share capital of Italian joint stock companies held by IRI, ENI and EFIM rose from 12. 5% in 1936to 22.7% in 1952, to 32.1% in 1972, and to 40.2% in 1983(Toninelli and Vasta 2010 Table 3.4). introduced in 1946), and, from the end of the 1960s, from an unprecedented wave of social struggles, a progressive expansion of welfare services to new social categories took place in Italy until a universalistic welfare system was introduced in 1978 (Franco 1993).
WL, however, no longer holds in recent years. Between 1980 and 2009, total government expenditure actually grew less than proportionately to GDP. From the beginning of the 1990sthe period during which budget deficits were reduced in anticipation of adhesion to the single European currencythere is no support for WL even for expenditure on economic affairs and on environmental protection and for expenditure on health and on social protection. Nonetheless, WL returns to be verified for spending in education and culture, even though evidence is much weaker than in the period prior to 1939.

Conclusions and policy implications
This paper has found a strong evidence in favour of WL in Italy in the years immediately after national unification, between 1862 and the end of the 19 th century. Here WL is confirmed for both total government expenditure and all the specific items of spending we have considered. The national income, and therefore the development process, appear to have played a fundamental role in the creation of a more evolved and modern central state.
Our impression is that, in the post-unitary period, the increasing incidence of state expenditure on GDP was a consequence of the efforts made by Italian governments to construct the framework of the new state, with the creation of indispensable organisational and physical infrastructures, the reduction of illiteracy and the formation of a national conscience among its citizens. A similar purpose was served by the power strategy precociously adopted by the Italian state in order to ensure its survival in the European geopolitics of the time. Thus, maybe paradoxically, evidence for WL seems to have been principally the consequence of an independent action of the Italian state, defined by the political objectives autonomously conceived by state policy-makers, and not of a response to a growing demand by its citizens for higher-category goods, as Wagner hypothesized.
Evidence for more recent years is different. In the post-Second World War years we find no support for WL for total government expenditure. Now WL holds only for some specific items of spending: capital expenditure, compensation of employees, and justice and national security. The former seems to stress the persisting strategic role of public investment in an economy, such as Italy, in which private big business is weaker than in the other major industrialized nations. Compensation of employees is a necessary consequence of the expansion of state administration. Justice and national security are fundamental services whose expansion is driven by the growing demand for regulation and the enforceability of contracts that arises as a society becomes more complex and the division of labour increases. Moreover, Italy's pro-American stance during the Cold War and the fear that the Communist Party might organise a revolution pushed the government to allocate a large proportion of government expenditure to national security. From the late 1950s to the early 1990s WL holds also for major economic and social components of expenditure: infrastructure, welfare and redistribution by the state, and only this latter evidence seems in line with Wagner's argument that public spending was boosted by the supposed superior income-elasticity of publicly provided goods and services.
Overall, investment in infrastructures, justice, national security, and welfare (to which WL applies for large part of the period under consideration) seems to have permanently played a central role in the action of the Italian state. However, this was not the case for investment in education and culture, that was substantial only until 1939. Although this may have served to strengthen the Italian state in the face of potential external and internal threats and to respond to the growing demand on the part of society for public services, it might not be true that it also was the most effective way to prompt economic growth. In fact, it is controversial whether welfare has an expansionary effect on the economy, while investment in education and human capital is one of the fundamental drivers of long term economic growth.
Certainly, as highlighted by Magazzino (2012), the relationship between economic development and long-term public spending is influenced by a number of variables, which are not considered in the Wagnerian perspective. For example, factors such as political decision-making mechanisms, the behaviour of stakeholder groups working within the economic system and the main features of Italy's economic developmenta late-comer nation which has created an industrial structure in many respects unique amongst the most industrialized economies, with a dominance of small firms and a nearly total absence of large firmsmay be considered elements explaining the evolution and the dimension of the Italian public sector.
In the light of our analysis, some policy implications could be suggested. Italy's public expenditure seems unbalanced. In particular, social spending (health and pensions) grew massively on the wake of the social struggles and the electorate's demand in the 1960s and 1970s and is very high compared to spending for education and scientific research. As a result, despite Italy's human capital has grown dramatically over the past thirty years 20 , in terms of both secondary and tertiary education the country has not managed to close the long-standing gap separating it from the other OECD economies (Sestito 2014).
The Italian state should find the capacity to set both objectives and strategies to prompt the longterm term growth of the economy (Mazzucato 2013). The creation of a national innovation system and massive investments in schools and university to improve the quality of human capital should play a key role in it. These must be complemented by a review of incentives for learning and teaching, the appreciation and reward of merit, a better system of continuous assessment, the adaptation of curricula, more attractive school settings (Visco 2014).
From a quantitative point of view, public investments in research and development must be increased, as these represent a fundamentally decisive lever for economic growth through technical progress, which generates and is incorporated in capital assets used in production processes. The component that has not yet been incorporated in these assetsinnovationreflects on the overall productivity of production factors. This can be said generally for all countries, but even more so for Italy, where the presence of small and medium enterprises often leads to an avoidance of investments in Research and Development, and more often than not their innovations are merely incremental, generating new assets for the companies but not for the market; ultimately this weakens the effect on the potential for growth (Visco 2014).

Economic actions and interventions
Economic affairs Section XVI -Economic works and interventions not attributable to specific areas etc.

Economic interventions
Section XIV -Trade and industry Environmental protection Section XIII -Agriculture and food Section XV -Interventions in depressed areas Section XVII -Regional and local finance Section XII -Transport, navigation and communications Transport and communications

Social actions and interventions
Social protection Section IX -Work and social security

Social interventions
Section X -Public assistance Health Section XI -Health and hygiene 13 Education and culture Recreational, cultural and religious activities Section VI -Education and culture Education and culture (excluding religion) Education Section VII -Universities and scientific research Source: RGS (2011). From 1862 to 1967, the reconstructions by functional category are based on the RGS volume (1969). From 1968, the data was taken from the General Financial Statement of the Italian State and from 2003, from the general "statements" of the RGS data warehouse.
Public spending for economic interventions 11 Public spending for social interventions 12 Public spending for education and culture 13