The misalignment of fiscal multipliers in Italian regions

ABSTRACT This paper estimates fiscal multipliers resulting from shocks to current public expenditure, total public revenues, and public investment in Italian regions by accounting for the structural heterogeneity between Northern and Southern economies. The estimation is carried out by estimating a Bayesian panel vector autoregression, where the structural shocks are identified using sign restrictions suggested by economic theory. The results shed new light on regional fiscal shocks’ magnitude and propagation. Moreover, a misalignment of fiscal multipliers is revealed, possibly policy relevant to local and central authorities.


INTRODUCTION
The Italian economy is characterized by a historical misalignment of the development process between Northern and Southern areas.Currently, the South of Italy has one of the highest levels of material deprivation in Europe, whereas the Centre-North has one of the lowest (Organisation for Economic Co-operation and Development (OECD), 2019, p. 138).It is not by chance that the word 'South' reappeared in the denomination of the Minister for the Territorial Cohesion of Italy by 2016 and that approximately 80% of both European Union (EU) and national resources for economic and social rebalancing actions are devoted to Southern regions. 1 At the same time, the flow of resources towards the South has always fostered Northern regions to claim fiscal autonomy, which was part of the political debate regarding the process of administrative decentralization over the last two decades.The latter implied that Italian regions now compete with the state in providing several public servicesfor instance, health, education, transport and energyand decide autonomously how to allocate expenditure that may be financed through their taxes.
In this context, it is of primary interest to understand how fiscal policy transmits to local economies and whether it is a source (engine) for economic divergence (convergence).This paper gauges the effects of fiscal policy in Italian regions by studying its cumulative impact on gross domestic product (GDP)the fiscal multipliers, local employment and price dynamics.The analysis builds upon a Bayesian panel vector autoregression (PVAR); fiscal multipliers and impulse response functions (IRFs) at regional and macro-area levels are obtained.Fiscal shocks are identified using sign and zero restrictions as in Pappa (2009) and Canova and Pappa (2007), which follow the standard Keynesian short-run implications of public intervention.The strategy separates current expenditure from public investment manoeuvres other than revenues shocks.No other papers implement this approach in the literature on regional economies.
The results in the paper are manifold.First, fiscal measures produce significant multipliers > 1 and persistently affect the regional output level in the years after the implementation.Second, on average public investment has the most potent dynamic multipliers, followed by government consumption and taxes.Third, fiscal multipliers are generally higher in the Centre-North than in the South, although those of regions within the same area present significant size differences.Those fiscal multipliers are robust to different computational methods and the fact that private agents may anticipate budgetary decisions.Remarkably, the positive effect of public investment on Southern GDP is also robust to a more loosely theory-driven identification method.Fourth, the fiscal measures' impact on macro-area price level is non-significant, while the effect on local employment is more diverse.In particular, shocks to current expenditure do not influence employment; tax shocks significantly do it in both macro-areas and public investment only in the South.
This paper enriches the existing literature by adding methodological insights to the PVAR literature in studying different macro-areas of the same country.It also extends the evidence on regional multipliers and the effects of fiscal policy in Italy; the resulting misalignment of policy transmission feeds the debate about the intervention of the Italian government in addressing the territorial divide.
The remainder of the paper is structured as follows.Section 2 reviews the literature on fiscal multipliers with a focus on Italy.Section 3 presents the Italian regional data with a focus on the territorial divide and regional fiscal policy.Section 4 shows empirical strategy, which consists of the PVAR model and the baseline identification scheme.Section 5 discusses IRF and fiscal multipliers of regions and macro-areas.Section 6 performs several robustness checks.Section 7 compares the results with other works in the literature.Section 8 has final remarks.

REVIEW ON FISCAL MULTIPLIERS
In the macroeconomic tradition, the measures of fiscal multipliers arise mostly from structural (micro-founded) dynamic stochastic general equilibrium models (DSGE), structural vector autoregressions (SVARs) and other structural equation methods. 2 With the former, fiscal multipliers depend on the functional forms and parameters, whose set is not fully estimable (e.g., preferences, real and nominal frictions).Those theories share the view that fiscal expansions cause output and employment increases.While the real business cycle (RBC) theory supports that the latter works through wealth effects (Baxter & King, 1993), the New Keynesian (NK) theoryby allowing for model rigiditiesclaims that those are due to short-run demand-side effects.Therefore, NK-DSGE models imply fiscal multipliers that are higher than RBC models.Moreover, under certain hypotheses on household preferences, they are often > 1 (as in, for instance, Galí et al., 2007;or Hall, 2009, to name a few).
Though SVARs soften the theoretical contamination of the multiplier, the latter depends on the identifying assumption of fiscal shocks.Those follow four main strategies: the Cholesky scheme (Fatás & Mihov, 2001), Blanchard-Perotti's (BP) method (Blanchard & Perotti, 2002), the proxy (narrative) approach (Mertens & Ravn, 2013;Ramey, 2011b;Stock & Watson, 2012), and the Mountford and Uhlig's ( 2009) method (MU) based on sign restrictions. 3 The first method imposes a causal-recursive structure to economic innovations such that zero restrictions rule out the contemporaneous automatic feedback from the business cycle to public expenditure.The BP method combines such a recursive scheme with a restriction on the contemporaneous response of taxes to output by imposing an external-estimated tax elasticity.The proxy approach relies on a partial measure of the structural shock of interest that is orthogonal to other economic innovations.MU discern sign-restricted fiscal shocks from business cycle and monetary policy shocks and use a penalty function to select structural models.While MU finds the impact tax multiplier larger than the one of spending (i.e., zero), both Cholesky, BP and proxy methods roughly point to a spending multiplier close to 1 and higher than a non-zero tax multiplier (Gechert, 2015).However, Caldara and Kamps (2017) show that none of those schemes is wholly neutral since fiscal multipliers depend on the assumed systematic component of government policy.
Another approach based on sign restriction is that in Canova and Pappa (2007) and Pappa (2009), who characterize the fiscal shock as a contemporaneous increase (decrease) of output and public deficit due to spending (revenues) shock.Their restrictions derive from typical theoretical prescriptions of both the RBC and NK framework.Chian Koh (2017) uses those restrictions in a panel VAR of advanced and developing economies to estimate fiscal multipliers.The present paper employs the same approach.Section 4.2 illustrates the scheme in detail, while section 6 proposes the MU approach as an alternative.
The empirical literature also shows that fiscal multipliers are context dependent; they change according to the business cycle phases (Auerbach & Gorodnichenko, 2012), the presence of hysteresis (DeLong et al., 2012), the specific fiscal/monetary policy (zero-lower-bound) nexus (Blanchard, 2019;Ramey & Zubairy, 2018), and private debt levels (Bernardini & Peersman, 2018).Concerning the sectional dimension, fiscal multipliers are heterogeneous across countries according to their level of development, institutions, exchange rate, openness to trade and public indebtedness (Ilzetzki et al., 2013).Moreover, Ramey (2011b) and Leeper et al. (2013) argue that not accounting for anticipation of agents to fiscal shocks may distort fiscal multipliers.Section 6 shows a robustness check on this issue.
Those characteristics imply a vast array of multipliers that differ by sign and magnitude. 4The latter also depends on the class of public expenditure to which the fiscal shock belongs; the literature focuses on the difference between public consumption, investment, and revenues.In a metaregression analysis, Gechert (2015) show that public spending multipliers are close to 1; the latter exceed those of tax cuts by 0.3-0.4units, while public investment exceeds that of public spending by 0.3-0.8units.While the relation between public consumption and tax multipliers finds a vast consensus in the literature (Caldara & Kamps, 2017), that between public consumption and investment multipliers is debated.On the one hand, Auerbach and Gorodnichenko (2012) find a peak multiplier of public consumption about half that of public investment, which equals 2; Deleidi et al. (2021a) identify a public investment shock in euro-area countries and find a multiplier close to 3.5 at the peak.Ellahie and Ricco (2017) find a similar result in a large-scale Bayesian VAR.On the other hand, many paperssuch as Perotti (2004), Ilzetzki et al. (2013) and Boehm (2020) show theoretically and/or find evidence that public investment is not much effective than public consumption in boosting economic growth.

Fiscal multipliers in Italy
Giordano et al. ( 2007) provide a SVAR-based measure of Italian multipliers relying on the BP scheme; their shock to government purchases has cumulative multipliers of 2.4, 3.0 and 1.7 at the fourth, eighth and 12th quarters, respectively.Cimadomo and D'Agostino (2016) find a total government spending well above 1 in the years of the Great Recession until 2013 through a time-varying coefficients' VAR.
The structural model of the Italian Ministry of Economy and Finance (MEF) estimates public consumption and investment multipliers of about 1.3 and 1.2 at the peak (MEF, 2017).Locarno et al. (2014) estimate Italian multipliers by a DSGE model and account for the high public debt level; the public consumption multiplier is 0.8 in the first year, while the tax-cut multiplier is 0.4 in the first year and > 1 in the long run.When accounting for the zero-lower bound, the spending multiplier increases to 1.4.De Nardis and Pappalardo (2018) using structural equations evaluate the one-year multiplier to 1.3 for intermediate consumption, 2.6 for public investment and 0.6 for labour tax.Deleidi (2022) find a similar relationship for multipliers of government consumption, investment and net tax, though, in the fifth quarter they are 2.8, 0.9 and −0.2, respectively.
While estimates of national fiscal multipliers for Italy follow mostly macro-methods, the literature on local multipliers also builds on micro-econometric approaches.In the input-output (IO) fashion, Benvenuti and Marangoni (1999) find that the Keynes-Leontief regional multiplier of investment ranges from 1.4 to 1.7 and is higher for Southern regions.Faggian and Biagi (2003) find that the Keynesian regional multiplier ranges from 0.9 to 2.0, where the latter includes both current and capital expenditure.Following the instrumental variables (IV) approach, Marrocu and Paci (2010) estimate a positive macro-area output elasticity by investment components, while Acconcia et al. (2014) estimate infrastructure multipliers > 1.5 at the provincial level.
Based on a structural equations' model, Piacentini et al. (2016) estimate macro-area multipliers due to fiscal tightening for the period 2011-13.They find Centre-North versus South cumulative multipliers equal to 0.3-0.7 for consumption spending, 1.5-1.9 for public investment, 0.1-0.4 for income tax and 0.4-0.4 for consumption tax.Finally, Destefanis et al. (2022) and Deleidi et al. (2021b) follow a PVAR approach to estimate fiscal multipliers using Italian regional data.While the latter assesses macro-area multipliers, Destefanis et al. (2022) obtain multipliers for single Italian regions by adopting the same model as the one here.Section 7 compares the results with those studies.

DATA
This paper employs annual regional data (NUTS-2) from 1995 to 2019 sourced by the Italian Institute of Statistics (ISTAT).Fiscal variables regard central and local administrations based on the ISTAT imputation of national data at a regional level.They are net interest rate expenses due to conceptual problems interpreting the territorial distribution of the benefits deriving from this type of expense.
The resources from local governments result from the regional financial statements, while central government resources are imputed according to the number of recipients in the region (Banca d'Italia, 2009).Notice also that the data do not differentiate the level of the administrative source, such as the one of the Agency for Territorial Cohesion. 5 The territorial divide of Italy into two macro-areas (Centre-North and South) is quite visible looking at the regional growth. 6The lines in Figure 1 represent the log-difference between the real per capita GDP of regions and the one of Italy relative to the same measure in 1995.For each region, Figure 1 tells how much its GDP distance from the national GDP has changed from 1995: being in the negative territory means that the regional-wide GDP is more distant from the national one.The blue solid lines represent the Southern regions, red dashed lines represent the Centre-Northern regions, and bold solid and dashed lines are areas and national medians, respectively.The macro-area divide started to increase from 2001: the gap steadily skirts the zero line for the median-Centre-Northern region, while it decreases for the median-Southern region from 1999 to the end of the sample period.In 2019, the GDP distance of southerners from the national GDP was 10 times higher than those in 1995.Though several Northern regions currently take up the negative part (Aosta Valley, Friuli Venezia Giulia, Liguria, Piedmont and Umbria), only one Southern region takes up the positive part (Basilicata).Overall, the Italian gap in the recent era started in 2001 and climbed during the years of the Great Recession.
The territorial divide is also visible in the fiscal policy; Centre-Northern regions, which also have a higher capacity to grasp taxes (MEF, 2020), can outperform Southern in terms of chance to boost local growth.On the other hand, austerity measures characterized the fiscal policy of the last decade.Figure 2 scatters the regional deficit-to-GDP (total government expenditure minus total revenues) and regional GDP growth in the two Italian macro-areas.The circles and the crosses distinguish the years before and after the Global Financial Crisis, with their respective dotted and solid fitting lines.These lines trace a descriptive relationship between regional economic growth and fiscal policy; regional deficits are higher when the economy slows down and lower when it grows, so those lines are downward sloping.However, the slope flattens in both the macro-areas in the after-crisis period, reflecting the implementation of the austerity measures through a progressive reduction of public expenditure at both national and regional levels (Deleidi, 2022;Deleidi et al., 2021b).The post-crisis flattening is more evident in the Centre-Northern than in the Southern area, where a flatter curve was already present before; this is because while the public capital expenditure reduced in the Centre-North since 2010, it started to decline from 2004 in the South (Banca d'Italia, 2018).
Those differences may imply a misalignment of the policy transmission between and within macro-areas.The analysis here addresses this issue by focusing on the The misalignment of fiscal multipliers in Italian regions 2075 following regional variables: GDP (Gdp),consumer price index (cpi), total employment on population (ula), current public expenditure (Gcurr) -final consumption of goods and services, total revenues (the sum of direct, indirect taxes and social contributions) net of transfers to households and firms (T ),public investment (Inp), and regional deficit (def = Gcurr + Inp − T ).Nominal variables are in real terms by using the GDP deflator.All the variables are expressed in logs, apart from ula and def .The analysis also employs the following variables at a national level: the deficit-to-GDP ratio (defita), the yield of the 10-year Italian government bond (int) and the public debt-to-GDP ratio (debt).Section 6 also uses the national forecasts of Italian government expenditure (G F ) to proxy fiscal expectations.
Appendix A in the supplemental data online summarizes and describes the data in detail.

Bayesian PVAR
The analysis borrows the model from Jarociński (2010) and estimates two PVAR for Italian macro-areas: a Centre-North model (12 regions) and a South model (eight ones). 7Since regional time-series data are only Figure 2. Regional deficit-to-gross domestic product (GDP) and regional GDP growth pre-and post-Global Financial Crisis.Note: Horizontal lines are macro-area deficit-to-GDP averages, whereas vertical lines separate recessions from expansionary periods.
available from 1995 at an annual frequency, the analysis exploits the cross-sectional dimension to deliver more robust estimates.The model assumes the data come from the same structural process as if regions belong to the same macro-area; that is, it accounts for cross-sectional heterogeneity (Canova & Ciccarelli, 2013, pp. 205-246).
That implies to assume that each model coefficient, say b r , equals b + v r , where subscript r refers to a region r, b is a common (macro-area-specific) mean, and v r N (0, r ).The Bayesian procedure treats b r as an exchangeable prior that quantifies the degree of heterogeneity in the dynamics.At the same time, this prior improves the quality of the estimate and allows to obtain region-specific IRFs.The model is a N-dimensional VAR of r = 1, . . ., R regions, l = 1, . . ., L lags and t = 1, . . ., T time periods, which takes the following form: where the vector u rt contains i.i.d.disturbances, which are N (0, S r ).y r,t is a vector of N ¼ 7 endogenous variables (Gdp, cpi, Gcurr, T , Inp, def and ula); w t is a vector of W = 3 exogenous variables (defita, int and debt); and z rt is a vector of region-specific constant terms.Since the data are at the annual frequency, the model includes one lag of the endogenous variables (L = 1).
Posterior distributions are obtained through the Gibbs sampler from 15,000 draws where the first 2000 are discarded and, of the remainder, one in every 10 draws is saved.Appendix B in the supplemental data online outlines the specification of the priors in detail.

Identification
The contemporaneous zero and sign restrictions in Table 1 identify three fiscal shocks: one to current public expenditure (CG), one to total net revenues (TT) and one to public investment (PI).Those restrictions are borrowed from Pappa (2009) and Canova and Pappa (2007).A positive CG shock financed by deficit spending increases both Gcurr and output via aggregate demand stimulus.Similarly, a deficit-financed PI shock increase increases public investment and output.For those spending shocks, revenues are left unrestricted.In contrast, a TT shock increases T and lowers regional deficit and output.The scheme also imposes zero restrictions according to the specific fiscal shock being orthogonal to other identified shocks in the economy; that way, the three shocks are distinct from the others.All the other variables are left unrestricted.
A few things are worth remarking on here.First, those restrictions are theory based and consistent with short-run implications of RBC and NK-DSGE models; that is, they are robust to a wide range of parameters and model choices.However, due to the absence of real wages and consumption, one should take the theoretical conclusions sceptically.
Second, the scheme distinguishes fiscal stimulus from automatic responses to business cycle innovations (for instance, unemployment benefits) through output restrictions (Pappa, 2009).Moreover, since Italian regions are small local economies in a monetary union, monetary policy decisions can be treated as exogenous (Canova & Pappa, 2007).
Third, alternative identifying methods are available in this setting.For instance, Destefanis et al. (2022), who estimate the same Bayesian PVAR for Italian regions, use a Cholesky identification scheme, though they do not estimate the TT shock.As for the latter, the BP (Blanchard & Perotti, 2002) scheme would be more problematic in this setting; tax elasticity to GDP is likely to change across regions, but it can be difficult to retrieve with such a small sample size and is subject to simultaneity bias (Mertens & Ravn, 2013).In general, while exclusion restrictions based on the timing of fiscal decisions are plausible with national-quarterly data and hard to justify with regional-annual data (Canova & Pappa, 2007), sign restrictions are more robust to small sample biases (Canova & Paustian, 2011).
Fourth, one competing scheme with sign restrictions is the one of MU (Mountford & Uhlig, 2009).The latter discern spending and net tax shocks from business cycle and monetary policy shocks while leaving output unrestricted.They use quarterly data for the United States and find tax-cut multipliers higher than that of government expenditure, which is about zero.Again, in the case of regional-annual data, the MU's business cycle shocka contemporaneous increase of Gdp and Tseems unrealistic. 8Nevertheless, section 6 provides a robustness exercise with the MU identification.
Lastly, the scheme here does not address the non-fundamentalness of fiscal shocks (Kilian & Lütkepohl, 2017;Leeper et al., 2013); that is, agents may anticipate the effects of any foreseen future intervention by the government.Even if annual data should mitigate the issuebecause agents are unlikely to foresee fiscal measures one year beforehand (Beetsma et al., 2008) the inclusion of agents' fiscal expectations may change the results.Yet, previous research shows that the distortion may be negligible: IRFs are not much different when including expectations, even if the 'true' shocks are non-fundamental (Perotti, 2014).Deleidi et al. (2021b), who work in the same Italian-regional framework of this paper, confirm that result by augmenting the SVAR with forecasts on Italian spending sourced by the OECD.Section 6 carries out the same exercise as a robustness check.The CG shock induces an upward tendency in employment and price levels, suggesting a demand impact.Yet, those increases are non-significant, although they are significant for some Centre-North regions (see Appendix C online).The impact increase in current expenses and GDP is higher in the Centre-North but is more persistent in the South.The TT shock in Figure 4 induces a significant reduction in employment accompanied and a nonsignificant one in prices.While current expenses have almost no impact, the employment impact of taxes is more evident.For both macro-areas, the TT impact on GDP is lower than that of the CG shock.The response

2078
Francesco Simone Lucidi tends quickly to zero in both areas and below zero after four years.The PI shock in Figure 5 does not imply significant price changes for the two macro-areas.Yet, some Southern regions experience a price increase, while some Centre-Northern a decrease.The GDP response is higher and more persistent in the South than in the North.Overall, the PI shock implies significant public investment and six-year-lasting GDP increases in the South.That increase also stimulates employment in the South, but only in a few regions of Centre-North.As for the latter, the GDP significantly increase in only three years.
Fiscal multipliers are the ratio between the cumulative sums of output responses over fiscal variable ones.Since the IRFs are log changes, the corresponding multiplier indicates the output elasticity.The multipliers in euro equivalentthe euro-change of the GDP for one eurochange of the fiscal variablecan be obtained through the ex-post conversion; the latter consists in multiplying the cumulative-IRFs ratio by the ratio of the sample averages of GDP and fiscal variable.Here those adjustment terms may significantly differ among regions: for instance, Y/Gcurr and Y /Inp are 3 and 62 in Lombardy and 1.5 and 15 in Calabria, respectively.That may induce an artificial heterogeneity of regional fiscal multipliers; therefore, those terms here are computed at the macroarea level.Section 6 checks if the resulting multipliers are consistent when using the 'ex-ante' strategy as in Gordon and Krenn (2010) and Ramey and Zubairy (2018).
Table 2 reports regions and macro-area multipliers.The macro-area multipliers are obtained by averaging regional multipliers weighted by the GDP size of each region within its area.The numbers in bold indicate significant output responses at the corresponding horizon for the specific shock.The fiscal multipliers are generally > 1 and higher in Centre-North than South at the macro-area level.At the impact, the PI multipliers (2.5 in Centre-North and 1.5 in the South) are higher than CG ones (1.7 and 1.3) which, in its turn, are higher than TT ones (1.6 and 1).While the TT multiplier peaks between the second and the fourth year and then goes to zero, the CG and PI ones persistently increase along the projected horizon.
Signs of macro-area heterogeneity are evident in the significance level of the multipliers.The TT multiplier is significant for four years in Centre-Northern regions and two in Southern (sizing −2.2 and −1.2, respectively); the opposite for CG and PI multipliers: those are significant for two years in Centre-North (sizing 1.6 and 3.8, respectively) and up to h = 4 in South (sizing 1.1 and 2.7, respectively).
Fiscal multipliers are also heterogeneous within macroareas.The impact CG multiplier ranges from 0.9 in Sardinia to 2.1 in Abruzzo and Basilicata in Southern regions, while among Centre-Northern ranges from 1.0 (Tuscany and Veneto) to 2.2 (Lazio).The €1 increase in taxes reduces GDP by about €0.9 in Trentino, €1.9 in Emilia-Romagna, €0.8 in Sardinia and Sicily, and €1.2 in Apulia at the impact.Seven out of 12 Centre-Northern regions present impact PI multipliers > 2, with the lowest value in Tuscany (1.3) and the highest in Trentino (7.8). 9 Basilicata is the only Southern region with an impact PI multiplier > 2. Calabria, Sardinia and Sicily have the lowest (about 1 and 1.1).However, Sicily and Apulia are the only regions with significant multipliers along the projected horizon.

ROBUSTNESS ANALYSIS
The discrepancy in sample means of output and fiscal variables is frequent among Italian regions.The exercise here ensures that the baseline ex-post conversion does not bias the size of fiscal multipliers.Borrowing from Gordon and Krenn (2010) and Ramey and Zubairy (2018), model ( 1) is estimated by expressing the regional variables in levels divided by the GDP trend.The latter derives from the one-sided Hodrick-Prescott (HP) filter. 10With the 'exante' transformation, the ratio of cumulative IRFs directly reveals the fiscal multiplier in euro-equivalents.Destefanis et al. (2022) adopt this strategy for their baseline results.
Figure 6 compares the macro-area multipliers obtained through the ex-post baseline (Bas) and ex-ante (ExA) transformations.It shows that the latter does not affect the impact multipliers, which have approximately the same dimension as those obtained with the ex-post conversion.Three things are worth recalling.First, with the ex-ante approach, TT and PI multipliers are significant for a shorter time horizon.Second, for h .1, the TT multipliers slightly change concerning the baseline model, while the PI multipliers are lower in both macro-areas.Third, the CG multipliers change their dynamics in both macro-areas: they significantly increase up the third year and then tend towards zero.Those differences are not surprising: one drawback of the ex-ante approach is the use of a potential output measure (the HP trend of GDP).The latter is sensitive to fiscal variables and business cycle fluctuations, especially in the presence of hysteresis (Coibion et al., 2017).Moreover, while the ex-post method assumes a constant elasticity, the IRFs obtained through the exante method produce a constant fiscal multiplier; that implies that fiscal multipliers in regions where unemployment is high are the same as for low-unemployment ones.Therefore, the output IRFs (not reported) go more rapidly towards zero and are similar across regions. 11 Fiscal measures have implementation lags from which agents may take advantage to anticipate the effect of the policy.Not accounting for such a fiscal foresight may alter the size of fiscal multipliers (Ramey, 2011b).By following the approach of Auerbach and Gorodnichenko  1) is then augmented by the growth rate of Italian government expenditure forecasts (G F ). 12 The identification now accounts for the fact that fiscal shocks must be orthogonal to fiscal expectations so that the fiscal shocks can be considered unexpected.Figure 7 compares the resulting macro-area multipliers (FFor) with the baseline.Impact multipliers are about the same as the baseline model.
Accounting for fiscal foresight in Centre-North does not affect CG and TT multipliers, while it lowers the PI ones, which pass from a peak of around 4.5 to around 4. As for the South, the PI and CG peak-multipliers increase by about 0.5 and 0.1, respectively.The TT multiplier decreases by about 0.8 concerning the baseline case.Overall, those differences do not alter the conclusions about the results in section 5.
One may ask how a more a-theoretical identification scheme changes the results.Table 3 shows the sign and zero restrictions to the three fiscal shocks identified by following the MU strategy (Mountford & Uhlig, 2009).With that scheme, signs and zeroes are the same as before, except for unrestricted output.A shock rising Gdp and T removes business cycle fluctuations from fiscal shocks.Since the national interest rate is exogenous to local economies, the IRFs would be unaffected by restricting it.Therefore, the identification of a monetary policy shock is unnecessary.Differently from MU, the identification here still imposes deficit-financed fiscal shocks. 13 Table 4 reports the implied fiscal multipliers.Many of the implied TT multipliers are positive, and none is significant.This result is not a surprise; Arias et al. (2018) point to a methodological issue in the penalty function approach that MU use to select structural responses. 14 Arias et al. (2018) propose a more parsimonious strategy to select meaningful structural responses, the importance sampler, that is also adopted here.The latter implies that the GDP responses to TT are non-significant anymore, like TT multipliers in Table 3 suggest.The CG multipliers are positive and significant in seven Northern regions.They reach the highest value in Lombardy (2.7 at the impact) and are non-significant when < 1.The CG multiplier dynamics of the Centre-North are almost the same as before.However, it becomes non-significant for the South, apart at h = 2; only two Southern regions, Campania and Sicily, present a positive and significant CG multiplier, though < 1.The contrary for the PI multiplier: it is significant only at h = 2 in Friuli, Trentino and Umbria, so the Centre-North multiplier is < 1 and nonsignificant.It is positive, upward sloping, and significant in six Southern regions for different periods; that implies an area multiplier > 1, which is significant from the second to the fourth year after the shock.This result reinforces that PI seems to be the most effective measure in stimulating the economy of the Southern area or, putting it the other way around, that public investment cuts in the South generate a more persistent depression.In half of the cases, the fiscal multipliers make no sense.For example, output falls after a CG shock in three Southern regions and a PI shock in five regions (Lazio, Tuscany, Veneto, Calabria and Sardinia).Those multipliers are at odds with the standard model predictions in the literature.
A similar result for a similar robustness check is in Pappa (2009).

RELATION TO OTHER STUDIES
The results of this paper are in line with the related literature.Canova and Pappa (2007) and Pappa (2009) find heterogeneous local multipliers (both CG, TT and PI) across American states and European countries, such as the regional ones here.As in those papers, price responses to fiscal shocks here are asymmetric, small and often nonsignificant.They also find sensitive local employment to fiscal stimulus, although reactions are small in magnitude.Caldara and Kamps (2008) show that this result is robust for a wide class of VAR-identifying schemes, ranging from BP to MU.On average, the TT shock here affects macroarea employment more than CG and PI shocks, although with some differences among regions.Caldara and Kamps (2008) explain that that result comes from the identification: while the recursive and BP approaches suggest that Ricardian Equivalence is a good approximation of economic reality, the sign-restrictions approach suggests that taxes are distortionary.Nevertheless, the size and the dynamics of the TT multiplier are in line with those obtained through other sign restrictions and proxy identifications (Caldara & Kamps, 2017;Mertens & Ravn, 2013;Mountford & Uhlig, 2009).Differently from Pappa (2009), Perotti (2004), Ilzetzki et al. (2013) and Boehm (2020), PI shock here may have long-lasting effects on GDP, which are remarkably higher than those of CG shocks.This result is in line with other studies such as Auerbach and Gorodnichenko (2012) and Boehm (2020).
One may think of Centre-North and South multipliers as upper and lower bounds of Italian national multipliers; in this view, the results align with other national estimates.The one-year CG multiplier is lower than that of    The misalignment of fiscal multipliers in Italian regions 2083 REGIONAL STUDIES Giordano et al. (2007) (which equals 2.4), but it has similar dynamics over the projected horizon.They find a similar response to employment, and a short-living effect on inflation, while their positive shock to taxes has a positive and non-significant impact on GDP.Concerning other studies on Italy, De Nardis and Pappalardo (2018) find similar multipliers of CG and PI; the latter equal 2.6 at the impact and reaches the value of 3 within three years.Locarno et al. (2014) and MEF (2017) find a similar relationship between TT and CG multipliers, though the ones here are smaller.The one-year multipliers of Deleidi ( 2022) are similar for PI and CG but lower for TT.Again, this might come from the difference between the sign restriction and the BP approaches.
Concerning macro-area estimates here, multipliers in Centre-North are generally higher than those in the South.This result is in contrast with Piacentini et al. (2016) and Marrocu and Paci (2010).However, the latter find mixed evidence accounting for local administration's efficiency.Chian Koh (2017) argues that the difference can be due to lower efficiency in public expenditure management of developing countries concerning advanced economies.The macro-area gap here is in line with that in Deleidi et al. (2021b).The latter find PI multipliers > 1 and higher than the CG multipliers.As in this paper, they find Centre-North PI multipliers generally higher than those in South (respectively, 3.5 and 2.0).Their average CG multipliers also are similar to the one here (1.7 in Centre-North and 1.3 in South), and, as in this paper, the inclusion of fiscal foresight does not alter the key results.Finally, Destefanis et al. (2022) estimate region-specific fiscal multipliers with a similar model structure to this paper.Their results are different from the baseline here: they find positive and significant multipliers of CG only in Liguria, Tuscany, Abruzzo, Campania, Apulia and Sicily.At the same time, significant PI multipliers are positive only in Marche, Lazio, Abruzzo and Campania.They use a recursive identification, where expenditure is predetermined; in fact, their results are much more compatible with the ones here when using the MU identification.Moreover, they use the ex-ante approach to get regional multipliers.

CONCLUSIONS
The analysis here shows that fiscal multipliers of regions and macro-areas in Italy to government consumption, investment and revenues are > 1.Though the lack of evidence on real wages and consumption, the size of the implied multipliers seems close to the NK class of model-based estimates; yet the persistent effects of public expenditurein particular, that related to public investmentplace the theoretical implications here near the old Keynesian theory.Cumulative regional multipliers implied by tax changes are also high, but their effect vanishes quicker than government consumption and investment multipliers.Given the potential relevance of regional taxes on production, further research should address the separation of those revenue shocks by tax type.
While the fiscal shocks' transmission is similar for Italian regions within the same area, they differ between macro-areas.On balance, IRFs are more persistent in the Southern area, where fiscal multipliers are remarkably smaller.The latter also holds by accounting for agents' expectations on budgetary decisions and using different computation methods.The most factual finding of the paper concerns the positive cumulative GDP impact of public investment in the South; the latter also endures atheoretical identifying schemes in which output responses are unrestricted.
The misalignment of the effects of fiscal policy in Italian macro-areas confirms a critical aspect of the analysis.Since the late 1990s, a prolonged course of economic depression has affected the Southern area of Italy, one of the poorest areas in continental Europe.Moreover, a prolonged period of public expenditure cuts exacerbated the territorial divide; the available data show that those cuts were harsher in Southern regions.In light of the results in this paper, the disconnect between the Italian areas can be partly explained by the fiscal tightening hitting local economies that have characterized the years following the Great Recession and the Sovereign Debt Crisis.DeLong et al. (2012) suggest that spending cuts may exacerbate the economic scars left by the recession, especially in depressed areas that are more subject to hysteresis (Blanchard & Summers, 1986).At the same time, the fact that fiscal shocks generate long-lasting effects in the South supports the view that fiscal policy can be selffinancing in depressed areas.A temporary increase in government spending can lead to a permanent output increase at small financial costs.This paper points to this conclusion by looking at public investment as a more effective tool for Italian policymakers.
3. Most of these approaches are also available in a local projection setting.See Plagborg-Møller and Wolf (2021) for an overview.4. Section 2 in Deleidi et al. (2021a) provides a detailed table on the size of multipliers by methodology, country and spending typology.5.These are not harmonized to international standards and cannot be reconciled with Eurostat data (Galli & Gottardo, 2020).6.The rest of the paper often labels the regions as the following in parentheses.For Southern regions: Abruzzo (abr), Apulia (pug), Basilicata (bas), Calabria (cal), Campania (cam), Molise (mol), Sardinia (sar) and Sicily (sic).For Centre-Northern regions: Emilia Romagna (ero), Friuli Venezia Giulia (fvg), Lazio (laz), Liguria (lig), Lombardy (lom), Marche (mar), Piedmont (pie), Trentino (taa), Tuscany (tos), Umbria (umb), Aosta Valley (vda) and Veneto (ven).7. Lucidi et al. (2021) show that the models obtained with this approach perform better than a single countrywide PVAR for one-step-ahead forecasts.8. Pappa's (2009) scheme encompasses the fact that there might be other valid structural models where business cycle shocks exist.9.At h ≥ 2, Trentino and Aosta Valley present significant PI multipliers > 6. Appendix C in the supplemental data online shows that they flag when carrying out a model of only special administrative regions, which are Trentino and Aosta Valley, together with Friuli, Sardinia and Sicily.I thank an anonymous referee for this suggestion.10.Another way to estimate the euro-equivalent fiscal multiplier directly is to use variables in levels; in this medium-scale panel VAR that causes unstable estimates for several regions.11.I thank an anonymous referee for pointing this out.Regional and macro-area IRFs are available upon request.12.This variable is not available at regional level, and then here is taken to be equal across regions.13.A similar exercise can be found in Caldara and Kamps (2008).14.Section 5 of Arias et al. (2018) explains the problem in detail, while an application of MU is in the working paper version (Arias et al., 2014).

Figure 1 .
Figure 1.Regional and national real per capita gross domestic product (GDP) ratio relative to 1995.

Figures 3 -
Figures 3-5 compare macro-area IRFs for the CG, TT and PI shocks; those derive from the sum of medians and percentiles of each region's response weighted by the size of the region in the area in terms of its GDP.The bands represent 68% credible set from median responses, while red-dotted and blue-solid lines are the median responses of Centre-Northern and Southern regions; to save space, the region-specific IRFs are moved and commented in Appendix C in the supplemental data online.

Figure 4 .
Figure 4. Impulse response function (IRF) for a shock to revenues (TT) in Centre-North (CN) and South (S).

Figure 3 .
Figure 3. Impulse response function (IRF) for a shock to current expenses (CG) in Centre-North (CN) and South (S).

Figure 5 .
Figure 5. Impulse response function (IRF) for a shock to public investment (PI) in Centre-North (CN) and South (S).

Figure 6 .
Figure 6.Baseline (Bas) and 'ex-ante' (ExA) multipliers for South (S) and Centre-North (CN).Note: Circles and triangles indicate horizons at which the multiplier is significant in the 68% credible sets.

Figure 7 .
Figure 7. Baseline (Bas) and fiscal foresight (FFor) multipliers for South (S) and Centre-North (CN).Note: Circles and triangles indicate horizons at which the multiplier is significant in the 68% credible sets.

Table 1 .
Pappa (2009) to current public expenditure (CG), one to total net revenues (TT) and one to public investment (PI) shocks based onPappa (2009).The misalignment of fiscal multipliers in Italian regions 2077