This is an author version of the contribution - IRIS Uni Torino

This is an author version of the contribution - IRIS Uni Torino

This is an author version of the contribution published in: Questa è la versione dell’autore dell’opera pubblicata in: Journal of Law and Economics, ...

477KB Sizes 0 Downloads 3 Views

Recommend Documents

This is an author version of the contribution - IRIS Uni Torino
May 13, 2009 - Important Ingredient of the Alpine Liqueur "Genepi", JOURNAL OF AGRICULTURAL. AND FOOD CHEMISTRY, Volume:

Artemisia umbelliformis Lam - IRIS Uni Torino
The alpine wormwoods (Artemisia genipi Weber, A. glacialis L. and A. umbelliformis Lam.), commonly referred to as genepì

This is an author produced version of a paper published in Outlook on
In place of collective and state farms, independent farmers relying on family labour were expected to make food producti

This item is the archived peer-reviewed author-version of - Institutional
calculate estimates of imputed rent and analyse how these benefits are distributed over the population and how they help

This item is the archived peer-reviewed author-version of - Institutional
A priori, viewers may not consider Pottery Barn as a brand .... the Movies” event and cross-checked with those of the

This item is the archived peer-reviewed author-version of - irua
Sep 2, 2016 - 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29.

This item is the archived peer-reviewed author-version of - Institutional
example, reported that 51.5% of SNS users always or sometimes accept friend requests from strangers (O'Dea and Campbell,

This item is the archived peer-reviewed author-version of - irua
students in Beijing. Interviews were conducted in various locations, including coffee shops around the Wudaokou area, in

IRIS is the catalog of generally-accepted performance - IRIS (GIIN)
The IRIS catalog compiles standardized metrics that are generally accepted in various social and environmental sectors.

Herder und der Entwurf einer Poetik der Moderne - IRIS Uni Torino
Nach wie vor klassisch der Beitrag von Max Rouché, La philo- sophie de l'histoire ...... bliographie, introd. et notes p

This is an author version of the contribution published in: Questa è la versione dell’autore dell’opera pubblicata in:

Journal of Law and Economics, vol. 56(3), 2013, pp. 741‐776 DOI: 10.1086/671479

The definitive version is available at: La versione definitiva è disponibile alla URL:

Tax mix corners and other kinks Federico Revelli∗ June 1, 2011; revised: July 19, 2012; April 3, 2013


This paper models the local tax mix determination process in the presence of statewide fiscal limitations - the decentralized government finance archetype - and shows how excess sensitivity of local public spending to grants (the conventionally and somewhat misleadingly termed “flypaper effect”) arises in the constrained tax mix irrespective of whether lower or upper limits bind, and cannot in general be taken as a symptom of local government overspending. An empirical application to Italian province panel data provides consistent evidence of the role of corner solutions produced by two-sided tax limits in explaining the sensitivity of local public expenditures to grants. JEL classification: C23; C24; H71; H77 Key words: flypaper effect; excess sensitivity; tax mix; switching regression

∗ Department of Economics, University of Torino, Campus Luigi Einaudi, Lungo Dora Siena 100A, 10153 Turin (Italy); e-mail: [email protected] I gratefully acknowledge financial support from the Institut d’Economia de Barcelona (“IEB Research Grant on Fiscal Federalism”). I would like to thank Giuseppe Bertola, Massimo Bordignon, Margherita Borella, Kai Konrad, Antoine Loeper, David Wildasin, Stanley Winer, and particularly the Editor, Richard Holden, and an anonymous referee for deep and constructive comments. I am responsible for any remaining errors.




The overall size as well as the tax revenue bundle of the local public sector in multi-tiered structures of government are the outcomes of the decentralized decision-making process subject to the fiscal rules set by central (state) governments. As documented by Anderson [1] and Wolman et al. [39] for the US, and by Joumard and Kongsrud [24] and Sutherland et al. [37] for the OECD countries, top-down tax and expenditure limitations (TELs) are frequently so tight and pervasive as to jeopardize the very principle of local fiscal autonomy.1 This paper aims at investigating how statewide revenue raising limitation rules shape local governments’ budget constraints, focusing on the kinks that are typically generated by tax floors and caps.2 In particular, it evaluates the effects of tax limits on the determination of the local tax mix and on the response of local public expenditures to grants. As far as the latter issue is concerned, a vast literature most recently reviewed by Inman [22] has investigated and sought to explain the anomalously high response of local spending to grants relative to the response to private income - the so-called flypaper effect by which money from central government “sticks where it hits.”3 Two broad kinds of explanations of the flypaper effect have been offered in the literature (Hines and Thaler [20]). The first has to do with a variety of specification and estimation errors that applied researchers would have kept on making for decades. Those errors range from mistakenly treating matching grants as if they were lump-sum, to the omission of important variables - such as unobserved population characteristics 1 Nechyba [29] argues, though, that state command on local fiscal choices (in terms of income tax-funded grants and state-imposed caps on local property tax rates) arises in equilibrium as an optimal outside enforcement when a collusive agreement to simultaneously introduce local income taxes is not self-enforcing. Vigdor [38] sees statewide local tax limitations as a way of allowing absentee landowners and nonresident employees to contain tax exporting. Calabrese and Epple [8] provide a comprehensive political economy model of the emergence of tax limits in the presence of multiple policy instruments. 2 Vertical restraints - such as quantity floors and price ceilings - are ubiquitous in private sector networks too. Zanarone [41] discusses the rich theoretical literature on vertical restraints as coordination mechanisms, and analyzes empirically the role of public regulation in manufacturer-dealer contracts. 3 According to Inman [22], over 3,500 research papers exist documenting and seeking to explain the flypaper effect. Payne [30] offers an insightful wide-ranging review of the more recent research into the mirror phenomenon of crowd-out.


or spatial lags of other governments’ policies - that are simultaneously correlated with grants and local public expenditures. The second explanation relies on the argument that the political representation process is substantially richer than the one postulated by the standard neoclassical model: asymmetric information, loss aversion, fiscal illusion, separate mental accounting, special interest groups, and citizens’ inability to write complete contracts with their elected officials would be responsible for the lack of fungibility between public and private uses of money, and would cause the observed large flypaper effect.4 I model here for the first time the local tax mix determination process in the presence of statewide tax limitations - the decentralized government finance archetype - and show how the so-called flypaper effect arises in the endogenously generated constrained tax mix. In particular, local expenditures are shown to be predicted to display a one-for-one response to grants in the presence of binding limitations on all local tax revenue sources. Interestingly, a binding cap on just one of the available own revenue sources is enough to generate some form of flypaper effect, in the sense of an excess sensitivity (Flavin [14]) of local spending to grants. Importantly, the above results hold when either upper or lower tax limitations are binding: in fact, local authorities will display an excess sensitivity of public expenditures to grants irrespective of whether they are against lower or upper bounds. Finally, the reaction of local public spending to own tax base shocks will be a function of the (lower or upper) binding tax rate limits. An important corollary of the analysis concerns the very interpretation of the public spending behavior that is conventionally known as the flypaper effect. Since excess sensitivity of local public spending to grants should be predicted to arise - and generally tends to manifest itself - both when grants increase and when they decrease, the depiction of grants as sticky seems semantically dubious, and the flypaper effect label turns out to be a misleading one: an 4 Building on the insights of Hamilton [17] and Becker and Mulligan [4], Dahlby [12] recently relaunched the explanation of the flypaper effect based on the convex deadweight costs of taxation.


higher sensitivity of local public expenditures to grants than to own revenue sources cannot in general be interpreted as a sinister symptom of decentralized government overspending. While the existing literature seems to have almost universally overlooked the potential impact of tax and expenditure limitations on the sensitivity of local public spending to exogenous variations in grants, some recent papers have brought the fiscal limitations issue into the empirical investigation of the flypaper effect. Lutz [28] conjectures that previous evidence of a flypaper effect might have arisen from state constraints preventing local governments from selecting their preferred bundle of public goods, and provides evidence of equivalence between grants and income from a school finance reform in New Hampshire - “one of only five states with no state-imposed limitations on the taxing or spending power of local governments” (Lutz [28], p. 317). Brooks and Phillips [7] offer the first formal statement and explicit empirical test of the hypothesis that restrictive fiscal institutions might be responsible for the flypaper effect. They use data on the US Community Development Block Grant (CDBG) program and argue that state TELs may systematically force city governments to underprovide local public goods and increase the stimulative effect of federal grants on city spending. However, since they do not observe either the municipal tax bundle or whether a revenue raising constraint is actually binding in any given city, they have to rely on a state-level index of fiscal constraints and ignore both the municipal choice as to own revenue source diversification and the issue of endogenous selection of a city government into the fiscally constrained status. Interestingly, Brooks and Phillips [7] find a generally high sensitivity of spending to grants in a period of dramatic retrenchment, while they find only limited evidence of an effect of statutory state-level tax limitations on municipal governments’ response to the collapse in CDBG grants. Finally, Baicker [3] analyzes US states’ responses to federally mandated increases in public medical spending (mandated expansions in Medicaid coverage). She develops a theoretical model showing that states subject to binding tax limits (legal ceilings)


would have to reduce spending on other programs by more when faced with a mandatory spending increase than they would if they were able to raise taxes. While her empirical analysis reveals that all states - with or without limits offset the mandated Medicaid increase by reducing other public welfare spending, she recognizes that this might be due to the fact that she does not observe whether tax limits are actually binding. The paper concludes with an empirical application to panel data on Italian provincial governments’ budgets. The econometric set-up relies on sharply observed corner solutions produced by nationwide tax limits during the years 2000-2007, and exploits the exogenous features of state retrenchment following Italy’s adherence to the EU Stability and Growth Pact to estimate the effect of state grants on provincial expenditures. An attractive feature of Italian provinces is that their own tax revenue sources (a tax on vehicle registrations, a tax on electricity consumption for business uses, and a waste management surcharge) are subject to strict and frequently binding upper and lower tax rate limitations. The empirical analysis based on a panel data switching regression approach that allows for endogenous selection into the tax-constrained regime as well as for potential grant endogeneity offers evidence of excess sensitivity of local public spending to grants - a one-for-one response - in tax-constrained localities, irrespective of whether upper or lower tax limits bind. On the other hand, I show that authorities that are not fully constrained turn out to be able to smooth out their expenditure profile by offsetting state grant policy in a period of fiscal retrenchment through own tax changes, and that the impact of own tax bases on local expenditures depends on whether lower or upper limits are binding in the observed tax mix. The paper is organized as follows. Section 2 sets up a simple model for the analysis of the local tax mix determination process in the presence of upper and lower constraints on own tax instruments. Sections 3 develops the model’s empirical implications and outlines the econometric strategy. Section 4 presents the tax structure of the Italian provinces. Section 5 reports and discusses the


switching regression estimation results, section 6 deals with grant and regime selection endogeneity issues, and section 7 concludes.


Communicating vessels

Consider the two vessels in figure 1. Say that vessel vpn represents consumption of private goods out of community n private income in (n = 1, ..., N ), and vessel vgn represents consumption of local public services. The width of the vgn vessel relative to the vpn one can be interpreted as reflecting community n’s preferences for publicly provided - and possibly non-rival - services versus market-provided consumption goods. The structure depicted in (1.a) amounts to a perfect tax centralization arrangement, where expenditures on local public services are entirely funded by central government grants gn . Of course, nothing ensures that the allocation of resources to private and public uses reflects local preferences, or that the marginal benefit from private consumption equals the marginal benefit from public consumption. In the central picture (1.b), the two vessels are allowed to communicate via local tax revenues. In order for local public goods to be provided optimally, and given that the marginal rate of transformation between private and public goods is assumed constant and equal to one, the marginal utility in the two vessels has to be equalized. Just like communicating vessels, where the force of gravity requires hydrostatic pressure to be balanced out in the two vessels regardless of their relative sizes, the welfare optimization forces make resources flow from vpn to vgn at the tax rate τ n =

tn in .

Once the even equilibrium level

is attained in the two vessels, whether additional resources are poured into vpn or into vgn the same allocation of private and public consumption will result by the law of communicating vessels. In the bottom picture (1.c), local jurisdiction n is subject to a tax rate cap equal to h =

thn in ,

with the cap binding if thn < tn . The Samuelson condition

for optimal public good provision will not be satisfied if the tax cap is binding,


meaning that more resources ought to flow from vpn to vgn in order to equate the “pressure” in the two vessels. An additional unit of private income will raise the consumption level at rate 1 − h in vpn , and at the rate h in vgn . If additional grants are poured into (pumped out of) vgn , the level will rise (fall) in vgn only. The flypaper effect, so to say.


The one-tax case

Let the welfare of lower-tier jurisdiction n (n = 1, ..., N ) in a two-tier structure of government be expressed as a separable, concave function of public and private goods consumption: wn = v(zn ) + ρn u(cn ) = ln(zn ) + ρn ln(in (1 − τ n ))


where zn stands for expenditure on local public services, in is some meaningful measure of community income, τ n is the flat local income tax rate, and ρn is a (positive) parameter reflecting preferences for private consumption versus consumption of local public services, and let local authority n abide to a balanced budget rule: zn = gn + τ n in


where gn stands for (lump-sum) grants from the upper tier of government.5 Maximization of (1) subject to (2) leads jurisdiction n to select the optimal tax rate-spending pair (τ ∗n , zn∗ ) as a function of the assumed exogenous variables gn and in : τ ∗n =

µ ¶ gn 1 − ρn in

1 1 + ρn

zn∗ = gn + τ ∗n in =

1 (gn + in ) 1 + ρn

(3) (4)

Equations (3) and (4) generate the standard neoclassical and somewhat uncomfortable result that exogenous perturbations in in or gn should be predicted to have an identical effect on zn∗ :

∗ ∂zn ∂gn


∗ ∂zn ∂in


1 1+ρn .

When this does not

5 It is usually convenient to interpret all monetary variables in (2) as measured in per capita terms, thus implying that publicly provided services entering the welfare function (1) are private (rival) in nature, and to ignore the revenue-raising effort and public good provision on the part of the upper level of government.


happen, and in particular if a change in grants turns out in practice to provoke a much larger reaction in local public spending than a change in own resources does, a flypaper effect is said to exist (Hines and Thaler [20]). Consider now the consequences of the introduction of a nationwide tax rate limitation: l ≤ τn ≤ h


Quite straightforwardly, expenditures equal grants plus the maximum amount of revenues that can be collected locally (gn + hin ) if local government n is against the upper bound h, meaning that authority n would like to tax private income more than permitted. Similarly, expenditures equal grants plus the minimum amount of revenues that have to be collected locally (gn + lin ) if local government n is against the lower bound l, meaning that it would like to tax private income less than permitted. As formally shown in Appendix A, this implies the following: Proposition 1 Consider the constrained optimization problem given by (1), (2) and (5). Whether the upper or the lower tax rate limit binds, the sensitivity of local public spending to grants in the constrained optimum is 1. The sensitivity of local public spending to own resources in the constrained optimum equals the binding tax rate limit. In the above circumstances, the so-called flypaper effect is the result of tax limitations, and arises irrespective of whether a lower limit (i.e., local authorities wishing to tax less than permitted) or an upper tax limit (i.e., local authorities wishing to tax more than permitted) binds. Moreover, since “constrained” local public expenditures should be predicted - and generally tend - to respond onefor-one both when grants increase and when they decrease (Stine [36], Hines and Thaler [20], Gamkhar and Oates [15], Brooks and Phillips [7]), the flypaper effect label seems an inappropriate or even misleading one: the long studied anomalous response to intergovernmental grants appears to be best described as an excess sensitivity (Flavin [14]), and cannot in general be interpreted as a symptom of local government overspending. 8


The multiple-tax case

Let now lower-tier government n rely on M ≥ 2 distinct own tax revenue sources as well as on upper-tier government lump-sum grants. Denoting by τ m n the flat rate set on tax base m (im n ), the budget constraint and welfare function can be expressed as: zn = gn + τ 0n in = gn +


m τm n in



wn = ln(zn ) +



m m ρm n ln(in (1 − τ n ))


with ρm n capturing the contribution of tax base m to community n’s social welfare. While we do not wish to make too specific assumptions here, we can think of the M tax bases as income flows generated within the local community by different assets held by residents (say, land, physical capital, human capital) or as comprehensive income accruing to distinct groups of residents (e.g., depending on households’ income brackets or age structure), as long as local governments can discriminate among those income flows by employing a given set of tax 6 instruments (τ 1n , ..., τ M Raising revenues from the M −set of available tax n ).

bases leads to a welfare loss that depends in turn on the selected M −vector of tax rates, with the marginal welfare loss of taxing base m increasing in the m tax rate.7 The first order conditions for maximization of (7) subject to (6) require equalization of the marginal welfare contribution of an additional unit of own tax revenue spent on local public services to the marginal costs of raising revenues across all tax bases. This results in a vector of optimal tax rates and expenditure level exhibiting again the standard “communicating” feature, with private resources and state transfers being fungible. Let now central government impose the set of tax rate limitations (m = 6 A more general model would have the government design a tax schedule T(b ), n   is arbitrarily correlated with the income vector where the tax base vector bn = b1n , ..., bM n  1 in = in , ..., iM n , and the latter might not be taxed. As long as in is observed, a welfarist government would set T so as to offset any variation in bn that is not correlated with in . 7 The welfare formulation (7) is compatible with the classic Hettich and Winer [18], [19] principle of political cost minimization in the presence of M potential tax sources.


1, ..., M ): m lm ≤ τ m n ≤h


Depending on the size and contribution to welfare of the M tax bases, their m m m respective tax rates might be observed in left (τ m n = l ) or right (τ n = h )

corner solutions, resulting in a tax mix that can potentially display lower and upper limits binding at the same time on different tax rates. Consider now the effect of a grant change on local government spending behavior. As formally proven in Appendix A, the sensitivity of local public spending to grants in the constrained optimum turns out to be increasing in the number of tax rate limits that are binding. Moreover, such sensitivity equals one if all limits bind. Interestingly, the one-for-one response of spending to grants emerges either when all upper limits bind or when all lower limits bind, or for any of the admissible fully constrained tax mix outcomes, with some of the limits binding from above, and some from below. To see why this is the case, take the example of a two-tax environment and a double-sided limit on each tax rate. While representing a very simple case, it provides the basic intuition and can be generalized to tax rate vectors of any dimension. Suppose that, conditional on state grants and community tastes for public services, the local government wishes to set higher tax rates than the upper limits on both revenue sources. This implies that any subsequent grant change will leave the government at its upper-constrained tax mix, and spending will respond to grants one-for-one: any additional transfers will be spent on public services to get closer to the desired spending level; on the other hand, there is no way of offsetting a grant reduction by further raising local tax rates. Suppose now the government is at a lower-constrained tax mix, wishing to tax both bases less then permitted. In this case too, any grant change will cause spending to respond one-for-one: additional transfers will have to be spent, because there is no way of returning them back to taxpayers via lower taxes; and it cannot be optimal for a lower-constrained government to offset a marginal grant cut by raising taxes. Finally, consider a tax mix where the


local government wishes to tax one of the two revenue sources, say j, less than permitted by the limit, and tax base k more than permitted, thereby hitting the lower limit for j and the upper limit for k. Consider what effect a shock to grants would have in that case: the local government would ideally return any additional grant to j taxpayers via lower taxes, but that is not feasible since the tax rate on j is already against the lower limit. Moreover, since it is clearly not desirable to lower the k tax rate that is already below what welfare maximization requires it to be, the additional grant will be spent on local public services. Similarly, a fall in grants cannot be offset by further raising the upperconstrained tax rate k, and neither is it optimal to raise the lower-constrained tax rate j. Spending will react one-for-one to the grant reduction by falling by the same amount. In the general case of M tax sources, Appendix A shows that tax limits make the marginal cost of raising own revenues steeper and the response of local public spending to grants larger by restricting the number of available tax instruments, eventually degenerating into an infinitely high marginal cost of self-financing and a one-for-one local public spending reaction to grants when all tax limits bind. Along a similar line of reasoning, Appendix A shows that in a fully constrained tax mix, the sensitivity of local public spending to a tax base shock turns out to equal the lower or upper binding limit, as the intuitive representation in figure (1.c) suggested. The main predictions from the multiple-tax model are summarized in the following: Proposition 2 Consider the constrained optimization problem given by (6), (7) and (8). The sensitivity of public spending to grants is increasing in the number of binding rate limits, and equals 1 if all limits bind, irrespective of whether they bind from above or from below. The sensitivity of local public spending to own tax bases in a fully constrained tax mix equals the binding tax rate limit.



Empirical implications and econometric approach

The results in section 2 prompt the estimation of the sensitivity of local public expenditures to changes in exogenous revenue sources, while allowing for heterogeneous responses depending on the degree to which local governments face financing constraints. In fact, the empirical investigation of the behavior of local government spending in the presence of tax limitations bears a striking similarity with two well developed lines of research. The first concerns the inquiry into the role of financing and liquidity constraints in explaining the elasticity of investment to cash-flow in Q models of the firm (Bond and Meghir [5], Fazzari et al., [13], Hu and Schiantarelli [21], Kaplan and Zingales [25], Cummins et al. [10]).8 The second relates to the borrowing constraint interpretation of the excess sensitivity of private consumption to disposable income in permanent income/life cycle frameworks (Runkle [34], Zeldes [42], Jappelli et al. [23]).9 In the empirical investment and consumption literatures, the conventional approach consists in splitting the sample according to an a priori index of financing/liquidity constraint (typically related to the dividend payout or liquid assets to capital stock ratio for firms, and to the asset-income ratio for consumers), and compare the switching regression estimates of the sensitivity of investment (consumption) to cash flow (income) for the constrained and unconstrained subsamples (Fazzari et al. [13], Kaplan and Zingales [25], and Runkle [34]). Similarly, in order to test on panel data whether the local public spending response to revenue sources is affected by tax limitations, a time-invariant selection criterion can be employed to assign authorities to either of two subsamples based on whether they are consistently constrained (or not constrained) during 8 In

their flypaper effect review, Hines and Thaler [20] mentioned the liquidity constraint explanation of flypaper-like effects in the private sector. However, they did not consider the possibility that local tax and expenditure limitations might be the root cause of the flypaper effect. 9 Borge and Tovmo [6] test whether liquidity constraints imposed by balanced-budget rules affect the intertemporal spending behavior of Norwegian local governments, and find that departures from rational forward-looking public consumption smoothing can in part be explained by financing constraints.


the whole period of observation (t = 1, ..., T ): znt

= q0nt β1 + ζ 1n + η1nt

if Kn = 1



= q0nt β0 + ζ 0n + η0nt

if Kn = 0


where: q0nt = [gnt i0nt x0nt ], and xnt is a vector of local characteristics that can be thought to affect community n’s preferences for public services, such as socioeconomic and demographic structure, or ideological complexion. ζ 1n and ζ 0n are jurisdiction-specific effects comprising all time-invariant, unobservable characteristics that might be correlated with qnt , and Kn is the switching indicator. Clearly, a number of criteria might be employed to code in a binary way the observed tax mix of authority n in the presence of M constrained tax revenue sources, leading to different sample splits.10 For the moment, based on the realizations of the M tax rates, let Kn be somewhat arbitrarily defined as: ½ 1 all tax limits bind if (11) Kn = 0 otherwise According to the sample separation criterion (11), parameter vector β1 in equation (9) measures the response by authorities that are structurally constrained, in the sense that tax limitations are binding for all revenue sources and in all years, while β0 in equation (10) captures the response of spending by authorities that are not, in the sense that at least one of their M tax rates does not hit the limits. The theoretical prediction on the size of the grant coefficient in particular is: β 1g = 1 > β 0g . Authorities in the Kn = 1 regime would necessarily exhibit a one-for-one response to transfers, not being able either to return increased grant money back to residents via tax cuts, or to raise taxes to offset a fall in state transfers, irrespective of whether lower or upper limits bind. As for the effect of tax base changes, fully constrained authorities would have β 1m = hm (β 1m = lm ) when an upper (lower) tax limit is binding on i i tax base m. On the other hand, authorities in the Kn = 0 regime would be able to transfer any extra grant back to their own taxpayers by reducing the 1 0 In principle, finer partitions than the two-sample split can be constructed according to the degree to which tax limitations bind.


unconstrained tax rates; similarly, they would partly offset a grant reduction by raising their unconstrained local tax rates, meaning that β 0g < 1. The above considerations suggest that the behavior of expenditures in the Kn = 1 versus Kn = 0 groups ought to be mirrored by changes in the M −vector of local tax rates. Of course, for all the authorities in the Kn = 1 regime, ∆τ m n = 0 by the very definition of the sample separation criterion (11), and this is actually the force driving spending to follow grants so closely. As regards the Kn = 0 regime, local governments would raise (lower) tax rates where possible when grants fall (increase), trying to offset the grantor’s impact on the local public-private consumption mix. Similarly (as formally shown in Appendix A), a positive shock to tax base m fosters an increase in tax rate m in an attempt to re-establish the optimal public-private consumption mix by directing additional resources to public spending, and a fall in all r 6= m tax rates in order to even out the marginal costs of raising revenues across tax bases. A disadvantage of any separation rule inspired by a principle similar to (11), though, consists in the fact that it implies “freezing” the sample and renouncing to using information on governments that switch from one regime to the other over the period of observation (Hu and Schiantarelli [21]).11 An alternative empirical approach - based, among the others, on Bond and Meghir [5], Jappelli et al. [23], Zeldes [42] and Cummins et al. [10] - consists in allowing for a time-varying constraint status as in (12) below, where an authority is rated as unconstrained in year t (Knt = 0) if it can manoeuvre at least one of its tax instruments: Knt =


1 0


all tax limits bind in year t otherwise


However, whether an authority is at a tax mix corner solution in a given year might in principle be determined endogenously, say if unobserved shocks to expenditures push local authorities towards the (left or right) tax limits. With 1 1 Moreover, K might be correlated with spending. However, since selection effects can n only occur through correlation between Kn and the time-invariant authority-specific effects, any selection bias is cancelled by differencing them away, and a linear panel data fixed effects estimator can be applied to the two subsamples (Charlier et al. [9]).


Knt in equation (12) depending in a structural way on M distinct tax rate realizations originated from the two-sided constrained optimization problem, the reduced form of the binary selection index can be expressed in a stochastic way as a function of the vector of exogenous variables qnt : ∗ = q0nt δ + µn + εnt > 0] Knt = 1 [Knt


∗ ∗ > 0, Knt is an auxiliary latent variable According to (13), Knt = 1 if Knt

with no straightforward intrinsic meaning, εnt is normally distributed, and µn is a time-invariant, authority-specific effect. Upon modelling the selection process as in (13), it is possible to apply the Wooldridge [40] two-stage procedure for fixed effects panel data, with the selection equation (13) being consistently estimated in the first stage, and the spending equation for Knt = 1 being estimated in the second stage after correcting for selection bias.12


Local tax limitations in Italy

I analyze the local public spending and tax setting behavior of the Italian provinces by using panel data through the years 2000 to 2007, i.e., during the fiscal consolidation process following Italy’s adherence to the EU Stability and Growth Pact. The Italian system of local government is organized as a three-tier structure, with the 103 provinces constituting the intermediate level between regional (20 regions) and municipal (over 8,000 municipalities) tiers. Provinces have responsibility mainly in the areas of education (maintenance and repair of schools and office buildings), road cleaning and maintenance in non-metropolitan areas, environmental protection in terms of measurement and monitoring of air and water pollution, waste dumps and sewage systems, and finally in planning, coordinating and providing technical support to municipalities’ policies. Over

3 4

of current provincial expenditure is funded by grants from upper

levels of government (state and regions).13 State grants are either specific or 1 2 The 13 A

Wooldridge [40] procedure is discussed in detail in Revelli [32]. small fraction of grants - less than 1% for most provinces - is funded by the EU.


general, the latter aiming at equalizing inter-provincial structural differences in spending needs and fiscal capacity. Provinces are divided into two demographic bands based on a 400, 000 resident population threshold, and grants are computed according to band-average service cost indices and tax bases. In principle, state support only concerns a number of mandated provincial functions, while expenditures on non-mandated services must be entirely funded by own revenues. In practice, the presence of a number of distinct state grant programs and the complexity of the distribution rules tend to make the overall amount of grants the outcome of state-province bargaining.14 On the other hand, regional grants typically finance specific functions that were devolved to the provinces during the decentralization reform started in the late 1990s. While the devolution process follows nationwide rules, the regions have a considerable degree of discretion in the delegation of functions to the provinces as well as in the quantification of the resulting spending needs. The rest of current spending is funded by three own tax revenue sources: the vehicle registration tax, the electricity consumption tax, and the waste management tax. The vehicle registration tax represents over 50% of own tax revenues. All brand new vehicles and used vehicles in case of change of ownership are liable to the payment of the tax the first time they are registered in a provincial archive under a given owner’s name.15 As shown in table 1, central government establishes a lower and an upper bound on the vehicle tax parameters that provinces can set, with the upper bound corresponding to a 20% higher tax burden (raised to 30% in 2007) than the one corresponding to the lower bound. Consequently, the decision of each province consists in determining autonomously the surcharge rate τ v . Second, the electricity consumption tax is applied by provinces on business uses of electricity. As shown in table 1, provinces set a tax rate τ e between a statewide lower limit of 0.093 and an upper limit of 0.114 Euro per kWh. Electricity tax revenues make around 1 4 The

1 3

grant endogeneity issue is dealt with in section 6 below. total tax due is made of a lump-sum amount plus a variable component that is related to the size, power and destination of the vehicle. 1 5 The


of total own tax revenues. Finally, the waste management tax is a surcharge applied by provinces on the waste collection bill charged by the municipalities located in the province on all households and businesses. Table 1 shows that the surcharge τ w must lie between 1% and 5% of the municipal levy. Revenues from the waste management tax amount to less than 20% of total provincial own tax revenues. Table 2 reports the number of authorities hitting lower and upper limits respectively, the latter showing a considerable increase along the fiscal consolidation process, while table 3 rates the authority-year observations based on how severely they are affected by the tax limitations.16 More than half of the observations in the dataset (416 out of 720) correspond to fully bound instances, with all available tax sources being set at left or right corners, while in only 9 observations none of the constraints is binding. For about 40% of the observations either one or two tax limitations are binding. Interestingly, in over

1 3


the observed tax mix outcomes one lower and one upper limit are binding at the same time. In terms of the theoretical model of section 2, and while we cannot accurately measure the vector of income flows upon which the welfare of a community depends, it seems indeed plausible that the costs of raising revenues across the three available provincial tax sources are different. Taxes on business uses of energy, vehicle registrations and waste service consumption tend to have an heterogeneous impact on business income depending on firms’ sector of activity and factor input mix (large versus small, energy-intensive versus labor-intensive), as well as on the business versus household sector in general, or on wealthy versus poor households. This implies that different tax vectors will likely generate different welfare losses and degrees of political opposition to taxation, thus explaining the diversity of observed tax mix outcomes and the lack of fungibility between alternative tax revenue sources. 1 6 The data refer to the 90 provinces (out of 103) for which all information from 2000 to 2007 is available.



Empirical implementation


Time-invariant sample separation

The sample is first split based on a time-invariant indicator Kn that equals 1 if province n is constrained on all own tax revenue sources for the entire period of observation, and equals 0 if the authority is never observed to have all constraints binding. Application of the splitting criterion (11) yields Kn = 1 for 24 provincial authorities, and Kn = 0 for 20 authorities in the 2001-2006 period, the rest of the observations being discarded (to be used later on) because of changing regime during the period. This leaves us with 264 observations.17 Of the 24 structurally capped authorities, 17 were at the upper bounds on all three own tax rates for the entire period, five were hitting two upper bounds and one lower bound, one was at one upper and two lower bounds, and one province was consistently at the three lower bounds. On the other hand, the authorities in the Kn = 0 regime have one to two binding constraints. We first estimate the switching regression model (9)-(10)-(11) as a single equation, with Kn working as a switcher, thus allowing us to test the difference between the β 0 and β1 coefficient vectors: £ ¤ znt = q0nt β0 +Kn ×q0nt (β1 −β0 )+ζ 0n +η0nt +Knt × (ζ 1n + η1nt ) − (ζ 0n + η 0nt ) (14)

where znt equals current spending per capita in real terms, and the vector of

explanatory variables qnt includes grants per capita (all current financial transfers from upper levels of government), proxies for private income flows in the province (tax base indicators for the three own provincial revenue sources and provincial gross domestic product at market prices), and a set of provincial characteristics: population size to control for economies of scale in service provision; demographic composition of the resident population (share of the population aged 0 to 4 years and aged over 65 years); a binary election year indicator to 1 7 In order to preserve the size of the K = 1 sample, it seems sensible to exclude for now n the last year in the sample because of the vehicle tax cap relaxation that occurred in 2007 (from 20% to 30%). Similarly, the first year (2000) is excluded since several provinces became consistently capped from the year 2001 on.


allow for opportunistic policy manoeuvring prior to elections;18 and a rightwing control dummy to capture a partisan cleavage in spending policy between right-wing and left-wing governments. As for tax bases, since official figures are not formally reported either by provincial governments or by the Ministry, they need to be recollected from other sources. The national motorvehicle registry system (PRA, Pubblico Registro Automobilistico) publishes annual data for all new vehicle registrations by province, and the electricity grid company (TERNA, Rete Elettrica Nazionale) releases yearly data on domestic and business electricity use by province. As for provincial surcharges on municipal waste collection bills, average city waste collection tax payments are available from the national statistics institute (ISTAT, Istituto Nazionale di Statistica). In fact, while constituting reasonable proxies for the level of resources available within provinces, none of those measures accurately reflects the actual tax bases. First, the nationwide vehicle registration tax on which provinces set the 0-20% surcharge is based on a non-linear formula related to the type, engine power and destination of the vehicle. I take the total number of registered vehicles in a province and transform it into a monetary tax base by multiplying it by the average baseline registration tax payment. As for electricity, the available total business consumption data do not equal the actual tax bases because large energy-intensive plants exceeding 300,000 kWh consumption per month are exempt from the provincial charge, meaning that we will overstate the tax base in provinces where larger plants are located. Finally, municipal waste collection tax payments are available for provinces’ main cities only, and actual provincial surcharge revenues tend to differ from theoretical revenues because of endemic low compliance. The fixed effects estimation results of equation (14) are reported in table 4, while table 5 reports the separate estimation results of equations (9) and (10) for the two subsamples. All equations include year dummies. Descriptive statistics 1 8 Provincial elections take place every five years with direct election of the president. The election schedule is asymmetric, meaning that provinces hold elections at different points in time.


and data sources for all variables are reported in Appendix B. Interestingly, all authorities exhibit what would be termed a flypaper effect according to conventional criteria in the literature. The results in tables 4 and 5 show that the grant effect is large and highly statistically significant. In fact, the Kn = 0 subsample is far from being unconstrained in practice, given that it groups authorities that are indeed against one or two tax limits. However, the results also show that authorities that are fully constrained react to grants to a significantly larger extent, actually on a one-for-one basis. The estimate of the effect of grants on spending is around 0.7 for the moderately constrained subsample (columns (4.2), (4.4) and (4.6) in table 4; columns (5.2), (5.4) and (5.6) in table 5), while the coefficient estimate virtually equals 1 for structurally bound provinces (columns (5.1), (5.3) and (5.5) in table 5). The difference among the two coefficients (over 0.2) is highly statistically significant. The large and significant difference in the response to grants in the two subsamples is robust to the introduction of various controls, none of which - including provincial GDP and demographic and political characteristics - contributes much to further explaining the pattern of spending. In particular, the estimated coefficients on the own tax bases all have the expected positive sign and generally plausible magnitudes that are compatible with the statutory tax rate admissible range, though they are hardly statistically significant.


The behavior of tax rates

The theoretical model of section 2 suggests that the one-for-one sensitivity of fully constrained authorities’ expenditures to grants arises from the fact that they cannot manoeuvre their own tax rates in an attempt to offset changes in transfers. The difference that emerges above as far as non-fully constrained authorities are concerned ought then to be due to their ability to purposefully use their tax policy to smooth out expenditures in front of year to year changes in state transfers. In fact, the authorities in the Kn = 0 sample changed their tax rates pretty frequently during the period, with over thirty tax rate increases over the 2001-2006 years. Interestingly, grants to those authorities fell by around 20

10% in real terms between 2001 and 2006. I can therefore investigate here if those tax rate changes can be explained as offsetting responses of own revenue raising policy to widespread state retrenchment. I take the provincial tax rates as the dependent variables and estimate the impact of grants and tax bases on the Kn = 0 sample, while controlling for a set of local characteristics as well as year and province fixed effects. The results are reported in table 6. With the only exception of the waste management tax rate in column (6.3), whose grant coefficient is estimated imprecisely, grants turn out to have a strong negative impact on own provincial tax rates, as suggested by the theoretical model (equation (27) in Appendix A). The fiscal policy changes observed over the decade seem therefore to be interpretable as offsetting responses to changes in state policy, suggesting that state retrenchment was responsible for the secular upward shift in provincial tax rates documented in table 2.19 The estimated coefficients on the provincial tax bases are more mixed. Vehicle registration and business electricity tax bases have insignificant effects on their respective and cross tax rates, while the waste management surcharge rate appears to be negatively affected by the municipal waste management charge base. The latter effect might seem at odds with the theoretical results in Appendix A, that predict positive effects of tax bases on respective tax rates (equation (30)). However, an increase in the municipal waste charge revenues on which provinces apply their own surcharge actually amounts to a fall in residents’ disposable income, and is likely to raise the marginal cost for provinces of collecting further revenues from the waste management bill - a sort of negative fiscal externality arising from tax base co-occupation by municipalities and provinces (Keen and Kotsogiannis, 2002). In fact, the results in table 6 suggest that an higher cost of raising revenues from the waste management surcharge twists instead the provincial tax mix towards increased reliance on business electricity taxes. 1 9 Similar results (negative and significant impact of grants on vehicle and electricity tax rates, and no effect on waste management tax rate) emerge when estimating the tax rate equations on the unbalanced panel of 304 moderately constrained observations of table 3.



Time-varying sample separation

Table 7 reports the estimation results of the spending determination model based on the time-varying index (12). I first use all available sample observations (as summarized in tables 2 and 3), and estimate a single equation that allows for interaction terms depending on the corresponding tax mix of provincial authorities, and a unique set of authority fixed effects. Column (7.1) presents the estimation results when pooling all observations irrespective of their tax mix. The estimated coefficient on grants is around 0.8, roughly the same figure as in table 4. The vehicle registration and electricity consumption tax base coefficients are significant, and their sizes are compatible with the statutory tax limits (reported in table 1), while that on the waste management base is virtually zero, again a similar result as in the restricted sample of table 4. The second column in table 7 shows the estimation results when grants are interacted with the dummy Knt as defined in (12). It turns out that authorities in the Knt = 1 regime - where all limits are binding - have a significantly higher sensitivity to grants than authorities that can manoeuvre at least one revenue source. However, the difference in behavior between the two regimes is considerably smaller than in the time-invariant sample split: fully constrained authorities’ expenditures are only estimated to have a 5% higher response to grants (standard error = 0.02) than non-fully constrained ones. This might be due to some of the two-limit authorities actually being pretty close to a fully constrained regime, and basically mimicking the behavior of fully-bound authorities. In order to verify if that is the case, and to further test whether the sensitivity of spending to grants actually increases with the number of binding limits, I partition observations into three groups: three binding constraints (Knt = 1), two binding constraints (K(2)nt = 1) and one or no binding constraint, the latter constituting the reference group. I then interact these group dummies with the grant variable, and estimate the equation with year and authority fixed effects in columns (7.3) and (7.4). In this case, it turns out that public expenditures in localities where two or three limits are binding are sig-


nificantly more sensitive to grants than where one or no limit binds, and the difference is around 0.2, a similar figure as in the time-invariant sample split of table 4. On the other hand, there seems to be no significant difference between authorities that have two or three binding rate limits. In columns (7.5) and (7.6), I proceed to a finer partition of the dataset in order to investigate if the specific features of the tax mix affect the response of provincial expenditures to grants. In particular, I test for different responses of authorities with an (hv , he , hw ) tax mix relative to those that are indeed fully constrained, but are against at least one lower bound: in most instances, the lower bound refers to the electricity tax, though there are cases (as shown in table 3) of binding lower constraints for vehicle and waste management taxes too. Of course, the theoretical model predicts that fully constrained authorities should exhibit a one-for-one response to grants, irrespective of whether they are hitting upper or lower bounds. The results in columns (7.5) and (7.6) suggest that all authorities in a fully constrained tax mix have a very large response almost one-for-one - to grants, while it does not make any difference whether they are against upper or lower bounds. In table 8, I present the estimation results of an expenditure determination equation that focuses on fully constrained authorities, with the aim of verifying whether those authorities respond differently to changes in tax bases depending on the specific features of their constrained tax mix. In fact, their expenditures’ response to tax bases should differ according to whether they are at lower or upper bounds, with the estimated coefficients equaling the respective binding tax rates (as shown by equations (32) and (33) in Appendix A). In order to have an as focused test as possible, I take the sample of the 238 observations at an (hv , he , hw ) tax mix, and the sample of 130 observations with (hv , le , hw ), the only tax mix difference among the two samples being that the former is upper-constrained on energy, and the latter is lower-constrained, as shown in table 3. These are the two most frequent tax mix outcomes in the dataset. Table 8 first reports estimates of the spending equation when pooling the two


samples in a single equation (columns (8.1) and (8.2)). The next two columns allow for heterogeneous effects from grants and energy tax bases by introducing interaction terms via a dummy that equals 1 if the observation corresponds to a fully upper-constrained tax mix.20 The results in the pooled sample of columns (8.1) and (8.2) show that the grant coefficient estimate is above 0.9. Energy and vehicle registration tax bases have positive significant effects on spending, and the estimated coefficients are within (vehicle tax) or close to (energy tax) the range of admissible rates. The energy tax base coefficient is in fact slightly below the lower bound of 0.093. The vehicle tax coefficient estimate is about half as it could be expected to be given that all authorities in the sample are at the 0.20 cap. The waste management tax base has no significant impact, a result in line with the evidence presented above and likely attributable to the poor measurement of the waste management tax base. Among the controls that are added along with tax bases in column (8.2), only provincial income shows some small positive effect. When allowing for heterogeneous responses as in columns (8.3) and (8.4), there is no evidence of significant differences among the two samples in their sensitivity to grants. Even when constrained at the lower energy tax limit, local authorities react to grants on an almost one-for-one basis. Authorities at the lower energy bound have a grant coefficient of 0.95, an even higher point estimate than the one obtained for fully upper capped authorities, though the difference among the two coefficients is not statistically significant. As for the estimated response of spending to the energy tax base, it turns out to be higher and significantly so when the upper limit he binds than when the lower limit le binds. Though somewhat stretched, the magnitude of the difference between the estimated energy tax base coefficients - 0.08 for lower-constrained authorities and 0.14 for upper-constrained in column (8.3); 0.07 and 0.13 respectively in column (8.4) - is compatible with the statutory limits. 2 0 No further significant differences emerge when allowing for heterogeneous effects from the other tax bases or control variables.



Endogeneity issues

One might wonder at this point whether the high sensitivity of local spending to grants - an almost one-for-one response for fully constrained authorities - is in fact determined by spurious correlation between expenditures and transfers due to omitted variables driving both.21 In the context we deal with here, shocks to provincial expenditures - due, say, to natural disasters or major infrastructure works - might simultaneously boost state grants to tackle rising spending needs, and induce local authorities to raise own taxes up to the rate limits. I tackle these endogeneity issues in the next two sections. First, in section 6.1 I exploit the institutional features of the Italian multi-tiered structure of government, and adopt an instrumental variables approach relying on within-province grant variability due to changes in variables that can be plausibly thought to have no independent effect on provincial spending. Second, I explicitly model endogenous selection into the fully constrained regime in section 6.2.


Discontinuity, ideology, and advocacy

As argued in section 4 above, a fraction of state general grants is distributed to provinces according to a formula where localities are split into two demographic bands (≷ 400, 000 inhabitants) in recognition of the specific features that tend to be typical of larger, metropolitan provinces relative to smaller, rural ones. The two-band system creates a discontinuity at the 400, 000 population threshold, with provinces happening to cross the threshold facing a different reference group against which their spending needs are evaluated.22 We therefore build a dummy variable equaling 1 if the population of a province exceeds the threshold in a given year, and use it as an instrument for grants. 2 1 Recent research actually finds little evidence of a flypaper effect when grant endogeneity arising from a number of sources is explicitly and properly accounted for (Knight [27], Gordon [16], Lutz [28]). 2 2 The argument is similar to Gordon [16] and Dahlberg et al. [11]. Gordon [16] exploits the infrequent updating of poverty data used in the US federal education grants to school districts (Title I), and uses a purely Census-determined grant change measure as an instrument for actual Title I revenue change. Dahlberg et al. [11] make use of a discontinuity in the grant formula for municipalities in Sweden, where localities with a net out-migration rate above a state-set threshold are entitled to extra grants.


Second, a far from negligible share of grants to the provinces (over 13 ) come from the regions, and are intended to fund specific administrative functions that the latter delegate to the former. While the delegation process abides to general national rules, the regions have a substantial degree of discretion in implementing and quantifying it, and it seems reasonable to allow the ideology of the regional governments to affect their policy design towards the provinces. In particular, based on the idea that the political complexion of the regional government might affect the size of grants flowing down to the provinces located within the region, while not directly influencing provincial expenditures, we use a right-wing regional government dummy as an instrument for grants.23 Finally, in spite of the undisputed national parliament supremacy in the Italian multi-tiered government structure, all laws concerning subnational administrative or financial issues, including local public service organization, management and financing, need to be preliminarily discussed in a state-local governments committee before final approval.24 The committee was established in 1996 to foster dialogue and cooperation between central and local governments. It can make recommendations, submit proposals or require amendments to state acts to be discussed in parliament. Besides central government representatives (typically the Finance, Regions, Infrastructures and Interior Ministers), the committee is composed of 14 representatives from the municipalities and 7 representatives from the provinces, and meets regularly during the year.25 The province representatives in the committee include the president of the national union of the Italian provinces26 (that is normally elected every five years among the province presidents) and six province presidents that are nominated by the union itself and sit in the committee for five years (unless they decade earlier from office). Province delegates tend to reflect the demographic (large metropolitan versus small rural), political (right-wing versus left-wing) and geographic 2 3 Arulampalam et al. [2] and Solé-Ollé and Sorribas-Navarro [35] show that alignment with grantors and vote swing can affect the size of grants. However, this does not appear to be the case as far as Italian provinces are concerned. 2 4 Conferenza stato-città enti locali ( 2 5 A parallel committee exists for state-regions issues (Conferenza stato-regioni ). 2 6 U.P.I., Unione delle Province d’Italia (


(North, South and Centre) diversity of the province universe.27 In practice, the union criteria for selecting its representatives in the state-local committee are hard to decipher, and plausibly seem to follow some crude rotation principle. Of course, and similarly to US state congressional delegations (Knight [27]), sitting in the state-local governments committee can enhance the “political power” of delegates to advocate their home province needs and interests. In the light of the virtually random process of committee member selection, we build an advocacy dummy variable equaling 1 if the president of a province sits in the state-local governments committee in a given year, and use it as an instrument for grants. The instrumental variables estimation results are reported in table 9. I focus here on the sample of authorities that are observed to switch from one tax mix regime to the other over the panel length. This allows me to test both the effect of grants on spending behavior conditional on observed tax mix regimes (table 9), and subsequently on the probability that an authority moves into a fully constrained tax mix (table 10). After excluding provinces that are never observed to hit all tax limits, as well as those that are observed in the Knt = 1 or Knt = 0 regime for less than two years, I end up with a balanced panel of 40 “switching” provinces over the eight years 2000-2007, with 188 observations in Knt = 1 and 132 observations in Knt = 0. Importantly, since I aim at estimating the response of authority n’s spending to grants in year t provided that authority n stays on the same portion of its budget constraint, i.e., it does not jump to a different segment of its kinky budget constraint by, say, raising a tax rate from a lower to an upper limit, I require the tax mix of authority n in year t to be identical as in year t − 1 in order for that observation to be selected into the Knt = 1 regime (Bond and Meghir [5]).28

Column (9.1) reports OLS estimates of the spending equation on the whole 2 7 Over the years, the commitee included delegates from huge metropolitan provinces (Milan and Rome, with around 4 million inhabitants) as well as small ones such as Brindisi in the South (<400,000 inhabitants), Trieste in the North-East (<250,000) and Rieti in the Centre (<150,000). 2 8 This implies that I cannot exploit here the changes in the K nt selection index that are generated by the exogenous vehicle tax rate cap relaxation that occurred nationwide in the last sample year.


sample of 40 switching provinces for comparison. The first stage results of the 2SLS approach are reported in the second column and reveal that two of the three instruments discussed above play a significant role in explaining grants. The population threshold dummy has no significant effect, though: this is likely due to the fact that only a tiny fraction of provinces actually crosses the 400, 000 threshold in the period considered. Moreover, since provinces tend to approach the population threshold towards the later sample years, the effects in terms of grant entitlements are likely to be observed after the end of the observation period. As far as the other instruments are concerned, provinces turn out to receive less grants when the regional government is from a right-wing coalition: switching from an extravagant left-wing regional government to a thrifty rightwing one is estimated to lead to a fall in grants to provinces located within the region of over 10% on average. On the other hand, it turns out that sitting in the state-local government committee indeed attracts significant additional resources to a province: on average, gaining a seat in the committee amounts to an almost 20% boost in grants to a province. Overall, the three instruments are valid and jointly significant in the first stage whether or not the other exogenous regressors are included, with an F test of over 11 in panel 9A, and over 14 in panel 9B. The second stage results are reported in column (9.3) for the whole sample, and in columns (9.4) and (9.5) for the fully constrained and moderately constrained samples respectively. When instrumented, the grant coefficient falls modestly in the overall sample, while remaining pretty close to 1 under the Knt = 1 regime. The gap among the estimates of the expenditure sensitivity to grants in the two regimes remains at around 0.2. Overall, allowing for endogeneity of grants still delivers an impressively high estimate of the spending sensitivity to grants in the Knt = 0 subsample (β 0g > 0.7), suggesting that binding tax limits might not be the sole source of excess sensitivity of local public expenditure to grants.



Endogenous selection

As far as the issue of endogenous selection into the Knt = 1 regime is concerned, table 10 reports the estimation results of the Wooldridge [40] two-stage approach. In order to focus on the forces pushing local authorities towards the upper tax rate limits, I perform the Wooldridge procedure on the 27 local authorities that: a) were not fully constrained for at least two periods, meaning that they could manoeuvre at least one of their own revenue sources, and: b) switched at some point to the fully upper-constrained regime (hv , he , hw ) and were observed there for at least two periods. Probit estimation of the first-stage binary selection equation (13) - column (10.1) - reveals that grant policy indeed played an important role in driving selection into the fully upper-constrained regime: the effect of grants on the probability of being in a (hv , he , hw ) tax mix is negative and highly statistically significant. On the other hand, as far as the stochastic component of equation (13) is concerned, the second stage estimation results suggest that the selection process can be considered exogenous with respect to the local public spending pattern. The Wooldridge [40] variable addition test reported at the bottom of table 10 is far from statistical significance, and column (10.2) reveals that performing the Wooldridge [40] correction has a negligible impact on the estimation results, including in particular the excess sensitivity of local public spending to grants. Finally, as argued above, one might believe that treating grants as exogenous in the selection equation be illegitimate, due to the possibility of shocks (say, a plant relocation) driving local authorities to corner solutions and simultaneously soliciting grants from upper level governments. In order to test whether this is the case, I regress grants on the three instruments discussed in section 6.1 (as well as all other exogenous variables in equation (13)) and include the residuals from that estimation in equation (13). The t statistic of the estimated coefficient on the residuals is a test of grant exogeneity in the selection equation (Rivers and Vuong [33]). The t statistic of the estimated coefficient on the residuals from the grant equation takes on the value of around 1 (p value of about 0.3),


suggesting that grants can legitimately be treated as exogenous in the selection equation.


Concluding remarks

By explicitly incorporating the corner solutions that are typically produced by statewide limitations on local tax rates, this paper has modelled the local tax mix determination process, and demonstrated that the so-called flypaper effect arises in the constrained tax mix. In particular, the paper has shown that local expenditures should be predicted to display a one-for-one response to grants in the presence of binding limitations on all local tax revenue sources. Interestingly, the above result holds when either upper or lower tax limitations are in place, and it turns out that a binding limitation on just one of the available own revenue sources is enough to generate a sort of excess sensitivity of local spending to grants. Moreover, such excess sensitivity should be expected to arise - and generally tends to manifest itself - both when grants increase and when they decrease. On the other hand, the response of public spending to a shock to a local tax base whose rate is constrained turns out to be a function of the (lower or upper) binding rate limit. The key empirical predictions of the model in terms of local public spending and tax rate sensitivity to grants and own tax bases have been tested on panel data on the Italian provinces through the years 2000s. The Italian provinces’ data allow us to exploit the sharp corner solutions generated by central government upper and lower limitations on own sources of tax revenue (a tax on vehicle registrations, a tax on electricity consumption for business uses, and a waste management surcharge). I have employed a switching regression approach where local authorities are assigned to either of two subsamples based on the intensity to which central tax limits bind, and estimated the response of local public expenditures to grants and own tax bases in the two subsamples. Whether the sample is split according to a time-invariant criterion or to a time-varying one, the empirical evidence consistently suggests that the reaction


of local spending to grants is significantly higher - actually, one-for-one - for fully constrained authorities than for authorities that can manoeuvre at least one of the local tax instruments. Those results turn out to be robust when endogenous selection into the constrained sample is controlled for, and when we allow grants to be determined endogenously. In particular, we employ a set of powerful institutionally-driven instruments related to exogenous discontinuities in the grant distribution formula, ideological complexion of upper level (regional) governments, and political power of province delegates sitting in the state-local governments advisory committee. While the results on the response of local public expenditures to own tax bases are less clear-cut, they generally provide a coherent picture of provincial expenditures being tied to the evolution of local resources via the binding upper or lower tax rates. Finally, the empirical evidence on the negative effect of grants on local tax rates for authorities that are not fully constrained provides support to the view that the fiscal policy changes observed over the decade can indeed be interpreted as offsetting responses to changes in state policy in a period of retrenchment, while fully constrained authorities sort of mechanically react to grants one-for-one. Overall, the results in this paper suggest that statewide limitations on local governments’ tax policies - or, more generally, the ample role of diverse forms of central command, including tax base assessment, general revenue limitation, and local public service mandates - ought not to be ignored when investigating the extent to which central funding crowds out local tax effort, and when interpreting the empirical evidence on local government response to state grants. In particular, the empirical phenomenon that has been documented for decades and has been conventionally - and somewhat incorrectly - interpreted as arising from stickiness of federal transfers (the so-called flypaper effect) seems instead to be best described as an excess sensitivity of local public spending to grants that cannot in general be taken as a symptom of decentralized government overspending.


References [1] Anderson, N., Property tax limitations: An interpretative review, National Tax Journal 59 (2006) 685-694. [2] Arulampalam, W., Dasgupta, S., Dhillon, A., Dutta, B., Electoral goals and center-state transfers: A theoretical model and empirical evidence from India, Journal of Development Economics 88 (2009) 103-119. [3] Baicker, K., Government decision-making and the incidence of federal mandates, Journal of Public Economics 82 (2001) 147-194. [4] Becker, G., Mulligan, C., Deadweight costs and the size of government, Journal of Law and Economics 46 (2003) 293-340. [5] Bond, S., Meghir, C., Dynamic investment models and the firms’ financial policy, Review of Economic Studies 61 (1994) 197-222. [6] Borge, L-E., Tovmo, P., Myopic or constrained by balanced-budget rules? The intertemporal spending behavior of Norwegian local governments, FinanzArchiv: Public Finance Analysis, 65 (2009) 200-219. [7] Brooks, L., Phillips, J., An institutional explanation for the stickiness of federal grants, Journal of Law, Economics and Organization, 26 (2010) 243-264. [8] Calabrese, S., Epple, D., On the political economy of tax limits, Institut d’Economia de Barcelona Working Paper No. 2010/14. [9] Charlier, E., Melenberg, B.,Van Soest, A., An analysis of housing expenditure using semiparametric models and panel data, Journal of Econometrics 101 (2001) 71-107. [10] Cummins, J., Hassett, K., Oliner, S., Investment behavior, observable expectations, and internal funds, American Economic Review 96 (2006) 796810.


[11] Dahlberg, M., Mork, E., Rattso, J., Agren, H., Using a discontinuous grant rule to identify the effect of grants on local taxes and spending, Journal of Public Economics 92 (2008) 2320-2335. [12] Dahlby, B., The marginal cost of public funds and the flypaper effect, International Tax and Public Finance 18 (2011) 304-321. [13] Fazzari, S., Hubbard, R., Petersen, B., Financing constraints and corporate investment, Brookings Papers on Economic Activity 1 (1988) 141-195. [14] Flavin, M., The adjustment of consumption to changing expectations about future income, Journal of Political Economy 89 (1981) 974-1009. [15] Gamkhar, S., Oates, W., Asymmetries in the response to increases and decreases in intergovernmental grants: Some empirical findings, National Tax Journal 49 (1996) 501-512. [16] Gordon, N., Do federal grants boost school spending? Evidence from Title I, Journal of Public Economics 88 (2004) 1771-1792. [17] Hamilton, J., The flypaper effect and the deadweight loss from taxation, Journal of Urban Economics 19 (1986) 148-155. [18] Hettich, W., Winer, S., Economic and political foundations of tax structure, American Economic Review 78 (1988) 701-712. [19] Hettich, W., Winer, S., A positive model of tax structure, Journal of Public Economics 24 (1984) 67-87. [20] Hines, J., Thaler, R., Anomalies. The flypaper effect, Journal of Economic Perspectives 9 (1995) 217-226. [21] Hu, X., Schiantarelli, F., Investment and capital market imperfections: A switching regression approach using US firm panel data, Review of Economics and Statistics 80 (1998) 466-479.


[22] Inman, R., Flypaper effect, in: S. Durlauf and L. Blume (Eds.) The New Palgrave Dictionary of Economics Online, Palgrave Macmillan, 2009. [23] Jappelli, T., Pischke, J-S., Souledes, N., Testing for liquidity constraints in Euler equations with complementary data sources, Review of Economics and Statistics 80 (1998) 251-262. [24] Joumard, I., Kongsrud, M., Fiscal relations across government levels, OECD Economics Department Working Paper 375 (2003). [25] Kaplan, S., Zingales, L., Do investment-cash flow sensitivities provide useful measures of financing constraints? Quarterly Journal of Economics 112 (1997) 169-215. [26] Keen, M., Kotsogiannis, C., Does federalism lead to excessively high tax rates? American Economic Review 92 (2002) 363-370. [27] Knight, B., Endogenous federal grants and crowd-out of state government spending: Theory and evidence from the federal highway program, American Economic Review 92 (2002) 71-92. [28] Lutz, B., Taxation with representation: Intergovernmental grants in a plebiscite democracy, Review of Economics and Statistics 92 (2010) 316332. [29] Nechyba, T., Local property and state income taxes: The role of interjurisdictional competition and collusion, Journal of Political Economy 105 (1997) 351-384. [30] Payne, A., Does government funding change behavior? An empirical analysis of crowd-out, Tax Policy and the Economy 23 (2009) 159-184. [31] Pudney, S., Modelling Individual Choice. The Econometrics of Corners, Kinks and Holes. Blackwell, Oxford, 1991. [32] Revelli, F., Tax mix corners and other kinks, Institut d’Economia de Barcelona, Working Paper No. 2010/50. 34

[33] Rivers, D., Vuong, Q., Limited information estimators and exogeneity tests for simultaneous probit models, Journal of Econometrics 39 (1988) 347-366. [34] Runkle, D., Liquidity constraints and the permanent income hypothesis: Evidence from panel data, Journal of Monetary Economics 27 (1991) 7398. [35] Solé-Ollé, A., Sorribas-Navarro, P., The effects of partisan alignment on the allocation of intergovernmental transfers. Differences-in-differences estimates for Spain, Journal of Public Economics 92 (2008) 2302-2319. [36] Stine, W., Is local government revenue response to federal aid symmetrical? Evidence from Pennsylvania county governments in an era of retrenchment, National Tax Journal 47 (1994) 799-816. [37] Sutherland, D., Price, R., Joumard, I., Sub-central government fiscal rules, OECD Economic Studies 41 (2005) 141-181. [38] Vigdor, J., Other people’s taxes: Nonresident voters and statewide limitation of local government, Journal of Law and Economics 47 (2004) 453-476. [39] Wolman, H., McManmon, R., Bell, M., Brunori, D., Comparing local government autonomy across states, Working Paper No. 35/2008, George Washington Institute for Public Policy, Washington. [40] Wooldridge, J., Selection corrections for panel data models under conditional mean independence assumptions, Journal of Econometrics 68 (1995) 115-132. [41] Zanarone, G., Vertical restraints and the law: Evidence from automobile franchising, Journal of Law and Economics 52 (2009) 691-700. [42] Zeldes, S., Consumption and liquidity constraints: An empirical investigation, Journal of Political Economy 97 (1989) 305-346.


Appendix A: Model The one-tax case Consider an unconstrained government with welfare function (1), budget constraint (2), and optimal spending and tax rate given by equations (3) and (4) respectively. Straightforward derivation of (3) and (4) shows that the effects of gn and in on τ ∗n and zn∗ are respectively: ρ 1 ∂τ ∗n =− n <0 ∂gn 1 + ρn in


ρn gn ∂τ ∗n = >0 ∂in 1 + ρn i2n


∂z ∗ 1 ∂zn∗ = n = >0 ∂gn ∂in 1 + ρn


Consider now the constrained optimization problem given by (1), (2) and (5). h i Letting λ0 = λh λl ≥ 0 be the vector of Lagrange multipliers, the Lagrangian function is:

L(τ n , λ) = wn + λh (h − τ n ) + λl (−l + τ n ) and the necessary Kuhn-Tucker conditions are:

∂L(τ n ,λ) ∂τ n


= 0, ∇λ L(τ n , λ) ≥ 0,

λ∇λ L(τ n , λ) = 0. Trivially, the optimal level of spending equals zn∗ (h) = gn + hin if local government n is against the upper bound h (λh > 0; λl = 0), and zn∗ (l) = gn + lin if it is against the lower bound l (λl > 0; λh = 0). With τ ∗n stuck at either the lower or the upper limit, the effects of gn and in on zn∗ are easily found as: ∂zn∗ (h) ∂z ∗ (l) = n =1 ∂gn ∂gn


∂zn∗ (h) =h ∂in


∂zn∗ (l) =l ∂in


Whether the upper or the lower tax rate limit in (5) binds, the sensitivity of local public spending to grants in the constrained optimum is 1. The sensitivity of local public spending to own resources in the constrained optimum equals the binding tax rate limit. 36

The multiple-tax case Consider the constrained optimization problem given by (7), (6) and (8). Letting h i λ0 = λh1 ... λhM λl1 ... λlM ≥ 0 be the vector of Lagrange multipliers, the Lagrangian function is:

L(τ n , λ) = wn +



λhm (hm − τ m n)+


λlm (−lm + τ m n)



and the necessary Kuhn-Tucker conditions are: ∇τ L(τ n , λ) = 0, ∇λ L(τ n , λ) ≥ ³ ´ ³ ´ lm hm 0, λ∇λ L(τ n , λ) = 0. Denoting by Ln ≡ τ m > 0 and Hn ≡ τ m >0 n |λ n |λ the sets of tax rates hitting the lower and upper bounds respectively, and ³ ´ lm by Un ≡ τ m = λhm = 0 the set of tax rates lying strictly between the n |λ

bounds, the constrained tax rate mix (m = 1, ....M ) is:  lm     hm  if     m∗ τ n (l, h) = P r  ρm    n  ψ n 1 + ρn  − im Γn    n  r∈Un 

m ∈ Ln m ∈ Hn


m ∈ Un


£ £ ¤ ¤ where: l0 = l1 , ..., lM , h0 = h1 , ..., hM , and: ψn ≡


1 P


Γn ≡ gn +


irn +

r∈Un r6=m




lr irn +



hr irn



From (23) and (6), the level of spending in the constrained optimum is: Ã ! P m P mm P mm ∗ in + l in + h in (26) zn (l, h) = ψ n gn + m∈Un

zn∗ :



gn and irn are easily found to have the following effects on τ m∗ n (m ∈ Un ) and ∂τ m∗ ∂τ m∗ ρm n n n = = −ψ n m < 0 if r ∈ Un r ∂gn ∂in in


ρm ∂τ m∗ n n r = −ψ n m l < 0 if r ∈ Ln r ∂in in


ρm ∂τ m∗ n n r = −ψ h < 0 if r ∈ Hn n ∂irn im n



∂τ m∗ ρm n = ψ n n 2 Γn > 0 m ∂in (im n)


∂zn∗ ∂z ∗ = rn = ψ n > 0 if r ∈ Un ∂gn ∂in


∂zn∗ = lr ψ n > 0 if r ∈ Ln ∂irn


∂zn∗ = hr ψ n > 0 if r ∈ Hn ∂irn



Appendix B: Data description Table A1 Variables used in the analysis: descriptive statistics

Vehicle registration tax rate (%) Electricity consumption tax rate (€ per kWh) Waste management tax rate (%) Vehicle registration tax base per capita (€) Electricity consumption tax base per capita (kWh) Waste management tax base per capita (€) Income (GDP per capita ,000 €) Population (,000) Aged 0-4 share Aged 65+ share Real current spending per capita (€) Real grants per capita (€) Election year (%) Right-wing control (%): province Right-wing control (%): region

obs. 720 720 720 720 720 720 720 720 720 720 720 720 720 720 136

mean 17.7 0.104 4.5 154.3 122.1 219.9 20.2 567.9 4.4 20.4 146.1 118.2 15.4 33.6 50.4

s.d. 7.2 0.010 1.1 46.4 35.9 89.9 5.1 631.6 0.6 3.1 46.1 44.7

min 0 0.093 1 51.9 53.4 37.4 9.5 89.0 3.0 12.0 56.9 36.6

Table A2 Variables used in the analysis: data sources

Vehicle registration tax rate & base Electricity consumption tax rate Electricity tax base Waste management tax rate Current spending Grants Election year Right-wing control Income Population & demographics

Data source Automobile Club Italy - PRA Italian Government, Ministry of Finance Terna Rete Elettrica Nazionale Italian Government, Home Office Italian Government, Home Office Italian Government, Home Office Italian Government, Home Office Italian Government, Home Office National Statistics Institute National Statistics Institute


max 30 0.114 5 359.1 279.0 516.6 34.1 4061.5 6.3 27.5 291.9 249.0

Table 1 Lower and upper tax rate limits

Vehicle registration tax rate τ v (% surcharge on national rate) Electricity consumption tax rate τ e (Euro per kWh) Waste management tax rate τ w (% surcharge on municipal levy)

lv hv le he lw hw

2000-6 0 20 0.093 0.114 1 5

2007 0 30 0.093 0.114 1 5

Table 2 Number of authorities (N = 90) at lower (l) and upper (h) limits




l hv le he lw hw

τe τw

2000 25 55 66 16 3 66

2001 15 65 54 29 2 64

2002 9 71 43 39 2 66

2003 7 72 37 45 3 65

2004 7 73 34 47 3 65

2005 5 77 27 52 2 66

2006 4 79 18 59 3 66

2007 3 43 15 64 3 68

Table 3 Tax limitation intensity Fully constrained (h,h,h) (h,h,l)

τv τe τw obs.

hv he hw

hv he lw

hv le hw

total (h,l,l)

lv he hw

238 6 130 0 Moderately constrained (h,h)

τv τe τw

hv he τ w∗



τv τe τw

hv τ e∗ τ w∗




hv le lw

lv he lw

lv le hw

lv le lw








hv τ e∗ hw

τ v∗ he hw

hv le τ w∗

lv he τ w∗

hv τ e∗ lw

lv τ e∗ hw

τ v∗ he lw

τ v∗ le hw

lv le τ w∗

lv τ e∗ lw

τ v∗ le lw

60 (h)



3 (l)








τ v∗ he τ w∗

τ v∗ τ e∗ hw

lv τ e∗ τ w∗

τ v∗ le τ w∗

τ v∗ τ e∗ lw

τ v∗ τ e∗ τ w∗







304 720


Table 4 Time-invariant splitting criterion (Kn )


(4.1) 0.844~ (0.035)

Kn × grants

(4.2) 0.722~ (0.046) 0.253~ (0.069)

(4.3) 0.838~ (0.034)

0.203~ (0.060) 0.057 (0.054) 0.024 (0.015)

(4.4) 0.744~ (0.047) 0.205~ (0.070) 0.224~ (0.088) 0.096 (0.084) 0.019 (0.020)

264 44

264 44

264 44

tax basev tax basee tax basew income population age 0-4 share age 65+ share election right-wing observations authorities

264 44

(4.5) 0.843~ (0.035)

0.215~ (0.065) 0.058 (0.056) 0.023 (0.015) -0.001 (0.001) -0.088 (0.077) 0.561 (6.412) -3.272 (3.334) -0.634 (1.333) 2.375 (2.823) 264 44

(4.6) 0.759~ (0.051) 0.200~ (0.074) 0.183∗ (0.098) 0.049 (0.092) 0.016 (0.020) -0.002 (0.002) -0.227 (0.134) -2.965 (7.784) -2.992 (5.226) -0.664 (1.775) 6.627 (6.615) 264 44

Notes: Dependent variable: real current spending per capita. Fixed province and year effects included; year effects interacted with the switching indicator Kn in columns (4.2), (4.4) and (4.6); exogenous variables interacted with the switching indicator Kn in columns (4.4) and (4.6); Kn defined in equation (11). Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


Table 5 Time-invariant splitting criterion: separate equations








Kn = 1

Kn = 0

Kn = 1

Kn = 0

Kn = 1

Kn = 0

144 24

120 20

144 24

120 20

0.975~ (0.048)

0.722~ (0.050)

0.949~ (0.049) 0.121 (0.079) 0.025 (0.064) 0.036∗ (0.021)

tax basev tax basee tax basew

0.744~ (0.050) 0.224∗ (0.095) 0.096 (0.091) 0.018 (0.021)

income population age 0-4 share age 65+ share election right-wing observations authorities

0.959~ (0.051) 0.133 (0.091) 0.036 (0.068) 0.037∗ (0.022) 0.001 (0.001) -0.032 (0.101) 9.940 (10.434) 0.959 (4.599) -0.172 (1.962) 1.089 (2.993) 144 24

0.759~ (0.055) 0.183∗ (0.105) 0.049 (0.099) 0.016 (0.022) -0.002 (0.002) -0.227 (0.144) -2.965 (8.400) -2.992 (5.639) -0.664 (1.915) 6.627 (7.247) 120 20

Notes: Dependent variable: real current spending per capita. Fixed province and year effects included. Kn defined in equation (11). Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


Table 6 Tax rates
















-0.070 (0.028)

-0.008 (0.003)

-0.003 (0.002)

120 20

120 20

120 20

tax basev tax basee tax basew income population age 0-4 share age 65+ share election right-wing observations authorities

-0.066 (0.032) -0.013 (0.061) 0.097 (0.057) -0.004 (0.013) 0.001 (0.001) -0.071 (0.085) 1.074 (2.173) -1.096 (1.424) 0.889 (1.113) 2.292 (4.041) 120 20


-0.009 (0.003) -0.004 (0.006) -0.004 (0.006) 0.004~ (0.001) -0.001 (0.001) -0.013 (0.008) 0.001 (0.212) 0.013 (0.139) -0.035 (0.108) -0.883∗ (0.393) 120 20

-0.005∗ (0.003) -0.013~ (0.005) -0.001 (0.005) -0.004~ (0.001) 0.001 (0.001) -0.013∗ (0.007) 0.228 (0.176) -0.050 (0.115) 0.033 (0.090) -0.015 (0.327) 120 20

Notes: Dependent variable: tax rates on vehicle registrations, electricity use and waste management charge. Fixed province and year effects included. Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


Table 7 Time-varying splitting criterion


(7.1) 0.864~ (0.016)

(7.2) 0.834~ (0.020) 0.051~ (0.020)

(7.3) 0.686~ (0.037) 0.203~ (0.038) 0.162~ (0.035)

(7.4) 0.683~ (0.038) 0.210~ (0.039) 0.170~ (0.035)

0.057∗ (0.030) 0.101~ (0.032) -0.001 (0.008)

0.055∗ (0.030) 0.103~ (0.032) 0.001 (0.008)

0.066∗ (0.030) 0.107~ (0.031) 0.004 (0.008)

720 90

720 90

720 90

0.068∗ (0.033) 0.109~ (0.033) 0.007 (0.008) 0.001∗ (0.000) 0.027 (0.019) 1.566 (3.003) -1.920 (1.515) 0.911 (0.952) 1.173 (1.748) 720 90

Knt × grants K(2)nt × grants K(h, h, h, )nt × grants tax basev tax basee tax basew income population age 0-4 share age 65+ share election right-wing observations authorities

(7.5) 0.686~ (0.037) 0.205~ (0.040) 0.162~ (0.035) -0.003 (0.025) 0.065∗ (0.030) 0.109~ (0.032) 0.004 (0.008)

720 90

Notes: Dependent variable: real current spending per capita. Fixed province and year effects included. Knt defined in equation (12); K(2)nt = 1 if two constraints bind in year t; K(h, h, h)nt = 1 if three upper constraints bind in year t. Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


(7.6) 0.683~ (0.038) 0.211~ (0.041) 0.170~ (0.035) -0.003 (0.025) 0.067∗ (0.033) 0.111~ (0.034) 0.007 (0.008) 0.001∗ (0.000) 0.027 (0.019) 1.502 (3.012) -1.942 (1.520) 0.926 (0.956) 1.146 (1.753) 720 90

Table 8 Time-varying splitting criterion: upper vs. lower constraints grants

D(he )nt × grants tax basev tax basew tax basee

(8.1) 0.936~ (0.020)

(8.2) 0.944~ (0.020)

0.099∗ (0.042) 0.017 (0.012) 0.084∗ (0.039)

0.113∗ (0.046) 0.020 (0.012) 0.088∗ (0.042)

368 63

0.002∗ (0.001) -0.005 (0.038) 4.799 (4.421) -1.560 (2.036) 0.445 (1.255) 2.925 (2.125) 368 63

D(he )nt × tax basee income population age 0-4 share age 65+ share election right-wing observations authorities

(8.3) 0.947~ (0.025) -0.013 (0.026) 0.104∗ (0.043) 0.016 (0.012) 0.084∗ (0.041) 0.059∗ (0.031)

368 63

(8.4) 0.950~ (0.026) -0.010 (0.026) 0.107∗ (0.047) 0.019 (0.012) 0.075∗ (0.046) 0.057∗ (0.034) 0.002∗ (0.001) -0.006 (0.038) 1.475 (4.729) -2.077 (2.053) 0.404 (1.258) 2.671 (2.152) 368 63

Notes: Dependent variable: real current spending per capita. Fixed province and e year effects included. Knt defined in equation (12); D(h )nt = 1 if upper limit on electricity tax is binding in year t. Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


Table 9 Grant endogeneity: IV approach (9.1) OLS


0.945~ (0.020)

(9.2) 2SLS: first stage

Panel 9A


(9.4) (9.5) 2SLS: second stage

Knt = 1

Knt = 0

0.926~ (0.059)

0.949~ (0.069)

0.750~ (0.163)

0.947~ (0.055)

0.951~ (0.075)

0.797~ (0.178)

320 40

188 40

132 40

Instruments: -12.127 (11.690) -16.750~ (3.789) 24.163~ (6.789) 11.42 (0.00) 2.50 (0.29)

discontinuity ideology advocacy F(3, 270) test (p value) χ2 (2) overid. test (p value)


0.931~ (0.020)

Panel 9B

Instruments: 0.259 (11.802) -18.372~ (3.796) 29.538~ (6.691) 14.32 (0.00) 1.39 (0.49)

discontinuity ideology advocacy F(3, 261) test (p value) χ2 (2) overid. test (p value) observations authorities

320 40

320 40

Notes: Fixed province and year effects included. Panel 9A: exogenous regressors not included; Panel 9B: exogenous regressors included. Knt defined in equation (12). Standard errors in parentheses. ~ : p-value < 0.01; ∗ : p-value < 0.10.


Table 10 Wooldridge two-stage approach (Knt = 1)

grants tax basev tax basew tax basee income population age 0-4 share age 65+ share election right-wing

(10.1) First stage: Knt Probit (balanced) -0.017~ (0.007) 0.004 (0.013) -0.002 (0.003) -0.011 (0.009) 0.001 (0.001) 0.006 (0.022) -1.123 (1.244) 0.789 (0.628) -1.063∗ (0.543) 0.879 (0.896)

(10.2) Second stage: znt Wooldridge correction 0.956~ (0.035) 0.172∗ (0.070) 0.010 (0.017) 0.074∗ (0.030) -0.002 (0.002) -0.203 (0.122) 1.167 (8.375) 0.856 (3.836) 1.758 (1.950) 4.048 (3.140) -1.12 (0.79)

216 27

147 27

Wooldridge t test (p value) observations authorities

Notes: Knt defined in equation (12). Standard errors in parentheses. < 0.01; ∗ : p-value < 0.10.



: p-value

Figure 1: