Explaining Fiscal Balances with a Simultaneous Equation Model of Revenue and Expenditure: A Case Study of Swiss Cantons Using Panel Data

r 2010 Public Financial Publications, Inc.

Explaining Fiscal Balances with a SimultaneousEquation Model of Revenue and Expenditure: ACase Study of Swiss Cantons Using Panel Data

JAYA KRISHNAKUMAR, MARC-JEAN MARTIN, and NILS SOGUEL

Empirical literature on the analysis of the efficiency of measures for reducingpersistent government deficits has mainly focused on the direct explanation ofdeficit. By contrast, this paper aims at modeling government revenue andexpenditure within a simultaneous framework and deriving the fiscal balance(surplus or deficit) equation as the difference between the two variables. Thissetting enables one to not only judge how relevant the explanatory variables arein explaining the fiscal balance but also understand their impact on revenue and/or expenditure. Our empirical results, obtained by using a panel data set onSwiss Cantons for the period 1980–2002, confirm the relevance of the approachfollowed here, by providing unambiguous evidence of a simultaneous relation-ship between revenue and expenditure. They also reveal strong dynamiccomponents in revenue, expenditure, and fiscal balance. Among the significantdeterminants of public fiscal balance we not only find the usual businesscycle elements, but also and more importantly institutional factors such as thenumber of administrative units, and the ease with which people can resort topolitical (direct democracy) instruments, such as public initiatives andreferendum.

INTRODUCTION

Persistent budget deficits became an issue of growing concern to governments during the

1990s. Various measures were taken to balance the budget, such as linear and/or targeted

reduction in spending, increase in revenue, and fiscal regulations like budget constraints.

Jaya Krishnakumar is Professor of Econometrics at the University of Geneva, Boulevard Carl-Vogt 102,

1211 Geneve 4, Switzerland. She can be reached at [email protected].

Marc-Jean Martin works as Project Manager at the Statistical office of the Canton of VaudFSCRIS, Rue

de la Paix 6, 1014 Lausanne, Switzerland. He can be reached at [email protected].

Nils Soguel is Professor of Public Finance at the Swiss Graduate School of Public AdministrationFID-

HEAP, Maladiere 21 Lausanne-Chavannes VD 1022, Switzerland. He can be reached at nsoguel@

idheap.unil.ch.

Krishnakumar et al. / Explaining Fiscal Balances with a Simultaneous Equation Model 69

mailto:[email protected]






This concern for debt sustainability and the consequent diversity of government re-

sponses triggered a large amount of theoretical and empirical research. Theoretical con-

tributions focused on explaining recurring deficits by processes ranging from the

expression of public choices to the provision of public services. Empirical investigations

focused on testing the theoretical assumptions and identifying the measures that are most

successful in limiting the deficits and could, hence, be recommended to policy makers.

Empirical approach has mostly been concerned with direct modeling of fiscal balance

(deficit or surplus) making it the dependent variable in a regression equation (see the

literature review in the next section). However, we propose a different approach in this

paper by simultaneously modeling revenue and expenditure, and then deriving the net

impact of the explanatory variables on fiscal balance. Joint modeling is more appropriate

for two reasons. First, it allows one to take into consideration structural restrictions

based on theory as well as the interdependence between revenue and expenditure. Fur-

ther, it enables us to understand the effect of different explanatory variables on revenue

and/or expenditure separately. Thus it provides a deeper explanation of how deficits are

formed and a more substantive empirical basis for concrete policy recommendations for

governments.

Our empirical context relating to the 26 Swiss Cantons offers a quasi-experimental

design to test our simultaneous modeling strategy because there are considerable eco-

nomic, demographic, institutional, and budgetary variations among them. These large

institutional and fiscal differences are primarily due to a high degree of federalism com-

bined with a direct democracy (initiative and referendum) system whose modus operandi

is different from one Canton to another. At the same time these Cantons can be jointly

considered for the analysis as the context is identical (homogenous definition and mea-

surement of variables, same currency, and same legal system at the federal level).

Cantons of course differ in size.1 They have the basic power to tax income and capital.

On average, these two resources make up for about 45 percent of the total revenue of the

Cantons. However, the tax base and the tax burden vary a great deal from one Canton to

another.2 Some Cantons give voters the possibility to participate in the budgetary pro-

cess with referendum on the budget draft, the budget deficit, or the tax rate.

The paper is organized as follows. The next section briefly reviews the economic

literature on deficits. The third section presents the single-equation approach and the

simultaneous approach for analyzing the fiscal balance. After giving the definition of the

variables and the functional form, the fourth section discusses some special features of

our data set. It also examines relevant modeling strategies in presence of panel data and

nonstationarity and explains the particular choice made in our application. The fifth

1. According to the Federal Statistical Office, in 2007, the population of the Canton of Appenzell

Innerrhoden is 15,471 inhabitants compared with 1,397,567 of Zurich.

2. For example, in 2004, the tax burden was about three times higher in the fiscally least attractive

canton (Obwalden) compared with the most attractive (Zug), according to the Federal Department of

Finance.

Public Budgeting & Finance / Summer 201070

section discusses our empirical results concerning the impact of various determinants on

the Cantons’ revenue, expenditure, and balance. We conclude the paper with some policy

recommendations based on our findings.

PREVIOUS LITERATURE ON PUBLIC DEFICIT

Fiscal imbalances and public debt have long been analyzed from a normative perspective

in the political economy literature.3 This approach is grounded on the fiscal position of

the median voter, which could often be assimilated to the position of a hypothetical

benevolent social planner.4 The median voter trades off the marginal utility of the pub-

licly provided services against their marginal cost.5

Subsequently a positive approach emerged in the context of public choice.6 It is now

generally agreed that particular political agents with specific interests distort choices that

would otherwise be made by the benevolent social planner. Hence, one needs to combine

economic theory with the analysis of collective choice under the influence of groups with

diverging interests in order to fully understand budget deficits and public debts.7

There are several theoretical models of budget deficits that incorporate these compo-

nents. Alesina and Perotti8 and Pinho9 group these models into seven broad categories.

� Models based on the theory of political business cycle and naıve voters with fiscal

illusion, which predict higher deficits before elections than in postelectoral

periods.10

3. Robert J. Barro, ‘‘On the Determination of the Public Debt,’’ Journal of Political Economy 87

(1979): 940–971 and Robert E. Lucas and Nancy Stokey, ‘‘Optimal Fiscal and Monetary Policy in an

Economy without Capital,’’ Journal of Monetary Economics 12 (1983): 55–93.

4. Alberto Alesina and Roberto Perotti, ‘‘The Political Economy of Budget Deficits,’’ IMF Staff

Papers 42, no. 1 (1995).

5. Anthony Downs, An Economic Theory of Democracy (New York, NY: Harper and Row, 1957);

Richard Musgrave, ‘‘Principles of Budget Determination,’’ in Public Finance: Selected Readings, eds. Helen

Cameron and William Henderson (New York, NY: Random House, 1966), 15–27; and Randall G. Hol-

combe, An Economic Analysis of Democracy (Carbonale: Southern Illinois University Press, 1995).

6. See Alex Cukierman and Allan H. Metzler, ‘‘A Political Theory of Government Debt and Deficits in

a Neo-Ricardian Framework,’’ The American Economic Review 79, no. 4 (1989): 713–732; Nouriel Roubini

and Jeffrey Sachs, ‘‘Political and Economic Determinants of Budget Deficits in the Industrial Democ-

racies,’’ European Economic Review 33 (1989): 903–933; and Alberto Alesina and Guido Tabellini, ‘‘A

Positive Theory of Fiscal Deficits and Government Debt,’’ Review of Economic Studies 57, no. 3 (1990):

403–414.

7. Piergiuseppe Fortunato, ‘‘Voting over Redistribution: The Occurrence of Polarized Outcomes,’’

University of Bologna Working Paper No. 421 (Bologna: Dipartimento Scienze Economiche, 2001).

8. Alesina and Perotti (1995).

9. Maria Manuel Pinho, ‘‘Political Models of Budget Deficits: A Literature Review,’’ University of

Porto Working Paper No. 138 (Porto: Faculdade de Economia, 2004).

10. William D. Nordhaus, ‘‘The Political Business Cycle,’’ Review of Economic Studies 42, no. 2 (1975):

169–190.


� Models based on the partisan cycles theory predicting that deficits are higher when

left-wing party is in office and lower when right-wing party is in power.11

� Models of debt as a strategic variable used by incumbent policy makers if they expect

to be defeated in the next election to affect the fiscal condition that the future gov-

ernment would have to compromise with (tax revenues committed to debt service12).

� Models of redistribution conflicts: a conflict may arise because a current selfish gen-

eration can vote for policies that shift the burden of taxation to future generations;

other possible conflicts may also occur between better-off and worse-off individuals

within the same generation; between employed and unemployed individuals, between

regions, between labor and capital, between organized and unorganized groups.

� Models of conflicts among political parties predicting that the deeper the conflict

the greater the difficulties to limit the deficits.13

� Models of geographically dispersed interests explaining aggregate deficit through

geographic distribution of costs, benefits, and decision power.14 Political repre-

sentatives of regional constituencies may overestimate the benefits of public ex-

penditure in their region relating to cost if the financing costs are borne by all the

taxpayers and not only those living in the region.

� Models emphasizing the effects of budgetary institutions in fiscal outcome. Budget

institutions include all the laws, rules, and regulations according to which budget is

drafted and implemented.15

There is a fairly large body of literature devoted to testing the empirical validity of the

above theoretical models. Many studies do it on the basis of an estimated relationship

between public debt (dependent variable) and relevant explanatory factors, for instance,

business cycle indicators. Barro’s16 analysis suggests that countercyclical government spend-

ing accounts for a large portion of the size of the public debt, whereas Lowery17 found that

deficits are strongly related to unemployment levels but not inflation. However, these vari-

ables did not adequately explain the financial position of OECD countries, especially the

11. Douglas Hibbs, The Political Economy of Industrial Democracy (Cambridge: Harvard University

Press, 1987).

12. Torsten Persson and Lars E. O. Svensson, ‘‘Why a Stubborn Conservative Would Run a Deficit,’’

The Quarterly Journal of Economics 104, no. 2 (1989): 324–345.

13. For example, Roubini and Sachs (1989).

14. Barry Weingast, Kenneth Shepsle, and C. Christopher Johnsen, ‘‘The Political Economy of Benefits

and Costs: A Neoclassical Approach to Redistributive Politics,’’ Journal of Political Economy 89, no. 4

(1981): 642–664.

15. James Poterba, ‘‘Budget Institutions and Fiscal Policy in the U.S. States,’’ American Economic

Review 86, no. 2 (1996): 395–400 and Juergen von Hagen and Ian Harden, ‘‘Towards Greater Fiscal

Discipline: National Budget Processes and Fiscal Performance,’’ European Economy Reports and Studies 3

(1994): 311–418.

16. Barro (1979).

17. David Lowery, ‘‘The Keynesian and Political Determinants of Unbalanced Budgets: U.S. Fiscal

Policy from Eisenhower to Reagan,’’ American Journal of Political Science (1985): 665–694.


persistence of the budget deficits during the 1990s.18 Tests were thus extended to political

and institutional models. More specifically Krol,19 Poterba,20 Alesina and Perotti,21 and

Kirchgassner22 concentrate on empirical tests of the influence of budgetary institutions.

Persson and Tabellini,23 Poterba and von Hagen,24 Drazen,25 and Imbeau and Petry26

offer a survey of various models explaining the size of deficits and, often, of the level of

spending. As exposed by Pinho,27 despite the quantity of work carried out, empirical

evidence is not always consistent with the theoretical models either because the tested

explanation is rejected or because the effect cannot be identified.

From an econometric point of view, most researchers attempt to explain the level of

deficits using a single-equation model. Some studies28 use either revenue or expenditure

as the dependent variable. In the Swiss case, Kirchgassner and Pommerehne29 use a

reduced form specification based on a seemingly unrelated regression model for revenue,

expenditure, transfers, and deficits, whereas Feld and Kirchgassner30 and Vatter and

Freitag31 model revenue and expenditure separately.

18. Alesina and Perotti 1995 and James Poterba, ‘‘Do Budget Rules Work,’’ in Fiscal Policy: Lessons

from Economic Research, ed. Alan J. Auerbach (Cambridge: The MIT Press, 1997), 53–86.

19. Rober Krol, ‘‘A Survey of the Impact of Budget Rules on State Taxation, Spending, and Debt,’’ The

Cato Journal 16, no. 3 (1997): 295–307.

20. Poterba (1997).

21. Alberto Alesina and Roberto Perotti, ‘‘Budget Deficits and Budget Institutions,’’ in Fiscal Insti-

tutions and Fiscal Performance, eds. James M. Poterba and Jurgen von Hagen (Chicago, IL: Press for

National Bureau of Economic Research, 1999), 13–36.

22. Gebhard Kirchgassner, ‘‘The Effects of Fiscal Institutions on Public Finance: A Survey of the

Empirical Evidence,’’ in Political Economy and Public Finance, eds. Stanley L. Winer and Hirofumi Shibata

(Cheltenham: Edward Elgard, 2002), 145–177.

23. Torsten Persson and Guido Tabellini, ‘‘The Political Economics and Macroeconomic Policy,’’ in

Handbook of Macroeconomics, Vol. IC, eds. John B. Taylor and Michael Woodford (Amsterdam: Elsevier,

1999), 1399–1482.

24. James Poterba and Jurgen von Hagen (eds.), Fiscal Institutions and Fiscal Performance (Chicago, IL:

Press for National Bureau of Economic Research, 1999).

25. Allan Drazen, The Size of Government and the Number of Nations. Political Economy in Macro-

economics (Princeton: Princeton University Press, 2000), 675–733.

26. Louis Imbeau and Francois Petry (eds.), Politics, Institutions and Fiscal Policy: Deficits and Sur-

pluses in Federal States (Lanham: Lexington Books, 2004).

27. Pinho (2004).

28. For example, John G. Matsusaka, ‘‘Fiscal Effects of the Voter Initiative: Evidence from Last 30

Years,’’ Journal of Political Economy 103 (1995): 587–623 or A. Alesina, N. Roubini, G. D. Cohen, Political

Cycles and the Macroeconomy (The MIT Press, London, 1997).

29. Gebhard Kirchgassner and Werner Pommerehne, ‘‘Public Spending in Federal States: A Compar-

ative Econometric Study,’’ in Budgetary Policy Modelling: Public Expenditures, eds. P. Capros D. Meulders

(London: Routledge, 1997), 179–213.

30. Lars P. Feld and Gebhard Kirchgassner, ‘‘The Political Economy of Direct Legislation: Direct

Democracy and Local Decision Making,’’ Economic Policy 33 (2001): 329–367.

31. Adrian Vatter and Markus Freitag, ‘‘Die Januskopfigkeit von Verhandlungsdemokratien,’’

Schweizerische Zeitschrift fur Politikwissenschaften 8, no. 2 (2002): 53–80.


To our knowledge, Lowry and Alt32 is the unique attempt to simultaneously

model revenue and expenditure so far. The authors use their model to test various

hypotheses about political party behavior based on the reduced form equation

for expenditure, but do not go on to derive a surplus/deficit equation. Other sections

of literature also examine both revenue and expenditure, but rather from a descriptive

point of view. Some of these studies carry out appraisals of deficit reduction programs

recently implemented in industrialized countries.33 Many others investigate the inter-

temporal relationship (Granger causality34) between revenue and expenditure. This

section of literature, so-called the revenue–expenditure nexus, is compressively sur-

veyed by Payne.35 Recent contributions to this literature introduced some exogenous

variables in the cointegration relation between revenue and expenditures but treating

them as control variables rather than looking at their effect on the endogenous

variables. Finally, mention should be made of the contributions which analyze

reduced form equations of revenue resulting from different joint models of revenue

and expenditure.36

Thus, even though we are not the first ones to consider bidirectional causality between

revenue and expenditure, our study provides a complete implementation of this idea

using an appropriate structural econometric model for analyzing deficit.

MODELING FISCAL BALANCE

Fiscal balance (surplus S or deficit D) is defined as the difference between revenue R and

expenditure E (S5R�E,D5E�R). In our model, we adopt the convention of mod-

eling S with the understanding that when S is negative we are in fact talking of a deficit.

All factors that influence revenue and/or expenditure determine its sign and size. One can

thus distinguish three groups among the explanatory variables of fiscal balance: first,

variables that influence revenue but not expenditure (XR); second, variables that influ-

ence expenditure but not revenue (XE); third, variables that influence both revenue and

32. Robert Lowry and James Alt, ‘‘Divided Government, Fiscal Institutions and Budget Deficits,’’

American Political Science Review 88, no. 4 (1994): 811–828.

33. Alesina and Perotti (1995); Alberto Alesina and Roberto Perotti, ‘‘Fiscal Adjustments in OECD

Countries: Composition and Macroeconomic Effects,’’ National Bureau of Economic Research Working

Paper No. 5730 (Cambridge: National Bureau of Economic Research, 1996); Poterba (1997); and Alberto

Alesina, Roberto Perotti, and Jose Tavares, ‘‘The Political Economy of Fiscal Adjustments,’’ Brooking

Papers on Economic Activity 1 (1998): 197–266.

34. Clive W. J. Granger, ‘‘Investigating Causal Relations by Econometric Models and Cross-Spectral

Methods,’’ Econometrica 37, no. 3 (1969): 424–438.

35. James E. Payne, ‘‘A Survey of the International Empirical Evidence on the Tax-Spend Debate,’’

Public Finance Review 31, no. 3 (2003): 3002–3024.

36. See Walter Hettich and Stanley Winer, Democratic Choice and Taxation (Cambridge: Harvard

University Press, 1999) for a broad survey of the contributions that relate to the political economy of public

revenue.


expenditure (XRE):

St ¼ aþ XR0tdþ XE0tlþ XRE0tbþ et ð1Þwhere St denotes the surplus for period t and XR0t, XE

0t, and XRE0t denote row vectors of

explanatory variables, a, d, l, and b their corresponding coefficient vectors and et rep-resents a random disturbance term.

The underlying structural model for the above equation comprises two equations, one

linking revenue and the other linking expenditure to appropriate explanatory variables.

The general form of the simultaneous equation model can be written as

Rt ¼ aR þ EtgR þ XR0tdR þ XRE10tb1R þ XRE00tb0R þ eRt

Et ¼ aE þ RtgE þ XE0tlE þ XRE10tb1E þ XRE00tb0E þ eEt

�ð2Þ

where gR6¼0 and gE 6¼0. However, the model may also feature two independent equations

(gR 5 0 and gE 5 0) or be a triangular system (gR 5 0 or gE 5 0). Among the exogenous

variables, one may find variables only influencing revenue (XR) or expenditure (XE);

variables jointly influencing revenue and expenditure, but with a noncompensating

strength so as to have a nonzero impact on fiscal balance in the end (XRE1) and even-

tually, variables jointly influencing revenue and expenditure but compensating each

other out so that they end up having no impact on the fiscal balances (XRE0). eRt and eEtare random disturbance terms such that E[eRt eEt] 6¼0.

The reduced form of the simultaneous equation model explains revenue and

expenditure only in terms of exogenous variables. Because revenue and expen-

diture are endogenous in the model, they will be (except in special cases) linked to

all the exogenous variables in the reduced form because of the mutual influence via gRand gE.

Rt ¼ fR þ XRE10tj1R þ XRE00tj0R þ XR1t$R þ XE0tyR þ BRt

Et ¼ fE þ XRE10tj1E þ XRE00tj0E þ XR0t$E þ XE0tyE þ BEt

�ð3Þ

where fR ¼ ðaR þ gRaEÞ=ð1� gRgEÞ, . . ., BEt ¼ ðeEt þ gEeRtÞ=ð1� gRgEÞ, j0R ¼ j0E,$R1=$E1 ¼ . . . ¼ $RH=$EH ¼ 1=gE, yR1=yE1 ¼ . . . ¼ yRH=yEH ¼ gR, and E[zRt zEt] 6¼0.Thus, the relationships between the structural and reduced form coefficients may lead to

additional constraints among the parameters. These constraints have to be taken into

account while estimating the model.37

37. Even in the simple case where the underlying structural model features no relationship between

revenue and expenditure (gR 5 0 and gE 5 0 in equation (2)), OLS estimation of equation (1) (model for

fiscal balance with a single equation) does not tally with the difference between the parameters of the

revenue and expenditure equations if there are determinants specific to revenue or expenditure. Put dif-

ferently, the OLS estimation of equation (1) provides coherent results only if all determinants are common

to revenue and expenditure.


Subtracting expenditure from revenue results in the surplus model derived from the

simultaneous equation system:

St ¼ fS þ XRE10tjS þ XR0t$S þ XE0tyS þ Bt ð4Þwhere fS ¼ f1R � f1E, etc.

38

From this derivation, one can see why joint modeling of revenue and expenditure is

more insightful than directly modeling surplus or deficit. First, this approach not only

makes it possible to obtain the impact of an explanatory variable on the deficit, but also

enables to identify its impact on the two major components of deficit, namely revenue

and expenditure. Secondly, it allows to test the existence of the simultaneous relationship

between the above two components of deficit. Thirdly, it produces more accurate results

by taking into account the constraints on the parameters of the deficit equation resulting

from the underlying structural model.

SPECIFICATION OF THE MODEL

Endogenous Variables

Various models rooted in the median voter theory predict that revenue and expenditure

are interdependent by nature.39 Governments tailor the provision of public services to

meet the citizens’ demand. Citizens trade off the marginal benefit of each service with its

marginal cost when they express their demand. This echoes the common sense according

to which people make decisions regarding revenue under the constraint of the expen-

diture to be financed (what must we pay for?), whereas decisions regarding expenditure

are made under the constraint of revenue (what can we afford?).

From a purely practical point of view, other reasons explain why it is necessary to

introduce the level of revenue as an explanatory variable for expenditure and the level of

expenditure as an explanatory variable for revenue. For instance, the public’s concern

about deficits and debt forces governments to secure the right balance between revenue

and expenditure to avoid snowballing of debt and interest payments eating up revenue.

Furthermore, any surplus is likely to be targeted by various interest groups. Should

revenue be higher than what is expected in a fiscal year, the leeway increases together

with the possibility of financing additional spending or tax cuts. In contrast, when rev-

enue tends to be lower than what is forecasted, governments introduce, sometimes al-

most immediately, mitigating measures to avoid worsening of the deficit. Besides these

discretionary measures, governments earmark the revenue and the expenditure they pay

for, increasing the dependency between both sides of the budget. Indeed, earmarking is a

38. XRE0t does not appear in equation (4) as, by definition, it has no net influence on the balance:

j0R ¼ j0E implies in term of parameters of equation (2) b0R � gRb0E ¼ b0E � gEb0R.39. Musgrave (1966) and Allan H. Meltzer and Scott F. Richard, ‘‘A Rational Theory of Size of

Government,’’ Journal of Political Economy 89 (1981): 914–927.


common practice to increase the acceptance of new taxes. Further, earmarking arises

when the provision of various public services is directly paid for using levies that are in

turn calculated to cover the cost of provision.

Concretely speaking, our simultaneous equations system explains the per capita rev-

enue and expenditure (denoted as REVENUE and EXPENDITURE40) of the 26 Swiss Cantons.

This model is used to derive the equation for the fiscal balance, which is also the per

capita borrowing requirement in the case of deficit or the amount of debt the Canton is

able to pay back in the case of surplus. REVENUE and EXPENDITURE are expressed in per

capita terms because there is a large variation in the population size of Cantons: Zu-

richFthe biggest CantonFhas 135 times more inhabitants than the smallest Canton,

Appenzell Innerrhoden. The values are also in real terms because we use annual ob-

servations over the period 1979–2002.41 Heteroskedasticity of errors is thus largely at-

tenuated as compared with using nominal figures.

Explanatory Variables

Drawing from the previous literature on public deficit and based on the typology pro-

vided by Garand and Kapeluck,42 our explanatory variables belong to different cate-

gories: institutional, political, economic, and structural.43

Lagged endogenous variables, that is REVENUE(� 1) and EXPENDITURE(� 1), are used

to take into account the inertia of revenue and expenditure. Incremental budgeting

(according to which previous budget components are increased or decreased by small

increments in the current year’s budget44) may also explain the frequently observed

inertia of revenue and expenditure. These variables therefore reflect the institutional

framework within which the budgetary process takes place. Existing constitutional and

legal commitments also prevent revenue and expenditure from being annually re-exam-

ined. Higg’s45 contention that expenditure or revenue does not return to its previous level

after a significant increase due to an exceptional event must also be considered. There are

several arguments that favor this contention: (a) bureaucrats resist downsizing of the

State, (b) during crisis periods, expertise of civil servants and politicians increases leading

to enhanced productivity and rise in marginal benefit associated with public services, and

(c) during crisis, citizens’ perception of the utility of the State changes; they show pref-

erence toward public services.

40. We write the names of variables in small capital letters so that they can be spotted easily in the text.

41. The implicit deflator of GDP is used to deflate all the variables expressed in monetary terms,

including revenue, expenditure, and fiscal balance. Appendix A provides some descriptive statistics.

42. James Garand and Branwell Kapeluck, ‘‘Understanding Surpluses, Deficits, and Debt in the

American States, 1950–1998,’’ in Politics, Institutions and Fiscal Policy: Public Deficits and Surpluses in

Federated States, eds. Louis Imbeau and Francis Petry (Lanham: Lexington Books, 2004), 49–87.

43. Appendix A provides summary statistics for all the variables.

44. Aaron Wildavsky, Budgeting: A Comparative Theory of Budgetary Processes (Boston, MA: Little

Brown, 1975).

45. Robert Higgs, Crisis and Leviathan (New York, NY: Oxford University Press, 1987).


As for additional institutional determinants, the possible right of citizens to initiate a

change in the legislation or oppose a spending proposal through a fiscal referendum may

also have an impact on the government’s fiscal position. In other words, financial REF-

ERENDUMs give citizens a veto right on spending proposals that exceed a given amount.46

Their influence can be direct, because voters may refuse the expenditure that is put to the

ballot. It can also be indirect in that it moderates the scope of the proposed projects,

because the promoters wish to avoid the project going up for vote or, if put to vote, have

it rejected by the voters. Each Swiss Canton is free to introduce the option for refer-

endum on expenditure issues in its legislation and to design it (e.g., the minimum amount

to trigger a referendum) in its own way. As a result, the degree of pressure on expenditure

varies among the Cantons. Here, we use the index developed by Frey and Stutzer47 to

measure the stringency of REFERENDUM.48 Another index devised by these authors has

also been used to evaluate how easy it is for citizens to put forward INITIATIVES for new

laws or change existing laws.49 The easier it is (for instance, lower number of peo-

pleFi.e., signaturesFrequired in support of the proposal, longer time span to collect

signatures), the more difficult it becomes for the lobbies to manipulate the political

decision process to suit their interests.50 If, as generally assumed, citizens favor solutions

that are cheaper than those proposed by the government, easier ways of introducing an

initiative must negatively influence expenditure. However, Matsusaka51 does not exclude

the possibility that citizens’ proposals can trigger higher expenditures than what would

otherwise have been spent. Hence, INITIATIVES may also positively influence expenditure.

As initiatives also deal with Cantonal tax laws, revenue may be positively or negatively

affected. Therefore, the expected impact of the variable INITIATIVES on fiscal balance

remains undecided.

46. Hanspeter Kriesi, Le systeme politique Suisse (Paris: Economica, 1995).

47. Bruno S. Frey and Alois Stutzer, ‘‘Beyond Outcomes: Measuring Procedural Utility,’’ Oxford Eco-

nomic Papers 57 (2005): 90–111.

48. The legal prescriptions regarding referendums to prevent new State expenditure differ from Canton

to Canton. The referendum can be either mandatory or optional. A mandatory referendum is automatically

organized when a new spending proposal exceeds a fixed limit. An optional referendum takes place only if a

citizen or a group of citizens collects a required number of signatures within a given time span. The financial

limit, the required number of signatures, and the time span are Canton specific. The variable REFERENDUM

was constructed following Frey and Stutzer (2005). The stringency of the mandatory referendum depends

only on the financial limit (per capita) and is evaluated on a 6-point scale (1 is a high barrier, 6 is a low one).

The stringency of the optional referendum is evaluated on the same scale for each dimension (financial

limit, signature, and time span). The three resulting indices are aggregated using a simple average. Then the

variable REFERENDUM is the maximum of the index for the mandatory referendum and the composite index

for the optional referendum.

49. The number of signatures required to launch an initiative and the time span within which the

signatures have to be collected vary from Canton to Canton. The variable INITIATIVE is constructed in the

same way as optional referendums.

50. Mutsusaka (1995).

51. Ibid.


The size of the administration, particularly the number of DEPARTMENTS, is a variable

whose influence has not been tested so far. However, its possible impact is grounded

either on the spendthrift behavior that Niskanen52 credits government with or on Ve-

lasco’s53 idea that the government’s budget can be considered as a common good pulled

by various interest groups. Therefore, the larger the number of departments, the higher

would be the expenditure and hence the worse the budget balance.

As for economic factors, GROWTH rate is expected to affect the budget balance through

the automatic stabilizers and discretionary countercyclical decisions that should increase

spending/reduce revenue during recession and improve the balance with lower spending/

higher revenue during economic expansion. The percentage change in GDP is commonly

used as a measure of business cycle activity. However, as there is no official data avail-

able on Cantonal GDP, growth rates in Cantonal income have been used here. The

UNEMPLOYMENT rate is also often used to reflect the economic situation. Over the time

period of our study (1980–2002), it can, however, only be used in addition to GROWTH,

firstly because joblessness only appeared in Switzerland since the early 1990s (before that,

job losses were compensated by the departure of new immigrant workers) and secondly

because unemployment benefits, being taxable, prevent the shrinkage of the tax base. So

UNEMPLOYMENT has no impact on revenue and is expected to influence expenditure neg-

atively, thus deteriorating the deficit.

We consider many political variables to test if preferences regarding the role of the

State actually make a difference. Several variables can be and have been used in this

regard. The Cabinet’s political leaning supposedly reflects individuals’ preferences when

members are democratically elected. Following Hibbs,54 right-wing elected officials

demonstrate more fiscal conservatism than their left-wing counterparts. Consequently,

the more the number of right-wing officials in the Cabinet, the lesser the deficit or the

more the surplus should be. This assumption is, however, subject to discussion. In fact,

although the left-wing officials are reputed to favor expenditure, they also favor taxes

and thus revenue.55 Consequently, a Cabinet dominated by right-wing officials may

eventually have a negative influence on revenue and expenditure, while the influence on

surplus is uncertain. The political leaning is defined by the number of right-wing mem-

bers in Cantonal Cabinets (RIGHT-WING).

The size of the COALITION, measured by the number of parties represented in the

Cabinet, is a frequent variable in empirical models (though not in models estimated using

Swiss data). The higher the number of parties, the lesser the chances of government’s

52. William A. Jr. Niskanen, Bureaucracy and Representative Government (Chicago: Aldine-Atherton,

1971).

53. Andres Velasco, ‘‘Debts and Deficits with Fragmented Fiscal Policymaking,’’ Journal of Public

Economics 76 (2000): 105–125.

54. Hibbs (1987).

55. Louis M. Imbeau and Genevieve Tellier, ‘‘The Political-Economy of Budget Deficits in the Ca-

nadian Provinces, 1968–2000,’’ in Politics, Institutions, and Fiscal Policy: Deficits and Surpluses in Federated

States, eds. Louis Imbeau and Francis Petry (Lanham: Lexington Books, 2004), 89–111.


actions being concerted and concentrated. This would reduce the Cabinet’s ability to

trade off between various spending requirements56 and increase its ability to agree on a

tax pressure. Therefore, a larger coalition should theoretically lower the revenue, but

increase the expenditure, and thus worsen the fiscal balance (lower surplus or higher

deficit).

By the same reasoning, if the Cabinet’s composition differs from that of the parlia-

ment, it reduces the Cabinet’s ability to deal with the parliament in trading off between

various expenditures and getting support for a revenue increase. Therefore, lack of

CONCORDANCE should also increase expenditure and lower revenue.57

According to Nordhaus’s58 theory of electoral cycles, politicians will seek election or

re-election by cosseting their clientele via public budgets. Hence, during the election year,

the revenue would be lower and expenditure higher, thus deteriorating the fiscal balance.

The dummy variable ELECTION-YEAR, which takes the value 1 during general election

years and 0 at other times, is used to test this phenomenon.

Finally, structural characteristics, reflecting differences in preference between groups

of population, often constitute important elements to be modeled in structural equations.

For instance, one usually considers that expenditure for ELDERLY people is greater than

that for other categories of population, whereas this group adds relatively less to the tax

base. Therefore, the share of ELDERLY in the population should increase expenditure and

negatively influence revenue and budget balance. On the contrary, one may advocate

that the elderly are more conservative than others and thus more likely to favor a

rigorous fiscal policy.59

In Switzerland, the attitude regarding the role of the State clearly differs between the

linguistic regions. Inhabitants of the German-speaking Cantons are assumed to be

more distrustful of State intervention than their French- and Italian-speaking

counterparts (especially the Canton of Geneva, e.g.). The latter categories are said to

hold the doctrine of balanced budget in lesser esteem.60 A dummy variable (LANGUAGE)

taking the value of 1 for French- or Italian-speaking Cantons, and 0 for others is used to

test this assumption, together with a dummy for GENEVA the most ‘‘spendthrift’’ of all

Cantons. LANGUAGE and GENEVA are expected to increase revenue, expenditure, and

deficits.

56. Stephen A. Borrelli and Terry J. Royed, ‘‘Government Strength and Budget Deficits in Advanced

Democracies,’’ European Journal of Political Research 28 (1995): 225–230 and Velasco (2000).

57. The concordance index is the percentage of the seats in the parliament that are occupied by members

of parties represented in the government.

58. Nordhaus (1975).

59. Garand and Kapeluck (2004) and Lars P. Feld and John G. Matsusaka, ‘‘Budget Referendums and

Government Spending: Evidence from Swiss Cantons,’’ Journal of Public Economics 87, no. 12 (2003):

2703–2724.

60. Francesc Pujol and Luc Weber, ‘‘Are Preferences for Fiscal Discipline Endogenous,’’ Public Choice

114, no. 3 (2003): 421–444.


Choice of the Model

While working with time series, one should first investigate the nature of the processes (in

terms of stationarity) before deciding on the appropriate specification of the model.

According to Hsiao,61 it is possible to formulate a simultaneous equation model with the

variables in levels if the variables of each equation are either stationary or cointegrated

(and if any variable that is present in the equation but does not enter the cointegrating

relation is also stationary). Appendix C reports the results of Maddala and Wu’s62 unit

root tests and Appendix D the results of Pedroni’s63 cointegration tests. The relatively

few observations per Canton motivate the use of such panel unit root and cointegration

tests instead of testing each Canton separately. Further, the small size of our sample also

leads us to assume identical coefficients for all Cantons. In line with the existing lit-

erature, the unit root tests show that revenue and expenditure are I(1) and that all the

other variables (including the balances) are I(0). According to Engle and Granger,64 if a

linear combination of I(1) variables is stationary, then they are said to be cointegrated.

Pedroni’s65 test confirms cointegration between revenue and expenditure. Thus we can

keep them in levels in our equations.

Stochastic Assumptions

Three statistical methods can possibly be used to address the issue of Cantonal heter-

ogeneity: (a) the White correction for heteroskedasticity, (b) the estimation of different

error variances for different Cantons, and (c) the inclusion of Canton-specific fixed

effects.66 Although Wald test results indicate that models with fixed effects must be

preferred to those without, we could not include them in our model as our equations

contain explanatory variables that do not vary (or vary little) over the considered time

61. Cheng Hsiao, ‘‘Cointegration and Dynamic Simultaneous Equations Model,’’ Econometrica 65,

no. 3 (1987): 647–670.

62. Gangadharrao S. Maddala and Shaowen Wu, ‘‘A Comparative Study of Unit Root Tests with Panel

Data and a New Simple Test,’’ Oxford Bulletin of Economics and Statistics 61, no. 1 (1999): 631–652.

63. Peter Pedroni, ‘‘Critical Values for Cointegration Tests in Heterogeneous Panels and Multiple

Regressors,’’ Oxford Bulletin of Economics and Statistics 61 (1999): 653–671.

64. Robert Engle and Clive Granger, ‘‘Co-Integration and Error Correction: Representation, Estima-

tion and Testing,’’ Econometrica 55, no. 2 (1987): 251–276.

65. Pedroni (1999).

66. Other models, such as the random effects model (REM) or the random coefficient model (RCM)

also incorporate specific effects to reflect the heterogeneity of behavior although they make different

assumptions regarding the type of heterogeneity. However, the fixed effects model (FEM) is theoretically

more appropriate than REM when all the individuals within a population are studied (cf. Marc Nerlove

and Pietro Balestra, ‘‘Formulation and Estimation of Econometric Models for Panel Data,’’ in The

Econometrics of Panel Data: Handbook of Theory and Applications, eds. Laszlo Matyas and Patrick Sevestre

(Dordrecht: Kluwer Academic Publishers, 1996), 3–22. In this case, the Hausman test confirms that FEM is

preferable to REM. Moreover, various RCM trials by the authors gave sufficiently unstable results to

discard them without further ado.


span and hence their coefficients will not be estimated in a fixed effect estimation pro-

cedure. For instance, LANGUAGE does not vary over time and thus cannot be added along

with fixed effects and DEPARTMENTS is also almost collinear with the Cantonal fixed

effects and produces unstable coefficient estimates.67 Finally, the relatively small number

of observations available compared with the number of parameters to be estimated can

only give rise to models with a limited number of statistically significant variables. All

these considerations made us go for the White correction procedure rather than adding

more parameters characterizing heteroskedasticity.68

EMPIRICAL RESULTS

Our sample is a balanced panel data set and contains annual observations relating to all

the 26 Swiss Cantons over the period 1979–2002, that is, 624(26 � 24) observations per

variable. Their summary statistics are given in Appendix A, while Appendix B provides

information on how inconsistencies or particularities that perturb the considered long-

run relation were rectified in revenue and expenditure data.

Table 1 presents the results of 3SLS estimation of our model. The derived surplus

equation is based on the difference between the reduced form equations for expenditure

and revenue as explained in the third section. For the sake of comparison, we also

present, in the last column, the results obtained by directly estimating a single-equation

model that explains the surplus as a function of the same explanatory variables.69

The models in Table 1 include most of the considered explanatory variables. However,

some variables are dropped, essentially from the revenue equation, either because they

should not have any influence according to theory or they did not turn out be significant.

Appendix E shows the results for a model incorporating the full set of explanatory

variables. While the t-tests show that some coefficients are not individually significant,

the F-ratios indicate that they are jointly significant. This is probably due to the relatively

high number of the coefficients to be estimated compared with the number of available

67. The values, based on the criterion of David A. Belsley, Edwin W. Kuh, and Roy E. Welsch,

Regression Diagnostics: Identifying Influential Data and Sources of Collinearity (New York, NY: John

Wiley, 1980), range from 20 to over 80 for the majority of the matrices composed solely of the fixed effects

and of one of the proposed variables. Now, Belsley and colleagues consider that problems of high mul-

ticollinearity are likely to start from the value of 20 for the matrices comprising all the explanatory variables

introduced in a model. Therefore, it is not surprising that the values of coefficient estimates changed

significantly depending on the variables included in the individual fixed effect models.

68. Our results showed that the coefficients obtained with alternative techniques are quantitatively and

qualitatively close to those presented.

69. REFERENDUM could potentially be endogenous like other variables that Cantons may influence

(INITIATIVE and DEPARTMENTS). Hence, we use the lagged value as an instrument for REFERENDUM, as done by

Feld and Matsusaka (2003). The same procedure is adopted for INITIATIVE but not for DEPARTMENTS,

because this variable is almost invariant over the considered period and thus is independent with regard to

endogenous variables.


TABLE1

Estimated and Derived Models of Revenue, Expenditure, and Budget Balance Per Capita

for the 26 Swiss Cantons Over the Period 1980–2002

Dependent Variable

Simultaneous, Two-Equation Model

Direct, Single-EquationModel for ComparisonbStructural Modela

DerivedModel

REVENUE EXPENDITURE SURPLUS SURPLUS

REVENUE 0.306(9.80)

EXPENDITURE 0.121(3.66)

GROWTH 23.5 � 10.9 26.8 28.9(3.42) (� 1.57) (3.57) (3.78)

UNEMPLOYMENT F 24.4 � 22.3 � 7.60(3.20) (� 3.06) (� 0.84)

REFERENDUM F � 22.7 20.7 15.7(� 2.17) (2.53) (1.68)

INITIATIVE r 19.5 � 17.8 � 5.43(1.68) (� 1.82) (� 0.44)

DEPARTMENTS F 11.26 � 10.3 � 10.1(2.15) (� 2.33) (� 1.92)

RIGHT-WING � 0.882 � 1.38 0.625 0.280(� 1.20) (� 1.61) (0.68) (0.27)

COALITION r r r rCONCORDANCE r � 2.20 2.01 2.07

(� 1.61) (1.59) (1.28)ELECTION-YEAR r 41.2 � 37.6 � 28.7

(1.65) (� 1.50) (� 0.91)ELDERLY r 11.5 � 10.5 � 6.39

(1.72) (� 1.45) (� 0.80)LANGUAGE r r r rGENEVA r 262 � 239 � 159

(3.37) (� 2.54) (� 1.48)REVENUE(� 1) 0.882 F 0.636

(25.8) (16.3)EXPENDITURE(� 1) 0.685 � 0.625

F (21.5) (� 16.4)SURPLUS(� 1) 0.660

(16.6)CONSTANT 134 198 � 83.5 � 121

(1.80) (1.35) (� 0.64) (� 0.81)Adjusted R2 0.99 0.99 0.34 0.56

Notes: The numbers shown in brackets beneath the estimated parameters represent the t-stat values.

F, no influence according to theory.

r, removed for reasons discussed in the text.a3SLS estimation with lagged value of REFERENDUM and INITIATIVE as instruments and White correction (see also

footnote 69 on page 81).bEstimation with instrumental variables and with lagged value of REFERENDUM and INITIATIVE as instruments and

White correction.


observations. At the same time, using the explained lagged variable as an explanatory

variable reduces or even masks the significance of other explanatory variables.

The results obtained confirm the existence of a simultaneous relation between revenue

and expenditure. However, the relation is stronger in the direction of revenue to ex-

penditure, with greater flexibility of expenditure as compared with revenue (0.306 against

0.121). This greater flexibility may also explain why revenue displays greater inertia

than expenditure as can be seen from the lagged variables’ coefficients (0.882 against

0.695).

Recognizing the strong correlation between explained variables and their lagged vari-

ables (more than 0.98), one may wonder if the whole influence can be attributed to lagged

variables excluding any simultaneous relationship between revenue and expenditure. The

simultaneous relationship cannot be refuted at a conceptual level as for instance some

laws explicitly require or justify matching revenues (or expenditures) for providing spe-

cific public services (or collecting specific revenues). Further, the fact that a net inertia

appears both in the derived model and in the comparison model (with the budget balance

as unique explained variable) is ample evidence that both an autoregressive and a si-

multaneous relationship exist between revenue and expenditure. Indeed, the coefficients

of the lagged variables REVENUE(� 1) and EXPENDITURE(� 1) are 0.636 and � 0.625 in the

derived model and the coefficient of SURPLUS(� 1) is 0.660 in the comparison (direct)

model. Note that this value is in line with the findings of Dafflon and Pujol70 who are the

only ones so far to use this lagged variable as an explanatory factor of Cantonal bal-

ances.

Other variables have a more limited explanatory power on revenue, expenditure, or

balance. Appendix F emphasizes this phenomenon by presenting estimated coefficients in

terms of standard deviations of explanatory and explained variables. Note that they

influence the fiscal balance more than both revenue and expenditure. For instance, even

though GROWTH has an almost equal impact on REVENUE and BALANCE in absolute terms,

it is not the case in relative terms: an increase of one percentage point of GROWTH implies

an increase of CHF 23.5 per person (0.014 standard deviation) over the period, but an

improvement of CHF 26.8 per inhabitant (0.102 standard deviation) in the surplus

(surplus increase or deficit reduction). Besides, the influence of specific explanatory

variables on fiscal balance does not qualitatively differ between the simultaneous and the

comparison (direct) models. In spite of this, large quantitative differences often arise,

mainly because weakly significant explanatory variables are removed from the revenue

equation. On the other hand, the coefficients of the ‘‘common’’ variables (common to

both revenue and expenditure equations) are not very different from those of the direct

model.

More fundamentally, the variables benefit from higher levels of significance in the

simultaneous model than in the direct model. In fact, the levels of significance are so low

70. Bernard Dafflon and Francesc Pujol, ‘‘Fiscal Preferences and Fiscal Performance. Swiss Cantonal

Evidence,’’ International Public Management Review 2, no. 2 (2001): 55–83.


in the direct model that they would not have been retained if not for comparison

purposes.

The finding that revenue is more sensitive to GROWTH than expenditure is entirely

plausible; a large share of revenue is represented by taxes levied on individuals’ income

and firms’ profit, which are quite sensitive to economic situation, whereas expenditure,

which is mainly on education, social security, and health care, is more sensitive to

demographic and political considerations.

As expected, UNEMPLOYMENT increases expenditure due to rising social security spend-

ing during downturns and deteriorates the balance according to the simultaneous model.

It must be pointed out that the variable is insignificant in the direct model. Moreover, its

influence is also weaker than in the derived one (� 7.6 instead of � 22.3). Thus, one can

perceive the benefit of simultaneous modeling over direct modeling. Indeed, the estimates

provided by the direct model are based on the implicit assumption that UNEMPLOYMENT

impacts revenue positively. Introducing the variable in the revenue equation of the si-

multaneous model allows us to show it. The UNEMPLOYMENT’s coefficient would amount

to 18.8 in the revenue equation and to 26.4 in the expenditure equation. So the net impact

(� 11.8) would be fairly close to the one given by the direct model (� 7.6). However, the

positive influence of unemployment on revenue cannot be justified because unemploy-

ment benefits cannot exceed the wages of laid-off persons.

REFERENDUM has a negative influence on public spending, which is almost fully carried

over to fiscal balances. As it correlates quite strongly with LANGUAGE (� 0.75), part of its

effect may well reflect the difference in preferences between German-speaking Cantons

on the one hand and French- and Italian-speaking Cantons on the other. This expla-

nation seems realistic because LANGUAGE is often positively and significantly correlated

with revenue and expenditure in the earlier studies of the Swiss Cantons. However, the

sign of LANGUAGE was found to be so unstable in many of the models that we tested and

its level of significance was so low that it was eventually removed from the model shown

in Table 1.71 For the same reasons, INITIATIVE was also removed from the revenue equa-

tion. Nonetheless, it positively influences public spending and shows that successful

initiatives can help in triggering additional public services rather than eliminating the

existing ones or changing the tax burden.

As theory predicts, the number of DEPARTMENTS increases expenditure and worsens

fiscal balance. Thus, the restructuring efforts taken over the period delivered the expected

results (214 departments in the 26 Cantons in 1980 and 168 in 2002).

The so-called culture of consensus in Switzerland probably explains why COALITION

happens to be systematically insignificant in all the models we tested. This result cor-

roborates the findings of previous studies for Switzerland.

The RIGHT-WING leaning of the Cabinet negatively impacts both revenue and expen-

diture. Therefore, its effect on fiscal balance depends on the respective magnitudes of its

influence on the two variables. But it also depends on the influence of revenue on

71. When included, LANGUAGE is always insignificant whatever be the model (see Appendix F).


expenditure and that of expenditure on revenue. Our empirical estimations show that

fiscal balance deteriorates when the Cabinet shifts to the right.72 However, the confidence

interval of the variable includes zero (� 1.17 to 2.42). Therefore, it is statistically im-

possible to exclude zero or even positive impact. Just like the findings regarding COALI-

TION, this result comes as a surprise in comparison to the results obtained from other

countries.73 However, previous studies for Switzerland also report a dampening influence

on expenditure and revenue and no significant influence on the balance.

As expected, a higher degree of CONCORDANCE between the Cabinet and the parliament

lessens public spending and thus improves fiscal balance. Only Vatter and Freitag74

tested this variable so far; however, they report it as insignificant.

We find that the coefficients for ELECTION-YEAR are sufficiently stable in all the variants

explored to keep this variable in the expenditure equation. On the contrary, these co-

efficients are not stable enough in the revenue equation. The reason for this could be that

one can more easily manipulate the level of spending than that of revenue. Additionally,

an improvement in public services may produce an electoral benefit for elected officials

or parties in power. This result confirms that the Swiss political system, despite its

characteristics of concordance and consensus, is open to vote-catching behavior. How-

ever, such behavior hardly explains the size of the fiscal balance, as shown by the stan-

dardized coefficient.

The influence of the share of ELDERLY people in the population is in line with the-

oretical expectations: negative on expenditure and positive on fiscal balance. However,

just as UNEMPLOYEMENT, it sometimes has a positive influence in the revenue equation.

Such unlikely impacts lead us to remove it in this particular equation.

The dummy variable representing GENEVA turns out to be constantly significant with a

positive influence on expenditure and a negative one on balance. This demonstrates that

other explanatory variables in the models cannot adequately explain the big and repeated

fiscal imbalances suffered by the Canton over the period. The variable does not show

such stability and significance in the revenue equation.

CONCLUSIONS

The main aim of this study has been to highlight the simultaneous relationship between

government revenue and expenditure and the need to model them jointly for a proper

72. Switzerland is a country of coalition and consensus in which, unlike other countries, there is no

marked changeover of political power between parties. Thus, the variable RIGHT-WING quantifies more the

effect of a greater or lesser shift to the right or to the left than the effect of right-wing as opposed to left-

wing government. For all the Cantons and for the entire period, the average value for this variable is 78

percent and the value for the few governments with a left majority does not exceed 40 percent.

73. Louis M. Imbeau, ‘‘The Political-Economy of Public Deficits,’’ in Politics, Institutions and Fiscal

Policy: Deficits and Surpluses in Federal States, eds. Louis Imbeau and Francis Petry (Lanham: Lexington

Books, 2004), 1–19.

74. Vatter and Freitag (2002).


identification of the determinants of public deficits. Direct modeling of surplus/deficit

completely ignores any potential endogeneity of revenue and expenditure. Nor does it

enable a finer analysis of the underlying mechanism through which an explanatory

variable influences surplus/deficit, for instance, it does not allow for the possibility that

an explanatory variable only affects revenue without affecting expenditure or vice versa.

Our empirical results confirm the relevance of the simultaneous approach. It is first

verified by the existence of a cointegrating relation between the two variables and their

lags. Secondly, the fact that some of the other explanatory variables of both revenue and

expenditure are highly significant unambiguously demonstrates that we are not in pres-

ence of a tautological relationship that would have meant a coefficient of 1 for revenue in

the expenditure equation (and vice versa) and 0 for the other variables.

A good example of the benefit resulting from being able to take into account the

restriction of a structural model is the derived impact of UNEMPLOYMENT on fiscal bal-

ance. Should one refer to a direct model of its influence on balance, then one would

implicitly accept that this net impact is a consequence of a hypothetical and unlikely

positive influence of revenue. The simultaneous framework allows us to model its in-

fluence only on expenditure because no influence is theoretically expected on revenue due

to unemployment benefits.

The low level of significance of some variables in the (direct) balance equation would

imply that we would have retained a different set of variables if the selection of variables

were driven by the results of direct models of fiscal balances. This would also have led to

different recommendations for policy makers compared with those based on our simul-

taneous model of expenditure and revenue. This leads us to conclude that the simultaneous

approach should be systematically adopted in future works devoted to public deficit/surplus

analysis for all countries. In light of our results, this statement is even more relevant when it

comes to examining the influence of variables that only affect either expenditure or revenue.

It is indeed for these variables that we see an important difference between the direct

estimation of the deficit equation and the indirect derivation of the same from the structural

model. Our joint approach enables us to arrive at conclusions based on proper inference

using a more appropriate representation of reality.

The explanatory power of the inertia of revenue and expenditure indicates a strong

dynamic component in the levels (and probably in the components) of revenue and

expenditure. The strength of this dynamic component shows how difficult it is, at least in

the short run, to counteract public deficits and advocates for a conservative fiscal policy

which aims to avoid deficits rather than first letting them accumulate and then trying to

reduce their size. This type of policy implies planning ahead for funding future expen-

ditures in addition to keeping the surplus made in the economic growth period for

compensating for the deficit in a recession.

The results on growth rates and unemployment rates show that business cycle con-

tributes significantly to the fiscal position of the Swiss Cantons. These variables can

hardly be influenced to balance the budget and particularly so in our empirical context as

the Swiss Cantons work like small open economies.


Fortunately, there are some variables that the Cantons do have control on. For

instance, our results show that the Cantons can improve their fiscal position (i.e., in-

crease the surplus or lower the deficit) by cutting down the number of departments in

their administration. However, this solution is now almost exhausted because the num-

ber of remaining departments is already quite small (seven or even five).

Regarding the effect of direct democracy instruments, the present findings show that

when fiscal referendum is more easily triggered (with lower thresholds of financial

amount and number of signatures, and a greater time span), it reduces expenditure and

deficits. Hence, it can be recommended that Cantons should strengthen their referendum

possibilities to benefit in terms of deficit reduction. As for the right to put forward

initiatives for new laws, the present results point in the opposite direction: the easier it is

to use this right, the higher would be the expenditure and thus the deficit.

NOTES

The authors thank Alexandre Trechsel for sending data related to referendums and initiatives for theperiod 1970–1996 and Alain Schoenenberger, Pascal Sciarini and Pascal St-Amour and two anon-ymous referees for their helpful comments.

This research was carried out as part of the project ‘‘Analysing the relationship between revenueand expenditure: impact on fiscal balance and econometric modelling’’ (12-67064.01) funded by theSwiss National Science Foundation.

APPENDIX A

TABLE A1

Summary Statisticsa

Domain Unit Minimum Maximum Mean Median

Standard

Deviation

REVENUE <1 CHF/inhab. 3,415 20,734 6,754 6,060 2,671

EXPENDITURE <1 CHF/inhab. 3,291 20,296 6,845 6,124 2,773

DEFICIT < CHF/inhab. � 2,513 1,297 � 90 � 38 432

GROWTH < Percentages � 1.33 4.55 1.67 1.31 1.65

UNEMPLOYMENT < 1 Percentages 0b 7.80 1.80 1.20 1.79

REFERENDUM [0–6] 0 6.00 3.96 4.25 1.32

INITIATIVE [0–6] 1.67 6.00 4.46 5.00 1.31

LANGUAGE 0 or 100 0 100.0 26.9 – 44.4

DEPARTMENTS < 1 5.00 13.00 7.66 7.00 2.33

COALITION < 1 1.00 5.00 3.31 3.00 0.92

RIGHT-WING [0–100] Percentages 33.3 100.0 78.3 80.0 16.0


APPENDIX B: ADJUSTMENTS ON REVENUE AND EXPENDITURE DATA

In this appendix, we explain how certain adjustments had to be made to the revenue and

expenditure data to remove their variation due to ‘‘exceptional’’ eventsFpolitical de-

cisions that induced a major change in their value in a particular year. However, we have

only taken off the direct impact and not tried to simulate how the Cantons would have

behaved (before, during, and after) if these events had not taken place. Moreover, it is

also very difficult to simulate the financial influences of these adjustments on other

variables.

Thus, revenues were adjusted to eliminate the direct influence of the privatization of

the Cantonal bank of Appenzell Ausserrhoden in 1996 (CHF 180 million, which rep-

resents over 57 percent of the Canton’s revenue; percentages are expressed as a function

of the adjusted levels), the conversion of the Cantonal bank of St. Gallen into a public

limited company in 2000 (CHF 500 million, 16 percent) and privatization of the same

bank in 2001 (CHF 160 million, 5 percent) and two exceptional bequests in Tessin in

1988 (6 percent) and in 1994 (almost 5 percent). The revenue was also adjusted to take

into account the collection of extraordinary revenue by Glarus in 2000 in relation to the

tax on the profit of legal entities (CHF 35 million; 10.5 percent)75 and the decline

observed for Zug in 1999.76

Similarly, expenditure was adjusted to neutralize the direct effects of the capitalization

of the Cantonal Bank of Geneva in 1993 (CHF 147 million, over 2.5 percent) and the

stabilization efforts by the same bank through the Revaluation Foundation in 2000

TABLE A1 (Continued)

Domain Unit Minimum Maximum Mean Median

Standard

Deviation

CONCORDANCE [0–100] Percentages 53.3 100.0 86.5 88.4 10.3

ELECTION-YEAR 0 or 1 0 1.00 0.23 – 0.42

ELDERLY [0–100] Percentages 10.3 21.0 14.6 14.6 2.1

Notes:aAll data are from official and centralised sources, namely the Swiss Federal Statistical, Swiss Federal Department

of Finance, except REFERENDUM, INITIATIVE, and DEPARTMENTS. The latter variables are provided by the Database

on Swiss Cantons and Towns (http://www.badac.ch). As we use a dynamic specification with a one-year lag,

summary statistics are for the period 1980–2002.bNo one among the 12,965 residents of the canton of Appenzell Rh. was reported to be unemployed in 1980.

75. While the profit of legal entities generally fluctuates between 20 and 30 million, this revenue had

risen to 60 million for that year. Thus, it may be estimated that the revenue surplus is between 30 and 40.

76. This decline is due to the significant fall in the contributions recorded in 1999. It would appear that

this decline is not real but can be explained by posterior accounting during the closing of accounts. Thus,

the sum of 47 million (6.7 percent) was added to the revenue, calculated through the linear interpolation

between the contributions for 1998 and 2000.


(CHF 19 million, 0.3 percent), 2001 (CHF 194 million, 2.5 percent), and 2002 (294

million, 3.7 percent), the transformation of the Cantonal bank of St. Gallen into a public

limited company in 2000 (CHF 255 million expenditure, over 8.5 percent), the bank-

ruptcy of the Cantonal bank of Solothurn in 1996 (CHF 34 million, 2.7 percent), in 1997

(CHF 34 million, over 2.5 percent), and in 1998 (CHF 125 million, 9 percent), the

recapitalization of the Cantonal bank of Jura in 1997 (CHF 42 million, 7.5 percent) and

the termination in 1994 of the special construction program in the Canton of Jura dating

from 1985 whose smoothing was planned over 25 years (CHF 41 million, almost 9

percent). Further, the expenditure series was adjusted on the basis of complementary

services provided to the pensions and invalidity insurance schemes by Berne in 1986

(CHF 242 million, over 6 percent), the unexplained CHF 36 million break (6 percent;

deviation from the linear interpolation between 1984 and 1986) in the series for Uri’s

expenditure clearly evident in 1985 following the subtraction of transfers relating to the

national roads, the expenditure arising from storms in Obwalden in 2000 (over CHF 16

million, or 7.7 percent), the costs that arose in Valais due to the 2001 storms (over CHF

175 million, or almost 8.5 percent) as well as the re-establishment of hydroelectric power

stations at a cost of CHF 100 million (almost 5 percent).

Finally, it seemed appropriate to rectify two more details in the posting of operations

concerning unemployment insurance. The first one involves the failure to post a loan of

around CHF 104 million granted by Aargau to its unemployment insurance fund in 1993

(i.e., almost 4 percent of the Canton’s expenditure). The second relates to the posting of

two reimbursements in 1994 that were not actually made by Solothurn. The first point

amounts to approximately CHF 63 million and the second approximately 32.5 million.

APPENDIX C

TABLE A3

Unit Root Tests of Maddala and Wua

Variablesb

Levelb Differencec,d

Conclusion

Test With

Constant

Alternative

Tests

Test Without

Constant, or Trend

Alternative

Test

SURPLUS

w2(52) 183.9 t.i. t.u t.i. I(0)

P-value 0.000

REVENUE

w2(52) 52.3 t.i. 214.6 t.i. I(1)

P-value 0.461 0.000

EXPENDITURE

w2(52) 61.5 t.i. 273.2 t.i. I(1)

P-value 0.172 0.000


APPENDIX D


Variablesb

Levelb Differencec,d

Conclusion

Test With

Constant

Alternative

Tests

Test Without

Constant, or Trend

Alternative

Test

GROWTHe

Z(26) � 3.918 t.i. t.u t.i. I(0)

P-value 0.000

UNEMPLOYMENT

w2(52) 195.2 t.i. t.u t.i. I(0)

P-value 0.000

ELDERLY

w2(52) 95.4 47.8 t.u 200.2 I(0)

P-value 0.000 0.640 0.000 or I(1)f

Notes:aResults obtained with STATA 8.2.bSome variables, namely the political ones, cannot be I(1) or more. Therefore, they are not mentioned in the table.ct.i. Indicates that the test is theoretically irrelevant for the considered series.dt.u. Indicates that the test is unnecessary: I(0) series are always I(0) in difference.eAugmented test of David Dickey and Wayne A. Fuller, ‘‘Distribution of the Estimators for Autoregressive Time

Series with a Unit Root,’’ Journal of the American Statistical Association 74 (1979): 427–431 over the period 1976–

2002, that is, 27 observations. The test is used because the concerned variables do not vary in space. Alternative

tests are based on equation in levels with trend and in difference with a constant.fThe ELDERLY variable is I(0) according to the tests with a constant and I(1) according to the tests with a trend, that is,

according to the less likely specification. Graphical analysis clearly allows one to conclude that the variable is I(0).

TABLE A4

Cointegration Tests of Pedronia,b

Panel

v-Stat

Panel

q-Stat

Panel

pp-Stat

Panel

adf-Stat

Group

q-Stat

Group

pp-Stat

Group

adf-Stat

Potential combinations in simultaneous models without lagged variables (1979–2002, i.e., 24

observations)

REVENUE

EXPENDITURE

2.817*** � 3.132*** � 3.384*** � 2.971*** � 1.134 � 2.626*** � 2.498***

EXPENDITURE

REVENUE

4.585*** � 3.228*** � 3.147*** � 3.078*** � 1.061 � 2.261** � 2.697***


APPENDIX E


Panel

v-Stat

Panel

q-Stat

Panel

pp-Stat

Panel

adf-Stat

Group

q-Stat

Group

pp-Stat

Group

adf-Stat

Potential combinations in simultaneous models with lagged variables. Revenue equation (1980–

2002, i.e., 23 observations)

REVENUE

EXPENDITURE

REVENUE(� 1)

9.548*** � 7.976*** � 11.175*** � 7.729*** � 7.086*** � 14.253*** � 9.384***

Potential combinations in simultaneous models with lagged variables. Expenditure equation

(1980–2002, i.e., 23 observations)

EXPENDITURE

REVENUE

EXPENDITURE(� 1)

5.176*** � 5.642*** � 9.021*** � 8.654*** � 3.954*** � 10.244*** � 9.839***

Notes:aResults from a routine developed by Pedroni for RATS (6.0). A single, double, or triple asterisk indicates that no

cointegration is rejected at 90%, 95%, or 99% threshold.bNo simulation study exists that would indicate a preference for one statistics out of the seven computed. There-

fore, cointegration is either accepted or rejected when a majority of the computed tests point toward the former or

the latter hypothesis.

TABLE A5

Unrestricted Structural Models of Revenue and Expenditure Per Capita Derived and Direct

Models for Budget Balance Per Capita Based on Observations on 26 Swiss Cantons Over

the Period 1980–2002

Dependent Variable

Simultaneous, Two-Equation ModelDirect, Single-Equation

Model for ComparisonbStructural Modela Derived Model


REVENUE 0.311

(9.96)

EXPENDITURE 0.127

(3.50)

GROWTH 24.1 � 10.3 26.6 28.5

(3.48) (� 1.49) (3.55) (3.70)

UNEMPLOYMENT F 27.7 � 25.1 � 8.90

(3.55) (� 3.39) (� 0.95)

REFERENDUM F � 37.6 34.1 23.7

(� 2.67) (3.01) (1.72)



Dependent Variable


Model for ComparisonbStructural Modela Derived Model


INITIATIVE 6.08 14.3 � 8.65 � 1.32

(0.47) (1.10) (� 0.70) (� 0.10)

DEPARTMENTS F 11.5 � 10.4 � 10.2

(2.15) (� 2.33) (� 1.89)

RIGHT-WING � 1.59 � 1.74 0.441 0.363

(� 1.53) (� 1.71) (0.39) (0.32)

COALITION � 14.2 1.28 � 11.3 � 3.30

(� 0.86) (0.08) (� 0.73) (� 0.21)

CONCORDANCE 0.708 � 1.93 2.26 1.91

(0.52) (� 1.37) (1.39) (1.15)

ELECTION-YEAR 15.2 43.0 � 28.2 � 28.3

(0.58) (1.71) (� 0.91) (� 0.89)

ELDERLY � 1.84 11.7 � 12.0 � 5.76

(� 0.25) (1.68) (� 1.38) (� 0.71)

LANGUAGE 0.0256 � 0.74 0.691 0.425

(0.07) (� 1.51) (1.58) (0.86)

GENEVA 0.969 304 � 276 � 171

(0.01) (3.72) (� 2.28) (� 1.56)

REVENUE(� 1) 0.878 0.630

(24.2) (16.9)

EXPENDITURE(� 1) 0.677 � 0.615

(21) (� 17.0)

SURPLUS(� 1) 0.657

(16.2)

CONSTANT 150 303 � 167 � 170

(0.81) (1.53) (� 0.85) (� 0.82)

Adjusted R2 0.99 0.99 0.33 0.57

Notes: The numbers shown in brackets beneath the estimated parameters represent the t-stat values.

F, No theoretical influence.a3SLS estimation with lagged value of REFERENDUM and INITIATIVE as instruments and White correction.bEstimation with instrumental variables and with lagged value of REFERENDUM and INITIATIVE as instruments and

White correction.


APPENDIX F

TABLE A6

Standardised Coefficients of the Simultaneous Equation and Direct Single Equation Models

(Coefficients Expressed in Terms of the Standard Deviations of Dependent and Explan-

atory Variables)

Dependent Variable


Model for ComparisonStructural Model Derived Model


REVENUE 0.295

EXPENDITURE 0.125

GROWTH 0.014 � 0.006 0.102 0.110

UNEMPLOYMENT F 0.016 � 0.092 � 0.031

REFERENDUM F � 0.010 0.061 0.047

INITIATIVE r 0.010 � 0.056 � 0.017

DEPARTMENTS F 0.009 � 0.056 � 0.054

RIGHT-WING � 0.0002 � 0.0002 0.001 0.0003

CONCORDANCE r � 0.008 0.048 0.050

ELECTION-YEAR r 0.006 � 0.037 � 0.028

ELDERLY r 0.009 � 0.050 � 0.031

GENEVA r 0.018 � 0.107 � 0.071

REVENUE(� 1) 0.870

EXPENDITURE(� 1) 0.676

SURPLUS(� 1) 0.686a 0.657

Notes: F, No theoretical influence.

r, Removed for empirical reason.aThe net influences of REVENUE(� 1) and of EXPENDITURE(� 1) compensate each other almost completely. Thus, the

standardised influence of these variables is estimated on the basis of their net influence, namely 0.636 �REVENUE(� 1)� 0.625 � EXPENDITURE(� 1)5 0.625 � SURPLUS(� 1)10.011 � REVENUE(� 1).


Documents

Explaining Fiscal Balances with a Simultaneous Equation Model of Revenue and Expenditure: A Case Study of Swiss Cantons Using Panel Data