Fiscal illusion, fiscal consolidation and government expenditure composition in the OECD: a dynamic panel data approach.

Transcription

1 Fiscal illusion iscal consolidation and government expenditure composition in the OECD: a dynamic panel data approach. Ismael Sanz Francisco J. Velázquez European Economy roup -UCM and FUNCAS* September 2002 Abstract Following the present atmosphere o budgetary cuts in the OECD countries we analyze the eects o iscal consolidation in the composition o government expenditures by unctions. We modiy a standard median voter demand model to incorporate a orm o iscal illusion based on the idea that voters-taxpayers may not be ully aware o the true composition o government expenditures because all types o expenditures are not equally visible. Then we exploit the panel structure o the dataset 26 OECD countries over the period by MM estimation o a dynamic model taing into account unobserved country eects and possible endogeinity o the explanatory variables. Under the assumption that governments now the relative visibility o each type o expenditure the pattern o the last three decades indicates that deense non-productive economic services and interest payments are the less visible expenditures. On the other hand education and housing seem to be the more visible expenditures. However income appears as the most important determinant o composition o government expenditures increasing health education social security transport and communication and public services. Other inluencing actors are relative public/private prices age structure o the population and land area. JEL Classiication: D72 D78 H50 C23 Keywords: Composition o government expenditures Median-voter model Dynamic panel data Determinants o public expenditure Ageing. ( * Ismael Sanz (ecap2zl@sis.ucm.es Francisco J. Velázquez (avel@ccee.ucm.es rupo de Economía Europea.Facultad de Ciencias Económicas y Empresariales Universidad Complutense de Madrid Campus de Somosaguas Madrid

2 Acnowledgements We are grateul to Norman emmell Richard Kneller John Ashworth Carmela Martin and Andreas Stephan or their helpul and interesting comments and suggestions on this paper. Part o this research was undertaen while the irst author was a Visiting Research Fellow o the School o Economics o the University o Nottingham. We are grateul or the hospitality received and all the research acilities provided. Financial support rom the European Economy roup (Universidad Complutense de Madrid FUNCAS and Science and Technology Minister o Sapain (CICYT proect SEC C03-0 is grateully acnowledged. We would lie to than also the participants in the Belgirate Meeting o the European Public Choice Society April 2002 and Oviedo meeting o the V Encuentro de Economia Aplicada June 2002 or their comments and suggestions. 2

3 . Introduction As OECD member countries have recently begun to reduce their budgets governments will be orced to mae spending decisions. Fiscal discipline will require cuts in government expenditure leading to trade-o between dierent unctions and aecting government expenditure composition. The pattern o the manner government expenditures size either in expansions or contractions have aected its allocation could shed some light on how iscal discipline might inluence public spending composition in the uture. The composition o the adustment matters in determining the success o a stabilization plan. Thus Zaghini (200 and Ardagna (200 show that iscal consolidations that concentrate on the expenditure side are more eective at achieving household's welare increases and long-lasting reduction o public liabilities than tax-based adustments. Indeed Zaghini (200 shows that policymaers have recently shited their priorities rom a past policy o deicit inancing to one o expenditure reduction policy. Furthermore Alesina and Perotti (995 show that adustments among expenditures are also relevant: those based on social transers and the wage component o public consumption are more persistent than the ones based on investments reduction and earned income tax increases. However many articles argue that iscal adustments tae place reducing investments because as pointed out by Kamps (985 Roubini and Sachs (989 and Oxley and Martin (99 political reasons mae it easier to diminish this type o expenditure. Thus 3

4 Henreson (988 in the case o Sweden and Haan et al. (996 and Sturm (998 or the OECD member states show that iscal adustment aects investments particularly. Through its impact on the composition o public expenditures iscal consolidation can also aect economic growth. Some endogenous growth models incorporate the composition o government spending among other iscal variables which are capable o yielding steady state eects (Devaraan et al. 996 Kneller et al 999. Moreover Davoodi and Zou (998 and Xie Zou and Davoodi (999 show that there is an optimal composition o government expenditures in which the optimal shares o each component equal its growth elasticity divided by the sum o all the unction's growth elasticities. Thereore the reduction o government expenditures can modiy the composition o government expenditures approaching or deviating rom its optimal structure. Nevertheless eects o iscal consolidation on the structure o government expenditures by unctions have received limited attention unlie its impact on the distribution o expenditures by economic type. Now in this article we explore how iscal consolidation has aected each o the unctions o the government expenditures. For this purpose we will modiy a standard median voter model incorporating a type o iscal illusion based in a analogous way to the case o investments on the idea that voters may not be aware o the true composition o government expenditures because all types o expenditures are not equally visible. We will also control or other relevant 4

5 actors in the determination o the composition o government expenditures observed in Sanz and Velazquez (2002. In order to investigate all these aspects in section 2 we review the existing literature on the eects o iscal consolidation on each o the unctions o government expenditures. In section 3 we go on to modiy the standard median voter model including a orm o iscal illusion that allows us to analyze the composition o government expenditures. In section 4 we exploit the panel structure o the dataset by estimating a three stage least square static model each o the composition o government expenditure in 26 OECD countries over the period Thus we will tae into account the unobserved heterogeneity between countries as well as possible endogeneity and simultaneity o the explanatory variables. However adapting to changed median voter demand may be a process requiring a slow adustment in the public expenditure allocation. Hence in section 5 we will analyze the determinants o government expenditure composition in a dynamic model ramewor. For this purpose we will use the eneralized Method o Moments (MM suggested by Arellano and Bond (99 which besides speciic country eects also taes account o endogeneity o the lagged dependent variable. Finally in section 5 we draw the most signiicant conclusions. 2. Determinants o the unctions o government expenditures The literature analyzing government expenditure determinants has ocused on 5

6 explaining the size o the public sector or one o its components independently (Tridimas 200. Many o the studies examining the composition o public expenditures has concentrated in testing Wagner s law or each component (Courais et al. 993 Bairam 995 Chletsos and Kollias 997 Biswal et al.999 and Kolluri et al As these authors argue the use o dissagregated public spending oers a better explanation o the eects o each determinant in the changes o government expenditures. However Wagner-type model may be too simplistic (eorgaopoulos and Loizides 994 so that it may be necessary to loo or determinants other than income to explain public expenditures. Furthermore the composition analyzed is typically that o economic type (consumption transers and investment or civilian and military expenditure. Thus Mauro (998 explores how corruption aects the composition o government expenditures by unctions ater controlling or other actors (such as income per capita and population age structure. He inds that corruption reduced public spending on education and to a lesser extent on health. Neither deense nor transportation displays any signiicant relationship with corruption 2. Some other studies have adopted a consumer demand ramewor in which government maximizes a utility unction and allocation o public consumption expenditures depend on prices o public services and total expenditures (see Tridimas 200 or a survey. Thereater Wagner's law postulates that as per capita income rises in industrializating nations their public sectors will grow in relative importance (Bird 97 see emmell 993 and Peacoc and Scott 2000 or a survey on Wagner's Law. 2 However upta et al. (200 ind that corruption is associated with higher military spending as a share o both DP and total government spending. 6

7 these studies tested and generally do not reect the constraints o the theory o demand. Tridimas (200 elaborates a model where the structure o public consumption expenditure is not only the outcome o an optimizing policy but also o the optimization o individual voter utility unctions. Hence the allocation o expenditures depend on prices o public services and private consumption goods total expenditure the distribution o voter incomes and the expected change in voter support. The scarcity o analyses o the composition o government expenditure leads us to explore the eects o various determinants on less aggregated government expenditure which hopeully will lead to more reliable conclusions. Furthermore we will tae into account all the components o public spending and ocus particularly on the eects o one o its determinants: total government size. I the reduction or expansion o the public sector size modiies some speciic unctions to a greater extent than others then the composition o government expenditures will be altered. The relevance o this result arises rom the act that most iscal consolidations have been achieved by reducing government expenditures instead o increasing revenues (Zaghini 200. Some articles have already analyzed the eect o the reduction o government expenditures in an individual unction. Thus Cashin et al. (200 and Baqir (2002 show or a sample o developing and developed countries during that when governments are aced with the need to cut overall expenditures education and health are relatively more protected. Hics and Kubisch (984 also ind that social spending 7

8 may be more resilient to cuts in real government expenditures than deense and that production and inrastructure expenditures are the most vulnerable. In act Looney (997 suggests that military expenditure ace hard budgetary constraints while Davoodi et al. (999 and upta et al. (200 ind that IMF-supported programs are associated with lower military expenditures. On the other hand Ravallion (2002 inds that iscal consolidation in Argentina in the decade o the 80's and the 90's led to more than proportionate cuts in education health and social security spending. Along these lines Jonain and Stephens (999 reveal that countries o Central America have changed their composition o government expenditures ater entering into ormal agreements o adustments and stabilization policies with international organizations. In particular these countries have reduced the shares devoted to human and physical inrastructure while at the same time increased the shares allocated to deence transers and interest payments. Now in this section we will use the set o variables pointed out by Sanz and Velazquez (2002 as determining the composition o government expenditures or controlling other actors aecting public spending other than the size o public sector. In this respect the unctions will be arranged in the order introduced by Oxley and Martin (99 Saunders (993 Kneller et al. (999 and shown in Table. This classiication is very similar to the one used in the Classiication o Functions o the overnment (COFO United Nations : public services deense health education housing transport and communications social security and other expenditures. Some variables appeared to be common determinants o the eight 8

9 dierent unctions o public spending considered: income relative public/private prices population its density and age structure institutional actors and interrelations among unctions relecting a certain degree o substitution or complementary. [Table ] 3. Theoretical model The base model under which the iscal consolidation and other demographic determinants o public spending by unctions are considered is the median voter model developed rom the studies o Borcherding and Deacon (972 and Bergstrom and oodman (973. In this model it is assumed that citizens vote by means o a maority system and that the size o the public sector is the single issue to be decided. The size o government expenditure will be a unction o the median voter's preerences and income which will be by assumption the median income together with his perception o the public/private sector price ratio. In act the unction o the median voter's demand or government expenditure will be expressed by: i ay α i P ; i 2 N. β gi ( where: i : quantity o public goods and services demanded by the median voter-taxpayer i. P gi : tax-price as perceived by the median voter-taxpayer i. Y i : median voter-taxpayer's i income by assumption the median income. 9

10 α y β: income and price elasticities. N: population. Now the price to be paid or public goods and services may be speciied as: P gi T CN i η (2 where: T i : median voter-taxpayer i share in total tax revenue. C: cost o a unit o public goods and services. η: degree o congestion o public goods and services. As Bocherding and Deacon (972 we assume non-discrimination in taxation 3 (T i /N then P gi CN (η- and substituting (2 in ( we would get: α β β ( η i ayi C N (3 However the voter-taxpayer's i consumption o public goods and services is not ust the level o per capita public spending but: i η N (4 where: : total public spending in real terms demanded by the median voter-taxpayer i. Note that i η is zero it is a case o a pure public good or service whereas i it is unity it will be a private one. Furthermore emmell et al. (999 argue that with this speciication the assumption is that the public/private sector price ratio remains constant 3 This is an assumption o convenience based on data availability that should be modiied wherever better inormation is available. 0

11 over time. To enable this ratio to be modiied in the course o time relative prices (C/Px where Px is the price o the private sector are included. Thereore taing this circumstance into consideration substituting (4 in (3 and aggregating to express the demand unction or the total government expenditure we get the expression 4 : ay α i P β r N (5 where: Y: total income o the country : population elasticity and; ( β ( η η α This is the standard median voter model. Now we will also incorporate demographic actors aecting the utility unction o the median voter. To be speciic we assume that the age structure o the population land area and population density o the country in which the median voter lives relects to some extent characteristics o the median voter. That is we assume that a country with higher mean age will have also an older median voter. Comparing across time as population in a country gets older also does the median voter a actor which will inluence his demand or public services. Certainly some public services such as education and pensions are directed at speciic age groups o population. Thus Biswal et al. (999 and Visco (200 suggest that ageing o the population and the gain in lie expectancy may increase demand or pensions health and social services. In act the age structure o the population was already taen into account in the empirical analysis o 4 We have taing into account that the data relating to total public spending in real terms in the theoretical model would be the nominal expenditure E divided by the tax price P gi. The data that is observed however is the nominal expenditure divided by the unit cost o C.

12 Borcherding and Deacon (972 who pointed to school-aged population rates as relevant in exploring the determinants o cross-sectional variations o per capita public spending and Bergstrom and oodman (973 who include the share o population over 65 as a actor inluencing the expenditure decision o a municipality. The distance o the median voter to the public services will also increase his demand or certain types o public services (Ellis-Williams 987. In act Mueller (989 points out that the actual deinition o public goods and services and externalities connotes geographic proximity. Particularly the distance may aect his demand or transport and communications expenditures. Thus we assume that the larger is the land area o a country the higher is the distance o the median voter to the public services. In act Borcherding and Deacon (972 and Bergstrom and oodman (973 point out that urbanisation and land areas are relevant or the demand o citizens or public services and include these variables in their estimations. As population is already taen into account in the standard median voter model the inclusion o land area relects also the eect o population density which is not ully captured in η. The congestion parameter captured by η relects the divisibility o public service consumption and thereore the amount o public services captured by the median voter assuming non-discrimination in expenditure (Borcherding and Deacon 972. However rural population may be discriminated as the amount o services captured by them may be less than or urban areas population rom a given level o expenditure. 2

13 In short ageing and distance to the public services could create an excess demand or services such as social security health or transport and communications so that more o this public expenditure is demanded than would be in the absence o these demographic actors. Thus observed expenditure will be greater or certain inds o unctions than predicted by a simple standard median voter model. Thereore we assume that the median voter maximises a utility unction which not only depends on public good and services and consumption o a composite private good X but also on the age and distance to the public services o the median voter relected by demographic actors o the country in which he lives Z i subect to a budget constraint such: max i ( X Z U s. t. P X Y i i i gi i i i (6 The inclusion o demographic actors in the utility unction o the median voter still maes that the problem has the orm o an ordinary two-good consumer problem where the price or i is P gi. The demand or i will be o the orm: i ( Y P N L ; 2 3. i gi (7 where: N N 2 N 3 : population in the median voter's country in the age interval o 0-5 years 6-64 years and over 64 respectively. L: land area o the country where the median voter lives. I we suppose again constant elasticities then we could express the demand as: 3

14 3 α β i ayi Pgi where: ( N L l ; i... N. (8 l: age group and land area elasticities. Following the same steps as the restricted model we arrive to: ay α i 3 β Pr N N L l (9 Note that the standard median voter model includes also per capita central government grants to median voter's municipality in the right hand side o the budget constraint equation (6. However as this model will be used or explaining consolidated government expenditures including all levels o administration spending transers between all levels o administration oset each other. Thereore results should be taen with some caution since the median voter model perorm less adequately in explaining public spending or higher levels o administration than or lowest level governments (Turnbull and Mitias 999. In act Oates (985 states that higher levels o administration are less responsive to voters than lower levels o government. To apply the median voter model to the composition o government expenditures it will be useul to assume that the demand or each o the unctions o government expenditures is also decided in a median voter model ramewor. In act the median voter may have dierent demand unctions or each o the types o spending considered according to its income price perception and demographic actors. We assume that there is correlation between these determinants and the median voter relative preerences or each o the eight unctions considered. Functions will be 4

15 arranged in a order very similar to the one used in the Classiication o the Functions o the overnment (COFO United Nations : public services deense health education housing transport and communications social security and other expenditures (mainly interest payments. Voters choose by maority rule the allocation o government expenditures between these eight dierent unctions. That means that total government size is decided simultaneously since it is the sum o each o the unctions. In this set up median tastes dominate i the number o voters is large relative to the number o issues and the preerence peas o the citizenry are normally distributed over the issue space (Tulloc 967. In addition Turnbull and Dounddourias (994 show that median voter demand model perorms stronger when issues are multi-dimensional than in single-dimensional settings: 3 8 α β a Yi Pr N N L ;...8. and C 8 ; P r C C Px where: (0 : quantity o public goods and services devoted to unction demanded by the median voter-taxpayer i. C : cost o a unit o public goods and services o unction. P r : public prices or the unction relative to private prices. : eight unctions considered; 5

16 αβ and l: income price population age structure and land area elasticities o unction. ( β ( η η α η: degree o congestion o public goods and services allocated to unction. Ideally we would speciy a cost unit C or each type o expenditure and the median income. However these data are not available or every country o the OECD and on yearly basis. Thereore we will assume that public delator is the same across unctions so that C C and hence P r P r. Thus will relect the opportunity cost o not having assigned those resources to another unction. We also assume that Y i equals the mean income per capita in every country Y. Now we will consider that voter-taxpayers may not be aware o the true government expenditure because public spending is not totally visible. Further all types o spending are not equally visible. Hence the true composition o government expenditure is a unction o the perceived composition ( and a perception parameter (: ˆ...8; ˆ 8 8 ˆ ( Where measures the degree visibility o unction. A negative value o will indicate that the perceived expenditure in unction is lower than the true expenditure. This will be a sort o iscal illusion aecting the structure o expenditures. For these 6

17 unctions where the median voter perceives the true total government expenditures o this unction will tae a value 0 i.e. is totally visible. The percentage o the expenditure in unction perceived by the median voter / is decreasing in this type o public expenditure. This will also lead to a mismatch between total government expenditures perceived by the median voter and true total public spending as long as total perception parameter is negative: 8 8 ˆ ˆ ; (2 We do not now which characteristics may be lagging behind the visibility o unction. However we hypothesise that governments are ully aware o the dierent visibility o each type o spending as well as the total perception parameter and beneit rom this asymmetric inormation. overnments devoting more resources to more visible expenditures may be seen as oering more public services than those allocating public expenditures to less visible type o spending even when having the same total government expenditure. Moreover governments acing the need or budgetary cuts will reduce the less visible expenditures. For example Kamps (985 Roubini and Sachs (989 and Oxley and Martin (99 Haan et al. (996 and Sturm (998 claim that public investment is less visible than public consumption or transers. Hence when iscal consolidation tae place this reduces investments because political reasons mae it easier to diminish this less visible expenditure. Those expenditures related to long-term proects or with higher degree o pure public good nature such as deence or transport and communications may have 7

18 8 lower visibility than social security education health or general public services. We do not now the total perception parameter but we will hypothesise that governments reveal the relative visibility o each type o expenditure when expanding or contracting total government expenditure. I we want to ocus on the composition o government expenditures rather that in absolute terms rom (2 we get: That is the residual o the variance o government expenditures devoted to a type o spending / that can not be explained by the demand o the median voter / is expected to give indications o how governments beneit rom this asymmetric inormation. Replacing by in (9 by in (0 and taing into account the assumptions o P r P r and Y i Y we get: Hence observed elasticity o each actor is aected by visibility o each type o spending. In addition elasticities in equation (4 are relative to that o the total government expenditure. A zero value should not be interpreted as that the determinant does not aect the demand or the public good but that it does not do so in a way signiicantly dierent rom the rest o government expenditure. A positive and signiicant relative elasticity is thus relecting that this unction is more elastic than total government expenditure. Finally a negative and signiicant relative ( ( ; ; / 3 L N N P Y a a l l r β β α α (3 ; 2...8; ; ˆ ˆ 8

19 elasticity relects is inelastic compare to total government expenditure. Functions with low visibility will have a positive coeicient associated with total government expenditure (- and thereore when total government expenditure decreases (increases unction contracts (expands more than demanded by the median voter. For these unctions with higher visibility (- will be negative indicating that this type o spending will be protected either in contractions or expansions o government expenditures. As unctions will react dierently to government contractions and expansions considering iscal consolidation process will produce a dierent composition o government spending as the standard median voter model. The use o disaggregated expenditures allows each component to have a dierent elasticity in respect o one o the determinants speciied. In act Courais et al. (993 Bairam (995 and Chletsos and Kollias (997 claim that disaggregated expenditure plays an important role leading to more reliable results. Observed response o total government expenditures to a change in a determinant ceteris paribus will be the weighted mean o observed responses to each o its component where the weight is the share o each unction in total government expenditures. From (4: 8 ( a / a Y α α P r β β N 3 N L l l (5 9

20 20 and taing derivatives in both sides yields: It can be argued that population age structure and land area may not be relevant or some unctions i.e. 0 or l0 or some. However as long as these variables are relevant or some any other unction then 0 or l 0 and ( 0 or (l-l 0. That is increased or decreased expenditures needs on other unctions may have a signiicant eect on the share o the rest o the unctions. Finally expressing it in logarithmic orm rearranging terms and considering that the model will be estimated or a panel structure o 26 OECD countries over the period we get: (6 23. ; ; ; ; ; ; l l β β α α

21 2 Where is the country t is the year s the sub-periods considered u is a dummy than taes value or country and 0 otherwise u s is a dummy taing value or country and sub-period s and 0 otherwise and εt is the classical disturbance term. Country ixed eects will be useul to capture preerences and institutional and historical actors speciic to each country and relecting preerences o the median voter in the determination o the share o each unction in total government expenditures. These dummies could be assumed to be constant over large periods (u i related to political systems such as the electoral rule 5 and regime type (Persson and Tabellini 999. We also include country dummies interacted with time dummies (u s to capture those preerences and politico-institutional actors such as ideological preerences (Borge and Rattso 995 Haan and Sturm 997 Sturm 998 and Merriield 2000 or government cohesion (Roubini and Sachs ( [ ] ( ( ( (7 ; ln ln ( ln( ( ln( ( ( ( ln( ( ln( ( ln ln 3 3 t K S s s K l t t t r t t t u u Y L N N N P Y a a t t t ε α α β β α α

22 which could change over large periods 6. As noted earlier we would lie to tae into account the dierent relative prices or each o the unctions and the median income o each country. I we had the delator or each o the unctions and the median income omitting or simplicity the year and country terms we would be estimating: which is the same as: ( ( ( ( ( (9 ln ln ( ln( ( ln( ( ( ( ln( ln ( ln ln( ( / ln( ln 3 3 t l l i i i r i Y L N N N P Y Y C C Y a a E E ε α α β β α α β α α 5 Though the median voter demand model assumes maority rule Downs (957 has shown that in a representative democracy a two party system electoral competition produces essentially the same outcome as the median voter theorem. ( ( ( ( (8 ln ln ( ln( ( ln( ( ( ( ln( ln( ln( ( / ln( / / ln 3 3 t l l i i i r r Y L N N N P P Y a a C E C E ε α α β β α α

23 This expression (9 diers rom (7 we are estimating only by the term [((β/(] ln(c /C and [((α-α(-/(] ln(yi/y. Thus assuming that C is the same or all unctions we do not incorporate any bias. Furthermore the bias will be smaller as the ratio (C /C approaches unity. Alternatively under the assumption that price-elasticity (β/( is close to one or all unctions we will incorporate a small bias in the estimation. Moreover i the ratio (C /C is constant across time the omitted term would only aect to the intercept since we will use panel data techniques. On the other hand i Yi is close to there will be small bias in the estimation which will be zero i the assumption Yi is true. Further i this ratio is constant over time the bias will inluence only the intercept. In act emmell et al. (998 ind that using median and mean income gives the same outcome in the case o local public spending in England and Wales both measures are equally valid representations o the decisive voter. Moreover Pommerehne (978 shows that the mean income its better than the median or representative government as is the case o most o OECD countries. However using the mean income instead o median income we deviate rom the median voter demand model including ad-hoc speciications (Mueller 989. The median voter model is the most widely used in the public choice literature. This model relects political choices o the composition o public expenditures since public expenditure is allocated through a political mechanism. Congleton and Shugart (990 and Congleton and Bennett (995 show that median voter models are stronger in explaining public spending on social security and highway than 6 However it will not be possible to iner rom the results what exactly those actors are which is beyond 23

24 interest group models. This does not mean that interest group or other institutional or redistributive actors are also important but that median voter model hold a slight advantage over models based on these other actors (Ahmed and reene 200. Moreover median voter model allows us to include the relative price o government goods and services which has been proved to be relevant in the determination o the composition o government expenditures (emmell et al In act Pitarais (200 points out that the allocation o consumption public expenditures depends also on the prices o private consumption goods typically ignored by traditional demand theory. Finally the logarithmic orm has been shown to be the most appropriate empirically erdtham (992 while conveniently relating the median voter consumption o government provided services and goods (which are unobservable because the degree o publicness is not nown to expenditures (which are observable. However this model is compatible with a number o other analyses o the allocation o public spending using dierent ramewors other than the median voter. Thus our analysis does also permit a test o unctions or which Wagner's law is particularly applicable. The parameter estimation o relative prices may be also interpreted as the extent to which unctions are exposure to Baumol's conecture 7. The population age group and land parameters may also relect economies o scale in the production o these unctions. Finally the visibility o each type o expenditure is compatible with two other theoretical model ramewors. On the one hand to a the scope o this article ust their signiicance. 7 Baumol (967 observed that productivity growth was lower in the public sector than in private sector while wage increase were similar. Hence the share o government spending in DP will tend to grow in nominal terms. 24

25 separately expenditure component public-choice analysis in which a leadership maximises a welare unction depending on private consumption a expenditure unction other public expenditures and state variables subect to constraint o resources (Hewitt Clements et al.998 Davoodi et al. 999 and upta et al On the other the perception parameter its also compatible to the above mentioned Tridimas (200 model. 4. Econometric analysis In order to test the visibility o each type o expenditure we will use data rom the OECD publication National Accounts. Volume II: Detailed Tables. This source is chosen inasmuch as it oers inormation on the consolidated spending o all levels o government and in addition it ollows the accrual criterion. Thereore this source includes imputed expenditures not necessarily taing place in the current period. Data rom national agencies OECD and World Ban country reports Eurostat: eneral overnment Accounts and Statistics and the IMF publication: overnment Finance Statistics is used on a supplementary basis so as to mae use o OECD data to obtain longer statistical series and supplement the inormative shortcomings o the basic sources 8. 8 Although IMF data covers a longer period o time it is not as a rule consolidated or all the Public Administrations. Thereore it has been necessary to separate transers between dierent administrative levels expenditures (see Easterly and Rebelo 993a Clements et al. 998 and Baqir 2002 or a discussion on the limitations o the data o this publication. Furthermore it uses the cash criterion. Hence its inormation has been used only to estimate the evolution o each unction or the years in which no OECD National Accounts OECD and World Ban country reports national agencies and Eurostat data was available. 25

26 The share o each unction among total government expenditures is calculated at current prices assuming as already mentioned that the delators across types o expenditure are the same. Relative prices are approximated as in emmell et al. (999 by the ratio o the public sector delator to the DP delator. Public sector delator is the result o the weighted mean o government investments delator public consumption delator and public transers the latter represented by the consumer price index all obtained rom the OECD: Economic Outloo and national sources. The per capita income (in Purchasing Power Parities o the 995 dollar and in real terms o that year population and government expenditures series are obtained rom the OECD: National Accounts: Volume I. Main Aggregates while the population age structure is taen rom the OECD: Labour Force Statistics. Some countries did not have data prior to 975 and have had to be completed using domestic sources. Panel data techniques are going to be used or the estimation o the expression (9 since country-speciic unobservables are liely to be correlated with the error term. The inclusion o a country dummy will thereby prevent bias in the estimations at the same time as capturing country eects or each o the sub-periods considered (the '70s '80s and the period available o the '90s that is according to the intervals dierentiated in Tanzi and Schunecht ( Thus we 9 These authors analyse with detail the evolution o the size and scopes o the public sector in developed countries and dierentiate three periods since 970. The irst period covers the decade o the seventies until 980 characterised by a rapid expansion o public expenditures. During the eighties this increase deinitely slowed down ollowing the scepticism about the role o the state emerged in the 70 s. In the irst hal o the 90 s public expenditure increase again as a share o DP until 995 the maximum o the whole period. 26

27 solve a shortcoming o some studies which include a country dummy or the whole period restricting these actors to be invariant over the whole period. In addition panel techniques combine utilisation o the time series and the cross-section variation. Thus studies analysing determinants o health and education expenditures contend that ailure to utilise cross-section variation overestimates the income elasticity (Culyer 988 erdtham 992 Mcuire et al. 993 erdtham et al. 994 Fernández and Rogerson 997 and Di Matteo and Di Matteo 998 Ounade and Karaus 200. On the other hand having multiple observations or each country enables to tae into account dynamics eects and country-speciic ixed eects (erdtham. 992 Hitiris and Posnett 992 and McCosey and Selden 997 Hitiris and Nixon 200. Note moreover that in act it is the estimation o a system composed o eight equations one or every spending unction so two problems arise. First o all these equations are related as median voter tae into account his preerences with regard to all the unctions when deciding the percentages allocated to each unction. Thus the shares devoted to each unction are simultaneously decided and thereore contemporary errors are correlated. In act as shown in (9 i one determinant increase the share o a unction it must decrease the participation o another so that the eect is oset. Moreover the economies o OECD member countries are increasingly interrelated hence common shocs could aect in similar ways to dierent unctions o government expenditures. 27

28 Secondly the inclusion o income and the share o government expenditures in the DP as determinants introduce endogeneity and simultaneity thereore biasing estimates. In act Devaraan et al. (996 and Bleaney et al. (999 show that the composition o public spending is a highly signiicant actor in economic growth 0. In addition the size and scope o total government expenditure is simultaneously decided. Thus government expenditures should be treated as endogenous. In act Pitarais and Tridimas (999 strongly reect the hypothesis o exogeneity o total public consumption expenditure in a study o the allocation o public consumption expenditure or the UK during the period in a Rotterdam model context. These authors conclude that insuicient attention to the issue o the endogeneity o total expenditure in demand equations may lead to biased estimates o the various elasticity coeicients. Moreover total public spending may respond to exogenous changes in one o its components (Davoodi et al Thus the recent reduction in interest payments (included in other expenditures may have contributed to reductions on the size o public spending. Thereore we proceed to estimate by the three stages least squares method. At the initial stage we instrumented income per capita and the share o government expenditures in DP using its lagged values and all the other regressors. In this way at a second stage we perormed an OLS-type two-stage estimation o the covariance 0 This potential source o endogeinity however is reduced i we bear in mind that the eect o the distribution o public spending is not immediate but taes several years to become apparent (Devaraan et al

29 matrix o the disturbance taing into account that it may not be diagonal i.e. contemporaneous errors may be correlated. Finally we perormed a LS-type estimation using the covariance matrix o the previous step. Hence we obtain the same are perorming Seemingly Unrelated Regressions (SURE obtaining the correlation matrix o the residuals between equations and perorming a Breusch-Pagan test or independent equations. In Table 2 we show the results obtained rom the three least squares estimation. Note as already mentioned that elasticities o each unction have to be interpreted in relation to those o total government expenditure. Thus irst o all we carried out a Hausman Test o the null hypothesis o no correlation between the unobservable eects and the explanatory variables. I the hypothesis is reected we choose the single unbiased estimator (within estimates or what amounts to the same the OLS including dummies per country. Hausman test conirms that country unobserved variables seem to be important when explaining government expenditures 2. In addition signiicance o country dummies interacted with time dummies suggest that considering country speciic actors as ixed over time may be inappropriate. [Table 2] Note that each equation has the same set o right-hand-side variables. Thereore the coeicients estimated are the same as i we regress independent equations or each unction. However we gain eiciency by estimating the third phase reducing standard deviations. 2 For the case o social security the null hypothesis is reected only at 0% signiicance. However we decided to introduce country dummies or this unction as well as at all events it impairs the eiciency o estimation but assures its consistency. Arellano and Bover test (990 also reects the absence o correlation between the regressors and the unobservable eects or most o the unctions. This test compares the coeicients in levels and irst dierences so that i these are dierent the hypothesis o absence o correlation between unobservable eects and explanatory variables is reected. 29

30 Many studies have already suggested the relevance o institutional actors determining a type o expenditure. Thus Visco (200 and Than Dang et al. (200 point out to individual country pension systems as the maor determinant in pension spending so that reorms to characteristics as the individual s level o beneit and eligibility could partly oset the impact o ageing on spending. Health expenditures relect a wide range o preerences and historical and institutional actors including the public-private mix decentralisation o public sector systems arrangements and lie style (Leu 986 erdtham et al. 994 Murthy and Upolo 994 Hitiris 999 Ounade and Karaus 2000 Snyder and Yacovelev 2000 Heshmati 200 and Than Dang et al Indeed the degree o decentralisation has also an important role on transport and communications expenditures (Randolph et al Falch and Rattso (997 show that political strength o the central government and ideological orientation are important in the determination o teacher wages and teacher employment and hence in education spending. Furthermore Fernandez and Rogerson (996 and Heinesen (200 point out that this dummy variable could also relect the preerences or private education in some OECD countries. Institutional actors also appear to be important or other expenditures unction. Thus Mongelli (997 predicted an harmonization in interest rates or Member States as a result o the European Economic and Monetary Union (EMU thus reducing interest payments or those countries which had higher inlation/devaluation ris premiums. In addition Maastricht criteria requires a public debt-to-dp ratio below 60% and a public deicit-to-dp ratio below 3%. Finally Clements et al. (998 ind that individual country-speciic actors play an important 30

31 role in determining government subsidies which may be partially considered as economic services. Secondly the Breusch-Pagan test reects the hypothesis o independent equations i.e. the disturbance covariance is not diagonal. This is consisting with median voter taing into account their preerences in respect o all unctions when demanding each one. This result can be suggesting that unctions prove to be complementary or substitutive to a certain degree. Indeed Tait and Heller (982 Heller and Diamond (990 and Clements et al. (998 ind that the signiicance o deence education social welare health and interest payments increases the magnitude o government subsidies and transers as a share o DP. Along these lines Looney (997 points out that deence expenditures increase interest payments and compete with inrastructure or unding. Moreover deence expenditure is identiied as competitor o education and health spending (Baqir In act Davoodi et al. (999 show that the share o nonmilitary expenditures increases in response to exogenous reductions in deence spending an eect which these authors associate with the so called "peace dividend". In contrast with these indings Marlow and Shiers (999 reveal that this expenditure and spending related to public order and security is complementary to that on education. In act Pradhan and Ravallion (998 ind that education increases demand or saety in the case o Brazil because it increases the nowledge about the true probabilities o crime. O the determinants we ind that the income elasticity is signiicant or every 3

32 unction conirming that is the most important determinant o the composition o public expenditures. It has a positive elasticity or public services housing social security education health and transport and communications. On the other hand increases in income per capita decrease the share o other expenditures and deence. As regards the public/private sector price ratio the estimation shows that it is signiicant or our unctions increasing the share o public services and social security while reducing the participation o housing and education. Deence public services transport and communications and housing reveal that they are the closest to have a more purely public nature showing a robust negative relationship with population. At the other extreme other expenditures reveals a positive relative elasticity. Apart rom these basic variables o the model the size o government expenditures land area and age structure o the population prove to be signiicant determinants in the composition o public spending. Total government expenditures appear to increase the share devoted to other expenditures and housing. On the other hand increases in total spending reduce the participation o education transport and communication and social security. Land area increases the needs or higher transport and communications deence general public services and housing at the cost o the participation o social security. This result may be suggesting that land area does not expand demand or social security hence reducing its share in total government expenditures. The eects o population density will be captured by the elasticity o 32

33 population less the elasticity associated with land area. The negative elasticity ound or transport and communications is based according to Randolph et al. (996 on the act that low population density may need higher expenditure or a given level o inrastructure service. Thus per capita inrastructure will be inversely related to population density 3. Moreover as shown by Strum (200 rural areas may demand more inrastructures since they need relatively more spending on inrastructure as compared to urbanised areas. Deence spending shows negative density elasticity. Public services also decreases its shares with population density. Alternatively population density has a positive inluence on the participation o social security. Overall results or population density may be also consistent with a supply view above all with the possibility o taing advantage o economies o scale with transport and communications (Randolph et al. 996 Fay 200. We only ind higher economies o scale in education and health comparing these unctions to social security expenditures (erdtham 992 and Taylor and Bradley Neither we ind any relative positive eect o population density indicating costs associated with congestion in the production o education as suggested by Marlow and Shiers (999 and corroborated by Baqir (2002. As regards population age structure the share o people older than 64 years raises the expenditure allocated to social security presumably through the increase in pensions. In this respect and i the pattern o recent years continues this increase will tae place at the expense o spending on transport and communications and deence. 3 In act our demand or a public expenditure unction can be also interpreted in per capita terms since dividing both and by N in (9 yields the same equation. 33

34 No signiicant eect is ound however o the older population on the health share. For its part the percentage o the population aged 0-5 has an even greater impact on the distribution o public spending by unctions. It has a positive eect on the share o the education health and social security unctions while other expenditures has a negative young population elasticity o great magnitude. 5 Dynamic adustment Adapting to changed median voter demand may be a process requiring a slow adustment in the public expenditure allocation. In act inertia in reormulating the budget results in composition o expenditure responding very slowly to changes in the demand. Hence we may analyse the determinants o government expenditure composition in dynamic model ramewor. In speciying a dynamic model we assume that actual government expenditure does not match the median voter desired level immediately but by a partial adustment process. The obstacles in a rapid adustment may be ound in the act that some o the expenditures are commitments as those related to social security or interest payments and most o the public employees and physical stoc are also ixed. Along these lines some authors have suggested that particular components o government expenditures tend to change reasonably slowly over time (Marlow and Shiers 998 Looney et al. 998 in the case military expenditures erdtham et al. 994 health Fay 2000 transport and communications Hanesen 200 education. The way governments beneit rom the dierent visibility 34