The Determination of Municipal Budget Incomes in Lithuania

Transcription

1 The Determination of Municipal Budget Incomes in Lithuania Mark William Shaw Chandler* Submitted for the conference: Government, Market and the Civic Sector: The Search for a Productive Partnership, 9 th NISPAcee Annual Conference March 2001 EuroFaculty Vilnius Centre Ekonomikos Fakultetas Vilniaus Universitetas Sauletekio al. 9, korp LT 2040 Vilnius Lithuania Tel.: Fax.: MARK.CHANDLER@EF.VU.LT

2 Abstract This paper estimates the appropriation of grants to Lithuanian municipalities in the national budget on the basis of public choice models developed by Alperovich (1984), Grossman (1994) and Worthington and Dollery (1998). It pools three years of observations by LSDV and SUR models and also estimates a weighted least squares version corrected for heteroskedasticity. The results overall are mixed and suggest rather weak predictive power of the public choice analysis. They also show that results from the LSDV approach are not robust and there appears to be significant spatial autocorrelation. JEL codes: H77, D78 Keywords: Intergovernmental grants, Public choice, Panels

3 1. Introduction The provision of grants from higher level governments to lower level governments has received consistent attention from economists over the last decade. The application of public choice models has led to a variety of results for a variety of Western economies. The gradual reforms in Eastern and Central Europe made possible the application of these models to a new set of states. These states have now evolved a degree of local government in the sense that separate institutions exist with separate democratic accountability to the national governments. Although these local governments are more centrally controlled than is usually the case in Western democracies they are properly viewed as independent units of government. In this paper I seek to evaluate whether the public choice approach sheds any light on the distribution of grants to Lithuanian municipalities from Lithuania s central government. 2. The Literature The initial approach of political scientists to intergovernmental grants focused on the needs approach. Stein (1981) modeled per-capita direct federal aid to general purpose governments in the U.S. His results for three separate years indicated that the proportion of the aged, dependent children, poverty and crime in the population and the tax burden were consistently significant positive determinants of aid while the education level and home ownership had negative impacts. When economists started working on intergovernmental grants they changed the approach radically. Downs (1957) argued that democratic governments designed policies to maximize their probability of reelection. Alperovich (1984) argued that this could be directly applied to the study of intergovernmental grants. Grants that are predetermined by legislative formula would not be under the discretion of politicians hence he focused on non-formula grants for the larger Jewish municipalities in Israel. The public choice variable Alperovich introduced was the proportion of municipal voters who voted for the party of national government. The other variables included a dependency measure, population, and the municipality s per-capita annual budget deficit. He presented a variety of specifications for two different years but in most cases the political variable was significantly positive, implying that Israeli governments rewarded their political supporters. For the 1978 data logarithmic specification he found a per-capita grants elasticity of 0.28 with respect to this variable. Grossman (1994) extended the analysis of Alperovich by including more political variables in the U.S. institutional environment and by including measures of special interest pressure. He split the political measure into 1

4 two variables, the percentage of votes cast for a Democrat governor in the state (the Democrats controlled the U.S. House of Representatives during the period) and the percentage of seats in the state house of representatives held by the Democrats. Grossman modeled interest group pressure through incorporating state and local government employment per capita and union membership per capita as two additional explanatory variables. His results for total federal grants to state governments in four separate years did not produce significant coefficients for the gubernatorial variable but did indicate a significantly positive coefficient on the percentage of Democrat seats in the state house of representatives for three of the four years. What these studies clearly lacked was any attempt to pool the data from different years. Worthington and Dollery (1998) pooled data for six Australian states over an 11 year period. They modeled three classes of per-capita federal grants using dummy variables for the states. The political variables related not only to votes in state elections, as in Grossman, but also the number of federal seats allocated to the state, the proportion of marginal federal seats in the state and dummy variables for election years. Worthington and Dollery (1998) contrasted ordinary least squares results with the results of a general least squares model corrected for heteroskedasticity and autocorrelation. Other than the dummies no variables were significant throughout all six models. The number of federal seats, proportion of state seats held by the federal government s party, and the proportion of marginal federal seats had significantly positive coefficients in the regressions of welfare spending grants but significantly negative for health grants. Overall the public approach seems most applicable to the welfare expenditure grants in the Australian case. 3. Lithuanian Local Government Lithuania has a three tier system of government, although the middle tier is only administrative regions appointed by the national government and has no separate democratic base. The bottom tier of local government consisted of 56 municipalities until changes introduced in 2000 increased the number to 60. These municipalities have municipal councils elected by their residents and are thus independent political units. Their independence, however, is restricted by the absence of a separate tax base. Both the form and level of taxation is set by the national government. On the expenditure side Lithuanian municipalities face such a large number of mandates from the centre that they often claim to have no discretion and even to have an enforced fiscal deficit due to the impossibility of providing all the services they must provide by law with the funds allocated to them. Martinez (1998) showed that overall transfers in Lithuania prior to the 1997 equalisation reform increased the disparity in revenue across municipalities in contradiction of one of their main goals. He also found that the 2

5 per-capita transfers were not significantly dependent on fiscal capacity, although there was a significantly positive effect from dependency and a city dummy. After 1997 a new equalisation system became law. There has been a gradual evolution in the formula. In 1997 the law established a formula for allocating funds to municipalities other than those covered by special assignments. Lithuanian municipalities formula grants were designed to cover shortfalls in tax revenue receipts, equalise planned revenue, and compensate for differences in cost factors. Changes in the cost factor indicators and their weightings produced a different distribution of funds from year to year. Hence there is scope for political factors to influence decision making in this sphere. We also observe changes from year to year in the number of special grants funded. The 1998 budget listed six special grants, the 1999 budget had three, and in the 2000 budget only one. This also implies a significant role for political decision making during the period. 4. Models The approaches discussed in the literature suggested a number of variables to include in a public choice model of municipal grants. I tested models of both total and per capita grants. For the total grant model the dependent variable is the total formula and non-formula grants allocated to a municipality in the budget. Population (POP) and its square (POPSQ) are important scale variables and give an indication of scale economies. The average earnings (EARN) are included in both public choice and needs based explanations but with opposite predictions. While a needs based approach predicts that towns with lower resident earnings receive higher grants to compensate, the public choice approach predicts that towns with higher earnings will have greater capacity to lobby for greater grants. Hence one test of the framework is the sign on the coefficient on this variable. An institutional variable of significance here is the percentage of local tax revenue kept by the municipality (IT). Towns that kept a higher percentage of their grant according to the formulae were more likely to be towns that qualified for higher grants to equalise revenue. Three key political variables are also suggested by the literature in the Lithuanian institutional environment. The first of these is the number of seats in the national parliament allocated to the municipality (SEATS). Although this is roughly proportional to population there is enough variation in the ratio of seats to population to expect that the more seats allocated to a municipality the greater influence it will have on national decision making. The second is the proportion of a municipality s electorate living in marginal seats (MARP). Since the Conservative Party formed the national government in the period under study I define a marginal seat as where the Conservative Party achieved votes of between 46 and 50 % in the second round of the previous parliamentary 3

6 election. 1 Governments that seek to maximize their probability of reelection would naturally concentrate resources on voters in marginal seats. The third political variable is the proportion of seats held by the Conservative Party in the municipal council (SHP). It measures the tendency, found in previous studies, for national governments to reward their supporters. The ethnic variable for the proportion of the municipality population belong to the Lithuanian ethnic group (LITH) tests the hypthesis that the government seeks to placate minorities by offering them more local funding. For the model of total grants, since some of the right hand side variables relate only to per capita tendencies, I decided to include some interaction terms with population in the model. The variables that related to per capita factors in the municipality are EARN, IT, MARP, SHP, and LITH. These are therefore included together with their interaction with population; POPEARN, POPIT, POPMARP, POPSHP and POPLITH respectively. 5. Empirical Specification I estimated the above model against data on all 56 of Lithuania s municipalities in the budget years 1998, 1999, and Since budgets are prepared and passed at the end of the year previous to their execution I used municipal earnings and population data for 1997, 1998, and The election data came from the parliamentary election of 1996 and the municipal elections of The sources of data were the budget laws for the grants and tax share data, Counties of Lithuania: Economic and Social Development published by Statistics Lithuania (various years) for population and earnings, and Savivaldybiu Tarybu Rinkimai published by the Chief Elections Commission for the municipal election results. I calculated SEATS and MARP from the announcement of the Chief Elections Commission On the boundaries of electoral wards. The most natural specification was to conduct a least squares dummy variable pooling (LSDV) of the three years of data, as described in Greene (2000: 560) and this is what I did first, using both municipality and year dummies (D99 and D00 for the years 1999 and 2000, respectively). 2 Due to the problems of heteroskedasticity that proved to be significant, however, it is necessary to attempt to specify the error variance. I used the squared errors from the LSDV regression and regressed them on the explanatory variables and their squares to obtain predicted error squares. I used these predicted squared errors in the general least squares estimation. An alternative method of pooling the data is to conduct seemingly unrelated regressions (SUR) [Judge, Hill et. al. (1988): 444). My use of a SUR model can be seen as a first attempt to incorporate spatial correlation. 4

7 6. Results Table 1 presents the results of the LSDV OLS and GLS models of total municipal grants for the pooled data. The LSDV model gives a mixed picture for the public choice approach. I do not list the results on the municipal dummy variables but two comments are relevant here. Firstly the data exhibited some perfect multicollinearity that resulted in a few of the dummies being dropped from the regression. Secondly, of the remaining dummies the vast majority were significant at the 1% significance level. The coefficients on the time dummies suggest growth in grants over time. The coefficients on earnings are not significant overall and hence do not give us any indication whether special interest or needs explanations are more powerful here. The positive coefficient on IT implies, as expected, that towns that kept more of their tax revenue were also awarded higher equalisation grants. The negative coefficient on the interaction of IT with population, however, indicates this was not true for larger towns. The political variables performed well overall in the LSDV. SEATS, however, is significantly positive only for the totals regressions. The proportion of municipal council seats held by the party of national government, SHP, is significantly negative across the models, indicating that the government used funds to attract new supporters rather than reward its previous supporters. The proportion of electors in marginal seats is significantly positive in both models indicating again that the political leverage of a municipality has an effect on their allocation. I evaluated the robustness of the LSDV results with a seemingly unrelated regressions approach. These estimates are in Tables 2 and 3 for each year with the associated results from an OLS regression for that year. The results are less encouraging about the efficacy of the public choice approach than the LSDV results were. None of the public choice variables is consistently significant across the SUR regressions and the proportion of voters in marginal seats is significant but with the wrong sign in Tables 2 and 3 also show the Moran z statistic for the OLS regressions, used to test for spatial dependence in the error term. These results indicate a start contrast between the models of total grants and of per capita grants. In each of the former the results reject the null hypothesis of no spatial autocorrelation at the 10% significance level while the null hypothesis can not be rejected in each of the per capita models. The Lagrange Multiplier test for a spatial lag, however, fails to reject the null hypothesis of no spatial lag in all cases except for 1999 total grants. Overall the assumption about the error term makes a large difference to the results obtained from the theoretical public choice model of municipal grant allocation by the Lithuanian parliament. Estimation based on the LSDV assumption that parameters are constant across municipalities contradicts the public choice approach with a significantly negative coefficient on the number of parliamentary seats in the municipality for the total grants case. 5

8 Correcting for heteroskedasticity with a GLS estimation of the model improves the significance of some of the variables, particularly in the per capita model. A first attempt at correcting for spatial error dependence with SUR also gave rise to patchy results. 7. Conclusions This paper applied the public choice explanation of grants from the national government to lower levels of government [Alperovich, Grossman, Worthington and Dollery] to grants to Lithuanian municipalities from It finds only weak support for the public choice approach. The positive finding of the LSDV model is not robust and largely disappears when applying SUR models. Moran s test for spatial error dependence indicates that this is significant for the models of total grants. More work is required to take account of the spatial nature of the data. A more explicit specification of spatial dependency would enable us to see if there are public choice effects that are not significant in the approaches used in this paper. 6

9 Table 1. LSDV Regression Results. (n=168) TOTAL GRANTS GRANTS PER CAPITA Variable OLS GLS OLS GLS Intercept *** *** ** ( ) ( ) ( ) ( ) D *** 32329*** *** *** (13390) ( ) ( ) ( ) D * 11403* ** (10450) ( ) ( ) ( ) EARN ( ) ( ) ( ) (0.9798) IT *** *** ( ) ( ) ( ) ( ) LITH 74079** 92037*** ** (33957) (23703) ( ) ( ) POP ( ) ( ) ( ) ( ) POPSQ *** 9.868E-05*** 1.33E E-07* (1.7E-05) (2.296E-05) (1.11E-07) (6.91E-08) MARP *** *** * ( ) ( ) ( ) ( ) SEATS *** *** (558651) (478822) ( ) ( ) SHP -3.7E+07** *** ** ( ) ( ) (51233) (29462) POPEARN * ( ) ( ) POPIT *** *** ( ) ( ) POPLITH ( ) ( ) POPMARP ** ( ) ( ) POPSHP ( ) ( ) R Adj. R F 12.47*** 31.12*** 16.16*** 17.21*** * Significant at the 10% level. ** Significant at the 5% level. *** Significant at the 1% level. 7

10 Table 2. Seemingly Unrelated Regression (Standard errors in brackets) Variable OLS SUR OLS SUR OLS SUR Intercept *** ** *** *** ( ) ( ) ( ) ( ) ( ) ( ) EARN ** * ( ) ( ) ( ) ( ) ( ) ( ) POP *** *** * ** ** ** ( ) ( ) ( ) ( ) ( ) ( ) POPSQ 2.26E-07*** 2.23E-07*** -1.78E E E E-08 (7.37E-08) (7.23E-08) (1.29E-07) (1.17E-07) (3.50E-08) (3.45E-08) IT ** ** * ** *** ( ) ( ) ( ) ( ) ( ) ( ) SEATS ( ) ( ) ( ) ( ) ( ) (3018.8) MARP *** *** ( ) ( ) ( ) ( ) ( ) ( ) SHP ** ** ( ) ( ) ( ) ( ) ( ) ( ) LITH ( ) ( ) ( ) ( ) ( ) ( ) POPEARN *** *** ** *** *** ( ) ( ) ( ) ( ) ( ) ( ) POPIT ** ** *** *** ( ) ( ) ( ) ( ) ( ) ( ) POPMARP *** *** * * ( ) ( ) ( ) ( ) ( ) ( ) POPSHP ( ) (0.2636) ( ) ( ) ( ) ( ) POPLITH * ** ( ) ( ) ( ) ( ) ( ) ( ) R Adj. R F 19.35*** 3.69*** 10762*** Moran z LM spatial lag test

11 Table 3. SUR Regressions of the Per Capita Grants Variable OLS SUR OLS SUR OLS SUR INTERCEPT *** *** *** *** *** *** EARN *** *** *** *** *** ** POP -9.20E-06** -8.12E-06** ** * ** * 3.65E E E E-06 POPSQ 4.45E-12*** 4.96E-12*** 1.66E-11*** 1.73E-11*** 1.22E-11*** 1.27E-11*** 1.49E E E E E E-12 IT *** *** *** *** *** *** SEATS * MARP SHP * *** LITH ** *** R Adj. R F 15.64*** 8.09*** 7.12*** Moran z LM spatial lag test

12 References Alperovich, Gershon (1984). The economics of choice in the allocation of intergovernmental grants to local authorities, Public Choice, 44: Chief Elections Commision (1997). Savivaldybiu Tarybu Rinkimai [Local Authority Council Elections] (Vilnius: Zenklas). Downs, Anthony (1957). An Economic Theory of Democracy (New York: Harper and Row). Greene, William H. (2000). Econometric Analysis (Upper Saddle River, NJ: Prentice-Hall). Grossman, Phillip J. (1994). A political theory of intergovernmental grants, Public Choice, 78: Judge, George G., Hill, R. Carter, Griffiths, William E., Luktepohl, Helmut, Lee, Tsoung- Chao (1988). Introduction to the Theory and Practise of Econometrics (New York: John Wiley). Martinez, Jorge (1998). Municipal finance, pp in World Bank. Lithuania: An Opportunity for Economic Success, Volume 2: Analytical Background (Washington, DC). Statistics Lithuania (various years). Counties of Lithuania: Economic and Social Development (Vilnius: Statistikos Centras). Stein, Robert (1981). The allocation of federal aid monies: The synthesis of demand- side and supply-side explanations, American Political Science Review, 75: Worthington, Andrew C. and Dolllery, Brian E. (1998). The political determination of intergovernmental grants in Australia, Public Choice, 94:

13 1 Second round elections for parliamentary seats took place where no candidate received more than 50% of the votes in the first round. I did not count seats as marginal if there was no second round. 2 The random effects model is not useful in this contexts since I have data on all Lithuanian municipalities rather than a sample. 11