In the latest of their distinguished contributions to South Asian studies, scholars Lloyd I. Rudolph and Susanne Hoeber Rudolph focus on this modern-day pursuit by offering a comprehensive analysis of India's political economy. India occupies a paradoxical plane among nation states: it is both developed and underdeveloped, rich and poor, strong and weak. These contrasts locate India in the international order.
The Rudolphs' theory of demand and command polities provides a general framework for explaining the special circumstances of the Indian experience. Contrary to what one might expect in a country with great disparities of wealth, no national party, right or left, pursues the politics of class. Instead, the Rudolphs argue, private capital and organized labor in India face a "third actor"—the state. Because of the dominance of the state makes class politics marginal, the state is itself an element in the creation of the centrist-oriented social pluralism that has characterized Indian politics since independence.
The rest of the paper is organised as follows. Section 2 lays out the model. Section 3 presents the econometric analysis of expenditure patterns of the state governments based on the predictions of the model.
Section 4 looks at the question of what determines the likelihood of having a coalition government. Section 5 concludes. Model Consider an endowment economy populated with a continuum of agents. There is borrowing and lending in the economy. A part of the endowment has to be used as a collateral for borrowing in the credit market. There is inequality in the initial endowment distribution and hence some of the agents might be credit constrained.
The economy is subject to aggregate shocks on endowments. However, at any given point the distribution of either agents is never degenerate. These changes in initial endowments affect the distribution of voter preferences to be discussed be- low. The state of endowments is always verifiable.
Agents All Agents are risk averse and altruistic. After receiving the endowment ei1 at the beginning of the period, agents face aggregate shocks that affect them differentially depending on severity of their credit constraints. Non credit constrained agents enter the credit market and trade to smooth consumption. Production of the public goods is financed by a distortionary tax on the non-credit constrained agents.
Examples of typical local pub- lic goods would be improved schools, introduction of meal schemes in the schools, improved health access,etc. These are visible and easily targetable expenditures and hence could be used for smoothing consumption by credit constrained agents.
Agents seek to optimise the expected value of life time consumption, where the expectation is conditioned on the distribution of shocks. Agents cannot enter into any contract before the realisation of shocks and hence there is no private insurance market.
The distribution of shocks has the same properties as the preference shocks described in the subsection that follows. The expectation is over the distribution of shocks. Agents who enter the credit market buy and sell bonds at the price R. Agents behave com- petitively in this market and hence take R as given.
Because borrowing needs collateral, the maximum an agent can borrow is given by the available endowment minus consumption , i. We could understand the agents in this economy as those in Kiyotaki and Moore - farmers and gatherers. Post shock some farmers need to go to the credit mar- ket but only few are left with any land to use as collateral in the credit market.
The remaining become almost or completely landless losing access to the credit market. Agents as voters care about ideology as well as economic policy. Having certain ideology would mean having specific preferences about social and economic justice and caring about the economic policy would imply caring about what kind of public goods are provided by us- ing taxes. Accordingly, credit constrained voters would care about economic policy more than ideology and vice versa.
Political Aspects There are two entities contesting an election, S and M , to form a government at the state level. The single party has an ideologically motivated election platform and the coalition has one promising provision of local public goods.
I will assume that fS is being expert in national issues and politics and fM as having expertise in assessing local public goods requirements. The policies that these parties choose will be aS and aM.
I will assume that the objective of both the political entities is to maximize the probability of winning. Usually in Downsian style models with or without probabilistic voting, we get a result of pol- icy convergence. In equilibrium, the competing candidates or parties choose the same policies and voters become indifferent between candidates see Persson and Tabellini for details. However, in this model we would expect differentiated platforms in equilibrium and voters to be not indifferent between candidates.
This approximates the reality where candidates rarely choose similar platforms and voters certainly seem to favour one candidate over other Krasa and Polborn There have been two ways in which such divergence has been achieved in theory. One way is to assume limited information on candidates in a Downsian setup and the other provided by Krasa and Polborn and Krasa and Polborn In the former paper, the authors specify conditions under which one could have a divergence and in the later they develop a model with multidimensional policy and a binary policy model which is capable of having con- vergence as well as divergence under clearly defined conditions.
In what follows, we adapt the model and an example economy from Krasa and Polborn to illustrate the choice of spending conditional on type of government.
The preference shocks basically act as a counterpart to the distribution of endowments shocks. Models in Downsian tradition with candidates without fixed characteristics satisfy UCR and that leads to similar policies being implemented in equilibrium. However, voter preferences here are non-UCR. Type S voters would primarily be not credit constrained and Type M voters be credit constrained. Though, there might be a certain number of voters who definitely belong to either of the groups, post shock realisation there are some voters who migrate to opposite groups depending on if they become credit constrained or not swing voters.
Policy Platform Equilibrium The above description of the game and voter preferences imply that policy platforms will not converge in equilibrium. The following proposition states this formally. Proposition 1. Proof of Proposition 1. This is because voters rank the entity implementing policy in accordance with its expertise higher than an entity implementing a policy not in accordance with its expertise See the preference description above.
Thus, the non-UCR preferences imply that there is no profitable unilateral deviation for either political entities ensuring aS , aM is the Nash Equilibrium of the political game. Voting equilibrium and Voting Rules Definition 2. A voting equilibrium for this economy is a list of allocations of endowments, debt and consumption of credit constrained and non credit constrained agents such that 1 Proposition 1 holds 2 agents maximise the utility given the distribution of shocks and the budget constraint and 3 given the credit limit based on the initial value of the endowment, the price of bonds clears the credit markets.
We use the above definition of voting equilibrium to derive the equilibrium voting rules. Given this, the voters will populate either groups credit constrained or not credit constrained depending on the following decision rules derived from the comparison of optimisation problem and the definition of the voting equilibrium. Proposition 2. Proof of Proposition 2. If there were no shocks, then given the endowments the agents would solve the utility maximisation problem for the optimal choice of consumption every period.
Such choice would depend on the endowment and hence would change from individual to individual. A short fall means that they become dependent on government expenditure to smooth consumption and therefore will vote based on economic policy than ideology. Accordingly, the follow- ing will hold about the nature of government in equilibrium: Proposition 3.
Stochastic Political Equilibrium : 1. Proof of Proposition 3. It follows from Proposition 1 that the type of government is condi- tional on the type of shocks realised. If the shocks are positive, we have a majority vote for a single party government and if the shocks are negative, the majority vote goes to coalition of regional parties. This emphasizes the role of credit constrained voters as swing voters and that the probability of having a single party or coalition government depends on the probability of type of shock.
Note that a positive probability for shocks implies that the presence of swing voters credit constrained or not depending on shocks and ensures that each type of voter group could end up as pivotal. Because we assume that the policies are implemented and in equilibrium the parties contesting elections choose differentiated policies, the nature of actual spending depends on who is in power. Once the type of government is determined based on the probability of shocks and existence of credit constrained voters, the spending policies are implemented by whichever political en- tity is voted into power.
If a coalition government is voted to power then we can expect the spending on local public goods like education and healthcare access to go up. If a single party government comes to power then spending policies will reflect the ideological preferences than being responsive to local public goods needs. In the empirical analysis that follows we test these implications of Propositions 1, 2 and 3. We test for differences in spending patterns conditional on the type of government as well as what affects the probability of having a particular type of government in the first place.
Econometric Analysis of the Spending Patterns In this section we test the implications of Proposition 3, using data on 17 Indian states for the period of This paper is definitely not the first attempt to do so.
There have been other studies on this issue, as mentioned above. However, they are not without problems. Chaudhuri and Dasgupta and Lalvani use different measures of political fragmen- tation and come to contradictory results in terms of education spending as well as current and capital account spending.
Saez and Sinha seems to be a more definitive analysis com- pared to these two studies. They improve on the earlier studies by including various measures of political fragmentation and confirm that coalition governments spend more on education. Though, this makes the tally in favour of positive effect of political fragmentation on educa- tion spending 2 versus 1, there are several counts on which even their analysis seems incom- plete.
First, they use only one econometric methodology to do so and hence do not provide the required robustness for the results. This constitutes a valid criticism because of the nature of data set being analyzed. Secondly, even though being econometrically more sophisticated than the other two papers, it does not control for GDP at all.
It only has state fixed effects. As much as controlling for unobserved heterogeneity is important, controlling for obvious differences is es- sential for a complete understanding of the underlying economic processes. The econometric analysis in this paper proposes to address these issues by using per capita state GDP as an addi- tional control along with multiple regression specifications.
Accordingly, I analyse expenditure on education, health, irrigation, agriculture and social services. It includes data on state expenditure under various heads and data on various political variables on 17 Indian states.
The coverage in POLEX is limited to the states for whom data is consistently available for the period It does not contain the state GDP data, though. The data on per capita state GDP at constant prices for the states was calculated from the series available in the Handbook of Statistics on Indian Economy maintained by the Reserve Bank of India on line and then incorporated in the POLEX data set to create the one used for analysis in this paper.
The summary statistics for the resulting data set are given in Table 1 : 3. Econometric Issues Because of the nature of the data and smaller N and T 17 and 20 respectively the question of appropriate econometric method becomes pertinent. Maharashtra Saez and Sinha above follow these methods. A standard modelling practice un- der this methodology is to use a fixed effect model with panel corrected standard errors and a lagged dependent variable to account for dynamics. Studies based on such data sets are not limited to political science, however.
Daron Ace- moglu and his coauthors have used such data sets in a series of papers. For example, Acemoglu et al. Much of this analysis is in the mean regression framework and the data is in the TSCS form. Alexander et al. Though, they do not contradict the findings in later, Alexander et al. Their basic argument for using quantile regression is thus, that it allows heterogenous marginal ef- fects across the conditional distribution and that it affords random coefficient interpretation allowing for slope heterogeneity arising out of non-Gaussian distributions2.
The debate is far from settled and hence in this paper, we follow the TSCS consensus method- ology, usual panel data methods often preferred by economists as well as the quantile regression approach to analyse the effect of political cohesiveness on state government spending in India.
In a separate subsection we also analyse the issue of what determines the probability that a state government has a given type of government. All this analysis is guided by the theoretical predictions of the model in the earlier section above.
Use of multiple methods and specifica- tions to test the hypothesis about effect of type of government on expenditure on local public goods serves as built in robustness check for the results.
Political Parties Given that much of the analysis that follows concentrates on political variables, this section describes the players in Indian state level politics briefly. In addition to these national parties there are parties that are dominant in the state regional parties. For example, Shivsena in Maharashtra, Telgu Desam 2 According to the authors, the distributions of two commonly used numerical measures of democracy is bimodal.
Such an alliance would imply a coalition government in the concerned state. Much of the socialist policies implemented after independence could be attributed to the Congress. India not only saw a rising dominance of government in the economy through industrial licensing and a significant public sector under its rule but also went through a brief period of emergency under Prime Minster Indira Gandhi during s. The balance of payment crisis of only made such reforms necessary rather than a political choice.
The current Prime Minister Dr. Manmohan Singh was the Finance Minster during the crisis and oversaw the initial reforms. The BJP had been around in some form or other since independence but rose to promi- nence only around s. The main reason for its rise was the Ramjanmabhoomi movement targeted at the majority Hindu population. Its right wing Hindu revivalist ideology has enabled it to come to the power a couple of times in the center and many more times in the states.
It continues to be the main opposition force to the current ruling Congress party and its allies at the center. The regional parties seem to have organized under several political platforms like linguistic based one or anchored in region specific political social history like the Shiv Sena in Maharash- tra or Telgu Desam in Andhra Pradesh mentioned earlier.
In general specific socio-economic and political conditions within the states seem to give rise to such parties. Regression Specifications As mentioned earlier, a somewhat standard practice under the TSCS methodology is to use a fixed effects model with panel corrected standard errors and lagged dependent variable to account for dynamics Bartels n.
They suggest that the best estimator in this case is to use pooled feasible generalised least squares estimator PFGLS with a distinct AR 1 process for error in each state. However, if T is not much larger than N , then it could lead to a finite sample bias and then it is advisable to at least use the errors still panel corrected but for only panel level heterogeneity. Further, keeping in lines with the suggestion of Cameron and Trivedi , I assume an AR1 process for the error term, while running the panel data fixed effects and random effects regressions.
Following Alexander et al. A detailed description of the results under quantile regression is given in a separate section below. Given the considerable number of specifications implemented and estimated for various expenditure categories, one has to use some rule to conclusively determine if a particular vari- able is a significant predictor of the given expenditure category.
Accordingly, if a particular explanatory variable is statistically significant for the given expenditure category in majority of the specifications, then I deem it as a robust specification. However, I do not distinguish or categorise the significance based on the power of the significance. In order to give a clearer picture of methodology implemented, I have included all the estimations regarding education expenditure in the main body of the paper.
A similar estimation exercise is conducted for other expenditure categories and the summary of results is included towards the end of next section. The detailed results and graphs are available in the appendix. Results In the analysis that follows the presence of a coalition or political fragmentation in general is captured by a coalition dummy variable and two index numbers capturing effective number of parties according to votes and seats effectvt and effectst respectively.
A higher numeric value for these indices signifies lower degree of political cohesiveness. As a variable capturing the effect of political fragmentation in general, these indices seem to be more reliable than the coalition dummy. Given this, the most robust results from all the analysis is about the relationship between education expenditure and the degree of political cohesiveness. From tables 2 and 3 , we can see that the a lower degree of political cohesiveness as measured by effective number of parties has an unambiguous positive and significant effect on education spending under all the specifications.
I repeat similar exercise for other expenditure categories and the detailed results are given in the appendix. The degree of political cohesiveness as measured by effective number of parties or the coalition dummy does not have any effect on health expenditure under any regression specifications. However, in all regressions but one, the variable election is significant and neg- atively related with health expenditure.
This suggests that not only there are political cycles in health spending but it does not seem to be a politically beneficial category of spending. For social expenditure, under all regression specifications except the panel data ones, BJP has a negative and significant effect.
I find similar effect of Congress as well. For irrigation expenditure, the results vary a lot according to specifications used. Hence, it is difficult to say anything conclusively. Having a low degree of political cohesiveness also has a negative effect on agricultural expenditure under TSCS and panel data specifications. Tables 4 and 5 show a comparative analysis for the TSCS specification and Table 6 gives the analysis under panel data specifications.
The analysis for the other expenditure categories is given in the appendix. I find that overall per capita real GDP is almost always significant. However, note that its co- efficient has a semi-elasticity interpretation as we used log transformed per capita GDP figures as a regressor. Additionally, we also find that inclusion of lagged dependent variable as a re- gressor reduces the coefficient on degree of political cohesiveness. This suggests some history dependence in all the expenditure categories and when the given regression does not account for it, the impact is absorbed by degree of political cohesiveness.
For panel data regressions, the indices capturing degree of political fragmentation are significant only under random effects columns 3 and 4 from Table 6 specification suggesting that the unobserved state character- istics relevant for political fragmentation may not be time invariant.
One could potentially include a lagged dependent variable as a regressor in the panel data regressions as well. But doing so, complicates the estimation process substantially because of endogeneity issues.
A way out is to use Arellano-Bond kind of an estimator, but this estimator was developed for short panels where the number of individuals on which we have observations are substantially greater than the time periods. There have been some simulation studies that have shown that application of this estimator to long panels type data leads to significant bias in estimation Baltagi Hence, I do not run regressions of these expenditures on their lagged values under panel data fixed and random effects estimation.
Quantile Regression A quantile regression is a good way of understanding the partial effect of an explanatory variable on various segments of a population Wooldridge
0コメント