40651 26/4/06 20:15 FJDS_A_168181 (XML)Journal of Development Studies,Vol. 42, No. 4, 640–661, May 2006The Political Determinants of FiscalPolicies in the States of India:An Empirical InvestigationKAUSIK CHAUDHURI* & SUGATO DASGUPTA***Indira Gandhi Institute of Development Research, Mumbai, India, **Jawaharlal Nehru University,New Delhi, IndiaFinal version accepted April 2005ABSTRACT Using data from the 14 major states of India, we investigate whether stategovernments’ ﬁscal policy choices are tempered by political considerations. Our principal ﬁndingsare twofold. First, we show that certain ﬁscal policies experience electoral cycles: stategovernments raise less commodity tax revenue, spend less on the current account, and incur largercapital account developmental expenditures in election years than in all other years. Second, weshow that coalition state governments raise less own non-tax revenues and spend less on thecurrent account than state governments that are more cohesive in composition. In sum, thedispersion of political power aﬀects government size.I. IntroductionWhat determines the ﬁscal policies of a government? While the traditional publicﬁnance literature answers this question from alternative perspectives, it is alwaysassumed that the government is a benevolent social planner, interested in maximisingthe representative citizen’s welfare. In contrast, the recent literature on politicaleconomy emphasises the institutional constraints under which policies areformulated. It argues that policy-makers are typically political parties or politicians.Naturally, the ﬁscal policies that are undertaken are tempered by political factors. This paper tests some of the political economy theories of government behaviourin the context of a developing country. Speciﬁcally, we examine the 14 major statesof India over 21 ﬁnancial years (1974–75 to 1994–95) and ask the following question:Does the proximity of a state legislative assembly election or the extent ofgovernment cohesion aﬀect ﬁscal policies in the states of India?Correspondence Address: Kausik Chaudhuri, IGIDR, Gen. Vaidya Marg, Goregaon (East), Mumbai400 065, India, Email: email@example.comSugato Dasgupta, Centre for Economic Studies and Planning, Jawaharlal Nehru University, New Delhi110 067, India, Email: sugatodasgupta@rediﬀmail.comISSN 0022-0388 Print/1743-9140 Online/06/040640-22 ª 2006 Taylor & FrancisDOI: 10.1080/00220380600682116
Political Factors and Fiscal Policies in Indian States 641 Our focus on India is interesting for two reasons. First, there is a vast literaturethat uses data from developed countries to test for the presence of electoral and otherpolitical considerations in government behaviour. This literature has scarcely beenextended to developing countries (Schuknecht (2000), Block (2002), Shi andSvensson (2002a, b), and Khemani (2004) are recent and notable exceptions). Thereis an obvious reason for the lacuna – democratic institutions have only very recentlytaken root in developing countries. India, on the other hand, has been a democracysince its independence in 1947, and periodic elections to the national and statelegislative assemblies have taken place since 1952. Therefore, the Indian experiencesheds light on political economy models in the context of a developing country withlong-standing democratic traditions.1 Second, it is commonly maintained that votersin India do not condition their votes on economic variables; rather, voting behaviouris exclusively based on caste allegiances. If this is true, political economy models,which invariably hinge on economic factors, are doomed to fail in the Indiancontext. Evidence to the contrary will indicate that voters in India, at least at themargin, are not blind devotees of primordial caste ties. Our use of state level data yields tangible beneﬁts as well. Most studies that seek toestablish electoral cycles in ﬁscal policies proceed one country at a time. Even when arelatively long country-speciﬁc history is observed, the number of election yearsremains small.2 We, on the other hand, obtain information on an aggregate of 68state legislative assembly elections, thereby circumventing the degrees-of-freedomproblem.3 In addition, the cross-section units in our data set (that is, the states ofIndia) are not too disparate entities. Similar arguments also apply when therelationship between government cohesion and ﬁscal policies is examined. We now brieﬂy outline the models that provide the theoretical rationale for ourempirical investigation. Beginning with Nordhaus (1975), several theoretical papershave explored the interactions between election timing and governments’ ﬁscal policychoices. We test the predictions of two inﬂuential models in this genre. In Rogoﬀ andSibert (1988), an opportunistic incumbent government seeks to remain in power. Inan election year, it lowers tax revenues and raises public expenditures to signal itscompetency to the electorate.4 In an interesting variant, Rogoﬀ (1990) allows thegovernment to incur expenditures of two sorts: public consumption and publicinvestment. Since the eﬀects of public investment are imperfectly observed by voters,election-year signalling shifts expenditures in favour of consumption and againstinvestment.5 A separate theoretical strand of the political economy literature emphasises theconnection between the dispersion of political power and budget deﬁcits. Alesina andTabellini (1990) consider a polity with two political parties that diﬀer in their pre-ferences over the composition of public spending. The party in power, fearing anelectoral defeat, issues public debt to constrain the spending capacity of its successor.Hence, the intertemporal sharing of political authority generates budget deﬁcits.6Alesina and Drazen (1993) explore the case of a coalition government that must raisetaxes to balance its budget. Each coalition partner wishes to avoid its share of the taxburden; the ensuing war of attrition generates rising public debt. Hence, the contem-poraneous sharing of political authority facilitates the creation of budget deﬁcits.7,8 Instead of focusing on the government budget deﬁcit, we consider its two maincomponents: expenditures incurred by the government and tax revenues collected.
642 K. Chaudhuri & S. DasguptaAt least two plausible theories have been proposed linking government characteristic(coalition or single-party) to expenditure and tax policies. First, in a coalitiongovernment, each coalition partner normally caters to a narrow constituency of coresupporters. A coalition partner therefore has an incentive to demand targetedprogrammes that are ineﬃciently large relative to ﬁnancing costs, which arecustomarily borne by the citizens in aggregate.9 In sum, coalition governments spendmore than single-party governments. If government spending is roughly matchedby tax revenue, coalition governments also raise more taxes than single-partygovernments. An alternative, albeit informal, argument (see, for example, Dutta(1996a) and Harrinvirta and Mattila (2001)) runs as follows: Coalition governmentsare frequently unstable; hence, the future consequences of current actions areseverely discounted. This general neglect of the future coupled with a desire to buymore time in oﬃce from supporters leads to lower taxes and higher expenditures. Weshall argue (see Section II for the theoretical framework) that the instabilityassociated with coalition governments may actually dampen public spending! Whenindivisibilities force the beneﬁts of public spending in a given year to be conferred ona subset of the coalition partners, the excluded parties in the government will bewilling to pay up their share of the ﬁnancing costs only if there are compensatingfuture gains to be had. The possibility of a government breakdown reduces theexpected value of future beneﬁts and therefore leads to a bargaining impasse.10 Our study contributes to the small but growing literature that uses data fromIndia to test political economy models of government behaviour. A branch of thisliterature explores the links between the timing of elections to the national legislativeassembly and central governments’ economic policies in India. Chibber (1995) showsthat central governments’ food subsidies are higher in election years than in non-election years. Karnik (1990) demonstrates that central government expenditure,aggregated over current and capital accounts, increases when an election isimpending. Sen and Vaidya (1996) fail to uncover electoral cycles in centralgovernments’ current account expenditure, capital account expenditure and currentaccount receipt; however, deﬁcit in the current account is shown to be higher inelection years than in non-election years. In a slightly diﬀerent vein, Chowdhury(1993) detects political surﬁng by central governments – that is, central governmentsstrategically call for national elections when the economy’s output growth turns outto be high. Besley and Burgess (2000, 2002), on the other hand, focus on the behaviourof state governments in India. Besley and Burgess (2000) demonstrate that partyideology aﬀects public policy. Speciﬁcally, the cumulative land reforms passed in agiven state-year depends on the four-year lagged state legislative assembly seat sharesof diﬀerent political groups. Besley and Burgess (2002) show that state governmentsare more responsive to falls in food production and crop damage through ﬂoods wherenewspaper circulation is higher and electoral accountability is greater. Our work is directly linked to two papers that we discuss below. Dutta (1996a)analyses state level data from India to determine whether the tax and expenditurepolicies of coalition governments are markedly diﬀerent from those of single-partygovernments. Coalition governments are shown to have higher levels of currentaccount expenditure and lower levels of non-tax revenue; on the other hand, thelevels of tax revenue and surplus (deﬁcit) in the current account budget areunaﬀected by government fragmentation. Khemani (2004) examines the impact of
Political Factors and Fiscal Policies in Indian States 643state legislative assembly elections on the ﬁscal policies of state governments in India.Her ﬁndings are in striking contrast to the predictions of Rogoﬀ and Sibert (1988)and Rogoﬀ (1990). During election years, Khemani concludes that state govern-ments do not provide tax rate cuts on products that are widely consumed; rather, taxbreaks are given to small groups of producers. Furthermore, the proximity of stateelections leaves state governments’ total expenditure unaﬀected. Instead, thecomposition of spending changes during election years, away from publicconsumption and in favour of investment spending. All of this is viewed as evidenceof election-year manipulation of ﬁscal policies to favour a narrow constituency ofpivotal voters. Our study diﬀers from these two papers in three ways. First, weexamine a substantially larger set of policy variables; the ensuing analysis is thereforemore detailed. Second, we provide a framework in which the policy eﬀects ofgovernment characteristic and election timing are jointly studied. It turns out thatboth of the political variables impact on state governments’ ﬁscal policies. Third, theeconometric model that we specify diﬀers somewhat from that of Dutta (1996a) andKhemani (2004).11 Empirical ﬁndings that survive these diﬀerences can therefore bedeemed to be reasonably robust. The principal ﬁndings of this paper are threefold. (1) State governments collect lesscommodity tax revenue in election years than in non-election years. (2) Spending onthe current account, which has a substantial government consumption component, islower in election years than in non-election years. In contrast, developmentalspending on the capital account, which has a large physical investment component,experiences an election-year spurt. Results (1) and (2) are broadly consistent with theﬁndings of Khemani (2004). (3) Coalition governments raise less own non-taxrevenues and spend less on the current account than governments that are morecohesive in composition. In other words, the dispersion of political power aﬀectsgovernments’ ﬁscal policy choices. The remainder of the paper is structured as follows. Section II presents atheoretical framework to demonstrate that the political instability associated withcoalition governments can actually dampen public spending. Section III provides adescription of the data set used in our analysis. Section IV presents both theeconometric procedures that we employed and the empirical results obtained. SectionV brieﬂy considers the robustness of the empirical results. Section VI concludes.II. A Theoretical FrameworkIn this section, we present a simple model to show that the political instability ofcoalition governments may actually dampen government spending. Consider, to thisend, a two-period discrete-time model in which a government forms in the ﬁrstperiod (that is, period 1). The government consists of two political parties, A and B,each political party representing exactly one quarter of the population of a nation. In period 1, the government has the option to undertake a new project with thefollowing features: (1) the total ﬁnancing cost of the project is c (each political party cin power has to pay up 4), and (2) aggregate beneﬁts from the project are V, whichare evenly distributed over both the time periods. In contrast to Baron (1991, 1998),project beneﬁts within a given period are however not divisible. Speciﬁcally, shouldthe project be undertaken, the period-1 beneﬁt from the project is V to one of the two 2
644 K. Chaudhuri & S. Dasguptapolitical parties in power. Without loss of generality, let party A be this period-1beneﬁciary. Once period 1 concludes, two possibilities arise. First, for exogenous reasons thegovernment terminates with probability p; post-collapse, the project is scrapped andthe continuation payoﬀs of the two erstwhile coalition members are normalised to 0.Second, if government collapse is averted, the project remains in place and party Breceives period-2 beneﬁts of V. 2 Suppose sanctioning of the project requires consent by both parties A and B. Then,for a ﬁxed V, when is the project undertaken? Notice that if party B agrees to ﬁnancethe project, its expected beneﬁt, assuming no discounting, is V Â ð1 À pÞ while its cost 2 cis 4. Hence, the project goes through only if c is weakly less than the threshold value of2V 6 (1 7 p). As government instability (that is, p) increases, the threshold valuedecreases. In other words, the overspending associated with pork barrel politics ismuted when government instability is enhanced. The intuition behind the result istransparent: party B agrees to ﬁnance the project in period 1 only because currentperiod cooperation leads to gains in the future. When government breakdown isimminent, the future is heavily discounted and cooperation becomes diﬃcult tosustain.12 We close the discussion of our model with a ﬁnal remark. Our model is written interms of public investment. A slight reinterpretation of the model allows us toconsider public consumption as well. Here, we can think of the political parties asrepresenting distinct geographical regions, the government potentially engaged inhiring a ﬁxed number of workers at an above-market wage, and certainindivisibilities in the hiring process.III. The DataThe data set for our study consists of annual observations. It spans 21 ﬁnancial years(1974–75 to 1994–95) and covers the 14 major states of India. India comprises 25states and seven union territories. In the ﬁnancial year 1994–95, the aforementioned14 states accounted for 83.1 per cent of India’s land area, 93.2 per cent of herpopulation, and 92.6 per cent of the net domestic product. The variables that we consider partition into three distinct categories: (1) Data onthe ﬁscal (revenue and expenditure) policies of state governments were assembledfrom various volumes of the Reserve Bank of India Bulletin, published by the centralbank of India. (2) State economic and demographic characteristics, which served ascontrol variables in the empirical analysis, were compiled as follows. The NationalAccounts Statistics (published by the Central Statistical Organisation) providedinformation on state domestic product and the proportion of state domestic productderived from the primary sector; state literacy rate data were obtained from theCensus of India and the National Sample Survey rounds (both published by theGovernment of India); while Press in India (published by the Ministry ofInformation and Broadcasting) recorded the per capita newspaper circulation datafor each state-year.13 (3) The political variables were gleaned from two sources. First,the dates of all state legislative assembly elections were taken from the book IndiaDecides. Thereafter, for each state-year, the ‘nature’ of the state government (details
Political Factors and Fiscal Policies in Indian States 645given below) was determined. This information was extracted from the publicationEncyclopaedia of India and Her States. How do we measure the ‘nature’ of state governments? The theory sketched inSection II maintains that ﬁscal policy choices are systematically inﬂuenced by the extentof government cohesion. At ﬁrst blush, this suggests a distinction between single-partyand coalition governments. Such a distinction misses an important point: coalitiongovernments per se do not aﬀect ﬁscal policy choices; rather, what matters is thebargaining between strategically important coalition partners with disparate interests.Following Dutta (1996a, b), we posit that a coalition partner is strategically importantonly if it is a pivotal member of the coalition; that is, if its withdrawal from thegovernment transforms a hitherto winning coalition (with majority support) into anonviable one. A coalition government, in turn, is presumed to distort ﬁscal policychoices should it contain two or more pivotal parties. Henceforth, we shall refer to sucha government as a fragmented coalition.14 For each of the 14 major states of India, column 1 of Table 1 shows the incidenceof fragmented coalition governments. The construction of column 1 proceeds asfollows. First, from the 21-year-long electoral history of each state (ﬁnancial year1974–75 to 1994–95), we calculated the number of months during which thegovernment had multiple pivotal parties. Second, the aggregate months wereconverted into equivalent years. For example, the state of Bihar witnessedgovernments of the aforementioned type for a sum total of 22 months. This is theequivalent of 1.83 years. Column 1 reveals an interesting ﬁnding: most of the 14chosen states have had some exposure to fragmented coalition governments. Table 2 shows the summary statistics of state governments’ ﬁscal policy variablesanalysed in this study. Notice that these variables are of three kinds: revenueTable 1. Incidence of fragmented coalition governments and left-wing governments in the states of IndiaStates Fragmented coalition govt. (in years) Left-wing govt. (in years)Andhra Pradesh 0.00 0.00Bihar 1.83 0.00Gujarat 5.08 0.00Haryana 5.08 0.00Karnataka 1.92 0.00Kerala 16.25 5.83Madhya Pradesh 0.00 0.00Maharashtra 2.08 0.00Orissa 0.17 0.00Punjab 2.75 0.00Rajasthan 0.67 0.00Tamil Nadu 0.00 0.00Uttar Pradesh 1.83 0.00West Bengal 0.00 17.83Note: The sample period covered is ﬁnancial year 1974–75 to 1994–95. ‘Fragmented coalitiongovt. (in years)’ shows the total number of years during which the government of the state wasa fragmented coalition. ‘Left-wing govt. (in years)’ shows the total number of years duringwhich a communist party headed the government of the state.
646 K. Chaudhuri & S. Dasgupta Table 2. Summary statistics of ﬁscal policy variables: (constant 1960–61 rupees per capita)Variables # Obs. Mean Std. dev.RevenueOwn tax revenue 21 35.94 18.50Commodity tax revenue 21 32.15 16.85Own non-tax revenue 21 13.26 8.29Current account expenditureTotal current acct. expenditure 21 74.89 30.58Non-developmental expenditure 21 23.14 11.36Developmental expenditure 21 50.89 20.41Grants to local bodies 21 0.85 0.80Economic services expenditure 21 22.03 11.53Social services expenditure 21 28.99 10.77Capital account expenditureDevelopmental expenditure 21 10.83 5.26Economic services expenditure 21 9.51 5.11Social services expenditure 21 1.38 1.00Education expenditure 21 0.19 0.20Note: The sample period considered is ﬁnancial year 1974–75 to 1994–95. All of the variablesare measured per capita in 1960–61 rupees. ‘# Obs.’ is the number of state-speciﬁcobservations for the variable. ‘Mean’ (‘Std. dev.’) computes the average (standard deviation)of the variable over the sample state-years.variables, and variables related to expenditure on the current and capital accounts.Below, we provide a brief overview of the ﬁscal structure of state governments. The total revenue of a state government has two components: total tax revenueand total non-tax revenue. Averaged over the sample state-years, total tax revenueaccounts for 67.8 per cent of total revenue. A state government’s total tax revenue is,in turn, decomposed into two parts: its share in the tax revenue of the centralgovernment (31.5 per cent), and revenue raised through state taxes (68.5 per cent).State governments levy taxes on agricultural income, property, and commodities.Commodity taxes yield 88.9 per cent of states’ own tax revenue. A stategovernment’s non-tax revenue derives from two sources: grants from the centralgovernment and own non-tax revenue (the three most important subcategories beinginterest receipts from loans issued by the state government, dividends and proﬁtsfrom public sector undertakings owned by the state government, and revenues fromstate lotteries). Our study pays special attention to the own tax revenue and ownnon-tax revenue components of total revenue. Expenditures incurred by state governments are on either the current account (71.1per cent) or the capital account (28.9 per cent). Current account expenditure is of threetypes: developmental spending (68.3 per cent), non-developmental spending (30.6 percent), and grants to local bodies and Panchayati Raj institutions (1.2 per cent).Developmental expenditure, which is largely devoted to the maintenance of existingassets, involves spending on social services (56.7 per cent) and economic services (43.3per cent). Non-developmental expenditure principally comprises interest payments onaccumulated debt and outlays on ﬁscal and administrative services. State governments’ capital account expenditure mainly entails the discharge ofinternal debt, the repayment of loans to the central government, and the provision of
Political Factors and Fiscal Policies in Indian States 647loans for assorted purposes. However, 39.8 per cent of capital account expenditure isdevoted to the creation of physical assets. This developmental component, inturn, involves spending on social services (12.2 per cent) and economic services(87.8 per cent). Our study focusses on capital account developmental expenditure.IV. Empirical ResultsDo political considerations (namely, the proximity of a state election and the extentof government cohesion) aﬀect the ﬁscal policy choices of state governments? Toanswer this question, we consider estimating a ﬁxed-eﬀects error-components modelof the form: Pst ¼ as þ dt þ bxst þ gFragmentst þ oElecst þ est ðs ¼ 1; . . . ; S; t ¼ 1; . . . ; TÞ ð1Þwhere Pst denotes a particular ﬁscal policy variable (for example, the log of percapita own tax revenue) in state s during ﬁnancial year t, as is a state ﬁxed eﬀect, dt isa year eﬀect, xst is a (k61) vector of explanatory variables that capture economicand demographic characteristics of state s in ﬁnancial year t, Fragmentst measuresthe proportion of ﬁnancial year t during which the government of state s had morethan one pivotal party, Elecst (henceforth, referred to as the election year dummy) is azero-one variable that equals one if ﬁnancial year t is an election year in state s, andest is the error term, presumed to be orthogonal to all of the regressors. The politicaltheories that we test deal exclusively with the signiﬁcance, sign and magnitude of theestimates of g and o. The treatment of elections in equation 1 merits scrutiny. First, some rule mustspecify when ﬁnancial year t is deemed to be an election year in state s. FollowingAlesina et al. (1993) and Reid (1998), ﬁnancial year t is called an election year instate s if a state legislative assembly election is held in the second half of ﬁnancialyear t or in the ﬁrst half of the next ﬁnancial year. A more serious problem pertains to the potential endogeneity of elections.15 Theconstitution of India mandates that a state legislative assembly have a normal termof ﬁve years from the date appointed for its ﬁrst sitting. Accordingly, we classify astate legislative assembly election as scheduled if it is held when the current assemblyis at least four years of age. In our data set, the 14 states have experienced anaggregate of 68 state legislative assembly elections; 47 of these elections are classiﬁedas scheduled. What accounts for the remaining 21 mid-term elections? Three circumstances lead to mid-term elections. First, a state government may losethe conﬁdence of a majority in the state legislature. The governor of the state, uponverifying that no claimant can form an alternative government commanding majoritysupport, conventionally calls for fresh elections. Second, the president of India, uponreceipt of a report by the governor of a state or otherwise, may be satisﬁed thatconstitutional breakdown has occurred at the state level. This leads to the temporaryimposition of President’s Rule and, eventually, fresh elections. Third, a state governmentmay voluntarily petition the governor of the state to hold mid-term elections. The third reason for mid-term elections is especially problematic. If the incumbentstate government strategically holds elections for electoral gains, then election timing
648 K. Chaudhuri & S. Dasgupta Table 3. Time duration between successive state legislative assembly elections # elects.; # elects.; # elects.; # elects.; # elects.;States 0–1 yr. 1–2 yrs. 2–3 yrs. 3–4 yrs. !4 yrs.Andhra Pradesh 0 0 1 0 4Bihar 0 0 1 0 4Gujarat 0 0 0 1 4Haryana 0 0 0 1 3Karnataka 0 0 1 0 4Kerala 0 0 2 0 3Madhya Pradesh 0 0 2 0 3Maharashtra 0 0 1 0 4Orissa 0 0 1 1 3Punjab 0 0 1 0 3Rajasthan 0 0 2 0 3Tamil Nadu 0 0 2 0 3Uttar Pradesh 0 1 2 1 2West Bengal 0 0 0 0 4Column Totals 0 1 16 4 47Note: The sample period considered is ﬁnancial year 1974–75 to 1994–95. For each of the 14states, the variable ‘# elects.; 0–1 yr.’ computes the number of state legislative assemblyelections for which the elapsed time since the immediately preceding state legislative assemblyelection is weakly less than 12 months. The other variables are constructed analogously.may be correlated with shocks to ﬁscal policies. Table 3 reports on the time durationbetween successive state legislative assembly elections. Table 3 is constructed asfollows. We proceeded one state at a time. For each state legislative assemblyelection held between ﬁnancial years 1974–75 and 1994–95, we calculated the numberof months that had elapsed between the election under scrutiny and the one directlypreceding it. Thereafter, we computed the total number of elections for which theelapsed time is less than 12 months, between 12 months and 24 months, and so on.Table 3 reveals two patterns of interest: (1) In aggregate, there are 47 scheduledelections and 21 mid-term elections in our sample. (2) Out of the 21 mid-termelections, 17 experienced an elapsed time less than 36 months. Yet, even thisclustering of mid-term elections near the beginning of a government’s term does notnecessarily preclude a strategic accounting of such elections.16 Following Khemani (2004) and Shi and Svensson (2002a, b), we therefore estimateequation 1 with Elecst equal to 1 only if a scheduled election takes place in state-year(s, t). In other words, our empirical work diﬀerentiates scheduled election years fromall other years.17 This estimation strategy does not guarantee that o measures thecausal eﬀect of scheduled elections on public policy. If governments that last for thefull term of ﬁve years are distinct from governments that dissolve rapidly, then ocould simply be reﬂecting policy diﬀerences between the diﬀerent governmenttypes.18 Such concerns about survivorship bias are partially allayed since we includeFragment (a government characteristic variable) as a regressor in equation 1. Inaddition, we add an extra regressor, called Match Dummy, to account for the extentof political aﬃliation between governments at the centre and the state. Speciﬁcally,Match Dummyst is a dummy variable that assumes the value 1(0) if the government
Political Factors and Fiscal Policies in Indian States 649in state s is politically aﬃliated with the central government for more (less) than sixmonths during ﬁnancial year t. We conclude this section with an important caveat. In equation 1, the followingfour variables are included in the vector xst: the log of per capita domesticproduct in state-year (s, t), the proportion of the domestic product in state-year(s, t) that is derived from the primary sector, the literacy rate in state-year (s, t),and the per capita newspaper circulation in state-year (s, t). These variables areintended to control for state speciﬁc factors that aﬀect both state governments’ﬁscal policy choices and government characteristics. Yet, concerns about theinterpretation of g remain warranted. Speciﬁcally, g measures the causal eﬀect ofgovernment fragmentation only if there is no omitted and time-varying statespeciﬁc variable that drives both government fragmentation and public policy.Since the above requirement is not guaranteed, our results should be viewed withcaution.19Empirical Results for State Governments’ RevenuesIn this subsection, we analyse the political determinants of state governments’ ownrevenue. Recall that state governments’ own revenue is the sum of own tax revenueand own non-tax revenue. Columns 1 and 3 of Table 4 present the regression resultsfor, respectively, state governments’ per capita own tax revenue and own non-taxrevenue.20 Note that in neither case is the coeﬃcient of the election year dummystatistically signiﬁcant at conventional levels. Averaged over the state-years in oursample, 88.9 per cent of state governments’ own tax revenue is obtained from taxeslevied on commodities. Given this, we next investigate whether political considera-tions inﬂuence state governments’ commodity tax revenue. Column 2 of Table 4presents the regression results. Here, the dummy variable for election year is Table 4. Impact of political factors on revenue Dependent variables (1) (2) (3) Own tax Commodity tax Own non-tax revenue revenue revenueShare primary sector 70.32 (71.16) 70.37 (71.56) 70.72 (70.86)State domestic product 0.54 (5.32) 0.56 (6.65) 1.23 (3.28)Literacy rate 0.01 (0.82) 0.00 (0.22) 70.04 (71.01)Newspaper circulation 70.12 (70.20) 0.51 (0.90) 0.30 (0.22)Match dummy 70.01 (70.35) 70.00 (70.18) 70.09 (72.05)Election year dummy 70.01 (70.68) 70.02b (71.78) 70.04 (71.44)Government fragmentation 70.01 (70.29) 70.00 (70.35) 70.11b (71.84)R-squared 0.98 0.99 0.84Number of observations 294 294 294Note: Each of the dependent variables is measured per capita in 1960–61 rupees. All of theregressions include state ﬁxed eﬀects and year dummies. The t-ratios, which areheteroskedasticity-robust and corrected for within-state clustering of the error term, are inparentheses; b ¼ signiﬁcance at the 0.10 level (two-tailed test).
650 K. Chaudhuri & S. Dasguptacorrectly signed and statistically signiﬁcant at the 0.10 level; per capita commoditytax revenue collected by state governments is approximately 2 per cent lower inscheduled election years than in all other years.21 The electoral cycle in commoditytax revenue is a robust feature of the Indian experience; Khemani (2004) detectsthis cycle as well despite diﬀerences in model speciﬁcation and a somewhat diﬀerentdata set. How does one interpret the electoral cycle in commodity tax revenue? Stategovernments’ commodity tax revenue derives from three sources: state excise dutieson the production of alcoholic liquor, a producer tax on inter-state trade of goods,and state sales tax. Khemani (2004: 139–40) ﬁnds that scheduled elections lower stategovernments’ revenues from state excise tax and trade tax, but leave revenues fromthe regressive sales tax unaﬀected. Since the ﬁrst two tax categories have a narrowbase, the evidence is viewed as inconsistent with the broad-based tax reductionspredicted by Rogoﬀ and Sibert (1988); instead, the targeted tax relief for a small setof individuals is regarded as pointing towards ‘a story of political purchase ofcampaign support’. Our evidence is more ambiguous. We ran separate regressionsfor state governments’ per capita sales tax revenue and the sum of state governments’per capita excise and trade tax revenue. In both regressions, the coeﬃcient of theelection year dummy is negatively signed but not signiﬁcant at conventional levels(the t-statistics are 71.34 and 71.28, respectively). The government fragmentation variable in Table 4 merits scrutiny as well. In eachof the three regressions reported, the coeﬃcient of Fragment is negative; statisticalsigniﬁcance at the 0.10 level obtains for state governments’ own non-tax revenue.How large in magnitude is this eﬀect? Contrast state-years of two sorts: one typewitnesses 12 months of fragmented coalition government, the other type experiences12 months of single-party rule. Per capita own non-tax revenue is 11 per cent lowerin a state-year of the ﬁrst kind than in a state-year of the second kind. Taken in total,our evidence suggests that fragmented coalition governments raise less revenue thansingle-party governments. In summary, political factors inﬂuence the revenue collected by state governments.The commodity tax revenue is reduced when state legislative assembly elections areimpending. Furthermore, own non-tax revenue declines when the state governmentis a fragmented coalition rather than a single party.Empirical Results for State Governments’ Current Account ExpendituresThis subsection explores the political determinants of state governments’ expenditureon the current account. To this end, column 1 of Table 5 reports the regressionresults for state governments’ per capita total current account expenditure. Contraryto the predictions of Rogoﬀ and Sibert (1988), there is no election year increase inpublic spending. Indeed, the coeﬃcient of the election year dummy is negative andstatistically signiﬁcant at the 0.10 level; per capita total current account expenditureby state governments is approximately 4 per cent lower in scheduled election yearsthan in all other years. How does one interpret the above result? Khemani (2004) also ﬁnds thatscheduled election years are associated with a decrease in state governments’ totalcurrent account expenditure. This, she argues, represents an election-year shift in
Table 5. Impact of political factors on current account expenditure Dependent variables (1) (2) (3) (4) (5) (6) Current Non-developmental Grants to Developmental Social Economic account exp. exp. local bodies exp. services exp. services exp.Share primary sector 0.17 (0.32) 0.92 (1.95) 1.92 (0.98) 70.08 (70.13) 70.59 (72.06) 70.29 (70.36)State domestic product 0.33 (1.91) 70.00 (70.07) 1.74 (1.30) 0.35 (1.84) 0.62 (5.22) 0.37 (1.32)Literacy rate 70.01 (70.47) 0.01 (0.64) 0.17 (3.97) 70.02 (71.79) 70.01 (70.47) 70.02 (71.30)Newspaper circulation 1.08 (1.08) 2.66 (1.62) 4.58 (1.04) 0.14 (0.88) 70.83 (71.56) 71.59 (71.65)Match dummy 70.02 (70.84) 70.03 (71.05) 0.11 (0.66) 70.02 (70.49) 70.07 (72.79) 0.00 (0.09)Election year dummy 70.04b (71.66) 70.05b (71.72) 70.16a (72.37) 70.03 (71.26) 70.01 (70.27) 70.01 (70.39)Government fragmentation 70.04b (71.84) 0.01 (0.16) 70.21 (70.63) 70.06b (71.88) 70.02 (71.00) 70.10 (71.53)R-squared 0.94 0.93 0.75 0.93 0.96 0.91Number of observations 294 294 294 294 294 294Note: Each of the dependent variables is measured per capita in 1960–61 rupees. All of the regressions include state ﬁxed eﬀects and year dummies.The t-ratios, which are heteroskedasticity-robust and corrected for within-state clustering of the error term, are in parentheses; a ¼ signiﬁcance at the0.05 level (two-tailed test), and b ¼ signiﬁcance at the 0.10 level (two-tailed test). Political Factors and Fiscal Policies in Indian States 651
652 K. Chaudhuri & S. Dasguptathe composition of spending, away from public consumption and towardsphysical investments. However, what drives the electoral cycle in total currentaccount expenditure? Columns 2, 3, 4, 5 and 6 of Table 5 report the regressionresults for, respectively, state governments’ per capita non-developmentalexpenditure on the current account, per capita grants to local bodies andPanchayati Raj institutions, per capita developmental expenditure on the currentaccount, and the two components of developmental expenditure (that is, percapita expenditure on social services and economic services). Our results showthat the electoral cycle in total current account expenditure is based on theelectoral cycles in non-developmental current account expenditure and grants tolocal bodies. In other words, during a scheduled election year, the incumbentgovernment does not tinker with subsidies or wages to public sector employees(these are items in the developmental expenditure category); rather, currentaccount expenditure is pruned at the expense of local bodies and by organisingthe functioning of government at a lower cost.22 The government fragmentation variable in Table 5 elicits two observations. First,the coeﬃcient of Fragment is negative and statistically signiﬁcant at the 0.10 level forper capita total current account expenditure. The magnitude of the coeﬃcientindicates that a state-year with 12 months of fragmented coalition governmentexperiences 4 per cent lower spending on the current account than a state-year with12 months of single-party rule. Second, the eﬀect of government fragmentation ontotal current account expenditure works through the developmental component.Column 4 shows that per capita developmental expenditure on the current account is6 per cent lower in a state-year with 12 months of fragmented coalition governmentthan in a state-year with 12 months of single-party government. In summary, the current account expenditure of state governments is inﬂuenced bypolitical factors. In striking contrast to the predictions of Rogoﬀ and Sibert (1988),current account spending by state governments is lower in scheduled election yearsthan in all other years. There is also some evidence that government fragmentationreduces total current account expenditure and developmental expenditure on thecurrent account.Empirical Results for State Governments’ Capital Account ExpendituresAs mentioned in Section III, state governments’ per capita total capital accountexpenditure consists of several items (for example, the discharge of internal debt, therepayment of loans to the central government, and the provision of loans for assortedpurposes) that are not readily manipulable. Accordingly, we emphasise here themanipulable component of per capita total capital account expenditure – namely, percapita capital account developmental expenditure. Column 1 of Table 6 reports theregression results for state governments’ per capita capital account developmentalexpenditure. Notice that in sharp contrast to the predictions of Rogoﬀ (1990), thecoeﬃcient of the election year dummy is positively signed and statistically signiﬁcant atthe 0.10 level; per capita capital account developmental expenditure by state governmentsis approximately 6 per cent higher in scheduled election years than in all other years. The above ﬁnding elicits two observations. First, Khemani (2004) detects anelectoral cycle in state governments’ per capita total capital account expenditure and
Table 6. Impact of political factors on capital account expenditure Dependent variables (1) (2) (3) (4) Developmental exp. Social services exp. Economic services exp. Education exp.Share primary sector 70.13 (70.12) 0.64 (0.47) 70.31 (1.20) 0.64 (0.36)State domestic product 0.60 (1.21) 0.66 (1.21) 0.63 (0.60) 2.23 (4.41)Literacy rate 70.02 (70.41) 70.09 (71.68) 70.01 (70.16) 70.12 (72.35)Newspaper circulation 0.10 (0.08) 1.72 (1.32) 70.17 (70.11) 7.77 (1.64)Match dummy 0.11 (1.49) 70.03 (70.41) 0.12 (0.08) 70.11 (70.72)Election year dummy 0.06b (1.69) 70.02 (70.33) 0.07 (1.61) 0.15a (2.09)Government fragmentation 70.09 (70.81) 0.01 (0.09) 70.11 (70.73) 70.27a (71.97)R-squared 0.75 0.73 0.72 0.74Number of observations 292 293 292 292Note: Each of the dependent variables is measured per capita in 1960–61 rupees. All of the regressions include state ﬁxed eﬀects and year dummies.The t-ratios, which are heteroskedasticity-robust and corrected for within-state clustering of the error term, are in parentheses; a ¼ signiﬁcance at the0.05 level (two-tailed test), and b ¼ signiﬁcance at the 0.10 level (two-tailed test). Political Factors and Fiscal Policies in Indian States 653
654 K. Chaudhuri & S. Dasguptaargues that the ﬂexibility of public investment spending allows the incumbent partyto provide election-year targeted beneﬁts to a select group of pivotal voters. If thisconjecture is accurate, then an electoral cycle should perforce crop up in thedevelopmental component of total capital account expenditure. We have shown thatthis is indeed the case. Second, in a panel study of 24 developing countries for the1973–92 period, Schuknecht (2000) ﬁnds prominent electoral cycles in publicinvestment. Our study extends this ﬁnding to a diﬀerent setting. Capital account developmental expenditure has two components: spending onsocial services and spending on economic services. Accordingly, we ran separateregressions with state governments’ per capita social services expenditure andeconomic services expenditure as regressands. For economic services expenditure,the coeﬃcient of the election year dummy is positive with a p-value of 0.108. Incontrast, for social services expenditure, the coeﬃcient of the election year dummy isnegative and far from achieving statistical signiﬁcance. This suggests that theelectoral cycle in capital account developmental expenditure stems from thebehaviour of economic services expenditure. We also estimated separate regressions for three expenditure subcategories ofsocial services (education, medical and public health, and water supply andsanitation) and two expenditure subcategories of economic services (agriculture andallied activities, and industry and minerals). An electoral cycle crops up only in thecase of education expenditure (column 4 of Table 6). The size of the electoral cycle ineducation expenditure is large: per capita capital account education expenditure bystate governments is approximately 15 per cent higher in scheduled election yearsthan in all other years. Does government cohesion aﬀect spending on the capital account? Table 6 showsthat the coeﬃcient of Fragment is negative and statistically signiﬁcant for stategovernments’ education expenditure. This eﬀect is substantial in size. Contrast state-years of two sorts: one type witnesses 12 months of fragmented coalitiongovernment, the other type experiences 12 months of single-party rule. Per capitacapital account education expenditure is 27 per cent lower in a state-year of the ﬁrstkind than in a state-year of the latter variety. We summarise the ﬁndings of Table 6 as follows. First, in contrast to thepredictions of Rogoﬀ (1990), there is no election-year decline in capital accountexpenditure. Indeed, we observe election-year spurts in capital account develop-mental expenditure and capital account education expenditure. Second, fragmentedcoalitions spend less on education than single-party governments.V. Robustness of the ResultsIn a seminal paper, Hibbs (1977) modelled political parties as partisan rather thanopportunistic. In other words, political parties were assumed to have intrinsicpreferences deﬁned on a policy space. Once in oﬃce, these political parties simplyimplemented their favoured policies.23 We, on the other hand, have allowed no rolefor government ideology as a determinant of ﬁscal outcomes (see equation 1). This isbecause most political parties in India are organised around linguistic, regional,religious and caste aﬃliations (Weiner, 1989); they simply do not have naturallocations on a conventional left–right scale.24
Political Factors and Fiscal Policies in Indian States 655 The above diﬃculties notwithstanding, we did experiment brieﬂy with governmentideology. Speciﬁcally, we distinguished between a left-wing government (that is, agovernment headed by a communist party) and all other government types.Column 2 of Table 1 details the incidence of left-wing governments in the 14 majorstates of India. Column 2 is constructed as follows. From the electoral history ofeach state, we ﬁrst computed the number of months during which a left-winggovernment was in place. Subsequently, the aggregate months were converted intoequivalent years. Column 2 indicates that only two of the 14 states, Kerala and WestBengal, have witnessed left-wing state governments. To study the impact of government ideology on ﬁscal policies, we constructed avariable, LeftWingst, that measures the proportion of ﬁnancial year t during whichthe government of state s was a left-wing government. Inclusion of this variable as aregressor yielded two conclusions. First, out of the 13 policy variables in Tables 4through 6, LeftWingst is statistically signiﬁcant at conventional levels in threecases. Speciﬁcally, the coeﬃcient of LeftWingst is 70.09 (t-statistic equals 71.95),70.52 (t-statistic equals 72.58) and 70.80 (t-statistic equals 74.25) for,respectively, non-developmental expenditure on the current account, social servicesexpenditure on the capital account, and education expenditure on the capitalaccount. Second, while the above ﬁndings are perhaps contrary to expectation, itturns out that the results for Elecst and Fragmentst are entirely insensitive to theinclusion of the ideology variable. In Section IV, the electoral cycle in expenditure variables was estimated byconsidering scheduled election years only. The logical justiﬁcation for thisapproach has previously been given. What happens, however, when the regressionsare re-run without diﬀerentiating between scheduled and mid-term elections?25Three observations are relevant here. First, the eﬀects of government fragmenta-tion are robust and do not depend on how the election year dummy is coded.Second, for the current account expenditure categories, there is no hint of anelectoral cycle: the t-statistic corresponding to the election year dummy rangesfrom a high of 0.99 (developmental expenditure) to a low of 0.01 (social servicesexpenditure). In other words, as in Section IV, our ﬁndings continue to violate thepredictions of Rogoﬀ and Sibert (1988). Third, in Section IV, we had notedelection-year spurts in capital account developmental expenditure and educationexpenditure, thereby contradicting the predictions of Rogoﬀ (1990). Now, theelectoral cycle in capital account developmental expenditure disappears. Summingup, insofar as expenditure variables are concerned, our evidence lends scantsupport to Rogoﬀ and Sibert (1988) and Rogoﬀ (1990) regardless of how theelection year dummy is coded. Finally, to allay concerns regarding the potential endogeneity of the (scheduled)election year dummy, we estimated panel regressions instrumenting Elecst inequation 1. Our instrument takes the value of 1 in the ﬁfth year after every statelegislative assembly election (mid-term or scheduled), and is 0 otherwise.26,27 Ourﬁndings can be summarised as follows. First, the eﬀects of government fragmenta-tion do not change when Elecst is instrumented. Second, consider the revenue resultsin Table 4. The dummy for election year remains negatively signed for own taxrevenue and commodity tax revenue. However, statistical signiﬁcance is no longerobtained for commodity tax revenue. Third, consider the current account
656 K. Chaudhuri & S. Dasguptaexpenditure results in Table 5. The election year dummy remains negatively signedfor all of the expenditure categories; statistical signiﬁcance at the 0.05 level obtainsfor total current account expenditure and current account non-developmentalexpenditure. Fourth, consider the capital account expenditure results in Table 6.Here, the election year dummy turns up positively signed for developmentalexpenditure and education expenditure; however, in neither case is statisticalsigniﬁcance obtained. While our results undoubtedly deteriorate when Elecst isinstrumented, note that we continue to ﬁnd no support for the predictions of Rogoﬀand Sibert (1988) and Rogoﬀ (1990).VI. ConclusionUsing state level data from India, we have studied whether the ﬁscal policies of stategovernments are systematically aﬀected by the proximity of scheduled state electionsand the extent of government cohesion. Our ﬁndings are twofold. First, stategovernments raise less commodity tax revenue, spend less on the current account,and incur larger capital account developmental expenditures in scheduled electionyears than in all other years. Consistent with Khemani (2004), we detect no election-year splurges in current account expenditure (as predicted by Rogoﬀ and Sibert(1988)) and no election-year contractions in capital account spending (as predictedby Rogoﬀ (1990)). Second, our evidence suggests that fragmented coalitiongovernments are smaller than their single-party counterparts. Relative to single-party governments, fragmented coalitions have lower levels of own non-tax revenueand current account expenditure. We note that many empirical questions remain to be explored. In this study,government expenditure was analysed. Variations in expenditure do not necessarilytranslate into corresponding movements in the provision of public goods. Hence,there is a need to estimate the impact of political factors on the supply of variouspublic goods. Frey and Schneider (1978a, b) and Schultz (1995) have independently suggested anamendment to the politico–economic models of government policy. Since policymanipulation is presumed to be costly (for example, there may be a loss inreputational capital), the extent of government manoeuvring depends on the value ofthe marginal vote. Thus, safe elections should not witness policy distortions. Indianstate level data provide an ideal testing ground for this intriguing observation (seenotes for details).28 Finally, we have presented but one half of the complete story. Speciﬁcally, whilegovernments’ ﬁscal policy choices were analysed, voter behaviour was unmodelled.Does the electorate condition its vote on ﬁscal variables? Some evidence, employingUS and OECD data, already exists. Peltzman (1992) and Lowry et al. (1998) ﬁndclear ﬁscal policy eﬀects in the vote data for US presidential, senatorial andgubernatorial elections. Besley and Case (1995b) show that US governors whose taxhikes are less than in adjacent states have a higher probability of re-election. Alesinaet al. (1998) use OECD data to demonstrate that government ﬁscal restraint is notsystematically followed by a popularity decline. Comparable work with Indian datais non-existent. In sum, the analysis of voter behaviour in India remains a fruitfuland unexplored research topic.
Political Factors and Fiscal Policies in Indian States 657AcknowledgementsThe authors would like to thank seminar participants at Indira Gandhi Institute ofDevelopment Research, Indian Statistical Institute (Delhi Centre), and University ofSydney for helpful comments. Thanks are especially due to Lekha Chakraborty,Pinaki Chakraborty, Saumen Chattopadhyay, Bhaskar Dutta, ShubhashisGangopadhyay, Jyotsna Jalan, Kirit Parikh and Arijit Sen for numerous helpfuldiscussions. Two anonymous referees suggested changes that greatly improved thequality of the paper. Of course, the usual disclaimer applies. S. Dasgupta’s researchwas supported by a grant from PPRU.Notes1. Khemani (2004) makes this point as well.2. For example, Alesina and Sachs (1988) study ten US presidential elections; Andrikopoulos et al. (1998) study ten national elections in Greece; Chowdhury (1993) studies seven national elections in India; Heckelman and Berument (1998) study ten national elections in Japan; while Schultz (1995) and Serletis and Afxentiou (1998) study, respectively, nine and twenty national elections in Great Britain and Canada.3. To alleviate the degrees-of-freedom problem, recent studies have examined cross-country evidence. Thus, Alesina and Roubini (1992), Alesina et al. (1993), De Haan and Sturm (1997), De Haan et al. (1999), Harrinvirta and Mattila (2001), and Perotti and Kontopoulos (2002) consider panels of OECD countries.4. Since this article repeatedly refers to Rogoﬀ and Sibert (1988), it may be worthwhile to provide a brief sketch of the mechanism at work in that paper. The set-up of the Rogoﬀ-Sibert model is as follows. An incumbent government, which may be intrinsically competent or incompetent, provides certain ﬁxed services to citizens. To pay for these services, revenues must be raised through any combination of non-distortionary taxes (that voters observe immediately) and distortionary taxes (that voters observe with a lag). To provide the given services, competent governments naturally require less aggregate tax revenue than incompetent governments. Suppose the incumbent government is competent. In an election year, it convinces the electorate of its competency by lowering the level of visible non-distortionary tax revenues. Why does an incompetent government not mimic the behaviour of a competent government, thereby enhancing its electoral prospects as well? This is because mimicking forces the incompetent government to raise very large revenues through distortionary taxes; the consequent damage to voter welfare serves as an eﬀective deterrent. The argument for electoral cycles in expenditures, omitted for brevity, is identical in spirit to that for the tax revenues case.5. Several studies furnish empirical support for the signalling models. Besley and Case (1995a), for example, examine the behaviour of governors in the US; state taxes are shown to be lower when the incumbent governor stands for re-election.6. De Haan and Sturm (1994) provide corroborating evidence: for member countries of the European Community, the growth of government debt is positively related to the frequency of government changes.7. Empirical evidence for this theory is mixed. Alt and Lowry (1994) and Poterba (1994, 1996) examine state governments in the United States; adjustments to ﬁscal shocks are demonstrated to be slower when a state government is ‘divided’ (that is, the executive and legislative branches are controlled by diﬀerent political parties). Roubini and Sachs (1989a, b) study the ﬁscal histories of several OECD countries; budget deﬁcits are shown to be correlated with the extent of government fragmentation. However, De Haan and Sturm (1994, 1997), De Haan et al. (1999), Volkerink and De Haan (2001), and Perotti and Kontopoulos (2002) argue persuasively that the Roubini-Sachs conclusions are fragile.8. While Alesina and Tabellini (1990) and Alesina and Drazen (1993) explore the link between budget deﬁcits and the fragmentation of political power, we alert the reader to a complementary strand of work (see, for example, Alesina et al. (1999), Harrinvirta and Mattila (2001), Perotti and Kontopoulos
658 K. Chaudhuri & S. Dasgupta (2002)) that uses cross-country panels to show that budget institutions (for example, ex ante spending limits or borrowing constraints) robustly predict budget deﬁcits and government spending. 9. Weingast et al. (1981) were the ﬁrst to argue that in a system wherein geography forms the basis for political representation, there is an inherent tendency for programmes to exceed eﬃcient scale. We adapt their insights to the problem at hand. To make matters concrete, consider a single-party government that represents exactly one half of the population of a nation. This government will implement a targeted programme should its beneﬁt–cost ratio exceed 0.5. Consider a two-party government in which each coalition partner represents one quarter of the population of a nation. A coalition partner will now seek to implement a targeted programme should its beneﬁt–cost ratio exceed 0.25. Hence, the fragmentation of a government leads to increased public spending (see also Perotti and Kontopoulos (2002) for an extended discussion).10. Boix (1997) analyses the privatisation experiences of various countries and hints at a similar treatment of coalition governments. In contrast to single-party governments, coalition governments are posited to generate gridlock eﬀects: ‘parties within the coalition are prone to veto each other’s projects whenever the resulting policies are believed to result in signiﬁcant costs for their corresponding constituencies’.11. Dutta (1996a), for example, does not include state ﬁxed eﬀects in his model and uses state speciﬁc time trends instead of year dummies. The set of controls in Khemani (2004) (state net domestic product, share of agriculture in state net domestic product, and proportion of state population that is rural) is diﬀerent from ours, the set of states considered diﬀers as well (she includes Assam in her list of 14 states while we include Haryana instead), and so on.12. Our model imposes period-by-period budget balance. Since unstable coalition governments spend less than stable governments, they raise less revenues as well. This revenue result will persist in a less restrictive model so long as expenditures and revenues move together (that is, are positively correlated). In our data set, the correlation coeﬃcient between state governments’ own tax revenue and current account developmental expenditure is 0.86, while that between state governments’ own tax revenue and capital account developmental expenditure is 0.27.13. For certain years, state literacy rate data were not available from either the Census of India or the National Sample Survey rounds. For these years, we have interpolated the data using a simple growth rate formula.14. In the Indian context, it is not uncommon to ﬁnd coalition governments with one pivotal party. For instance, the state of West Bengal has been ruled by a coalition called the Left Front since 1977. The Communist Party of India (Marxist), which is a member of this coalition, has won an absolute majority of legislative assembly seats in each of the state elections since 1977.15. The possibility of election endogeneity in a parliamentary system has been recognised by several researchers: Cargill and Hutchison (1991) explore this issue in the case of Japan, Chowdhury (1993) and Khemani (2004) focus on India, Heckelman and Berument (1998) study both Japan and the United Kingdom, while Reid (1998) looks at the Canadian situation.16. The authors thank an anonymous referee for emphasising this point.17. Section V discusses the sensitivity of the regression results to an alternative coding of Elecst.18. Khemani (2004) makes this point as well.19. The authors thank an anonymous referee for emphasising this point.20. The regressands are, in fact, the log of state governments’ per capita own tax revenue and per capita own non-tax revenue. For expositional ease, we henceforth suppress the qualiﬁer ‘log of.’21. States’ own tax revenue consists of commodity tax revenue and revenue from taxes levied on agricultural income and property. There is an electoral cycle for commodity tax revenue. Why does this cycle disappear for the broader category of own tax revenue (compare columns 1 and 2 of Table 4)? For many states, the direct tax component of own tax revenue has been left untouched for years. Thus, the Report of the Eleventh Finance Commission (p. 28) observes: ‘There are a few other taxes which the states can levy, but remain unexploited or under-exploited. Taxation of agricultural incomes is one of them and profession tax is another. The land revenue which has traditionally been the principal mode of taxing agriculture in the country has fallen into disuse’. At any rate, since one component of own tax revenue does not respond at all to electoral considerations, precise estimation of the electoral cycle for own tax revenue becomes diﬃcult. Economists at the National Institute of Public Finance and Policy (a government-funded institute dealing with public ﬁnance issues) provided two reasons for why state governments show little interest in manipulating the direct tax component of
Political Factors and Fiscal Policies in Indian States 659 own tax revenue: (1) the cost of administering such taxes is exceptionally high; and (2) states perceive their share of centrally-levied direct taxes to be the principal source of direct tax revenue.22. Non-developmental current account expenditure consists mainly of interest payments on accumulated debt and outlays on ﬁscal and administrative services. It is surprising that an electoral cycle crops up for non-developmental current account expenditure since its components appear a priori to be diﬃcult to manipulate. To explore this issues further, we ran a regression where the regressand was the sum of per capita expenditure on ﬁscal services and per capita expenditure on administrative services. The coeﬃcient of the election year dummy was negative (70.09) and statistically signiﬁcant (t-statistic equal to 72.24).23. Alesina (1987) demonstrates that the partisan assumption does not rule out election-year eﬀects on macroeconomic variables. Hibbs (1992) provides an in-depth review of the political economy literature with partisan political parties.24. The diﬃculty in assigning ideological locations to political parties is a common feature of developing country experiences. We are unaware of a single paper that uses data from developing countries and considers ﬁscal policy as partially determined by government ideology (see, for example, Schuknecht (2000), Block (2002), Shi and Svensson (2002a, b), and Khemani (2004)).25. Now, Elecst equals 1 if an election (scheduled or mid-term) takes place in state s during the second half of ﬁnancial year t or during the ﬁrst half of the next ﬁnancial year.26. Khemani (2004) brieﬂy experiments with this instrument as well.27. Following Davidson and MacKinnon (1993: 241–42), we also conducted a test to determine whether the (scheduled) election year dummy, Elecst, is endogenous in India. The exogeneity of Elecst could not be rejected.28. The electoral history of India divides into two distinct phases: a pre-1967 period during which state legislative assembly elections were dominated by the Congress party, and a post-1967 era that beheld lively inter-party competition for assembly seats. Electoral cycles in ﬁscal variables – shown, in this study, to be sometimes present in the post-1967 data – should disappear completely during the years of Congress party hegemony. More generally, a contrast of the two electoral phases can shed light on the beneﬁts and costs of political competition.ReferencesAlesina, A. (1987) Macroeconomic policy in a two-party system as a repeated game, Quarterly Journal of Economics, 102, pp. 651–78.Alesina, A., Cohen, G. and Roubini, N. (1993) Electoral business cycles in industrial democracies, European Journal of Political Economy, 9, pp. 1–23.Alesina, A. and Drazen, A. (1993) Why are stabilizations delayed? American Economic Review, 82, pp. 1170–88.Alesina, A., Hausman, R., Hommes, R. and Stein, E. (1999) Budget institutions and ﬁscal performance in Latin America, Journal of Development Economics, 59, pp. 253–73.Alesina, A., Perotti, R. and Tavares, J. (1998) The political economy of ﬁscal adjustments, Brookings Papers on Economic Activity, 1, pp. 197–266.Alesina, A. and Roubini, N. (1992) Political cycles in OECD economies, Review of Economic Studies, 59, pp. 663–88.Alesina, A. and Sachs, J. (1988) Political parties and the business cycle in the United States, 1948–1984, Journal of Money, Credit and Banking, 20, pp. 63–82.Alesina, A. and Tabellini, G. (1990) A positive theory of ﬁscal deﬁcits and government debt, Review of Economic Studies, 57, pp. 403–14.Alt, J. and Lowry, R. (1994) Divided government and budget deﬁcits: evidence from the states, American Political Science Review, 88, pp. 811–28.Andrikopoulos, A., Prodromidis, K. and Serletis, A. (1998) Electoral and partisan cycle regularities: a cointegration test, Journal of Policy Modeling, 20, pp. 119–40.Baron, D. (1991) Majoritarian incentives, pork barrel programs, and procedural control, American Journal of Political Science, 35, pp. 57–90.Baron, D. (1998) Comparative dynamics of parliamentary governments, American Political Science Review, 92, pp. 593–609.
660 K. Chaudhuri & S. DasguptaBesley, T. and Burgess, R. (2000) Land reform, poverty reduction, and growth: evidence from India, Quarterly Journal of Economics, 1l5, pp. 389–430.Besley, T. and Burgess, R. (2002) The political economy of government responsiveness: theory and evidence from India, Quarterly Journal of Economics, 117, pp. 1415–51.Besley, T. and Case, A. (1995a) Does electoral accountability aﬀect economic policy choices? Evidence from gubernatorial term limits, Quarterly Journal of Economics, 108, pp. 769–98.Besley, T. and Case, A. (1995b) Incumbent behavior: vote-seeking, tax-setting, and yardstick competition, American Economic Review, 85, pp. 25–45.Block, S. (2002) Political business cycles, democratization, and economic reform: the case of Africa, Journal of Development Economics, 67, pp. 205–28.Boix, C. (1997) Privatizing the public business sector in the eighties: economic performance, partisan responses and divided governments, British Journal of Political Science, 27, pp. 473–96.Butler, D., Lahiri, A. and Roy, P. (1996) India Decides, Elections 1952–1995 (New Delhi: Living Media India Limited).Cargill, T. and Hutchison, M. (1991) Political business cycles with endogenous election timing: evidence from Japan, Review of Economics and Statistics, 73, pp. 733–39.Chibber, P. (1995) Political parties, electoral competition, government expenditures and economic reform in India, Journal of Development Studies, 32, pp. 74–96.Chowdhury, A. (1993) Political surﬁng over economic waves: parliamentary election timing in India, American Journal of Political Science, 37, pp. 1100–18.Davidson, R. and MacKinnon, J. (1993) Estimation and Inference in Econometrics (New York: Oxford University Press).De Haan, J. and Sturm, J.-E. (1994) Political and institutional determinants of ﬁscal policies in the European Community, Public Choice, 80, pp. 157–72.De Haan, J. and Sturm, J.-E. (1997) Political and economic determinants of OECD budget deﬁcits and government expenditures: a reinvestigation, European Journal of Political Economy, 13, pp. 739–50.De Haan, J., Sturm, J.-E. and Beckhuis, G. (1999) The weak government thesis: some new evidence, Public Choice, 101, pp. 163–76.Dutta, B. (1996a) Coalitional governments and policy distortions: the Indian experience (mimeo), Indian Statistical Institute, New Delhi.Dutta, B. (1996b) Fragmented legislatures and electoral systems: the Indian experience (mimeo), Indian Statistical Institute, New Delhi.Frey, B. and Schneider, F. (1978a) An empirical study of politico–economic interactions in the United States, Review of Economics and Statistics, 60, pp. 174–83.Frey, B. and Schneider, F. (1978b) A politico-economic model of the United Kingdom, The Economic Journal, 88, pp. 243–53.Government of India (2000) Report of the Eleventh Finance Commission (For 2000–2005) (New Delhi: Government of India Press).Government of India, Ministry of Planning, Department of Statistics, various issues, National Accounts Statistics (New Delhi: Central Statistical Organisation).Government of India, Ministry of Information and Broadcasting, various issues, Press in India, Annual Report of the Registrar of Newspapers for India (New Delhi: Government of India Press).Grover, V. and Arora, R. (1998) Encyclopaedia of India and Her States (New Delhi: Deep and Deep Publications).Harrinvirta, M. and Mattila, M. (2001) The hard business of balancing budgets: a study of public ﬁnances in seventeen OECD countries, British Journal of Political Science, 31, pp. 497–521.Heckelman, J. and Berument, H. (1998) Political business cycles and endogenous elections, Southern Economic Journal, 64, pp. 987–1000.Hibbs, D. (1977) Political parties and macroeconomic policy, American Political Science Review, 71, pp. 1467–87.Hibbs, D. (1992) Partisan theory after ﬁfteen years, European Journal of Political Economy, 8, pp. 361–73.Karnik, A. (1990) Elections and government expenditures: the Indian evidence, Journal of Quantitative Economics, 6, pp. 203–12.Khemani, S. (2004) Political cycles in a developing economy: eﬀect of elections in the Indian states, Journal of Development Economics, 73, pp. 125–54.
Political Factors and Fiscal Policies in Indian States 661Lowry, R., Alt, J. and Ferree, K. (1998) Fiscal policy outcomes and electoral accountability in American states, American Political Science Review, 92, pp. 759–74.Nordhaus, W. (1975) The political business cycle, Review of Economic Studies, 42, pp. 169–90.Peltzman, S. (1992) Voters as ﬁscal conservatives, Quarterly Journal of Economics, 107, pp. 327–61.Perotti, R. and Kontopoulos, Y. (2002) Fragmented ﬁscal policy, Journal of Public Economics, 86, pp. 191–222.Poterba, J. (1994) State responses to ﬁscal crises: the eﬀects of budgetary institutions and politics, Journal of Political Economy, 102, pp. 799–822.Poterba, J. (1996) Budget institutions and ﬁscal policy in the US states, American Economic Review, 86, pp. 395–400.Reid, B. (1998) Endogenous elections, electoral budget cycles and Canadian provincial governments, Public Choice, 97, pp. 35–48.Reserve Bank of India, various issues, Reserve Bank of India Bulletin (Mumbai: Reserve Bank of India).Rogoﬀ, K. (1990) Equilibrium political budget cycles, American Economic Review, 80, pp. 21–35.Rogoﬀ, K. and Sibert, A. (1988) Elections and macroeconomic policy cycles, Review of Economic Studies, 55, pp. 1–16.Roubini, N. and Sachs, J. (1989a) Political and economic determinants of budget deﬁcits in the industrial democracies, European Economic Review, 33, pp. 903–38.Roubini, N. and Sachs, J. (1989b) Government spending and budget deﬁcits in the industrial democracies, Economic Policy, 8, pp. 100–32.Schuknecht, L. (2000) Fiscal policy cycles and public expenditure in developing countries, Public Choice, 102, pp. 115–30.Schultz, K. (1995) The politics of the political business cycle, British Journal of Political Science, 25, pp. 79–99.Sen, K. and Vaidya, R. (1996) Political budget cycles in India, Economic and Political Weekly, 30, pp. 2023–27.Serletis, A. and Afxentiou, P. (1998) Electoral and partisan cycle regularities in Canada, Canadian Journal of Economics, 31, pp. 28–46.Shi, M. and Svensson, J. (2002a) Political budget cycles in developed and developing countries (mimeo), Development Research Group, World Bank.Shi, M. and Svensson, J. (2002b) Conditional political budget cycles (mimeo), Development Research Group, World Bank.Volkerink, B. and De Haan, J. (2001) Fragmented government eﬀects on ﬁscal policy: new evidence, Public Choice, 109, pp. 221–42.Weiner, M. (1989) The Indian Paradox: Essays in Indian Politics (New Delhi: Sage Publications).Weingast, B., Shepsle, K. and Johnson, C. (1981) The political economy of beneﬁts and costs: a neoclassical approach to distributive politics, Journal of Political Economy, 89, pp. 642–64.