Tax Buoyancy Estimates for Indian States

Author(s): Indira Rajaraman, Rajan Goyal and Jeevan Kumar Khundrakpam

Source: Economic and Political Weekly , Apr. 22-28, 2006, Vol. 41, No. 16 (Apr. 22-28,
2006), pp. 1570-1573
Published by: Economic and Political Weekly

Stable URL: https://www.jstor.org/stable/4418117

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Economic and Political Weekly is collaborating with JSTOR to digitize, preserve and extend
access to Economic and Political Weekly

This content downloaded from

202.70.131.9 on Wed, 28 Jun 2023 15:02:32 +00:00
All use subject to https://about.jstor.org/terms
 Special artile ___

Tax Buoyancy Estimates for Indian States

With the introduction of a destination-based VAT in all but eight states starting April 2005,
there is need for a good baseline indicator of tax buoyancies in states in the period
immediately preceding. Thi s to provide such a base, with buoyancies
estimated over a 23-year span starting in 1980-81. If estimated over a sufficiently l
period of time, the buoyancy coefficient essentially estimates the underlying reve
generating properties of the system with endogenised tax policy. A log linear trend f
the entire period showed serial correlation, which is eliminated for all but one state, As
with the introduction of structural breaks. A third specification, including the log of
per cent share of industry in the domestic product, eliminates serial correlation for
Assam, and imnproves the goodness-of-fit for some other states. In all but six state
the sign of the change in the buoyancy coefficient at the break is positive. Where
buoyancy-enhancing break occurs in the late 1990s, the spurt in tax effort mnight h
an endogenous response to the expenditure shock from implementation of the high
salary scales recommended by the Fifth Pay Commission.

INDIRA RAJARAMAN, RAJAN GOYAL, JEEVAN KUMAR KHUNDRAKPAM

T ax buoyancy estimates, which measure the percentage Buoyancy estimates for tax revenues of states are estimated with
response of tax revenue to a one per cent change in therespect to the gross state domestic product (GSDP). The GSDP
tax base, usually proxied by the gross domestic product,estimates for states in India are available only at factor cost, not
are a routine requirement for fiscal projection purposes. The at market prices.4
elasticity of tax revenue is more stringently defined as the Section I presents the specifications estimated. Section II presents
underlying revenue response, holding constant all parameters of the buoyancy coefficients themselves from the results of the best
tax policy. In developing countries, where tax policy parametersspecification for each state, and compares these with the buoy-
are changed every year and sometimes in the course of the year, ancies projected in the report of the Twelfth Finance Commission
the elasticity of tax revenue is virtually impossible to estimate
for the period 2005-10. Section III concludes the paper.
with any appreciable degree of accuracy. In such a fiscal context,
where tax policy parameters are in a state of constant flux, the
buoyancy coefficient may provide the only feasible alternative
The Specifications Estimated
to estimating the underlying revenue-generating properties of the
system. If estimated over a sufficiently long period of time, theThe basic estimation procedure for tax buoyancies is through
buoyancy coefficient essentially estimates the revenue response a double log specification of the type given in equation (1) below,
with endogenised tax policy. The problem with estimation over which yields the buoyancy coefficient [3p
In (OTRt) = c1 + 1 (lnGSDPt) + ut ...(1)
a long period of course is the possible presence of structural breaks
due to regime changes in tax effort, which will lead to serial
where In (OTRt) = log of (nominal) revenue in year t
correlation in the residuals and thus a biased estimate of the In (GSDPt) = log of (nominal) GSDP in year t
buoyancy coefficient when a log linear trend is fitted over ccthe= intercept
entire sample period. = buoyancy estimate
This paper estimates buoyancies for Indian states with respect The residuals from estimation of equa
to their own tax revenues for the period since 1980-81, correlation
not for most states. This could be on account of an
including tax revenues received from the centre, and not incorrect functional form (fitting a linear specification to an
including their own non-tax revenues.2 Non-tax revenues ofunderlying non-linear relationship, for example, or where there
states display high volatility, with spikes resulting from an as-is a structural break), or because of omitted variables. The
sortment of accounting practices, which vary from state to state. Cochrane-Orcutt two-step estimator is a commonly used me-
Chief among these, but not the only one, is the practice with chanical way of correcting for autocorrelation, when the source
respect to the recording of non-tax revenues from lottery of the problem is unknown. In the context of buoyancy estimation
schemes.3 Tax revenues display greater stability year-to-year,
over a long period, clearly one source of the problem could be
and are the dominant source of revenue, accounting for 80 pera kink in the underlying relationship, which would generate serial
cent of own revenue collections in aggregate across states. correlation in the residuals when fitting a log linear trend over

1570 Economic and Political Weekly April 22, 2006

This content downloaded from

202.70.131.9 on Wed, 28 Jun 2023 15:02:32 +00:00
All use subject to https://about.jstor.org/terms
the entire sample period. Allowing for a structural break serves developing countries is the most tractable sector for taxation
three purposes: it solves one possible source of serial correlation, purposes, buoyancy estimation with respect to total domestic
it avoids the degrees of freedom problem that would arise if the product alone could carry an omitted variable problem that gets
buoyancy is estimated for the several periods separately, and most reflected in the residuals. The alternative that is sometimes
importantly it brings out regime changes in tax effort. adopted is to estimate buoyancies with respect to industrial sector
A second specification was therefore fitted, allowing for struc- value added alone.
tural breaks in the tax series, marking points where there have
been major alterations in the tax policy parameters, such that there II
is not merely a one-time change in levels, but also a change in Estimated Buoyancy Coefficients
the revenue-generating properties of the system. This is shown
below in equation (2). Table 1 lists the states, with the P-lcJ! of the LM test on the
In (OTRt) = oc, + (oc2-o)D residuals with specifications (1) and (2), and the break year in
+ [31 (lnGSDPt) + {(32-3i)D*(lnGSDPt)) + ut ...(2) the second specification. The LM test is performed for two lags.
Equation (2) has a dummy variable D which takes the value The null hypothesis in the LM test is that there is no serial
one for years after the single structural break in the estimation correlation in the series tested for. A low P-value indicates
period, zero otherwise. There is provision for both an intercept that the null hypothesis can be rejected with a low probability
change in levels, as well as a change in the slope. The coefficients of error, and so indicates the presence of serial correlation. A
attached to the dummy variable terms give the difference between high P-value indicates that the null hypothesis cannot be rejected,
the coefficient for the period when the dummy variable carries Assam shows serial correlation in the residuals even with equa-
the value zero, and the period for which it carries the value one. tion (2), with the LM test carrying a P-value of 0.12. When
After the structural break, the buoyancy coefficient is [2. estimated with specification (3), the P-value improved to 0.65.
The introduction of a structural break took care of serial This is shown in Table 2. along with some other states for
correlation in the residuals for most states, as measuredwhich the P-value showed a reduction in serial correlation with
by the
LM test. The year of the structural break was chosen from specification
all (3) relative to specification (2), and the goodness
the possible break years within the estimation period based of fitonimproved.
the LM test, and the significance of the coefficients of the
intercept and slope dummies. Where there was a break in the Table 1: Results of LM Test on Residuals for Alternative
1980s using the full sample period, as in the case of Karnataka, Specifications
Kerala, Meghalaya, Rajasthan, Tamil Nadu and Tripura, the State Period P-value for LM Test on Break Year
equations were re-estimated looking for another break in the Residuals (2 Lags)
second period using the same model. Among these six states, Equation (1) Equation (2)

a second break was found in the case of Karnataka, Rajasthan Andhra 1981-03 0.00 0.23 1994-95
and Tamil Nadu. Arunachal 1987-03 0.01 0.30 1994-95
Assam 1981-03 0.00 0.12 1995-96
Where serial correlation remained even after testing, for all
Bihar 1981-03 0.99 No break
possible structural breaks, a third equation was estimated Goa
with 1987-03 0.17 No break
sectoral shares in GSDP.5 It is a well-known feature of the tax
Gujarat 1981-03 0.03 0.96 1994-95
Haryana 1981-03 0.01 0.59 1995-96
system in all developing countries that industry is more amenable
Himachal 1981-03 0.05 0.37 1991-92
to taxation than agriculture and services. The share of industry
Jammu and Kashmir 1981-03 0.07 0.97 1994-95
in domestic product would by prior expectation, increase steadily
Karnataka 1988-03 0.15 0.82 1997-98
over time in such a setting, and therefore be serially correlated.
Kerala 1981-03 0.02 0.24 1986-87
Madhya Pradesh 1981-03 0.24 0.27 1993-94
The third specification tried, in cases where residual correlation
Maharashtra 1981-03 0.02 0.21 1997-98
persisted even in specification (2), is given below:
Manipur 1981-03 0.76 0.95 1996-97
ln(OTRt)=oc 1+(oc2-oc l)D+Pt(lnGSDPt) Meghalaya 1981-03 0.01 0.78 1986-87
+{ (P2-i)D*(lnGSDPt) ) +(lnprcntindsharet)+ut ...(3) Mizoram 1988-03 0.66 No break

An alternative method by which to correct for systematic Nagaland 1981-02 0.04 0.72 1992-93
Orissa 1981-03 0.17 0.56 1997-98
variations over time between the base used, GSDP in this case,
Punjab 1981-03 0.09 0.42 1997-98
and the true base, would be an error correction model, with the Rajasthan 1987-03 0.07 0.29 1996-97
one-period lagged value of the dependent variable included as Sikkim 1981-03 0.02 0.35 1997-98
Tamil Nadu 1988-03 0.00 0.18 1997-98
a regressor on the right hand side. That is difficult to do in the
Tripura 1981-03 0.03 0.53 1988-89
present case where, as will be seen, there are structural breaks Uttar Pradesh 1981-03 0.18 0.82 1994-95
in nearly all the states, which further occur typically in the 1990s, West Bengal -i981-03 0.00 0.46 1996-97
with short post-break estimation periods.
Notes: The estimation period does not start at 1980-81 for six states:
The data on own tax revenues of states are sourced from Arunachal, Goa and Mizoram on account of data unavailability for
RBI publications, for the period 1980-81 to 2002-03, and on earlier years; Karnataka, Rajasthan and Tamil Nadu, because the
GSDP, including sectoral shares, from the Central Statistical truncation at what was clearly the earlier of two structural breaks
improved the goodness of fit. The estimation period ends at 2001-02
Organisation. for Nagaland for data reasons. The R-bar Squared values are not
Although the augmented specification actually improved the reported, but were uniformly high even for equation (1).
quality of the estimation in only a small number of states, it is Source: Own tax revenue from RBI, State Finances, assorted issues; GSDP
figures from Central Statistical Organisation. The data for Bihar, Uttar
a useful supplement to the simpler specifications normally used
Pradesh and Madhya Pradesh are inclusive of the figures for
for buoyancy estimation. The sectoral share of industry in a Jharkhand, Uttaranchal and Chhattisgarh for the post-partition years
developing country increases over time, and since industry in (2000-03).

Economic and Political Weekly April 22, 20061571

This content downloaded from

202.70.131.9 on Wed, 28 Jun 2023 15:02:32 +00:00
All use subject to https://about.jstor.org/terms
 The final set of state own tax buoyancy coefficients, as es- special category, saw a decline in the buoyancy, at break years
timated here for the post-break period up to 2002-03, is shown ranging between the late 1980s and late 1990s.
in Table 3. The break year in many cases is in the late 1990s, The post-break buoyancies for the majority of states which
1996-97 or 1997-98, with a positive and statistically significant experienced an increase in buoyancy in recent years, falls in a
change in the buoyancy coefficient at the break. This spurt in fairly high range, between 1.01 for Nagaland, and as high as 2.00
tax effort is a plausibly endogenous response to the enhanced for Manipur. It is only the states which have seen a post-break
expenditure on salaries starting in the year 1996-97, with imple- decline where the coefficient has dipped below one. In states with
mentation of the salary scales recommended by the Fifth Pay no discernible break, like Goa, Bihar and Mizoram, the coefficient
Commission. is above one.
The advantage of identifying structural breaks rather than The a states are grouped by the assigned values for own tax
mechanical solution like the Cochrane-Orcutt two-step estimator,buoyancies in the report of the Twelfth Finance Commission.
to correct for serial correlation, is that it yields a handle onThe
themedian buoyancy assigned by the TFC is 1.20, whereas the
timing of changes in tax policy effort. median buoyancy as estimated here is 1.30. There is some
question as to whether the post-break surge in estimated buoy-
The direction of change at the structural break is not, however,
positive in all cases. West Bengal, Gujarat and Kerala among ancies, which in many states has held for only a five-or six-year
the major states, and Himachal, Meghalaya and Tripura, inperiod the going up to 2002-03, can be sustained going into the future.
This might be the reason for the more conservative buoyancies
Table 2: Results for Specification Inclusive
projected in the TFC report. There was also a fear that the switch
of Industry Share in Domestic Product to a VAT might be revenue-reducing, but preliminary indications
are that the VAT has been revenue-enhancing.6
P-value for LM test on Goodness of Fit -
Residuals AIC; SBC However, the TFC projected buoyancies are not uniformly
Equation (2) Equation (3) Equation (2) Equation (3) lower than the estimated buoyancies for all states. The highest
TFC buoyancies of 1.35 and 1.30 have been assigned to seven
Assam 0.12 0.65 -2.03;-1.83 -2.26;-2.02
Madhya Pradesh 0.27 0.36 -2.31;-2.12 -2.69;-2.44 states, whose buoyancies as estimated here are well below the
Rajasthan 0.29 0.64 -2.13;-1.93 -2.26;-2.01 projected values in all but one case. These seven states include
Tripura 0.53 0.73 -2.49;-2.29 -2.62;-2.37 two with estimated buoyancies below one, Gujarat and West
Notes: Equation (3) is estimated over the same periods and for the same Bengal. At the other extreme, the TFC buoyancies of 1.10 have
break years as equation (2). been assigned to five special category states, some of which, like
Source:lbid. The set of states is confined to those for which equation (3)
Arunachal and Manipur, have experienced among the highest
reduced serial correlation in the residuals, and improved goodness of
fit, relative to equation (2). buoyancies, albeit starting from a low level.

Ill
Table 3: Post-Break Own Tax Buoyancies and
TFC Projected Buoyancies Conclusion
State Post-Break Change of Estimated TFC
With the introduction of a destination-based VAT in all but
Period Sign at Break Buoyancy Projected
Coefficient Coefficient eight states starting April 2005, there is need for a good baseline
indicator of tax buoyancies in states in the period immediately
Goa 1987-03 No break 1.05 1.35
Punjab 1998-03 plus 1.61 1.35
preceding. This paper attempts to provide such a base.
West Bengal 1997-03 minus 0.76 1.35 When buoyancies are estimated over a 23-year span starting
Gujarat 1995-03 minus 0.95 1.30 in 1980-81, there is serial correlation in the residuals. The struc-
Himachal 1992-03 minus 1.07 1.30
tural breaks fall in the 1990s for the most part, and eliminate
Karnataka 1998-03 plus 1.23 1.30
Kerala 1987-03 minus 1.02 1.30 serial correlation for all but one state. A specification including
Haryana 1996-03 plus 1.35 1.25 the log of the sectoral share of industry in GSDP eliminates serial
Maharashtra 1998-03 plus 1.44 1.25
correlation in that one exception, and improves the goodness of
Andhra 1995-03 plus 1.51 1.20
fit for a few other states.
Assam 1996-03 plus 1.54 1.20
Bihar 1981-03 No break 1.12 1.20 The sign of the change in the buoyancy coefficient at the break
Jammu and Kashmir 1995-03 plus 1.55 1.20 is positive in all but six states. The set of six where there was
Madhya Pradesh 1994-03 plus 1.09 1.20 a negative change at the break includes the three states with (post-
Meghalaya 1987-03 minus 0.96 1.20
Orissa 1998-03 plus 1.36 1.20 break) buoyancy coefficients below one: Gujarat, Meghalaya and
Rajasthan 1997-03 plus 1.44 1.20 West Bengal. In all the rest, the post-break coefficients are
Sikkim 1998-03 plus 1.81 1.20 comfortably above one. In states which experienced a buoyancy-
Tamil Nadu 1998-03 plus 1.29 1.20
enhancing structural break in the late 1990s, the spurt in tax effort
Uttar Pradesh 1995-03 plus 1.36 1.20
Arunachal 1995-03 might have been an endogenous response to the implementation
plus 1.73 1.10
Manipur 1997-03 plus 2.00 1.10 of the higher salary scales recommended by the Fifth Pay
Mizoram 1988-03 No break 1.03 1.10
Commission, starting in the year 1996-97. There is some question
Nagaland 1993-02 plus 1.01 1.10
as to whether these enhanced buoyancies, which have prevailed
Tripura 1989-03 minus 1.04 1.10
typically for a post-break period of only five or six years going
Notes: In states where there is no break, thecan
up to 2002-03, start of into
be sustained the theestimation
future. period
dictated by data availability (see Table 1 and notes). The series in all
The median buoyancy assigned to states for the period 2005-
states terminates at 2002-03, except for Nagaland. For the four states
listed in Table 2, the coefficients reported here are for specification (3). 10 is 1.20 in the report of the Twelfth Finance Commission,
Source: Ibid. whereas the median buoyancy as estimated here is 1.30. The

1572 Economic and Political Weekly April 22, 2006

This content downloaded from

202.70.131.9 on Wed, 28 Jun 2023 15:02:32 +00:00
All use subject to https://about.jstor.org/terms
projections of the TFC for the period 2005-10 are clearly con- but the reports carry only projected buoyancies, which are often unrelated
to the historical values, and carry varying normative elements.
servative relative to the realised buoyancies in recent
3 Non-tax revenues from state lotteries are often reported gross, with payment
years. However, the cross-sectional pattern of TFC projected of prize money reported separately in revenue expenditure and not netted
buoyancies does not accord with the cross-sectional pattern ofreceipts. Other spikes result from the episodic routing through the
out of
buoyancies estimated in this paper. The highest TFC buoyancies budget of notional receipts on account of bunched interest and dues from
of 1.35 and 1.30 have been assigned to seven states, where in parastatals, against offsetting subsidies and otherexpenditures to parastatals.
all but one case, the buoyancies estimated here are well below Finally, loan waivers on state debt owed to the central government enter
non-tax receipts as an accounting entry.
the projected values. This set includes two of the three states
4 This will impart a slight upward bias to state-level buoyancy estimates,
which have estimated buoyancies below one, Gujarat and West if the share of indirect taxes in total tax collections (nationally) increases
Bengal. The TFC projected buoyancies underlie the deficit steadily over time.
grants awarded to states, and therefore carry a normative com- 5 In its estimates of aggregate buoyancy across states, the Shome Committee
ponent. What these comparisons show is that the sign and report (Government of India, 2001) attributes the fall from 1.12 in the
1980s to 1.04 over the period 1990-1999 to the sectoral shift towards
quantum of the normative component is not uniform, but variesservices.
across states. [31 6 This could be a one-time enhancement, and may not translate into a
buoyancy enhancement.
Email: indira_raja@yahoo.com
References
Notes Bhat, K Sham, G Kannabiran (1992): 'Measuring Elasticity and Buoyancy
of Tax Revenue in Tamil Nadu: A Divisia Index Approach', Prajnan,
[The paper does not represent the views of the organisations to which the 21:2, p 195.
authors belong. The authors thank Lant Pritchett and an anonymous refereeGovernment of India (2004): Report of the Twelfth Finance Commission
for useful advice, with the usual disclaimer.] (2005-10).
-(2001): Report of theAdvisory Group on Tax Policy and Tax Administration for
1 The classical procedure for cleaning out the incremental impact of changes the Tenth Plan (Shome Committee), Planning Commission, New Delhi.
in tax policy parameters through the proportionate adjustment method isKhadye, I K (1981): 'The Responsiveness of Tax Revenues to National
provided in Prest (1962) and Mansfield (1972). Sen (2003) offers a Income in India (1960-61 to 1978-79)', RBI Occasional Paper, 2:1.
possible method of correcting for projection errors in budget estimates of Mansfield, Charles Y (1972): 'Elasticity and Buoyancy of a Tax System:
total tax receipts, but the procedure remains dependent on official estimates A Method Applied to Paraguay', IMF Staff Papers, XIX:2.
of the impact of rate and base changes, which are mechanically drawn. Prest, A R (1962): 'The Sensitivity of the Yield of Personal Income Tax
Tanzi (1969 and 1976), provided an ingenious method by which cross- in the United Kingdom', Economic Journal, LXXII, pp 576-96.
sectional data from sub-national regions could be used to estimate the Purohit, M C (1979): 'Buoyancy and Income-Elasticity of State Taxes in
elasticity of a nationally-levied tax. Clearly, this method can be extended India', Artha Vijnana, 20:3, p 244.
to state-level taxes, provided data are available by administrative subdivisionRao, V G (1979): The Responsiveness of the Tax System in India, Allied
within each state, which is not presently the case. Publishers, Mumbai.
2 The Shome Committee (Government of India, 2001) provides buoyancy Reserve Bank of India State Finances: A Study of Budgets, assorted issues.
estimates in aggregate across all states for the same period. Estimates forSen, Pronab (2003): 'A Note on Estimating Tax Elasticities', Planning
earlier periods, either in aggregate or for individual states, are ij Purohit Commission, mimeo.
(1979), Rao (1979), Khadye (1981) and Bhat and Kannabiran (1992). Tanzi, Vito (1969): 'Measuring The Sensitivity of the Federal Income Tax
There are very many other studies of elasticity and buoyancy of tax from Cross-Section Data: A New Approach', The Review of Economics
revenues in India, but most relate to income taxes, which are levied only and Statistics, pp 206-09.
by the national government at the centre. State-specific estimates are among - (1976): 'The Sensitivity of the Yield of the US Individual Income Tax
the background estimation exercises performed for all finance commissions, and the Tax Reforms of the Past Decade', IMF Staff Papers; pp 441-54.

SPECIAL ISSUE

Measurement of Poverty
October 22, 2005

Poverty Lines versus the Poor: Method versus Meaning - Ashwani Saith

Of Calories and Things: Reflections on Nutritional Norms, Poverty Lines

and Consumption Behaviour in India - Pronab Sen

Are the 2000 Poverty Estimates for India a Myth, Artefact or Real? - Gurleen Popli, Ashok Parikh,
Richard Palmer-Jones

Fractions versus Whole Numbers: On Headcount Comparisons of Poverty

across Variable Populations - S Subramanian

For copies write to:

Circulation Manager
Economic and Political Weekly,
Hitkari House, 6th Floor, 284, Shahid Bhagatsingh Road, Mumbai 400 001.
email: circulation @epw.org.in

Economic and Political Weekly April 22, 2006 1573

This content downloaded from

202.70.131.9 on Wed, 28 Jun 2023 15:02:32 +00:00
All use subject to https://about.jstor.org/terms