An Examination of Revenue-Expenditure Linkage

A case study of Manipur

A research paper published in the book

PERSPECTIVE ON THE ECONOMY OF MANIPUR


Edited by:

Prof. E.Bijoykumar Singh
Dr. Damudor Nepram


Published By
AKANSHA PUBLISHING HOUSE
New Delhi


















































An Examination of Revenue-Expenditure Linkage
A case study of Manipur

Dr . Md. Samsur J aman
Lecturer in Economics
Jiri College, Jiribam
e-mail: [email protected],
[email protected]

1. Introduction
In undertaking fiscal adjustment policies the developing countries are facing
dual challenges. One arises from the increasing demand for public expenditures for
infrastructure and social sector investment, and the other arises from the lack of
capacity to raise revenue from domestic sources to finance the increased expenditure,
primarily due to narrow tax base. To boost competitive capacity of the country in a
rapidly globalizing world, the governments of developing countries have to invest a
large portion of their revenue in building physical infrastructures. In addition, the low
income developing countries also need to spend a major portion of their development
expenditures in providing social services to the poor such as health, education etc. On
the other hand, as Khattry (2003: 402) pointed out, the structural characteristics of
low income countries, combined with prevalence of unsophisticated tax
administration limit their ability to raise taxes from domestic sources, namely income
and domestic indirect taxes. Also, the existence of a large informal sector and the
underground economy constrains the governments capacity for revenue growth.

The purpose of this paper is to evaluate the relationship between revenue and
expenditure and its implication for managing the budget deficit. In order to do this a
three variable model is formulated comprising government expenditure, revenue and
NSDP. Firstly, existence of a long-run relationship among these three variables is
tested by using Johansen (1991, 1995) cointegration approach. Granger causality test
is applied on the corresponding Vector Error Correction Model to examine long-run
causal relationship between the variables.

The goal of fiscal policy is to enhance economic growth and employment and
to bring stability in economic outcome variables such as the real GDP growth rate.
Under the above circumstances, the nature and objectives of fiscal policy may differ
with the level of development of the countries. Long run outcome of expansionary
fiscal policy depends on the nature of distribution of public resources as the same
amount of public money can generate different growth pay-offs in different sectors,
and the overall growth of the economy depends on the combined growth of these
sectors.

Following the introduction in section 1, this paper is organized as follows.
Section 2 provides a brief discussion on fiscal reforms in Manipur and section 3
discusses the dynamics of revenue and expenditure relationships. Section 4 and 5
respectively provide a review of empirical findings of existing literature and data and
methodology of the study. Section 6 provides the results and policy implications and
finally Section 7 concludes the paper.

2. Fiscal reforms in Manipur
Government of Manipur initiated a reform programs under the behest of the
guide line of central government during late 90s and till date. One of the central
objectives of this reform programs was to readjust government spending and to
improve revenue collection.
The fiscal consolidation programs in Manipur aim at
i) to strive to remain revenue surplus by making a balance in revenue
receipt and expenditure and build up further surplus;
ii) to strive to bring down fiscal deficit to 3% of GSDP;
iii) to limit the amount of outstanding government guarantees as per the
provisions of the Manipur ceiling on state government Act., 2004;
iv) to follow a recruitment and wage policy in a manner such that the total
salary bill relative to revenue expenditure excluding interest payment
and pensions does not exceed 35%.
3. Dynamics of Revenue-Expenditure Relationships
There are four possible hypotheses regarding the relationship between revenue
and expenditure. Firstly, revenue-expenditure hypothesis can arise in two distinct
forms
1
. The first relationship is advocated by Friedman and his followers. According
to Friedman (1978) raising taxes in an attempt to reduce deficit will also cause

1
Martin et al (December 2004), discussed both the forms of the revenue-expenditure hypotheses.
However, the latter one is less likely scenario due to budget constraints.
expenditure to rise. Therefore it will not be possible to reduce deficit by increasing
taxes. The alternative version of revenue-expenditure hypothesis is popularized by
Wagner (1976) and Buchanan and Wagner (1977). According to them, reducing taxes
will cause expenditure to rise because of fiscal illusion of the population.

The third kind of relationship that may appear between these two variables defined as
fiscal synchronization hypothesis which suggests that revenue and expenditure are
determined simultaneously. This argument is mainly developed by Musgrave (1966)
and Meltzer and Richard (1981). According to them, government expenditure and
revenue are determined in the process of equalizing marginal benefit and marginal
cost of government services by the population of the country.

Finally, there is the view that there can be no relationships between revenue and
expenditure i.e. these variables are determined independent of each other. Darrat
(1998), Widavsky (1988), Baghestani and McNown (1994) support this argument.
This is a possible scenario when the government determines expenditure on the basis
of requirements of the citizen and imposes taxes up to a tolerable limit to the citizen
(Martin et al, 2004).

4. Review of empirical findings in the existing literature
Direction of causal relationship between revenue and expenditure and its
implication for budget deficit has not been empirically resolved. Though over the last
three decades a number of studies have been carried out in different countries to
explore the issue, findings vary from country to country and also within the country.

The study of Francisco et al.(2001) by using the Preliminary Granger causality
tests, which are based in the notion that lagged realisations of one variable help to
predict current values of another variable, offers two main results. First, the deficit is
not independent of fiscal pressure or the size of the public sector. Second, we find
only very weak evidence, if any, that revenues G-cause expenditures. However, these
results may be biased in that they do not consider the possibility of the existence of
long-run relationships among the variables involved. Once we consider this
possibility and develop the causality analysis with cointegrated variables, we find
clear evidence of public expenditures G-causing public revenues in the long run and
also, although less clear, evidence of long-run G-causality from revenues to public
expenditure. In other words, we obtain evidence of long-run bi- directional causality.
Furthermore, in the short run the direction of causality seems to hold only from
expenditures to public revenues in the whole sample.

The study of Barua(2005) found by examining the causal relationship between
revenue and expenditure in Bangladesh that there is no causal relationship between
government expenditure and revenue in the short run.

5. Data and Methodology
5.1 Data:
Government revenue and expenditure data are collected from various issues of
Finance Accounts of Manipur and Manipur Budget at a glance, Government of
Manipur. NSDP at current market price data are collected from Statistical Abstract of
Manipur. NSDP deflator is used to convert the nominal data into real data. Annual
data from 1972 to 2004 are taken. Real series (1972=100) of revenue, expenditure and
GDP data are transformed into their logarithmic form to test causal relationships
among the variables. However there is a debate whether nominal and real form of
revenue and expenditure would be appropriate to test the causality between the
variables. Martin et al (2004), in their study used revenue and expenditure data in real
terms assuming that government takes budgetary decision by taking account of the
expected level of inflation and because inflation affects actual level of expenditure
and revenue. Here we used real values of revenue and expenditure in order to
eliminate the inflationary effect from revenue and expenditure.
Definitions of the variables are as follows:
LRTEM = Log of Real Government Expenditure
LRTRM = Log of Real Government Revenue
LRNSDP = Log of Real Net State Domestic Product

5.2 Methodology
Granger Causality
Revenue (RTRM) G-cause expenditure (RTEM) if in a regression of
expenditure on its own lagged values (RTEM
t-1
), the inclusion of lagged values of
revenue (RTRM
t-1
) significantly improves the prediction of expenditure. However,
existence of Granger causality from revenue to expenditure does not imply that
expenditure is the result of revenue. Rather it implies that changes in revenue precede
expenditure, and past and present values of revenue provide important information to
forecast future values of expenditure, that are not incorporated in the past values of
RTEM. Granger causality form with two variables, expenditure and revenue can be
expressed in following ways,
RTEM
t
=

n
j
t t
n
i
t iRTRM iRTEM
1
1
1
1 c | o 1
RTRM
t
=

n
j
t t
n
i
t iRTEM iRTRM
1
1
1
1 c o ; 2
An unidirectional causality from revenue to expenditure will be found if

i | 0 and

i o = 0, that is the set of estimated coefficients of lagged RTRM are

significantly different from zero in RTEM equation and cluster of estimated
coefficients of lagged RTEM are not significantly different from zero in RTRM
equation.

Conversely, an unidirectional causality from expenditure to revenue will be
found if

i o 0 and

i | = 0, that is the set of estimated coefficients of lagged

RTEM are significantly different from zero in RTRM equation and cluster of
estimated coefficients of lagged RTRM are not significantly different from zero in
RTEM equation.

There will be bidirectional causality or feedback between revenue and
expenditure if both the conditions i 0 and i 0 simultaneously hold, that is the
set of estimated coefficients of lagged RTRM and lagged RTEM are significantly
different from zero in both RTEM and RTRM equation. Revenue and expenditure
will be determined independently if i =0 and i = 0, that is, there is no causal link
between these two variables.

Modern time series econometrics allows us to test joint significance of the
coefficients of lagged terms. However, validity of the test results depends on the
stationary properties of data i.e., if both revenue and expenditure series are stationary
in level forms we can use Granger causality test described above. When both the
series are non stationary in level forms but stationary after first differencing, that is
both are integrated of order one or I(1), we can use the differenced series to test
causality between the two variables. But if both the series are stationary after first
differencing and cointegrated, we cannot use the VAR in first differences to test the
causality between the two series. If the variables are integrated of same order we can
use cointegration test to check whether the variables are cointegrated or not. If the
series of revenue and expenditure are cointegrated, error correction representation of
cointegrated series can be estimated to examine causality between the variables.

Johansen (1991, 1995) cointegration test is used to examine cointegration
between revenue and expenditure. Johansen developed a multivariate technique to
evaluate the cointegration relationship among a group of variables.

Johansen developed two test statistics for cointegration test, namely trace
statistic

n
r l
i T
1
) 1 ln( max and maximum eigen value statistics
) 1 1 ln( max r T , where i is defined as the estimated values of
characteristics roots obtained from the estimated coefficient matrix and T is the
number of included observations. If the revenue and expenditure series are
cointegrated we can use error correction representation to test causality between the
two series.

After including real NSDP as an additional variable, VAR in first differences
with one lag of error correction terms can be represented as follows:

t t l t i j t i i t i t R LRNSDP LRTRM LRTEM LRTEM
n
l
n
j
n
i
c u o | o u

1 1 1
1 1 1
3
t t l t i i t i j t i t R LRNSDP LRTEM LRTRM LRTRM
n
l
n
i
n
j
c u o o ; u

1 2 2
1 1 1
4
t t i t i j t i l t i t R LRTEM LRTRM LRNSDP LRNSDP
n
i
n
j
n
l
e u n m u

1 3 3
1 1 1
5
Here R
t-1
is the lagged residual from the cointegrating regression, the
coefficient of which represents speed of adjustment to the long-run equilibrium.
The above formulation is very useful to identify the sources of causationwe
can evaluate short-run causal relationship between the variables by testing joint
significance of lagged dynamic terms in each of the above equations and we can also
derive information on long-run causality by testing significance of error correction
terms.

In a multivariate framework Granger causality test is also called block-
causality or block exogeneity test where cross equation restrictions are imposed to test
causality between the variables. In this paper, we applied Granger causality test or
block-exogeneity test to evaluate short-run causality between the variables. Weak
exogeneity test is also conducted to evaluate whether a variable can be treated as
exogenous or not, that is whether the variable adjusted towards long-run equilibrium
or not.

6. Econometric Results
At the outset of the cointegration test we have to check the stationarity
property of the variables. Augmented Dickey Fuller test (ADF) and PhillipsPerron
(PP) tests are used to test the null hypothesis of unit root. To test formally for the
presence of a unit root for each variable in the model, ADF and P-P test of the types
given by equation (7) and (8) were conducted. The ADF test is conducted using the
regression of the form:

k
i
t i t i t t u W W t a a W
1
1 1 0 p 6
where t W are the first difference of the series W
t
, k is the lag order and t stands for
time. Equation (7) is with constant and time trend.
P-P test involves computing the following OLS regression:
t t t u
T
t a W a a W )
2
( 2 1 1 0 7
where a
0
, a
1
, a
2
are the conventional least squares regression coefficients. The
hypothesis of unit root to be tested are H
0
: a
1
=1 and H
0
: a
1
=1, a
2
=0. Akaikes
Information Criterion (AIC) is used to determine the lag order of each variable under
study. Mackinnons (1991) tables provided the cumulative distribution of ADF and P-
P test statistics.
Test for stationarity indicates that the null hypothesis of a unit root cannot be
rejected for the levels of the variables except the variables LRTEM and LRTRM
under the assumption of constant and trend. Using differenced data, the computed
ADF and P-P test suggested that the null hypothesis is rejected for the individual
series at 1 or 5 percent significant level.

Table1 Unit root test Result
Test
Trend ssumption
Variables LRTEM LRTRM LRNSDP

ADF
Constant Level -2.02799(1) -1.998(1) -0.773(1)
First diff -5.412(1)* -5.467(1)* -3.9666(1)*
Constant and
trend
Level -4.3056(1)* -4.290(1)* -2.084(1)
First diff -5.202(1)* -5.266(10* -3.965(1)**

P-P
Constant Level -2.191(3) -2.276(3) -1.374(3)
First diff -6.151(3)* -5.803(3)* -5.575(3)*
Constant and
trend
Level -3.681(3)** -3.709(3)** -2.103(3)
First diff -6.466(3)* -6.102(3)* -5.540(30*

Unit root test results on the basis of Augmented Dickey Fuller test (ADF)
indicate that all the series are integrated of order one or I(1). The null hypothesis of
unit root could not be rejected for all the three series in the log level form and clearly
rejected in their first differences for both the models (constant and constant with
linear trend). However, ADF test rejects the null hypothesis of unit root (for the model
with constant and linear trend) only at the 5 percent level of significance for NSDP
series in the log level. Though for the model with a constant and linear trend there is
some doubt about the unit root property of the data in log level form, all the variables
are nonetheless stationary in their first difference form. So we can use Johansen
(1991, 1995) cointegration test to determine the underlying relationships between the
variables.

At the start of Johansen (1991, 1995) test we have to determine the appropriate
lag length for the VAR system as the test results can be sensitive to the choice of the
lag length. We estimated an unrestricted VAR model in level form of the series and
used Akaike Information Criterion (AIC) and Schwartz Bayesian Criterion (SBC) to
choose appropriate lag length p. It is observed that it takes 2 lags to get uncorrelated
and homoscedastic residual for the VAR system. The cointegration test is carried out
assuming linear trend in data, and both an intercept and a trend in the cointegrating
equation. Because some of the series seem to be trend stationary and a linear trend
term in cointegrating space minimizes the value of AIC. Both trace statistic (trace)
and maximal eigenvalue (max) statistics indicate that there is at least one
cointegrating vector between LRTRM, LRTEM and LRNSDP, we can reject the null
hypothesis of no cointegrating vector in favour of one cointegrating vector in both the
cases at 5 percent level of significance. We cannot reject the null hypothesis of at
most one cointegrating vector against the alternative hypothesis of two cointegrating
vectors, for both the Trace and Max-eigen test statistics. Therefore there is a long-run
equilibrium relationship between real government expenditure, real government
revenue and real NSDP. Long-run relationship between these three variables is
derived by normalizing on LRTEM, reported in Table 3.

Table 2 Johansen Cointegration Test Result
Cointegration with unrestricted intercepts and unrestricted trends in the VAR
Cointegration LR Test Based on Maximal Eigenvalue of the Stochastic Matrix
****************************************************************
List of eigenvalues in descending order:
.55782 .38875 .11074
****************************************************************
Null Alternative Statistic 95% Critical Value 90% Critical Value
r = 0 r = 1 25.2972 24.3500 22.2600
r<= 1 r = 2 15.2596 18.3300 16.2800
r<= 2 r = 3 3.6383 11.5400 9.7500
****************************************************************
Cointegration with unrestricted intercepts and unrestricted trends in the VAR
Cointegration LR Test Based on Trace of the Stochastic Matrix
****************************************************************
List of eigenvalues in descending order:
.55782 .38875 .11074
****************************************************************
Null Alternative Statistic 95% Critical Value 90% Critical Value
r = 0 r>= 1 44.1951 39.3300 36.2800
r<= 1 r>= 2 18.8979 23.8300 21.2300
r<= 2 r = 3 3.6383 11.5400 9.7500
****************************************************************

The estimates of normalised coefficients normalized on LRTEM are also
given in the same table in under brackets. Any deviation from long-run equilibrium
will be corrected by changes in government expenditure or NSDP because adjustment
coefficients of these two equations are significant.
Existence of a single cointegrating vector indicates that government expenditure,
government revenue and NSDP display long-run comovement, i.e. there is a long-run
equilibrium relationship between the variables. Whenever the gap between these three
variables widens above the long-run equilibrium, expenditure and NSDP will adjust to
restore the equilibrium. That is, short-run adjustments are done by changes in
government expenditure and NSDP to restore the long-run equilibrium.
Table 3 Cointegration Equation estimated using Cointegrated Vector in Johansen
Estimation (Normalized in Brackets) under the assumption with unrestricted
intercepts and unrestricted trends in the VAR

Vector 1
LRTEM 11.7180
( -1.0000)
LRTRM -11.8443
( 1.0108)
LRNSDP .91800
( -.078341)

However, the existence of long-run equilibrium relationship between these
three variables does not assure a balanced budget, rather the result indicates that the
gap between government expenditure and government revenue will persist at a long
run sustainable level. When budget deficit rises above the long run sustainable level,
expenditure is reduced to maintain the deficit at a tolerable limit. It is worth
mentioning that, government expenditure adjusts at a reasonable speed to the long-run
equilibrium. NSDP also converges to its long-run equilibrium level thorough a series
of partial short-run adjustments. Therefore, government expenditure must be reduced
to maintain budget deficit at a long run sustainable limit. The above finding appears
plausible. Because it is not easy to raise revenue by increasing the taxes on domestic
consumption and income due to various structural constraints and lack of
administrative capacity to collect potential revenue in a developing country like India.
On the other hand, expenditure growth is constrained by revenue shortfall and
increasing cost of deficit financing. Therefore, whenever there is a tendency of rising
deficit beyond its long run limit, expenditure adjusts to restore the equilibrium.
Granger causality test is applied on estimated VEC, where cross equation restrictions
are imposed on the lag differences in each of the equations of VEC incorporating the
cointegrating vectors. If lagged values of LRTEM are statistically significant in both
the LRTRM and LRNSDP equations, we can say that LRTEM G-causes LRTRM
and LRNSDP. The Granger causality tests or the block exogeneity test restricts all the
lag differences of LRTEM or LRTEM to be equal to zero in both the LRTRM and
LRNSDP equations. We can test these restrictions with the help of likelihood ratio
test, which follows chi-sq distribution.

Table 4 Preliminary Granger Causality Tests
1 2 3
Null Hypothesis Short run test
(Preliminary)
Long run test (VEC)
F-Statistic Probability Ch-sq-
Statistic
Probability
LRTRM does not G-Cause LRTEM
LRTEM does not G-Cause LRTRM
0.60162
0.05380
0.55537
0.94772
1.1848
1.5377
0.276
0.215
LRNSDP does not G-Cause LRTEM
LRTEM does not G-Cause LRNSDP
9.04634
3.59961
0.00104
0.04169
3.7437
0.13623
0.053
0.712
LRNSDP does not G-Cause LRTRM
LRTRM does not G-Cause LRNSDP
9.16233
2.44351
0.00097
0.10655
3.8140
0.37964
0.051
0.538

Results of the Granger causality test are reported in Table 4. Here it is clear
from column 3 of table 4 that we cannot reject the null hypothesis that LRTRM does
not Granger cause LRTEM which have a chi-square value of 1.18 with a probability
of 0.276 in the LRTEM equation, reported in row one of Table 4. Again the null
hypothesis that LRTEM does not G-cause LRTRM, cannot be rejected in the
LRTRM equation which have chi-square value of 1.53 with a probability of 0.215.
In both the equations, LRTEM and LRTRM, null hypothesis of no causal
relationship from LRNSDP to LRTEM and LRNSDP to LRTRM is rejected.
However null hypothesis of no causal relationship from LRTEM to LRNSDP and
LRTRM to LRNSDP are cannot be rejected in LRNSDP equation. The above
findings indicate that there is no causal relationship between government expenditure
and revenue in the long run. Granger causality test is also applied on the unrestricted
VAR model for each pair of the above three variables. The column 2 of table 4 shows
the calculated F-statistic of short run G-causality for the variables. The calculated F-
statistics for both the revenue and expenditure equations also suggest that there is no
causal relationship between revenue and expenditure in the short run. Therefore, we
can say that government expenditure and revenue are independently determined in the
short run. Budget deficit can be reduced either by reducing government expenditure
or by raising revenues in the short run. When both government expenditure and taxes
are raised together, rise in expenditure should be less than the rise in taxes to reduce
the budget deficit.

On the other hand, in the short run, there exist a bi-directional causality
between TEM and NSDP. The implication is that to increase TEM and TRM, the
NSDP must be boost up in the short run. Short run causality between the expenditure
variable and NSDP implies that changes in expenditure policies may affect real NSDP
growth (in the sense that changes in government expenditure precede changes in
NSDP) in the short run. Further, the causal relation between the variables is studied
by using the VEC causality analysis. The results are reported in table 4. From the
analysis, we noticed that there is no causality between revenue and expenditure
because both the hypothesis of TRM does not G-cause TEM and its reverse are cannot
be rejected. But from the analysis, it is further noticed that the hypothesis that NSDP
does not G-cause TRM and TEM are rejected. This indicates that in the long run
NSDP has impact on both TEM and TRM.

7. Conclusion
The paper examines causal relationship between real government expenditure
and revenue by using annual data from 1972 to 2004 period. Real NSDP is also
included in the model along with these two fiscal variables. Johansen (1991, 1995)
cointegration test suggests that there is a long-run equilibrium relationship between
government expenditure, government revenue and NSDP. Granger causality test
based on the estimated VEC suggests that there is no causal relationship between
government expenditure and revenue in the long run. It is felt that expenditure is the
key variable to reduce budget deficit because of structural constraints of the economy
in raising revenue. So, budget deficit should be reduced by reducing public
expenditure in unproductive sectors and at the same time ensuring effective utilization
of available resources such as to engender the productivity of labor and capital in the
economy. If the economy achieves enhanced economic growth it will be possible to
raise revenue from domestic sources.

However, any conclusion to be derived from these results should bear in mind
the limitations of the analysis, in particular the fact that this is based on past data and,
consequently, the future evolution of some structural factors (for example,
demographic rends) and their impact on public finances are not taken into account.

Preliminary Granger causality tests, offer two main results. First, the deficit is
not independent of fiscal pressure or the size of the public sector. Second, we find
only very weak evidence, if any, that revenues G-cause expenditures. However, these
results may be biased in that they do not consider the possibility of the existence of
long-run relationships among the variables involved. Once we consider this
possibility and develop the causality analysis with cointegrated variables, we find no
evidence of public expenditures G-causing public revenues in the long run and long-
run G-causality from revenues to public expenditure.

References
Barro, R. J., (1988), Government Spending in A Simple Model of Endogenous
Growth, Working Paper No. 2588, NBER.

Bose, N.; M. E. Haque, and D. R. Osborn, (2003), Public Expenditure and Economic
Growth: Disaggregated Analysis for Developing Countries, Economic and Social
Research Council, UK.

Cheng, B.S. (1999), Causality between Taxes and Expenditures: Evidence from
Latin American Countries. Journal of Economics and Finance, vol. 23, no.2, pp. 184-
192.

Chowdhury, O. H., Domestic Resource Generation and Public Expenditure, paper
presented at the Seminar on National Budget for 2005-06 and PRSP.

Engle R. F and C. W. J. Granger (1987), Co-integration and Error Correction:
Representation, Estimation, and Testing, Econometrica, vol. 55, no.2, 1551-1580.

Hussain M. H. (2005), On the Causal Relationship between Government,
Expenditure and Tax Revenue in Pakistan, Economics Working Paper Archive
(EconWPA), http://econwpa.wustl.edu:80/eps/mac/papers/0509/0509014.pdf.

Johansen S., (November, 1991), Estimation and Hypothesis testing of Cointegration
Vectors in Gaussian Vector Autoregressive Models. Econometrica, vol. 59, no. 6,
pp. 1551-1580.

Khattry, B. (2003), Trade liberalization and the Fiscal Squeeze: Implications for
Public Investment, Development and Change 34(3), pp. 401-424.

Maghyereh, A. & O. Sweidan, (2004), Government Expenditure and Revenues in
Jordan, What Cause What? Multivariate Cointegration Analysis, Social Science
Research Network Electronic Paper Collection.

Martin, M. J., J. Krishnakumar and N. Sougel, (2004), Application of Granger
Causality Tests to Revenue and Expenditure of Swiss cantons, Cahiers du
department d econometrie Faculte des sciences economiques et sociales Universite
de Geneve

Payne, J. E., (2003), A Survey of the International Empirical Evidence on the Tax-
Spend Debate, Public Finance Review, vol. 31 no. 3, pp. 302-324.