© Economic Literature volume IX April, 2010

Public Expenditure and Economic Growth in Nepal

Shashi K. Chaudhary

Abstract
This paper uses cointegration and error-correction models to analyze the causal relationship
between public expenditure and economic growth of Nepal. While conventional wisdom suggests
that public expenditures contribute positively to economic growth, this study provides strong
evidence rejecting the fact. Though cointegration analysis provides positive evidence for the
existence of a long-run relationship between public expenditure and real GDP, an indicator of
economic growth; however the long run causality test based on the standard t-test statistics from
the ECM indicates that there is a unidirectional causality from real GDP to public expenditure,
not vice versa. The short run causality test based on F-test statistics from the ECM indicates no
causality between real GDP to public expenditure. The pair-wise Granger causality test
confirms the absence of the short run causality between real GDP to public expenditure. Thus,
the results support the Wagner’s hypothesis, which states that the growth of public expenditure
can be explained as a result of the increase in economic activity. The findings suggest that the
increase in the public expenditure has no influence on the economic growth of Nepal.

Keywords: cointegration, economic growth, error correction model, public expenditure

1. INTRODUCTION

The relationship between public expenditure and economic growth is an important subject of
analysis and debate, especially for developing countries. A central question is whether public
expenditure increases the long run steady state growth rate of the economy. The general view is
that public expenditure, notably on physical infrastructure or human capital, can be growth-
enhancing although the financing of such expenditures can be growth-retarding, for example,
because of disincentive effects associated with taxation (Kweka and Morrissey, 2000:1). There
are considerable debates over the effects of the government spending on economic growth.
Besides, providing national defense and securities and transfer payments to maintain social
welfare and harmony, a government can provide economic infrastructure to facilitate economic
growth. Government expenditures on health and education can improve labor force productivity.
In addition governments can provide information, reduce risks and alter incentives. However, the
quantity of public goods provided by the government may be inefficient. There are also possible
negative impacts on economic growth induced by a government’s revenue raising and transfer
mechanism. Thus, government taxation may produce a misallocation of resources as well as
disincentives. Many policies contain incentives and disincentives for growth because they
increase or reduce rewards to human as well as physical capital (Albatel, 2000:173).
Nepal has been practicing expansionary fiscal policy since long time. The most important
objective of Nepalese fiscal policy is to attain a significant economic growth. In Nepal, the
public expenditure has remained the most important tool for the fiscal policy. During the period
between 1975 and 2006, social and political changes have accompanied by a sharp increase in
government spending. For example, while the ratio of total public expenditure to GDP was 9.13
percent in 1975, this ratio doubled in just thirty years, increasing to 20.16 percent in 2005.
During the mentioned period, the economic development efforts have yielded an average growth
of around four per cent per annum only. Therefore, the effectiveness of fiscal policies in Nepal
has been reported unsatisfactory. The part of the capital expenditure in the budget has fallen far
below the expected level which is the central matter of concern as the low level of capital
expenditure adversely hits the development activities. The issue of whether increasing public
expenditure is the cause of economic growth or economic growth is the cause of increasing
public expenditure is important for Nepal since the public expenditure has been increasing at
faster rate while the indicators of economic growth have not shown drastic variations. Many of
the economic indicators have remained almost constant showing as if there is no long run
relationship between public expenditure and economic growth.

This paper investigates econometrically the existence of a long-run relationship between public
expenditure and GDP, both measured in real terms, using data over the period 1975-2006. The
growth of real GDP has been used as a proxy for the economic growth. Further, according to
Wagner’s law, there is a long-run tendency for public expenditure to grow relative to some
national income aggregate such as gross domestic product (GDP). Hence, the long-run
relationship between public expenditure and GDP (if exists) has also been examined along the
lines suggested by Wagner’s Law.

2. METHODOLOGY

The present work follows a three step procedure. In the first step, the stationary properties of the
data series have been examined to determine the order of integration of the series. To this end,
tests for unit roots have been carried out using the augmented Dickey-Fuller (ADF) test. Tests
97
for unit roots in the levels of the series have been followed by tests for unit roots in the first
difference of the series. In the second step, test for cointegration among the variables involved
have been carried out, in the event that they are identified as I(1) in the first step, using the Engle
– Granger technique to define the number of the cointegrating vectors and report the estimated
relationships. In the third step, the causality dynamics between the variables by carrying out
Granger causality tests has been examined. All the necessary computations have been done using
Eviews 5.1 version.

2.1. Stationarity and Unit Root Test

The stationarity of the series {lnYRt} and {lnGRt} have been investigated; where YR is the real
GDP and GR is the real public expenditure, both expressed in logarithm. To determine the order
of integration of these processes, the augmented Dickey–Fuller (ADF) test has been used on the
level and the first difference of the series. The general form of ADF test can be written as
follows:
p
∆Xt = a0 + a2t + γXt-1 + ∑ βi ∆X t − i + εt (for levels) (1)
i =1

p
∆ ∆Xt = a0 + γ∆Xt-1 + ∑ βi ∆ ∆X t − i + εt (for first differences) (2)
i =1

where ∆Xt = Xt – Xt-1 and X is the variable under consideration, p is the number of lags in the
dependent variable, and is chosen so as to induce a white noise term and εt is the stochastic error
term.

2.2. Testing for Cointegration

The testing of cointegration is accomplished here by using the Engle-Granger two-step
cointegration procedure: firstly, the time series properties of each variable are examined by unit
root tests. Having tested the stationarity of each time series, two cointegration regressions (direct
and reverse) between variables are estimated using the OLS1. The second step involves directly
testing of the stationarity of error processes of two cointegration regressions estimated in
previous step.

1
The direct and reverse cointegration regressions for two time series Yt and Xt can be written as follows in log
linear form:
Yt = a0 + a1 Xt + u1t and Xt = b0 + b1 Yt + u2t

98
Engle and Granger (1987) have shown that if variables Yt and Xt are cointegrated, the residuals
from the equilibrium regression can be used to estimate the error correction model. If {Yt} and
{Xt} are CI (1.1), the variables have the error correction form
p p
∆Xt = α0 + αx ρt-1 + ∑ α1i ∆ X t − i + ∑ α 2i ∆Yt − i + εxt (3)
i =1 i =1

p p
∆Yt = β0 + βy ρt-1 + ∑ β1i ∆ X t − i + ∑ β 2i ∆Yt − i + εyt (4)
i =1 i =1

where ∆ is first difference operator on variables, εxt and εyt are white noise disturbances, and α’s
and β’s are all parameters. The residual, ρt-1 is the error correction term which is the lagged
residuals from the cointegration relations. Its magnitude, αx or βy implies the deviation from long
run equilibrium in period (t-1). The independent variables are said to long run ‘cause’ the
dependent variable if the error correction term, (ρt-1) is significant based on t-test statistics or
short run ‘cause’ if the coefficients of the lagged independent variables are jointly significant
based on F-test statistic in their first differences.

2.3. Testing for Causality

This paper uses Granger type causality methodology to determine the causality direction between
the two variables used in the study. The simplest Granger causality test (Granger, 1969) is:
p p
Yt = b0 + ∑ b1i Yt − i + ∑ b 2i X t − i + et (5)
i =1 i =1

p p
Xt = δ0 + ∑ δ1i X t − i + ∑ δ 2i Yt − i + ut (6)
i =1 i =1

where Yt and Xt are the variables under consideration; et and ut are white noise error terms.

The null hypothesis for equation (5) is that 'X does not Granger cause Y'. This hypothesis would
be rejected if the coefficients of the lagged Xs (summation of b2i as a group) are found to be
jointly significant (different from zero). The null hypothesis for equation (6) is that 'Y does not
Granger cause X'. This hypothesis would be rejected if the coefficients of the lagged Ys
(summation of δ2i as a group) are found to be jointly significant. If both of these null hypotheses
are rejected, then a bi-directional relationship is said to exist between the two variables Yt and
Xt.

99
2.4. Data
This study is based on the econometric analysis of secondary data from FY 1974/75 through FY
2005/06 and the frequency of the data is annual. The necessary data have been borrowed from
various issues of ‘Economic Survey’ (2005/06, 2006/07 and 2007/08) of MOF, G/N and ‘A
Handbook of Government Finance Statistics’ (2008), NRB. The nominal figures of GDP and
public expenditure have been deflated by the GDP deflator (1975=100) to express them in real
terms.
Using annual data is believed to be appropriate here because public expenditure and gross
domestic product is not very sensitive to seasonal and cyclical fluctuations. Hakkio and Rush
(1991) argue that increasing the number of observations by using monthly or quarterly data do
not add any robustness to the results in tests of cointegration.

3. EMPIRICAL RESULTS

In the light of econometric methodology presented in the previous section, the empirical results
have been discussed here.
Table 1 presents the results of unit root tests obtained using both the augmented Dickey-Fuller
test and Phillips-Perron (PP)2 test on individual series. The lag length selection was set automatic
based on SIC (Schwarz Information Criterion), max lag = 9 for ADF and Newey-West
Bandwidth for Phillips-Perron test. The results support the presence of unit roots in both the
series. The absolute calculated values are less than the corresponding McKinnon critical values
for the levels of the variables. Hence, the null hypothesis is not rejected at the levels for both the
series. However, the null hypothesis is rejected in favor of alternative hypothesis of series are
stationary when the first difference of the variables are taken. Thus, their first difference is found
to be stationary and hence lnYR and lnGR are both integrated of order one, I (1).

Table 1: Results of Unit Root Tests

ADF statistic PP statistic
Series Degree of Integration
Level First difference Level First difference
lnYR -1.013 -3.705* -1.006 -3.741* I(1)
lnGR -2.042 -4.830* -1.803 -4.739* I(1)

2
The findings of ADF tests have been confirmed by PP test.

100
* denotes 1% significant level based on Mackinnon’s critical values. In the regression with level, both
intercept and trend have been included while for first difference, only intercept has been included.

Having confirmed the existence of unit roots for the data series, the next step involves applying
Engle-Granger two-step cointegration procedure to check whether the two variables lnYR and
lnGR are cointegrated. The results of the ADF test applied to residuals of the cointegration
equations are presented in Table 2. Together with the results, the values of the slope coefficients
and Cointegration Regression Durbin Watson (CRDW) statistics are also presented.

Table 2: Results of Engle-Granger Cointegration Tests

Cointegration equation Slope CRDW Calculated ADF statistic for Residuals
0.688
lnYR = f(lnGR)3 (17.579*) 0.29** -2.869*

1.325
ln GR = f(lnYR)4 0.31** -3.815*
(17.579*)
* and ** indicate the statistical significance at 1 percent and 5 percent levels of significance. The figures
in brackets are t–statistics for the coefficients. The Mackinnon’s critical value of ADF statistic at 1
percent level of significance is -2.641672. The critical value of the CRDW statistic is 0.282 and 0.209 at
the 5% and 10% levels of significance respectively, tabulated in Table III of Engle and Granger (1987).

The results presented in Table 2 indicate that the estimated ADF statistics for the residuals are
greater than their corresponding critical values. Therefore, lnYR and lnGR are cointegrated. This
finding is also confirmed by the CRDW statistic5. The results indicate that the CRDW statistic is
statistically significant at 5 percent level. The observed CRDW of 0.29 and 0.31 are greater than
5 percent value of 0.282 indicating the presence of cointegration. Thus, the CRDW statistic
confirms the stationarity of the residuals of cointegration equations and hence implies a long-run
association between real public expenditure and real GDP.
Next is the Ganger causality test with error correction terms from the cointegration equations.
The optimal number of lags for each variable has been determined by resorting to AIC (Akaike
Information Criterion) and SIC criterion in order to determine which variable Granger causes the
other and provides the short-run dynamics adjustment toward the long-run equilibrium. Since,

3
The actual cointegration relation being calculated is:
lnYR = 4.39 + 0.68 lnGR
(13.20)* (17.58)* R2 = 0.911 DW = 0.29 F-statistic = 309.03*
4
The actual cointegration relation being calculated is:
lnGR = - 5.07 + 1.32 lnYR
(-6.56)* (17.58)* R2 = 0.911 DW = 0.31 F-statistic = 309.03*
5
Though ADF test is a more powerful test when compared with the use of the CRDW statistic, Engle and Granger
(1987) point out that for quick approximate results one could use the CRDW statistic.
101
the data taken are annual and the size is not very large, the maximum order of lag-length was
checked up to 4 lag. The resulted values are as follow:

p 1 2 3 4
AIC -5.237 -4.856 -4.574 -4.298
SIC -4.770 -4.196 -3.717 -3.242

The minimum AIC and SIC criterion values are achieved at the first lag. Therefore, the equations
(3) and (4) have been estimated at p=1.
The empirical results of the estimated error-correction models are presented in Table 3. It
presents the results of both the long run Granger causality test based on a standard t-test statistics
of the error terms, (ρt-1) lagged one period as well as the short run Granger causality test based on
a standard F-test statistics for the jointly significance of the coefficients of the explanatory
variables in their first differences.
The coefficient of the error term in equation (3.a) is statistically significant based on standard t-
test which means that the error term (ρt-1) contributes in explaining the changes in public
expenditure (lnGR). On the other hand, the coefficient of the error term in equation (4.a) is
statistically insignificant which means that the error term (ρt-1) does not contribute in explaining
the changes in real GDP. Therefore, there is unidirectional causality running from GDP (YR) to
public expenditure (GR) in the long run. Further, the growth of one year lag GDP has established
a positive relationship, with long-term public expenditure growth.
The coefficient of the error correction term is of particular interest because it represents the
direction and speed of adjustment as well as the stability of the system. The absolute value of the
coefficient has been found to be less than unity (0.3173), which indicates that the system is
stable. Further, the magnitude of the coefficient suggests that about 32 percent of any deviation
of the system from its long run equilibrium path is likely to be corrected within a year.

Table 3: Results of Error Correction Models

∆lnGRt = 0.0389 + 0.3173 ρt-1 + 0.5757 ∆lnYRt-1 - 0.2149 ∆lnGRt-1 + ut (3.a)
(1.62) (2.66)* (1.33) (-1.04)
R2 = 0.24 F-Statistic = 2.68***
∆lnYRt = 0.0396 - 0.0565 ρt-1 + 0.0073 ∆lnYRt-1 - 0.1804 ∆lnGRt-1 + et (4.a)
(2.68) (-0.77) (0.027) (-1.41)
2
R = 0.11 F-Statistic = 1.12

102
where ρt-1 = (lnYRt-1 - 0.9303 lnGRt-1 - 2.3207), is the error correction term. (.) indicates t-statistic. ***
indicates statistically significant at 10 percent level.

Although the coefficients of the explanatory variable in their first differences in equation (3.a)
are jointly statistically significant based on standard F-test statistics, none of the coefficients of
the explanatory variables based on t-test statistics are statistically significant within five percent
level. This does not mean that YR Granger causes GR, but rather a reflection of the significance
of the error term (ρt-1). Further, the coefficients of the explanatory variables in their first
differences in equation (4.a) are jointly statistically insignificant based on F-test statistics.
Therefore, there is no causality between real GDP and public expenditure in the short run.
In order to confirm the result of the short-run causality between real GDP and public expenditure
based on ECM estimates, a standard Granger causality test has been run based on chi-square (χ2)
statistics, which is presented in table 4.

Table 4: Results of ECM Pair-wise Granger Causality/Block Exogeneity Wald Tests

Null Hypothesis χ2 - statistic Probability Comment
ΔlnYR does not Granger cause ΔlnGR 2.006 0.1566 Accept
ΔlnGR does not Granger cause ΔlnYR 1.765 0.1839 Accept

Table 4 confirms the earlier finding based on ECM estimates of the absence of short run
causality between public expenditure and real GDP.

4. CONCLUSION

This paper has applied cointegration and error-correction models to test causal relation between
public expenditure and economic growth of Nepal. After a thorough examination of the time
dependence properties of the series, cointegration analysis has validated the existence of long-
run relationship between the variables. Though, cointegration analysis provides positive evidence
for the existence of a long-run relationship between public expenditure and real GDP; however
the long run causality test based on the standard t-test statistics from the ECM indicates that there
is a unidirectional causality from real GDP to public expenditure and the inverse relation does
not hold. The short run causality based on F-test statistics from the ECM indicates no causality
between real GDP to public expenditure. The pair-wise Granger causality test confirms the
absence of the short run causality between real GDP to public expenditure. Thus, the results
103
support the Wagner’s hypothesis, which states that the growth of public expenditure can be
explained as a result of the increase in economic activity. The findings suggest that the increase
in the size of public expenditure has no influence on the economic growth of Nepal.
This finding is surprising since the researcher had expected the existence of a bi-directional
causality between them. The budgetary operation in Nepal has been based on expansionary
policies since long back. It is believed that government can play influential role in economic
development of Nepal and therefore, the size of government should be increased. But, the
finding of this paper suggests that the increase in the public expenditure has no influence on the
economic growth of Nepal.

REFERENCES
1. Albatel, A.H. (2000). The Relationship between Public Expenditure and Economic Growth
in Saudi Arabia. Journal of King Saud University,12; 173-191.
2. Enders, W. (2008). Applied Econometric Time Series (2nd ed.). New Delhi: Wiley India Pvt.
Ltd.
3. Engle, R.F., & Granger, C.W.J. (1987). Cointegration and Error Correction: Representation,
Estimation and Testing. Econometrica,55: 251-276.
4. Granger, C.W.J. (1969). Investigating Causal Relations by Econometric Models and Cross
Spectral Methods. Econometrica, 37: 213-228.
5. Hakkio, C. S., & Rush, M. (1991). Cointegration: How Short is the Long Run? Journal of
International Money and Finance, December: 571-581.
6. Kweka, J.P., & Morrissey, O. (2000). Public Spending and Economic Growth in Tanzania,
1965-1996. CREDIT Research Papers, No. 00/6: University of Nottingham.
7. MOF (2006). Economic Survey, FY 2005/06. Kathmandu: HMG/N.
8. ____ (2007). Economic Survey, FY 2006/07. Kathmandu: G/N.
9. ____ (2008). Economic Survey, FY 2007/08. Kathmandu: G/N.
10. Nachane, D. M. (2006). Econometrics: Theoretical Foundations and Empirical
Perspectives. New Delhi: Oxford University Press.
11. NRB (2008). A Handbook of Government Finance Statistics. Kathmandu: Research
Department, Government Finance Division.

ììì

104