You are on page 1of 9

Energy Economics 26 (2004) 77–85

The impact of electricity supply on economic


growth in Sri Lanka
Risako Morimoto*, Chris Hope
Judge Institute of Management Studies, University of Cambridge, Cambridge CB2 1AG, UK

Abstract

Yang (Energy Econ. 22 (2000) 309) has found a bi-directional causal relationship between
gross domestic product and electricity consumption in Taiwan, ROC. This study applies
Yang’s model to examine the impact of electricity supply on economic growth in Sri Lanka.
Morimoto and Hope (An extended CBA model of hydro projects in Sri Lanka (2001)) have
found that the expected increase in economic output due to increased electricity supply
(parameter EO) plays a crucial role in their cost benefit analysis model. This study shows
that the application of Yang’s regression analysis is one possible approach to estimate a
better range for the parameter EO. The estimated figure is that an extra output of Rs. 88 000–
137 000 ($US1120–1740) for every 1 MW h increase in electricity supply.
䊚 2003 Elsevier Science B.V. All rights reserved.

JEL classifications: Q41

Keywords: Economic growth; Electricity; Gross domestic product; Sri Lanka

1. Introduction

An adequate and regular power supply may be one of the most crucial factors,
which supports economic growth in developing countries. According to a study on
the relationship between electricity use and economic development conducted by

*Corresponding author. Tel.: q44-1223-339639; fax: q44-1223-339701.


E-mail address: rm253@cam.ac.uk (R. Morimoto).

0140-9883/04/$ - see front matter 䊚 2003 Elsevier Science B.V. All rights reserved.
PII: S 0 1 4 0 - 9 8 8 3 Ž 0 3 . 0 0 0 3 4 - 3
78 R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85

Fig. 1. GDP and electricity demand growth rates in Sri Lanka (%).

Ferguson et al. (2000), there is a strong correlation between electricity use and
economic development.1 The Pearson correlation coefficient between growth in
annual electricity use and average annual economic growth for Sri Lanka during
the period 1971–1995 is 0.993 (Ferguson et al., 2000).2
Fig. 1 shows that there is a strong correlation between average annual growth
rates of gross domestic product (GDP) and electricity demand in Sri Lanka between
1984 and 1997 (CEB, 1999). Energy demand in Sri Lanka is mainly met by
hydropower so that electricity supply decreases severely when the country is hit by
serious droughts. This has led to a dramatic decline in its economic growth. In
particular, serious droughts in 1996 meant that Sri Lanka experienced a severe
power crisis that adversely affected the economy in 1996 (CEB, 1999). The GDP
growth rates in 1995, 1996 and 1997 are 5.45, 3.76 and 6.45%, respectively. This
evidence of the strong correlation between the growth rates of GDP and the
electricity demand in Sri Lanka enables one regression analysis that has a potential
to estimate the impact of electricity supply on economic growth to be applied. The
detail of the regression analysis applied to this particular case of Sri Lanka is
explained below.
Yang (2000) finds that there is a bi-directional causal relationship between GDP
and electricity consumption in Taiwan for the period 1954–1997. This study applies
1
Similar studies are, for example, Ramcharran (1990), Huang (1990), Mashi and Mashi (1996) and
Asafu-Adjaye (2000).
2
Their analysis uses electricity consumption in kWh as the energy use variable and GDP at purchasing
power parities in 1995 US dollars.
R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85 79

his model to examine the impact of electricity supply on economic growth in Sri
Lanka. Morimoto and Hope (2001) find that the expected increase in economic
output due to increased electricity supply (parameter EO) plays a crucial role in
their CBA (cost benefit analysis) model, which assesses the economic, social and
environmental impacts of dam projects in Sri Lanka. The model has used a mean
value of Rs. 26 000yMW h for the parameter EO, based upon an official calculation
by the Ceylon Electricity Board (CEB). This study shows that the application of
Yang’s regression analysis is also one possible approach to estimate a better range
for the parameter EO in the CBA model we have developed (Morimoto and Hope,
2001).

2. Methodology and estimation results

2.1. Methodology

Yang (2000) employs the Granger-causality test and his finding implies that
electricity shortage may have restrained economic growth in Taiwan over the study
period (1954–1997). He uses first differenced real GDP as a dependent variable,
with lagged first differenced electricity consumption and lagged first differenced
real GDP as independent variables in the model described in Eq. (1):

DGDPtsaq83is1biDGDPtyiq82is1ciDELECTtyiqut (1)

where DGDPt, first differenced real GDP in Taiwan at time t; DELECTtyi, first
difference of electricity consumption in Taiwan at time tyi and ut, error term at
time t.
Yang does not present the estimated coefficients of the equation in his article.
However, he estimates another model without DELECTtyi and obtains Akaike’s
final prediction errors (error values that are used to select appropriate independent
variables and lag specifications) for both models with and without DELECTtyi. He
finds that Akaike’s final prediction error for the equation with DELECTtyi is smaller
than the one without DELECTtyi, which accepts the hypothesis that electricity
consumption Granger-causes GDP.
Since Sri Lanka does not import or export electricity, the amounts of electricity
production and consumption are the same.3 Thus, in this article, annual electricity
production will be used instead of electricity consumption. The current electricity
consumption was not included in Yang’s model as his model was purely to
investigate the directions of the causality.4 However, according to Yang, the current
electricity consumption is included in the estimation of a production function in
research publications on Taiwanese economic growth. In the case of Sri Lanka, the
change in electricity production at time t (DELECTt) should also be included, as
3
See UN Energy Statistics Year book.
4
A time series is said to Granger-cause another time series Y if the prediction error of current Y
declines by using past values of X in addition to past values of Y.
80 R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85

Fig. 2. Electricity production in Sri Lanka (million kW h).

GDP growth at time t seems highly dependent on electricity supply at time t (Fig.
1). Hence, the following slightly modified equation is used in this study:

DGDPtsaq83is1biDGDPtyiq82is0ciDELECTtyiqvt (2)

where DGDPt, first differenced real GDP in Sri Lanka at time t; DELECTtyi, first
difference of electricity production in Sri Lanka at time tyi and vt, error term at
time t. Eq. (2) implies that change in real GDP at time t is a function of past
changes (with yearly lags up to ty3) in real GDP and of current as well as past
changes in electricity supply (with yearly lags up to ty2).

2.2. Results

Annual data for the period 1960–1998 for real GDP (billion Rs.) and for
electricity production (million kW h) in Sri Lanka are used (Figs. 2 and 3).5
Before estimating the model, the stationarity of dependent and independent
variables are examined in order to meet the condition of using Yang’s Granger-
causality model. Although both GDP and electricity production are non-stationary,
their first differenced values are stationary according to the result from the
5
The nominal GDP is transformed into real GDP in 1998 prices using GDP deflators, which is the
same transformation method as Yang’s (2000).
R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85 81

Fig. 3. Real GDP in Sri Lanka (billion Rs.).

Augmented Dickey–Fuller test for stationarity and unit roots.6 Moreover, the
cointegration test for the series of GDP and electricity production is also satisfied,
therefore the standard Granger-causality test can be applied.7
Eq. (2) is estimated by a standard ordinary least square (OLS) regression analysis
with 35 observations and the result is presented in Table 1, column 1. The value
for the model fit (adjusted R 2s70%) seems to be reasonable for this type of growth
models.8 All the coefficients of DGDPtyi are insignificant at the 5% level individ-
ually (t-tests) and simultaneously (F-test).9 Therefore, they are dropped from the
model, which is then re-estimated. The results are presented in Table 1, column 2.
All the coefficients of DELECTtyi in the model (2) are significant at the 5% level.
The result implies that current as well as past changes in electricity supply have a
significant impact on a change in real GDP. According to the specification in column
2 in Table 1, one unit change in DELECTt, DELECTtyi and DELECT ty2 separately
leads to changes in DGDPt of 38.2, 30.0 and 44.1 units, respectively. In other words,
6
The ADF test statistics show the values of 1.95 and 0.85 for the variables GDP and ELECT,
respectively, which are insignificant at the 5% level. However, the ADF test shows the values of
y4.68 and y5.97 for a first difference of GDP and ELECT, respectively, which are both significant at
the 5% level. See basic econometrics textbooks such as Greene (2000) for more details about the ADF
tests.
7
The unit root tests for residuals show the value of y2.5, which is insignificant at the 5% level.
8
Its diagnostic test also shows that there are no serial correlation or heteroscedasticity problems.
9
F-test for the null hypothesis of all the coefficients of DGDPtyi being insignificant simultaneously
shows the values of 0.72 with P value 0.56.
82 R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85

Table 1
OLS estimates of first differenced real GDP (DGDPt) in Sri Lanka for the period 1960–1998

Explanatory variables (1) Original model (2) DGDPtyi all (3) DELECTty3
eliminated added to (2)
Intercept 7072 7710 7138
(2.4)a (3.4)a (3.2)a
DGDPty1 0.15
(0.8)
DGDP ty2 y0.22
(y1.0)
DGDPty3 0.12
(0.8)
DELECTt 37.2 38.2 34.2
(4.6)a (7.2)a (5.8)a
DELECTty1 25.5 30.0 24.2
(1.8)b (4.9)a (3.4)a
DELECT ty2 46.4 44.1 37.7
(2.4)a (4.5)a (3.6)a
DELECTty3 22.0
(1.5)
Adjusted R 2 0.70 0.71 0.72
The sample size is 35. The figures in parentheses are t-values; DGDPtyi, first differenced real GDP
in Sri Lanka at time tyi; DELECTtyi , first difference of electricity production in Sri Lanka at time ty
i.
a
Significant at the 5% level.
b
Significant at the 10% level.

1 MW h increase in electricity supplies at time t, ty1 and ty2 results in economic


growth at time t of Rs. 38 200, 30 000 and 44 100, respectively. The coefficient of
DELECT ty2 is larger than the coefficients of DELECTt and DELECTty1, which is
surprising, as it is generally expected that the impact decays as time passes.
However, this result could be due to time lags in the system, so that increased
electricity supply takes some time to have its full effect on GDP growth. Yang
(2000) has also used time lags ty1 and ty2 for DELECT in his analysis, though
the comparison is unfortunately not possible, as his coefficient values were not
presented in his article.10

2.3. Implications for the parameter EO

On the basis of the above regression result, a value for the parameter EO
(expected increase in economic output) used in the CBA model in Morimoto and
Hope (2001) can be estimated. The standard error (S.E.) for the coefficients of
DELECTt, DELECTty1 and DELECT ty2 are 5.3, 6.1 and 9.7, respectively.
Therefore, the S.E. for their sum should be w5.32q6.12q9.72x1y2s12.6 with the
10
Yang (2000) has employed the Akaike final prediction error criterion for the selection of the
appropriate lag specifications for the variable DELECT.
R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85 83

assumption of these coefficients following independent normal distributions.11


Hence, the 95% confidence interval (CI) for a joint impact of DELECTt,
DELECTty1 and DELECT ty2 is w38.2q30.0q44.1x"1.96(12.6)s88–137. Then,
the 95% CI for the parameter EO may be expressed as the range Rs. 88 000–
137 000yMW h ($1120–1740yMW h).12 This range of values is much higher than
the CEB estimation of EOsRs. 26 000yMW h ($330yMW h), used in the CBA
model in Morimoto and Hope (2001).13
The range that Morimoto and Hope (2001) used for the parameter EO was
minimumsRs. 10 000yMW h ($200yMW h), most likelysRs. 26 000yMW h
($330yMW h) and maximumsRs. 118 500yMW h ($1500yMW h), respectively.
The regression results provide some support for this range since EOsRs. 88 000–
137 000yMW h ($1120–1740yMW h), estimated in this article, overlaps this range.
However, the estimates here allow the value to be much higher. The minimum value
of Rs. 10 000yMW h, which is much lower than the minimum estimated value, now
seems hard to justify. The maximum value of Rs. 118 500yMW h is not large
enough to cover the full range of the estimated results. It is possible to expect such
large economic growth due to increased power supply as predicted by the model,
since Sri Lanka is currently facing power shortages, which may slow down the
economic growth.
The above result shows that the coefficient of DELECT ty2 has the highest value,
hence it is necessary to check whether DELECTty3 also has a significant impact on
GDP growth. Thus, DELECTty3 is added to the model in order to check its impact
on DGDP. The result shown in column 3 of Table 1 indicates that the coefficient of
DELECT ty2 is again larger than the coefficients of DELECTt and DELECTty1;
however, the coefficient of DELECTty3 is the smallest and insignificant. The 95%
CI value for the parameter EO predicted by the model with this specification is Rs.
78 000–158 000yMW h, calculated using the same procedure as above; a very
similar value to the one without DELECTty3.14

11
However, if the distributions of each coefficient are correlated, S.E. (sum of the three coefficients)s
wS.E.(c0)2qS.E.(c1 )2 qS.E.(c2 )2 q2{cov(c0 , c1 )qcov(c0 , c2 )qcov(c1 , c2)}x1y2sw5.32q6.12q9.72q
2(0.25y3.8q25.1)x1y2s14.2. This calculation is carried out by the statistical package MICROFIT.
12
The value for the S.E. for the coefficients of DELECTt, DELECTty1 and DELECTty2 are 5.3, 6.1
and 9.7, respectively. Thus, the 95% CI values are calculated as follows: CI(DELECTt)s
38.2"1.96*(5.3), CI(DELECTty1 )s30.0"1.96*(6.1), CI(DELECTty2 )s44.1"1.96*(9.7). Hence, the
95% CI values for the coefficients DELECTt , DELECTty1 and DELECTty2 are w27.8–48.6x, w18.0–
42.0x and w25.1–63.1x, respectively.
13
The UK Electricity Pool uses the even lower value of Rs. 23 000yMW h as the value of lost load
in UK. The value of outage cost to Chilean industry lies between Rs. 6080yMW h (for 10% 1-month
equiproportional restriction) and 17 380yMW h (for 30% 10-month equiproportional restriction). See
Serra and Fierro (1997) for more details. The exchange rates of £1sRs. 184 and $1sRs. 79 are used
(Central Bank of Sri Lanka, 1998).
14
(34.2q24.2q37.7q22)"1.96(5.92q7.22q10.52q15.02 )1y2 s78y158. The value for the S.E. for
the sum of the coefficients of DELECTt , DELECTty1 , DELECTty2 and DELECTty3 is 14.5 if these
coefficients are correlated. This is calculated by w5.92q7.22q10.52q15.02q2(11.2q8.5y41.6q41.6y
59.5y65.9)x1y2.
84 R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85

3. Conclusion

The impact of electricity supply on economic growth in Sri Lanka is closely


examined in this study. The methodology is based on the research conducted by
Yang (2000) for Taiwan, using a simple regression analysis. The findings from the
Sri Lankan data imply that current as well as past changes in electricity supply have
a significant impact on a change in real GDP in Sri Lanka. The result can also be
used to estimate a better range for the parameter EO (increase in economic output
due to increased electricity supply in Sri Lanka) in the CBA model that we have
developed (Morimoto and Hope, 2001). The estimated figure for the parameter EO
using the regression analysis is that an extra output of Rs. 88 000–137 000 ($1120–
1740) for every 1-MW h increase in electricity supply in Sri Lanka, which seems
to support the actual range used in Morimoto and Hope (2001), though is much
higher. Having seen the serious power shortages in Sri Lanka and its adverse impact
on the country’s economic growth, using these regression results instead of the
original values might improve the result.
Finally, one should note that the above bi-variate relationship between GDP and
electricity production might be an overestimate, as the usual production function
also includes capital and labor. Hence, in future work, capital and labor parameters
could also be included in order to improve the model.

Acknowledgments

We would like to thank Mr S. Fernando and Mr N. Sandasiri from CEB, and the
Staff of the Munasinghe Institute for Development (MIND) and Lanka International
Forum on Environment (LIFE) for their help towards this study, especially for data
collection. Useful comments and suggestions from Prof. H.-Y. Yang (Department of
Economics, Shih-Hsin University, Taiwan) are also greatly acknowledged.

References

Asafu-Adjaye, J., 2000. The relationship between energy consumption, energy prices and economic
growth: time series evidence from Asian developing countries. Energy Econ. 22, 615–625.
Central Bank of Sri Lanka (CBSL), 1998. Annual Report, Central Bank of Sri Lanka (CBSL, Sri
Lanka).
Ceylon Electricity Board (CEB), 1999. Long Term Generation Expansion Plan 1999–2013 (CEB, Sri
Lanka).
Ferguson, R., Wilkinson, W., Hill, R., 2000. Electricity use and economic development. Energy Policy
28, 923–934.
Greene, W.H., 2000. Econometric Analysis. fourth ed.. Prentice Hall international Inc, London.
Huang, J.-P., 1990. Electricity consumption and economic growth: a case study of China. Energy Policy
21, 717–720.
Mashi, A.M.M., Mashi, R., 1996. Energy consumption, real income and temporal causality: results from
a multi-country study based on cointegration and error-correction modeling techniques. Energy Econ.
18, 165–183.
Morimoto, R. and C. Hope, 2001. An extended CBA model of hydro projects in Sri Lanka, Judge
Institute of Management Research Paper 15.
R. Morimoto, C. Hope / Energy Economics 26 (2004) 77–85 85

Pesaran, M.H., Smith, R.P., Akiyama, T., 1998. Energy Demand in Asian Developing Countries. Oxford
University Press, Oxford, UK.
Ramcharran, H., 1990. Electricity consumption and economic growth in Jamaica. Energy Econ. 12,
65–70.
Serra, P., Fierro, G., 1997. Outage costs in Chilean industry. Energy Econ. 19, 417–434.
Yang, H.-Y., 2000. A note on the causal relationship between energy and GDP in Taiwan. Energy Econ.
22, 309–317.