The World’s Largest Open Access Agricultural & Applied Economics Digital Library

This document is discoverable and free to researchers across the

globe due to the work of AgEcon Search.

Help ensure our sustainability.

Give to AgEcon Search

AgEcon Search
[Link]
aesearch@[Link]

Papers downloaded from AgEcon Search may be used for non-commercial purposes and personal study only.
No other use, including posting to another Internet site, is permitted without permission from the copyright
owner (not AgEcon Search), or as allowed under the provisions of Fair Use, U.S. Copyright Act, Title 17 U.S.C.
 The structure of consumer energy demand in Australia: an
application of a dynamic almost ideal demand system

Muhammad Akmal*

46th Annual Conference of the

Australian Agricultural and Resource Economics Society
Canberra, 12-15 February 2002

By parameterising the Almost Ideal (AI) demand system as a vector error correction model
(VECM), this paper estimates the structure of consumer energy demand in Australia. To this
end, domestic per person expenditure on energy is divided into the expenditure on electricity,
gas and a miscellaneous category, residual fuels. To close the system, non-energy household
consumption expenditure is introduced, resulting in a four-equation share system, which is
estimated using national-level quarterly data. The demand for electricity and gas is price and
income inelastic whereas that of other fuels is highly price elastic. Significant substitution
possibilities are found between electricity and other fuels and between gas and other fuels.
However, electricity and gas – which, together, account for more than 90 per cent of
household fuel expenditure – are estimated to be complementary fuels.

The paper is based on PhD work at the ANU. Comments from Prof. Ron Duncan, David Stern and
Trevor Breusch, all from ANU, are gratefully acknowledged.
*Economic consultant, ABARE, GPO Box 1563 Canberra, ACT 2601.
I Introduction
Despite having significant policy implications for issues ranging from competition
policy to environmental management, residential energy demand and, more precisely,
the estimation of demand elasticities for the various energy sources has not attracted
much attention in Australia. Not only is the literature on the subject very limited but
also electricity consumption has been the focus of attention.1 Hawkins (1975), for
instance, employed single equation methods to estimate the demand for electricity in
the Australian Capital Territory (ACT) and New South Wales (NSW). Donnelly
(1984) and Donnelly and Diesendorf (1985) also estimated an electricity demand
function for the ACT using single equation procedures. A number of other studies,
which belong to this class of specification and estimation, have modelled electricity
demand but only as a part of an aggregate measure of electricity (see, for instance,
Donnelly and Saddler 1984; Stromback 1986).

Rushdi (1986), on the other hand, has modelled the interrelated demand for
electricity, natural gas and heating oils in South Australia using a translog demand
system. However, to the best of this author’s knowledge, no study, at least in the
recent past, has made an attempt to determine inter-fuel substitution possibilities at
the national level using a systems approach. The objective of this paper is to model
the structure of consumer energy demand in Australia using national-level data with a
view to estimating substitution possibilities between electricity, gas and other fuels.

The Almost Ideal (AI) demand system, which is developed by Deaton and
Muellbauer (1980) and is probably the most extensively used system among the
family of flexible consumer demand systems, is chosen to represent the underlying
equilibrium structure of interrelated consumer demand. Domestic per person
expenditure on energy, in this regard, is divided into the expenditure on electricity,
gas and a miscellaneous category, residual fuels. In order to close the system, non-
energy household consumption expenditure is added as another commodity. The
resulting four-equation share system is implemented using national-level quarterly
data.

1
Considerable attention has, however, been paid to study the demand for energy at the level of specific
end-uses, such as cooking, cooling, space heating, and water heating (Goldschmidt 1988; Bartels and
Fiebig 1990; Fiebig et al. 1991; Bauwens et al. 1994; Bartels et al. 1996a; and Bartels and Fiebig
2000).

2
 In the initial estimation of the equilibrium relationships, as represented by the
static share system, significant autocorrelation was observed, indicating misspecified
dynamics. The model, therefore, was parameterized as a vector error correction model
(VECM), nesting the dynamic specifications of autoregressive error model of Berndt
and Savin (1975) and the generalized stock adjustment model suggested by Nadiri and
Rosen (1969).

The rest of the paper is organized as follows. Section II presents the AI model
specification as the underlying steady-state relationship and derives an appropriate
error correction representation that is consistent with the available data set. A brief
description of the data is given in the next section. Results are reported and discussed
in Section IV. The section also compares the results obtained in this paper with those
of the previous studies for Australia and North America. Finally, Section V
summarises the study and offers some concluding remarks.

II Methodology

Following Deaton and Muellbauer (1980), the underlying consumer preferences are
represented by the linear approximate AI demand system:
n
x
wi   i    ij log p j   i log  (1)
j 1 P

where i  j  4; 1 = electricity; 2 = gas; 3 = other fuels; 4 = non-energy good; wi is

the ith budget share; pi denotes the price of the ith commodity; x is per capita
expenditure on n commodities; log P   i wi log( p i ) and the s and s are the

unknown parameters. The adding-up, homogeneity and symmetry properties require

the following restrictions:
n n n

  i  1,
i 1
  i  0,   ij  0
i 1 i 1
(adding-up) (2)


j 1
ij 0 (homogeneity) (3)

 ij   ji (symmetry)

(4)

3
As the model is applied to a quarterly data set, three intercept dummies are also
included among the group of explanatory variables, treating the October-December
period as the reference quarter. There are 36 unknown parameters in the above
system, including 12 dummy variable coefficients, in the absence of symmetry
conditions. This number, however, reduces to 30 after accounting for the restrictions
implied by the symmetry property (12 = 21, 13 = 31, 14 = 41, 23 = 32, 24 = 42, 34 =
43). The adding-up restrictions further help reduce the number of unknown
parameters in the steady-state system to 21.

In order to facilitate the derivation of a VECM representation, the system in (1) is

written, using matrix notation, as:

W(t )  Z (t ) (5)

where W(t) is a 41 vector of expenditure shares; Z(t) is a k1 matrix of explanatory
variables which includes a unit variable and seasonal dummies and  is a 4k matrix
of the steady-state parameters. For the purposes of this study, the following
specification of the VECM is proposed:
~
 4W(t )  A 4 Z (t )  B(W(nt4 )  n Z (t 4 ) )   (t ) (6)

where 4 is defined as  4 y t  y t  y t  4 (yt is an auxiliary variable) which reflects the

fact that the specification will be applied to quarterly data; A and B are the matrices
which consist of short-run parameters;  implies that the intercept and the seasonal
dummies are excluded from the matrix Z; n reflects that the last element of the W and
 matrices is deleted due to the singular nature of the system; and  (t ) is a matrix of

disturbance terms which are assumed to be independently and identically distributed

normal variables. This specification has an interesting economic meaning. It allows
consumers to adjust their consumption expenditure in response to new information on
the explanatory variables as well as in response to the observed deviation from the
steady-state equilibrium.

It is desirable to test the restrictions implied by the different models which are
nested in this dynamic specification. Three such formulations are tested. The first is
an autoregressive error model, which is written as:

4
 W (t)  Z (t)   (t)
(7)
 (t)    (t  4 )   (t )

where  is 44 matrix of unknown parameters. The second formulation is the partial
adjustment model of the following type:

 4W( t )  C (Z (t )  W(t  4) )   (t ) (8)

Nadiri and Rosen (1969) suggested this formulation and used it to study interrelated
factor demands for US manufacturing. This procedure, which is essentially a
generalisation of Koyck’s single equation adjustment mechanism, permits dis-
equilibrium in one commodity market to affect the demand for other commodities.
Finally, the static model:

W( t )  Z (t )   (t ) (9)

is also considered as a special case of the dynamic formulation in (6). This static
representation assumes instantaneous adjustment and thus the estimated elasticities
are interpreted as the long-run elasticities.

Empirical estimates based on aggregate time-series data quite often reject the
symmetry and homogeneity restrictions. The violation of these fundamental economic
postulates is due, according to Anderson and Blundell (1982, 1983), to the fact that
proper attention is not paid to the dynamic structure of the models. This hypothesis is
considered in this study by testing the symmetry and homogeneity conditions, taking
both the static and the dynamic models as maintained hypotheses.

Demand elasticities

For the linear approximate AI model, the Hicksian own-price (  ii ) and cross-price

elasticities (  ij ) can be computed from:

 ii  1   ii / wi  wi , i  1,2,........,4 (10)

 ij   ij / wi  wi , i, j  1,2,.........,4; i  j (11)

The above elasticity estimates and, more precisely, the sign of  ij will help determine

the nature of the relationship between the different forms of energy. A positive sign

5
implies they are substitutes and a negative sign indicates that they are complements to
each other. The uncompensated own-price elasticities (  ii ) and cross-price elasticities

(  ij ) are obtained from:

 ii  1   ii / wi   i , i  1,2,.........,4 (12)

 ij   ij / wi   i ( w j / wi ), i, j  1,2,..........,4; i  j (13)

A positively (negatively) signed  ij implies, on the other hand, that the two fuels are

gross substitutes (complements). Finally, the expenditure elasticities (  i ) are

estimated from:

 i  1   i / wi , i  1,2,............,4 (14)

It should be noted that the predicted shares are employed in the estimation of the
above elasticities along with the estimates of the ijs and s. Further, because
parameter estimates and predicted cost shares have variances and covariances, the
elasticity estimates have stochastic disturbances as well. Since the elasticities are non-
linear functions of parameter estimates and fitted cost shares, the standard errors
cannot be calculated exactly. The variances of the elasticity estimates are, therefore,
computed from:

V ( ij )  V ( ij ) / wi2
V ( ii )  V ( ii ) / wi2
V ( ii )  V ( ii ) / wi2  V (  i )  2Cov ( ii ,  i ) / wi (15)
V ( ij )  V ( ij ) / wi2  V (  i )( w 2j / wi2 )  2Cov ( ij ,  i )( w j / wi2 )
V ( i )  V (  i ) / wi2

where V stands for variance and Cov indicates covariance. The estimated variances of
the estimated ij and  parameters and fitted cost shares are used while obtaining
estimates of the above variances.

III Data and estimation

The AI/VECM is estimated using national-level quarterly data – seasonally

unadjusted – spanning the period from the third quarter 1969 to the second quarter

6
1998. Total household consumption expenditure, household expenditure on energy,
and population were obtained from various issues of the ‘Australian National
Accounts: National Income, Expenditure and Product’ (ABS Catalogue No. 5206.0)
and the ‘Australian National Accounts: Quarterly State Details’ (ABS Catalogue No.
5206.0.40.001). Both nominal and constant values of expenditure at 1990 prices were
obtained. The break-up of the energy category into expenditure on electricity, gas and
other fuels was also obtained from the Bureau on request, as these data are not
published. The price deflators were constructed by dividing the nominal variables by
the corresponding real ones.

The expenditure shares of the three energy sources along with the total energy
expenditure share are plotted in Figure 1. The total energy expenditure share has
fluctuated significantly around an average (entire-period) share of 2.2 per cent during
the last three decades, primarily due to seasonal factors. The share peaks in the third
quarter, the coldest quarter, because of a significant increase in the consumption of
electricity, gas and other fuels, and also due to the relatively low non-energy
consumption during this period. The share falls sharply during the fourth quarter and
to a lower level during the first quarter, although the downward movement in the
share from the fourth quarter to the first quarter is relatively minor. This systematic
pattern of fluctuations holds for the entire period with a few exceptions.

Figure 1 Expenditure shares, percent

7
 Electricity, which accounts for almost 74 per cent of total energy consumption
expenditure, has more or less the same seasonal pattern. Its average share of total
household expenditure seems to have increased from around 1.5 per cent in the early
1970s to as high as 2 per cent during the late 1980s, primarily at the expense of the
other fuels. A declining trend in this variable is obvious during the 1990s, with the
share of electricity in overall consumption expenditure falling back to the level of the
1970s. The share of other fuels has fallen in a cyclical fashion to almost 0.25 per cent
in 1997 from around 0.5 per cent in 1969, mainly due to a substantial increase in the
real price of this variable, which occurred mostly during the 1980s. Natural gas, on
the other hand, has increased its share considerably during the last two decades.2

The average price level for the household sector increased by a factor of seven
during this period of almost three decades (Figure 2). The nominal prices of
electricity and gas increased by a factor of less than six. The real prices of electricity
and natural gas, as a consequence, declined by 16 per cent and 30 per cent,
respectively. The relative price of the residual energy category, on the other hand,
almost doubled as the nominal price of this fuel increased by a factor of roughly 14
during this period due primarily to the rising petroleum product prices.

Figure 2 Price indices, fuel and average consumer price

2
The expenditure on natural gas as a per cent of total energy expenditure rose from a little less than 14
per cent in the early 1980s to around 18 per cent in 1997.

8
 From this study’s point of view, however, it is of more interest to compare the
price path of one energy category with the others because a significant relative price
change is crucial to being able to make robust estimates of the substitution
possibilities between the different energy categories. Electricity and gas prices grew
at roughly the same rate up until the late 1970s. The electricity price index, however,
rose relative to that of natural gas at the beginning of the next decade. The gap
between the two indices has subsequently diminished owing to a slow down in
electricity price inflation during the last eight years or so. The price of the residual
fuels has not only fluctuated substantially but has also increased very significantly
relative to the other energy prices.

Most of the price increases in energy, and in the household expenditure items more
generally, occurred between 1978 and 1991, triggered by the second oil price shock.
Almost 84 per cent of the other fuels price rise, for instance, occurred during this
period. The price index of the non-energy category is not graphed because it is almost
perfectly described by the consumer price index, due to the overwhelming proportion
of the non-energy expenditure in total household consumption expenditure.

For the purposes of estimating the model, the non-energy share equation is
arbitrarily dropped and the remaining three equations are estimated simultaneously in
SHAZAM using the non-linear iterative seemingly unrelated regression procedure.
The estimates of parameters, log-likelihood values, and standard deviations are
invariant to the choice of which three equations are directly estimated.3 All
parameters of the non-energy equation are recovered with the help of demand system
restrictions.

IV Results

Before moving on to the main body of results, the results from cointegration and unit
root analysis should be presented. The unit root results for the four residuals, which
are obtained by estimating the static AI model, are given in the lower part of Table 1.
The statistics show that the null hypothesis of a unit root is consistently rejected even
after accounting for the fact that the co-integrating vector is unknown. The table also

3
For a proof, see Kmenta and Gilbert (1968) and Dhrymes (1973).

9
contains the unit root statistics performed on the main variables of the model. As
expected, the expenditure shares are all stationary because the shares are bounded.
Total household expenditure is also stationary in the sense that it has no stochastic
trend. The four price variables, on the other hand, are I(1) as they become stationary
after differencing once.

Table 1 Unit root analysis using the Phillips-Perron procedure

Variables Level First-differenced

w1 -9.654 -19.001

w2 -10.178 -11.253

w3 -10.298 -10.852

w4 -10.810 -13.548

log(p1) -1.153 -13.453

log(p2) -0.729 -10.260
log(p3) 0.675 -8.259
log(p4) 2.767 -5.207
log(x/P) -11.872 -20.209

Res1 -5.340 ..
Res2 -8.120 ..
Res3 -7.400 ..
Res4 -7.270 ..

Notes: [Link] unit root analysis takes into consideration the quarterly nature of the data by
incorporating quarterly dummy variables. 2. The 5 per cent critical t-test value for the residuals is 4.71
and the corresponding 1 per cent critical value for the usual variables is 3.96.
Source: Author’s estimations.

The results pertaining to the restrictions implied by homogeneity, autoregressive

error, partial adjustment and the static models are reported in Table 2. The maximised
value of the (log) likelihood function in the absence of any restrictions is 2247.89.
The symmetry restrictions when imposed reduce this value to 2243.93. Clearly, the
symmetry conditions are not rejected even at the 10 per cent level of significance. It is
interesting to note that these conditions are rejected at the 1 per cent level when the
static model is taken as the unrestricted model. The dynamic model with symmetry

10
imposed, therefore, is taken as the maintained model. The underlying expenditure
function associated with this symmetry imposed dynamic model, however, frequently
violates the curvature restrictions. Fortunately, it is strictly quasi-concave at the
sample means, as the eigenvalues associated with the Slutsky matrix are all negative.
The elasticities reported below are evaluated at the sample means.

Table 2 Tests of various models

Numbe Model log L Test LR test DF CV

r value 1%
1 Dynamic (No Symmetry) 2247.89 .. .. .. ..
2 Dynamic (Symmetric) 2243.93 2v1 7.92 6 16.8
3 Autoregressive error 2211.30 3v2 65.25 15 30.6
4 Partial Adjustment 2201.89 4v2 84.07 15 30.6
5 Static 2105.48 5v2 276.91 24 43.0

Notes: 1. v stands for versus. 2. LR stands for likelihood ratio. 3. DF denotes degrees of freedom and
CV 1 per cent means critical value at the 1 per cent level of significance.
Source: Author’s estimations.

Both the autoregressive error model and the partial adjustment model impose 15
restrictions on the parameters of the symmetric dynamic model. Clearly, the
restrictions are not the same as is evident from the different log L values. These
restrictions are rejected with overwhelming support from the data. The static model,
which imposes 24 restrictions on the structure of the maintained model, is also
rejected very decisively.

The regression results reported in the first two columns of Appendix Table 1
correspond to the dynamic model, which incorporates the restrictions of homogeneity
and symmetry. The results from the static AI model are also reported in the last two
columns for the sake of comparison. The short-run parameters are omitted due mainly
to space limitations and also because individual (short-run) parameters lack economic
interpretation of any significance. Most of the steady-state parameters are estimated to
be quite significant. Out of the six insignificantly estimated steady-state parameters,
two are actually intercept terms. Two income coefficients, 2 and 3, are also
insignificantly estimated. The sign of these coefficients, however, is not changed
under the dynamic specification relative to the static one where these parameters are
estimated very precisely.

11
 The coefficient of the income/expenditure variable is negative in the electricity
share equation and positive in the corresponding equation for the non-energy
expenditure, implying that electricity is a necessity and the composite good a luxury.
It would, however, be a too strong a conclusion to say that gas is a necessity and the
other fuels a luxury, as the respective coefficients are not significant, although they
are quite significant in the static model.

A significant upward shift in the shares of the three fuels during the third quarter
relative to the fourth one (the base quarter in this model specification) is obvious, as
the respective coefficients associated with the third quarter dummy are positive and
highly significant. The second quarter dummy also picks up an upward shift in the
energy expenditure shares. The degree of shift, however, is relatively minor due to the
fact that the second quarter is a warmer period. The summer factor, which is captured
by the first quarter dummy, seems to have no significant impact on the shares of
electricity and the other fuels as the respective dummy coefficients are not significant.

The top panel of Table 3 reports the Hicksian price elasticities along with the t-
scores. The diagonal elements in these four columns are the own-price elasticities and
the off diagonal ones are the cross-price elasticities. Out of a total of 16 elasticity
estimates, three are not significant at the 10 per cent level. The cross-price elasticities
between the different energy categories on the one hand and the composite good, non-
energy consumption, on the other, are all positive and mostly significant, implying
that energy and non-energy consumption are substitutes. The two consumption
categories may be gross complements as the corresponding Marshallian elasticities
reported in the lower part of the table are all negative, although mostly insignificant.

The (Marshallian) demand for the composite non-energy good is almost unit elastic
with respect to both income and own-price, indicating the dominance of this
commodity in the demand system. Electricity demand is price and income inelastic,
which is consistent with the existing Australian literature on electricity demand
estimation (Appendix Table A2). The corresponding two gas elasticities are fairly
similar in terms of magnitudes; however, the price elasticity is not significant. The
demand for the residual fuels, which are dominated by wood and heating oils, is
highly price elastic but the income elasticity of the energy source, though close to
unity, is not statistically different from zero.

12
Table 3 Demand elasticities at the means, quarterly data

Hicksian elasticities

Quantity Price of Expenditure

Electricity Gas Other fuels Non-energy
Electricity -0.6321* -0.1789* 0.1727* 0.6383* na
(6.20) (3.28) (4.05) (6.78) na

Gas -0.8717* -0.5846 0.7092* 0.7472*** na

(3.28) (1.52) (2.96) (1.83) na

Other fuels 1.2013* 1.0122* -2.3087* 0.0952 na

(4.05) (2.96) (4.70) (0.26) na

Non-energy 0.0106* 0.0026*** 0.0002 -0.0134* na

(6.78) (1.83) (0.26) (6.07) na

Marshallian elasticities
Electricity -0.6450* -0.1816* 0.1709* -0.1365 0.7922*
(6.35) (3.33) (4.00) (1.08) (14.00)

Gas -0.8854* -0.5874 0.7072* -0.0719 0.8375**

(3.34) (1.53) (2.94) (0.13) (2.18)

Other fuels 1.1848* 1.0088* -2.3110* -0.8924 1.0098

(3.98) (2.96) (4.69) (0.97) (1.25)

Non-energy -0.0057* -0.0008 -0.0021** -0.9953* 1.0040*

good (3.66) (0.58) (2.40) (350.32) (712.25)

Note: *-Significant at the 1 per cent level, ** significant at the 5 per cent level, *** significant at the
10 per cent level.
Source: Author’s estimations.

The cross-price elasticities – both compensated and uncompensated – between

electricity and the residual fuels and between gas and the residual fuels are positive
and highly significant, indicating that the fuels are substitutes. The cross-price
elasticities between electricity and gas – the two main fuels, which dominate the
domestic fuel expenditure with a share of more than 90 percent – are negative and are
significant at the 1 per cent level. This finding of complementarity is apparently
unexpected, as the two sources of energy appear to be good substitutes in the areas of
cooking and space and water heating. The single most important determinant of the

13
sign of the above-mentioned elasticities is the sign of the 12 parameter, which is
negative in this case and has remained negative in the face of different experiments.4

Appendix Table A2 presents, along with the present estimates, the elasticity
estimates from previous studies of Australian residential energy demand as well as the
estimates from two North American studies by Dumagan and Mount (1993) and Ryan
et al. (1996). Of the previous Australian studies, only Rushdi's study (Rushdi, 1986) is
comparable in scope as the other studies deal only with electricity demand. However,
as stated in the introduction, this study is national while the other studies deal with the
Australian Capital Territory (temperate climate similar to upland regions of some
northern Mediteranean countries or Central France), Tasmania (temperate climate
similar to South-West England or North-West France), South Australia
(Mediterranean climate) and New South Wales (mostly sub-tropical summer-rainfall-
maximum climate). The North American studies cover temperate regions with cold
winters in New York and Ontario, quite unlike any climates found in Australia. Only
the Dumagan and Mount study estimates income elasticities for all fuels, though
several of the Australian studies estimate income elasticities for electricity.

With a value of -0.65, the own-price elasticity of electricity estimated in this paper
is fairly similar to the previous estimates of the parameter for Australia, which vary
between -0.56 and -0.86. Similarly, this paper, like previous studies for Australia,
finds electricity and other fuels substitutes, with broadly similar magnitudes.
Likewise, the income elasticity of the fuel source estimated in this paper is in broad
agreement with previous estimates for Australia as all studies, with the exception of
Donnelly and Saddler (1984), find electricity a necessary fuel.

However, these similarities disappear almost completely for the remaining

parameters, partly because there is only set of existing estimates – Rushdi’s – for
Australia that are presented in the table. The present paper, for instance, finds strong
complementarity between electricity and gas, whereas Rushdi finds significant
substitution possibilities between the two energy sources for South Australia.
Similarly, gas demand, according to Rushdi’s estimates, is highly price elastic,
whereas in the present study the elasticity is insignificantly estimated.

4
The sign of this coefficient, for example, remained negative when the separability assumption was
invoked to reduce the number of parameters.

14
 As far the price elasticities are concerned, there are substantial differences
between the results of this study and the North American estimates. The reported
elasticities for North America are mostly very small (in absolute terms), especially for
New York with many close to or equal to zero. The income elasticities for the New
York study, however, are fairly similar to the present estimates, especially in the case
of electricity and gas.

V Summary and Concluding Remarks

Energy demand for electricity, gas and residual fuels in Australia was modelled
and estimated by parameterising the Almost Ideal demand system as a vector error
correction model. Domestic per person expenditure on energy was divided into the
expenditure on electricity, gas and a miscellaneous category, residual fuels. In order
to close the system, non-energy household consumption expenditure was added as
another commodity, resulting in a four-equation share system. The model was
estimated using a national-level quarterly data set spanning the period from the third
quarter 1969 to the second quarter 1998.

The underlying expenditure function frequently violates the curvature properties,

although it is strictly quasi-concave at the sample means where the reported
elasticities are evaluated.

The demand for electricity is price inelastic and that of other fuels highly price
elastic. In contrast, the corresponding elasticity for gas is insignificantly estimated. As
far the estimated income responses are concerned, electricity and gas are necessary
fuels, whereas the demand for other fuels is unit (income) elastic. The econometric
analysis found significant substitution possibilities, both net and gross, between
electricity and residual fuels and between gas and residual fuels. However, contrary to
expectations, the regression results suggest that Australian households consume
electricity and gas in a complementary fashion.

There could be two explanations of the complementary relationship between the

fuels. The first is a supply-side explanation. Traditionally, electricity has been the
main source of energy for the residential sector, followed by the residual category.
The development of natural gas fields during the late 1960s provided a new source of

15
energy. Although the gas transmission and reticulation system has expanded
significantly over time, a substantial fraction of homes is still not connected to the
grids. The consumers without a gas connection are expected to differ in their response
to, say, a relative fuel price change from those consumers with connections to gas
supplies. The distortion created by the absence of this factor from the demand analysis
might have resulted in the complementary relationship between electricity and gas.
An attempt was made to account for this factor by introducing the fraction of
households connected to gas reticulation systems as another explanatory variable. It
was, however, dropped later due to its insignificance.

The other explanation is a demand-side one. During the past three decades the
price of gas relative to that of electricity has decreased only moderately, not attracting
a significant fraction of households to gas consumption. Relatively slow expansion of
the gas transmission and reticulation system and a lack of enthusiasm on the part of
households caused by a relatively stable electricity-gas price ratio is expected to have
resulted in this unusual finding.

However, before drawing any conclusions regarding the nature of the relationship
between the two main fuels, it seems appropriate to test the robustness of elasticity
estimates using, for instance, different functional specifications. Also, the problem of
the gas supply constraint can potentially be controlled more effectively by pooling a
cross-section of different states. Access to reticulated gas is high in some Australian
states such as Victoria and South Australia, and low in others including Queensland
and Tasmania. By pooling the state-level data and introducing state-dummies, one
should expect to obtain theoretically correct signs of the two elasticities. Further, the
introduction of a cross-sectional dimension brings an additional source of price
variation and probably more variation in the electricity/gas price ratio, and thus a
better probability of obtaining theoretically correct signs of the two cross elasticities.

16
 References

Anderson, G.J. & Blundell, R.W., 1982. 'Estimation and hypothesis testing in
dynamic singular equation systems', Econometrica, 50 (6):1559-71.

Anderson, G.J. & Blundell, R.W., 1983. 'Testing restrictions in a flexible dynamic
demand system: an application to consumers' expenditure in Canada', Review
of Economic Studies, 50 (3):397-410.

Bartels, R. & Fiebig, D.G., 1990. 'Integrating direct metering and conditional demand
analysis for estimating end-use loads', Energy Journal, 11 (4):79-97.

Bartels, R. & Fiebig, D.G., 2000. 'Residential end-use electricity demand: results from
a designed experiment', Energy Journal, 21 (2):51-81.

Bartels, R., Fiebig, D.G. & Nahm, D., 1996a. 'Regional end-use gas demand in
Australia', Economic Record, 72 (219):319-31.

Bartels, R., Fiebig, D.G. & Plumb, M.H., 1996b. 'Gas or electricity, which is
cheaper?: an econometric approach with application to Australian expenditure
data', Energy Journal, 17 (4):33-58.

Bauwens, L., Fiebig, D.G. & Steel, M.F.J., 1994. 'Estimating end-use demand: a
bayesian approach', Journal of Business and Economic Statistics, 12 (2):221-
31.

Berndt, E.R. & Savin, N.E., 1975. 'Estimation and hypothesis testing in singular
equation systems with autoregressive disturbances', Econometrica, 43 (5-
6):937-57.

Deaton, A.S. & Muellbauer, J., 1980. 'An almost ideal demand system', American
Economic Review, 70 (3):312-26.

Dhrymes, P.J., 1973. 'Small sample and asymptotic relations between maximum
likelihood and three stage least squares estimators', Econometrica, 41 (2):357-
64.

Donnelly, W.A. & Diesendorf, M., 1985. 'Variable elasticity models for electricity
demand', Energy Economics, 7 (3):159-62.

17
Donnelly, W.A. & Saddler, H.D.W., 1984. 'The retail demand for electricity in
Tasmania', Australian Economic Papers, 23 (42):52-60.

Donnelly, W.A., 1984. 'Residential electricity demand modeling in the Australian

capital territory: preliminary results', Energy Journal, 5 (2):119-31.

Dumagan, J.C. and Mount, T.D., 1993. ‘Welfare effects of improving end-use
efficiency: theory and application to residential electricity demand’, Resource
and Energy Economics; 15(2): 175-201.

Fiebig, D.G., Bartels, R. & Aigner, D.J., 1991. 'A random coefficient approach to the
estimation of residential end-use load profiles', Journal of Econometrics, 50
(3):297-327.

Goldschmidt, P., 1988. 'Domestic appliance energy usage in Western Australia',

Energy Economics, 10 (2):155-62.

Hawkins, R.G., 1975. 'The demand for electricity: a cross-section study of New South
Wales and the Australian Capital Territory', Economic Record, 51 (133):1-18.

Kmenta, J. & Gilbert, R.F., 1968. 'Small sample properties of alternative estimators of
seemingly unrelated regressions', Journal of American Statistical Association,
63 :1180-200.

Nadiri, M.I. & Rosen, S., 1969. 'Interrelated factor demand functions', American
Economic Review, 59 (4):457-71.

Rushdi, A.A., 1984. 'Industrial demand for electricity in South Australia', Australian
Economic Papers, 23 (43):259-80.

Rushdi, A.A., 1986. 'Inter-fuel substitution in the residential sector of South

Australia', Energy Economics, 8 (3):177-85.

Ryan, D.L., Wang, Y and Plourde, A., 1996. ‘Asymmetric price responses of
residential energy demand in Ontario’, Canadian-Journal-of-Economics,
29(0): S317-23 (Special issue).

Stromback, C.T., 1986. 'Modelling electricity demand in Western Australia',

Australian Economic Papers, 25 (46):106-17.

18
Appendix Table A1 Regression results

Variables Dynamic model Static model

Value T-score Value T-score
1 0.0445* (5.40) 0.0385* (11.59)
2 0.0069 (0.61) -0.0084* (4.12)
3 0.0046 (0.28) 0.0181* (6.05)
4 0.9440* (76.98) 0.9517* (245.30)
11 0.0057* (3.70) 0.0062* (7.47)
12 -0.0030* (3.25) -0.0009*** (1.87)
13 0.0028* (3.58) 0.0031* (10.21)
14 -0.0055* (3.70) -0.0083* (9.94)
22 0.0014 (1.04) 0.0035* (4.79)
23 0.0024** (2.51) 0.0006** (2.40)
24 -0.0008 (0.55) -0.0032* (4.82)
33 -0.0031** (2.28) -0.0006** (2.46)
34 -0.0021** (2.15) -0.0030* (8.63)
44 0.0084* (4.07) 0.0145* (13.95)
1 -0.0034* (3.56) -0.0026* (6.78)
2 -0.0005 (0.43) 0.0013* (5.63)
3 0.0000 (0.01) -0.0017* (4.93)
4 0.0039* (2.82) 0.0030* (6.67)
11 0.0001 (0.37) 0.0001 (0.70)
12 -0.0008** (2.03) -0.0005* (5.27)
13 -0.0001 (0.09) -0.0004** (2.53)
14 0.0008*** (1.83) 0.0008* (3.75)
21 0.0004 (1.36) 0.0006* (3.28)
22 0.0007 (1.61) 0.0006* (5.27)
23 0.0010 (1.50) 0.0015* (10.74)
24 -0.0021* (4.52) -0.0026* (12.74)
31 0.0039* (12.84) 0.0041* (22.69)
32 0.0027* (5.52) 0.0020* (18.49)
33 0.0012 (1.58) 0.0020* (13.73)
34 -0.0077* (15.85) -0.0081* (39.55)
Log L 2243.9280 2105.4760

Note: *-Significant at the 1 per cent level, ** significant at the 5 per cent level, *** significant at the 10
per cent level.
Source: Author’s estimates.

19
Appendix Table A2 Residential energy demand elasticities, a comparison

Study Country Functional Region Period Marshallian Elasticities

specification covered  11  12  13 1y  21  22  23 2y  31  32  33 3y
Hawkins Australia single equation NSW, cross-section, -0.55 - - 0.93 - - - - - - - -
(1975) (linear) ACT 1971

Donnelly Australia single-equation ACT 1964-82 -0.77 - 0.42 0.69 - - - - - - - -

(1984) (log-linear
dynamic)
Donnelly Australia single-equation ACT 1964-82 -0.86 - 0.46 0.32 - - - - - - - -
(1984) (linear,
dynamic)
Donnelly Australia log-linear TAS 1961-80 -0.56 - 0.31 1.13 - - - - - - - -
and Saddler (static)
(1984)
Donnelly Australia several single- ACT 1964-82 -0.76 to - - - - - - - - - - -
and equation -0.81
Diesendorf
(1985)
Rushdi Australia static translog SA 1960-82 -0.69 0.49 0.20 - 1.68 -1.53 -0.15 - 1.34 -0.29 -1.06 -
(1986)
Dumagan USA dynamic logit New 1960-87 -0.07 0.02 0.00 0.72 0.02 -0.23 0.05 0.78 0.00 0.06 -0.66 0.86
and Mount system York
(1993)
Ryan et al. Canada translog Ontario 1962-89 -0.23 0.14 0.04 - 0.19 -0.25 0.10 - 0.10 0.20 -0.47 -
(1996)

This Study Australia Dynamic AI National 1970-98 -0.65 -0.18 0.17 0.79 -0.89 -0.59 0.71 0.84 1.18 1.01 -2.31 1.01
model

Notes: 1. ACT= Australian Capital Territory, NSW = New South Wales, SA = South Australia. 2. The Ryan and Wang elasticities correspond to 1984, the rest are evaluated at the respective
sample means. 3.  ij = elasticitiy of the ith fuel source with respect to the price of the jth fuel, where i,j= 1, 2, 3 (1=electricity, 2=gas, 3= other fuels). 4.  iy = income elasticity of the ith fuel.