F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 187
Ž .
Fig. 1. Decomposition results OECD-ROW given by the RLM and the LDM. Plot 1 refers to RLM with purchasing power GDP, plot 2 refers to LDM with purchasing power GDP, plot 3 referes to RLM
with exchange-rate-converted GDP, and plot 4 refers to LDM with exchange-rate-converted GDP.
given by the RLM are larger than the corresponding estimates given by the LDM, irrespective of how GDP is measured. It seems that the LDM, which contains
logarithmic terms in its formulae, gives more stable decomposition results. In contrast, the RLM tends to introduce greater ‘overlaps’ among effects such that
the estimated effects are larger in absolute terms and the degree of cancellation among effects is greater when they are added up to give the actual total change.
This is illustrated in a numerical example in the Appendix A, which shows that, when the amplitude of variations in explanatory variables increases, the RLM
yields less stable decomposition results as compared to the LDM. It may, therefore, be suggested that the results given by the LDM are more robust than those given
by the RDM.
5. Impacts of the choice of GDP measure
Many studies have investigated the problems of using exchange-rate-converted GDP to compare the level of economic activities across countries and in energy-
Ž .
GDP correlation analysis e.g. David, 1972; Ang, 1987; Birol and Okogu 1997 . It is a well-known fact that exchange-rate-converted GDP tends to exaggerate the
income differences between the developing and industrial countries. Gross domes- tic product with adjustment for purchasing power parity, although still subject to
error, is closer to the true product output level, particularly for the developing countries. Our study provides some interesting results related to the impacts of
F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 188
Ž .
Fig. 2. The estimated intensity effects ⌬C given by the LDM using different GDP measures.
int
switching from exchange-rate-converted GDP to purchasing power GDP in decom- position analysis.
According to the decomposition formulae, the way GDP is measured affects all Ž
the effects in the RLM but it only affects two GDP-related effects i.e. intensity .
and income effects in the LDM. Fig. 2 compares the estimates of ⌬C given by
int
the LDM using the two different GDP measures. For ROW-FSUrCEE, the estimates are about the same since, as already mentioned, the relative size of their
GDP is quite independent of GDP measure. However, in the case of OECD- FSUrCEE, the absolute value of ⌬C
computed from purchasing power GDP is
int
more than twice that given by exchange-rate-converted GDP. For OECD-ROW, the two measures yield estimates which are opposite in sign. Thus very different
conclusions may be reached based on different GDP measures.
6. Structural comparability
The study of the effect of structural change, in particular the structure of production, has been one of the objectives of decomposition analysis. In fact it is
the main objective in many decomposition studies on industrial energy demand that appeared in the 1980s. More specifically, the objective then is to study how
changes in the product mix of industry affect trends in industrial energy demand. Structural comparability is seldom an issue in chronological decomposition for a
specific country as the data needed have generally been collected and presented in a standard format. However, the problem of incompatibility is often encountered in
F.Q. Zhang, B.W. Ang r Energy Economics 23 2001 179᎐190 189
cross-country studies as there are often variations among countries in data collec- tion and presentation. Hence, adjusting the original data to make them compatible
across country is a complication in cross-country decomposition studies. In the Ž
. study of CO
emissions by Proops et al. 1993 , the input᎐output tables of
2
Germany and the UK were both aggregated to give consistent production sectors. Ž
. Similarly, in the study by Chung 1998 , the data for South Korea were modified to
make them compatible with those for China and Japan. Generally, as the level of disaggregation increases the need for data adjustments becomes greater. The study
by Proops et al. involved 47 economic sectors and that by Chung 45 industrial sectors. In our study, this problem did not arise as structural change involves fuel
mix and only six fuel types are studied.
7. Conclusion