Energy Elasticities And Market Saturation

Power Efficiency Guide

Ultimate Guide to Power Efficiency

Get Instant Access

Much as country-specific structural characteristics should be recognized in the estimation of energy elasticities, temporal characteristics related to a given economy's position on the development curve must be recognized as well. For instance, energy use in many of the most energy-intensive sectors may eventually be constrained by market saturation, with respect either to energy-related capital goods or to the energy service itself. Thus, caution must be exercised when applying elasticities estimated on the basis of the existing economic structure to make forecasts stretching into the distant future when the economy may have matured considerably.

For instance, price elasticities may be very different when the market for various pieces of energy-using capital equipment in the domestic sector (e.g. ownership of central heating, electric appliances or vehicles) is saturated. Thus, the response of a household to a change in fuel prices would be reflected, even in the long term, more in adjustments of the variable input (the fuel) and less in changes in the fixed input (the boiler or vehicle) which would be determined by replacement demand. Similarly income elasticities may be very different once the saturation point is reached. Thus, in terms of demand for energy services themselves, increases in real income may have little effect on fuel use in domestic heating once a satisfactory level of physical comfort has been realized. Analogously, with respect to demand for energy-

using capital, changes in real income may have little effect on car ownership, and thus to some extent fuel use, once the market is completely saturated. Indeed, once saturation is complete, energy elasticities in many sectors will probably be asymmetric with little potential for increased consumption but significant potential for downward adjustment, indicating relatively higher downward income elasticity and upward price elasticity. These questions are discussed explicitly in Hodgson and Miller, but are of considerable relevance to the chapters by Sterner and Franzen as well as Grubb.

It should be recognized, moreover, that in a number of important sectors the saturation rate is not a fixed parameter but is instead contextual. For example, households with a saturated demand for domestic heat may find that their demand becomes relatively more price inelastic because warmth becomes accepted as a necessity. Another example is in the transport sector where a number of models indicate that it is average commuting times and not ownership rates per se which represent the fundamental determinant of saturation rates for travel to work, although these averages and the distances travelled have been getting longer. This underlines the importance of distinguishing between the energy service (mobility) and the means by which it is realized (fuel and capital). It would also indicate that public investment in transport infrastructure can be better understood as a significant determinant of transport mode splits (road, rail, air, water) and not purely as a reaction to consumer demand for different modes. Given the different energy intensities of different modes this would indicate that long-term transport-related energy elasticities depend, implicitly or explicitly, upon government policy.



National attempts to address the problem of global warming have significant international repercussions, in both economic and environmental terms. As such it is vitally important to determine the sources of consequent reductions in energy consumption in the economy. Decomposing changes in energy use into those associated with changes in the level of output (activity effects), those associated with a changing sectoral composition of output (sectoral effects) and those associated with increased energy efficiency (intensity effects), it is possible to distinguish between the sources of changes in energy consumption (Schipper and Meyers 1992). Thus, it is quite possible to have decreased energy consumption in a growing economy with no increase in energy efficiency. This would arise if there was a fundamental transformation away from more energy-intensive sectors as the economy matured (Solow 1987). The input-output analysis of Proops et al. (1993) indicates that this has been the case for the UK between 1971 and 1982.

The environmental and economic significance of such a development path depends upon the tradeability of the affected sectors. Thus, it is important to distinguish between changing sectoral composition of output which reflects a demand response and that which reflects a supply response. Given the pure common property nature of the environmental consequences of CO2 emissions such a distinction is of considerable significance. For instance, policies which reduce domestic energy consumption in the non-tradeable transport sector would certainly have benign environmental consequences; policies which result in decreased domestic energy consumption in highly tradeable manufactures may conceivably have malign environmental consequences, depending on international trade flows and production technology elsewhere.

This emphasizes not only the importance of international policy co-ordination, but also the importance of the careful interpretation of the true causes which lie behind changed energy consumption in national economies. For this reason, the causes behind changes in energy consumption in the fastest-growing economies may be of particular significance. In this light, the chapter by Pesaran and Smith on energy elasticities in East Asian economies and many of the results cited in the survey by Manning and Atkinson are of considerable interest.



All the chapters in the book are concerned ultimately with the modelling of CO2 abatement policies. Chapters 2-4 are concerned with estimating long-term price and income elasticities of the demand for energy and substitution elasticities for the demand for different fuels to meet the energy demand. Chapter 5 is a detailed treatment of the response of UK manufacturing to particular energy price shocks showing how the changes in boiler stock may affect the response. Chapters 6 and 7 show how the elasticity approach is incorporated into national and international models designed to assess policy responses as well as to develop energy scenarios. Chapters 8 and 9 discuss the specification and construction of models incorporating energy demand equations which provide estimates of energy price elasticities for nine OECD countries and for the UK. The models are designed to assess GG abatement policy and Chapter 9 in particular looks at the effects of different estimated price elasticities in relation to the EU's proposed carbon/ energy tax. Chapters 10 and 11 go beyond the price elasticities to consider technological aspects and the potential for energy saving. Chapter 12 reviews the published estimates of the costs of GG abatement, suggesting why in many cases these may have been overestimated. Chapter 13 is a pointer to the way modellers are dealing with the effects of technological change in economic models; it expounds the idea that price responses might be asymmetrical as a result of induced technical change. This direction of research is taken up in the conclusions in Chapter 14: modelling of long-term GG abatement is seen as requiring both an economic and an engineering component if the full range of policy options is to be addressed. Models should then be in a better position to explore policies which could yield net economic benefits if there is a move towards a low-carbon economy.


1 The Council of Economic Advisors' Report to the President in 1990 was influential in promoting the idea that the costs of abatement were unimaginably large. The Report quoted estimates of costs to the US economy of $3.6 trillion and suggested that US growth could be cut in half. However, the Report used US experience following the oil shocks as an illustration of the effects of abatement policies and relied on studies of the effects, e.g. Jorgenson and Hogan (1990), which also depended heavily on data dominated by the 1973 and 1979 oil price shocks. Moreover, the $3.6 trillion estimate does not seem quite so large if reinterpreted as a reduction in the US growth rate 1990-2010 from 2.52 per cent per annum to 2.46 per cent (Barker 1991).

Part I

Estimating long-term energy elasticities

Chapter 2

Alternative approaches to estimating long-run energy demand elasticities An application to Asian developing countries

M.Hashem Pesaran and Ron Smith


Long-run energy elasticities have been estimated using time series for individual countries, cross-sections over countries and pooled data for panels of countries. This chapter reviews what is known about the properties of these estimators, emphasizing the potentially misleading results that can be obtained by pooled estimators in dynamic heterogeneous panels. The issues are illustrated by an application to aggregate energy demand in ten Asian countries over the period 1973-90. Although these results can only be tentative because of the problems caused by aggregation they illustrate the methodological problems raised by the theoretical discussion.1


Long-run energy elasticities have been estimated from dynamic time-series models for individual countries, from cross-section models across countries and from pooled models on panels of data: time series for a number of countries or regions, such as US states. The estimates obtained from the various methods have tended to differ substantially. Bohi (1981) reviews and highlights the differences in the estimates of the energy demand elasticities in the early literature. A more recent review can be found in Hawdon (1992). The purpose of this paper is to review the properties of the various estimators, cross-section, time-series, pooled, used to estimate elasticities and illustrate the issues with an application to energy demand in ten Asian developing countries.

The differences between the estimates produced by the various procedures is not a new problem, and it is often argued that cross-sections provide better estimates of long-run effects than time series do. For instance, Baltagi and Griffin (1984) argue that time-series estimates of energy demand elasticities are misspecified by omitting the long distributed lags typical of energy demand, e.g. because of the time it takes consumers to adjust their capital stockā€”to replace cars or heating appliances etc. Thus time series will underestimate the total effect of the price change, as it works slowly through the system. They show through Monte Carlo simulations that for equations typical of estimated energy demand, where long distributed lags on price are important, the time-series estimates are biased downwards but cross-section estimates produce more sensible long-run estimates. They also give references to the large literature trying to interpret the relationship between cross-section and time-series estimates.

The issue is important, because it is now quite common to have panels in which both N (the number of groups) and T (the number of time periods) are quite large. With such panels it is possible to compare the properties of the various procedures employed to estimate long-run effects and that is the objective of this chapter. Quah (1990) calls such panels fields, to distinguish them from the small T panels typical of microeconometrics. A comprehensive survey of the literature on panel data is provided by Hsiao (1986) and Matyas and Sevestre (1992).

In this chapter we consider panels in which both N and T are large, although some of our results have implications for small T panels. We assume that the panels are heterogeneous, in the sense that parameters vary randomly across groups, and are dynamic, in the sense that each equation includes a lagged dependent variable. There are four procedures that can be applied to such panels. The data are averaged over groups and aggregate time series are estimated; they are averaged over time and cross-sections are estimated on group means; they are combined imposing common slopes but allowing for fixed or random intercepts and pooled regressions are estimated; or separate regressions are estimated for each group and the coefficients are averaged over groups. This last procedure we shall refer to as the 'mean group estimator'.

In the static case, where the regressors are strictly exogenous and the coefficients differ randomly and independently of the regressors across groups, all four procedures provide consistent (and unbiased) estimates of the coefficient means. It often seems to be assumed that a similar result holds for dynamic models, i.e. all four procedures give consistent estimators. This is not the case; in particular aggregating and pooling can produce highly misleading estimates of long-run effects in heterogeneous dynamic panels. The panel literature tends to assume homogeneity of slope coefficients, but this seems implausible for many energy applications; thus our results are of some practical importance. Griliches and Mairesse (1990) discuss heterogeneity in production function estimates.

The chapter is in two parts: first, Sections 2.1-2.5 review the theoretical issues; and second, Section 2.6 contains the empirical application. Sections 2.2-2.4 summarize the detailed technical results in Pesaran and Smith (1995), and assume that the underlying relations are correctly specified. Section 2.5 discusses the consequences of certain types of misspecification in the context of estimating short- and long-run energy demand elasticities. In the second part, Section 2.6, the theoretical issues are illustrated by an application to aggregate energy demand equations estimated for ten Asian developing economies over the period 197390. Given the limited econometric evidence available on energy demand in developing countries, the time-series data and the empirical results discussed in Section 2.6 may be of general interest to energy economists, although the highly aggregative nature of the analysis means that the elasticity estimates should be treated as preliminary and with due caution.


Initially, we assume static relationships with coefficients that are constant over time but differ randomly across groups, and that the distribution of the coefficients is independent of the regressors. Suppose the parameters of interest are the averages over groups of the coefficients. More specifically, suppose xit and are k* 1 vectors with where the sit are serially uncorrelated with zero means and constant variances. We assume that the coefficients vary according to fit- 0 + Vi (2.2)

where ft is fixed and the ni have zero means and a given covariance matrix. We also assume that {xit}, {eit} and {ni} are independently distributed. There are four procedures that could be applied to such data fields.

Cross-section estimators

The data could be averaged over time periods and a cross-section regression run on these 'time averages' defined by

The cross-section regression is given by which yields the estimates r - (x^r (1*4

This is known as the 'between' estimator in the panel literature.

Aggregate time-series estimators

The data could be averaged over groups and an aggregate time-series regression run on the group means (aggregates for each period). This is the dominant practice in macroeconomic analysis of long-run relationships. The procedure involves forming group averages for each period

Was this article helpful?

0 0

Post a comment