مدل سازی تقاضای انرژی صنعتی سازمان همکاری و توسعه :واکنش های نامتقارن قیمتی و تغییر فنی صرفه جویی در انرژی

The industrial sector embodies a multifaceted production process consequently modelling the ‘derived demand’ for energy is a complex issue; made all the more difficult by the need to capture the effect of technical progress of the capital stock. This paper is an exercise in econometric modelling of industrial energy demand using panel data for 15 OECD countries over the period 1962–2003 exploring the issue of energy-saving technical change and asymmetric price responses. Although difficult to determine precisely, it is tentatively concluded that the preferred specification for OECD industrial energy demand incorporates asymmetric price responses but not exogenous energy-saving technical change.

Given the importance of the global environmental agenda, never before has it been so important to understand the determinants of industrial energy demand in the developed world in order to assist international policy makers in their deliberations. They require sound and dependable models to support their projections of future industrial energy demand to underpin policy; for example the allocation of emission trading permits. However, the industrial sector embodies a multifaceted production process so that modelling the ‘derived demand’ for energy is a complex issue; made all the more difficult because of the need to capture the effect of technical progress of the capital stock and its subsequent effect on improved energy efficiency and hence energy consumption. Consequently, an understanding of this issue is vitally important — whatever modelling approach is adopted. This paper is an exercise in econometric modelling of OECD industrial energy demand in a panel context in order to explore the relationship between energy-saving technical change and asymmetric price responses since, as far as is known, modelling of the industrial sector has not been undertaken in this way before. In particular an attempt is made to determine whether an industrial energy demand model that incorporates either asymmetry in the price response or exogenous energy-saving technical change or both is accepted by the data in order to better understand the determination of OECD industrial aggregate energy consumption. The sharp increases in crude oil prices during the early 1970s stimulated a significant interest in energy demand research. This interest was maintained with the further increases in crude oil prices in the late 1970s and early 1980s followed by the collapse in the mid-1980s. The effect of these changes on the real OECD industrial energy demand price is illustrated in Fig. 1 along with the index of production and energy consumption.1 It can be seen that OECD industrial energy demand was rising consistently until the early 1970s and the first crude oil price hike, but since then has fluctuated with total consumption in 2003 for the countries in the sample being below that in 1974. At first sight this would appear to suggest an asymmetric price response with the large increases in the real energy price causing a significant reduction in consumption that was not reversed as prices subsequently eased.Most of the earlier studies of industrial energy demand followed the seminal work of Berndt and Wood (1975) and concentrated on factor substitution and subsequently inter-fuel substitution models. However, these models were based on a ‘strict’ neoclassical production and cost structure (normally represented by the translog function) that were often at odds with the data and, as Waverman (1992) states, the results from such models were “based mainly on intuition and thus incorrect” (p. 23). More recently, as Table 1 illustrates, a number of studies of industrial energy demand published since 1990 have continued to employ factor substitution models but in addition a number of studies have used a single equation approach often with a constant elasticity of demand (linear in logs) function. This procedure has become standard in energy demand estimation given its simplicity, straightforward interpretation, and limited data requirements and, as noted by Pesaran et al. (1998), it generally outperforms more complex specifications across a large variety of settings. Table 1 also illustrates that all cited studies assume that the estimated elasticities are symmetric; however, they do differ in terms of the country or countries, data frequency and period, the dynamic specification, the econometric technique used and the allowance for technical progress (or the underlying energy demand trend). For example Hunt and Lynk (1992) estimated a cointegrating error correction model (ECM) for the UK manufacturing sector with a deterministic trend using annual data from 1952 to 1988. Hunt et al. (2003a) and Dimitropoulos et al. (2005) used the structural time series model (STSM) to capture a non-linear underlying trend with an autoregressive dynamic lag (ARDL) model with UK quarterly and annual data respectively. Whereas Jones, 1995 and Jones, 1996, estimated a dynamic linear–logit factor substitution model using annual data but for the USA and the G-7 countries. Chang and Martinez-Chombo (2003), on the other hand, used a cointegrating ECM with time varying parameters to estimate electricity demand in Mexico using annual data but made no allowance for exogenous energy-saving technical change — instead implicitly assuming all technical progress is induced through the price effects. Medlock III and Soligo (2001) utilised a log quadratic specification to estimate the relationship between energy and income for a panel of mixed OECD and non-OECD countries by allowing for non-constant income elasticity with no deterministic trend or time dummies in their preferred model. Kamerschen and Porter (2004) estimated an electricity demand relationship using an adjustment factor on USA energy prices to reflect consumer expectation of future prices and included a deterministic trend as a measure of technical progress. Casler (1997) and Dahl and Erdogan (2000) estimated factor substitution models for the USA and Turkey respectively, but neither included a time trend in their model. Not surprisingly therefore, as Table 1 illustrates, there is a fairly wide range of estimates for the long run OECD aggregate industrial energy demand price elasticity whereas, when estimated, the OECD aggregate industrial energy demand long run income elasticity is about 0.7.2As noted, all the studies cited in Table 1 used a symmetric price elasticity approach, but Table 1 also illustrates that there is some variation in the way technical progress, or the more general underling energy demand trend, is captured. This reflects the long-running debate about whether a deterministic trend is an appropriate specification in such circumstances. A number of studies, (including Beenstock and Willcocks, 1981 and Jones, 1994) have attempted to determine whether a simple deterministic trend should or should not be included in the estimation to capture the effect of technical progress but more recently Hunt et al., 2003a and Hunt et al., 2003b have argued that it is unrealistic to expect a simple deterministic time trend to capture technical progress and other important exogenous factors (such as government policies, important changes in economic structure etc.) so that they allow the underlying energy demand trend (UEDT) to be stochastic. In summary, there is no consensus on how to estimate industrial energy demand, in particular how the effect of technical change (and possible other important exogenous factors) is captured. Therefore the principle suggested by Hunt et al., 2003a and Hunt et al., 2003b is followed in that any estimated ‘general’ model should be as flexible as possible and any restricted version is only accepted if supported by the data. Consequently, for the framework adopted here, the general model allows for both asymmetric price responses and energy saving technical progress. Nevertheless, it could be argued that asymmetry is less likely for industrial sector energy demand than for whole economy energy or oil demand; for example, the introduction of more fuel-efficient cars is unlikely to be reversed (fully) by an oil price decline. However, if an energy price rise does stimulate the installation of more efficient capital in the industrial sector it is arguably still unlikely to be reversed if the price rises again. Therefore, by utilising the ‘general to specific’ approach alluded to above it allows the data to determine whether the asymmetry is statistically important (relative to the energy saving technical change dummies) and if so whether the responses are ‘weaker’ than for other sectors of the economy. Furthermore, to date as far as is known, there has been no attempt to model asymmetric price responses for the industrial sector as an alternative way to capture induced technical change. 3 Therefore in order to undertake this exercise for the OECD industrial sector the approach adopted here follows and extends the approach applied to the whole economy in a series of papers by Gately and Huntington (2002), Griffin and Schulman (2005), and Huntington (2006). Gately and Huntington (2002) – hereafter GH – estimated aggregate energy (and oil) demand functions by allowing for asymmetric price elasticities4 in a similar way to that originally used in the agricultural supply literature5 and applied in the energy field, for example by Dargay and Gately, 1995a, Dargay and Gately, 1995b and Dargay and Gately, 1997, and Dargay (1990) – but not for industrial energy demand in the OECD. However, Griffin and Schulman (2005) – hereafter GS – suggest that that price asymmetry methodology popularised by GH is merely acting as a proxy for energy-saving technical change. To explore this, GS included time dummies as a proxy for induced technical progress arguing that this better represents the underlying demand trend than the asymmetries. Huntington (2006) in response showed that if the restrictions are actually tested on the GS results then it is possible to conclude that both asymmetric price responses and the exogenous time dummies have a role to play. The use of ‘top–down’ econometrics in this way arguably gives valuable insights into the aggregate effects of ‘macro’ trends in response to changing macro economic variables. However, unlike more ‘bottom–up’ engineering approaches it does not capture the intricacies of various new and developing technologies for the many industrial sub-sectors. In forecasting the future, it is usually preferable therefore to combine both ‘top–down and ‘bottom–up’ techniques, however for the remainder of this paper the focus is on the ‘top–down’ aspect. Therefore the question that this paper attempts to answer is whether OECD industrial sector energy demand is best modelled by the use of ‘time dummies’, ‘asymmetric price responses’ or ‘both’ when modelled using panel econometric techniques. The plan of this paper is therefore as follows: Section 2 details the methodology used in the study; Section 3 presents the data and estimation results; with Section 4 providing a summary of the findings and some general conclusions.

This paper is an exercise in estimating a panel data model of OECD industrial energy demand for 15 countries with data covering the period 1962–2003 based on the models and procedures developed by GH, GS, and Huntington (2006). The results, discussed in the previous section, show that there are mixed messages. In particular, unlike Huntington (2006) for the whole economy energy (and oil) demand, it is not possible to conclusively conclude that both asymmetric price responses and time dummies have a role to play; that is they are ‘complements’ rather than ‘substitutes’ as GS imply. Although the tests suggested by Huntington (2006) do support a similar conclusion for OECD industrial energy demand, the estimated price coefficients are not well determined. The coefficients are not in line with economic theory; with the coefficients on the price-max and price-rec variables being either positive and/or insignificantly different from zero leaving the price-cut variable the only significant price variable. Therefore Model III is rejected on economic grounds, hence the idea that asymmetry and time dummies are ‘complements’ is rejected leaving the choice between the ‘substitutes: Model I and Model II. However, the choice between Model I and Model II is not an easy one; although all price variables have the right sign the total price variable in Model II and the price-cut variable in Model I are not significant and the relative sizes of the coefficients on price-max and price-cut are not as expected.21 For pragmatic reasons Model I is therefore chosen over Model II on the grounds that, despite the relative sizes of price-max and price-rec not being as expected they are statistically significant22 — whereas the total price term in model II is not. Although it should be stressed that this is nowhere clear cut and further research is needed to try and disentangle this complicated relationship. Taking Model I as the preferred model suggests that the estimated long-run income elasticity of OECD industrial energy demand is 0.8. Furthermore the estimated long run elasticity of OECD industrial energy demand with respect to a price rise above the previous maximum and with respect to a price rise below the previous maximum are − 0.5 and − 0.6 respectively, whereas the estimated long run elasticity of OECD industrial energy demand with respect to a price cut is − 0.3 (although this is based on the short run coefficient which is not statistically significant from zero). The estimated income elasticity is therefore very close to the previous estimates presented in Table 1 for the UK. A comparison of the estimated price elasticities is more difficult given that all the cited previous studies of industrial energy demand assumed symmetric elasticities, however, the estimated price elasticities for a price-max and price rise are generally towards the more central range of the previous estimates for the total price elasticity. This exercise shows that when estimating energy demand models and considering the important issue of energy-saving technical progress (and other exogenous trends) a general flexible approach should initially be adopted. The chosen model should be the one that is accepted by the data while at the same time conforming to economic theory — but this should be estimated and tested rather than imposed at the outset. However, this exercise also illustrates that even then a clear favoured statistical model may not be found without the recourse to economic intuition and theory. In conclusion, it has been shown that econometric modelling of OECD industrial energy demand is not an easy task and further research is needed before ‘definitive’ estimates are obtained. Nevertheless this exercise has illustrated the importance, when modelling industrial energy demand in a panel context, of using a general flexible framework allowing for asymmetric price responses and time dummies to capture the underlying energy demand trends driven by technical progress and other exogenous factors. Although the results are not conclusive they do show that assuming a specific model or imposing, rather than testing, particular assumptions can be equally misleading and wherever possible the data should be allowed to determine the model — but guided by economic intuition and theory. However, the exercise has also exposed a number of weaknesses that need further research. A number of restrictions imposed by the GH and GS panel framework are adopted here, in particular the assumption that the slope and time coefficients are constant across a wide range of countries.23 Given the diverse nature of the countries used in this study it could be argued that these restrictions are unrealistic.24 Each country's share of industrial output in GDP is likely to be different and also involve different industrial structures, institutions and socioeconomic patterns. Therefore the imposition of the same pattern of underlying energy demand trend across each country via the fixed time effects appears particularly restrictive (at least without formal testing). Furthermore, although this is a panel of developed OECD countries, each country is likely to be at different stages of development. Given these factors it is not surprising that it is difficult to obtain statistically sound and economically consistent estimates with sensible elasticities that apply across all countries with an identical underlying energy demand trend. Future research will therefore aim to investigate these matters further, by testing pooling restrictions across countries 25 and, if as expected the pooling restrictions are generally rejected, estimate industrial demand models for each country separately, again starting with a general specification that allows for asymmetric responses but with a non-linear underlying trend to capture energy-saving technical change and other exogenous effects and test the models accordingly.