Prices that clear the air: energy use and pollution in Chile and Indonesia.

INTRODUCTION

The best way to "clear the air" is to make polluters face the marginal social costs of their actions. A policy which could implement such a principle is a pollution tax equal to the marginal social damages of emissions. Such corrective taxes to protect public goods - such

However, such first-best policy instruments are not always feasible. It may be very costly to monitor emissions from every polluter, particularly if there are many of them (as with motor vehicles, residences and manufacturing firms). Such costs are onerous in general, but particularly for developing countries, with scarcity of institutional and technical capacity for monitoring and enforcement. Even industrialized countries generally try to economize on monitoring, for instance by setting standards for emission rates, by type of equipment or activity. However, studies show that such policies should be complemented by taxes on inputs and/or outputs in the polluting activities?

When first-best instruments such as emission taxes are not available, indirect instruments, such as presumptive taxes levied on polluting goods and inputs may be attractive alone or in combination with other instruments. Taxation of fuels - or energy - is such an indirect instrument that is potentially attractive for air pollution control, because energy consumption is a proxy for the utilization of polluting equipment.(3) Thus, if individuals and firms are induced to economize on energy use or to switch to cleaner fuels, their emissions will fall. A good illustration of this is the U.S. Environmental Protection Agency's standard reference on emission modelling, AP-42, which projects emissions by multiplying a constant (an "emission factor") by the amount of fuel used (U.S. EPA, 1986).(4)

To use energy prices-or taxes-to control pollution, policy makers would want estimates of the responsiveness of emissions to price changes. The main objective of this paper is to demonstrate how these can be provided. We first present an empirical framework for estimating the necessary parameters (section 3), and then apply it to two cases for which fuel pricing could be attractive strategies: manufacturing in Chile and Indonesia (sections 4 and 5). Chile and Indonesia are different in many respects - such as income levels and fossil fuels endowments - but have in common a sizeable manufacturing sector and a recognition of air pollution problems.(5) We report results from both countries to highlight the approach and a range of results, rather than merely point estimates for a country. The model combines econometric estimates on how fuel demand responds to price changes with engineering estimates of the link between input use and emissions. The main finding is that the responsiveness of emissions makes energy prices - or taxes - powerful indirect policy instruments.

The relevance of our findings for control costs and strategy analysis is not limited to the case in which fuel prices are used as policy instruments. Interfuel substitution will be among the responses firms can make to reduce emissions even in a first-best world. Thus, estimates of substitutability in demand (as provided here), as well as of technical control options (filters, say), are relevant in general for analysis of emission control costs (see also Kopp 1992, which proposes a general framework and eclectic use of knowledge from technical studies as well as econometric models). A recent study (Eskeland and Feyzioglu, 1997) parallels this study by showing the role of demand relationships in an air pollution control program for the transportation sector. Eskeland and Feyzioglu (1997b) demonstrates unfortunate consequences when regulation - rather than market based instruments - are used in demand management.

2. RELATIONSHIP TO THE LITERATURE

Energy Demand and Fuel Substitutability. This analysis contributes to the body of quantitative evidence regarding the price responsiveness of energy demand in developing countries. Since the first oil price shock in 1973, there has been substantial research on the price sensitivity of demand in OECD countries.(6) The evidence from aggregate data suggests that there is significant potential for energy conservation and interfuel substitution. A casual illustration may be the contrast between manufacturing in OECD countries, which faced increased energy prices from 1973 onwards, and in Mexico, which was shielded from these world price movements. In OECD, energy consumption did not increase between 1971 and 1988, while manufacturing output increased by 62 percent. This implies a 38 percent reduction in energy intensity (Bacon, 1992). In Mexico, the manufacturing sector did not reduce its energy intensity in this period (Ten Kate, 1993). Quantitative studies using econometric techniques also indicate price-responsiveness in manufacturing input demand. Using annual observations across Canadian provinces, Fuss (1977) finds relatively large own price elasticities for fuels, in the range of -0.7 to -2.9. His estimate is -0.5 for aggregate energy. Using cross-country data, Pindyck (1979), finds corresponding estimates of -0.7 to -2.2 (fuels), and -0.8 (aggregate energy).

Studies from developing countries are less numerous, but growing. Pindyck (1979) describes the high elasticities he finds for Spain, Greece, Turkey, Brazil and Mexico as "consistent with the expectation that energy demand in the industrial sector should be more price elastic in developing countries due to a greater ability to substitute low-priced labor" (p. 255). There are also demand studies for the countries of the present paper. Pitt (1985) uses pooled cross section data from Indonesian manufacturing. With only three years of data (1976-78) and two aggregate inputs (energy and labor) he finds own price elasticities for aggregate energy between -0.07 and -0.8, with the higher elasticity (in absolute value) for the more energy intensive sectors. Moss and Tybout (1994) uses plant-level data from Chile, and finds substantial variability in fuel use per unit of output among firms and over time, even within subsectors. In another study, Guo and Tybout (1993) estimates inter-fuel substitution possibilities for four subsectors of Chilean manufacturing from plant observations. They find significant substitutability for some sectors, little in others. Another branch of the literature focuses on energy conservation - built on prospective analysis of technical opportunities (new capital, often) or on econometric estimation. Jaffe and Stavins (1994) examine determinants of diffusion for energy-saving technologies. A recent econometric study by Kaufmann (1994a), shows that expected energy prices have a significant effect on energy demand in the United States. Bhatia (1987) surveys the literature on energy demand in developing countries, but does not report econometric results. This paper contributes by using more recent data, providing results for the manufacturing sector as a whole, and separating heavy fuels, light fuels and electricity. These features make the results more relevant for a policy problem such as pollution control.

Modelling Emissions. This study also makes an explicit link between pricing policy and emission outcomes. Until now, economic models have linked emissions to energy use, focusing mostly on climate change (often limited to C[O.sub.2] emissions) aggregating over fuel types as well as different economic activities.(7) We contribute to this literature by allowing for the substitution among fuels with differing emission characteristics, and by studying several pollutants. Since emissions of different pollutants are produced - and controlled - as joint products, multipollutant treatment is essential (See Eskeland, 1997).

Studies that examine the link between emissions and fuels based on sample tests and engineers' estimates are rare - they are available only for a few developed countries, such as the U.S. The key result from these studies is that emissions are proportional to fuel use, with emission factors for each fuel determined by the characteristics of fuels and equipment.(8) It is precisely this property that we exploit in these case studies.

Our approach thus gives an improved understanding of the link between pricing and emissions. By disaggregating energy, it provides a broader set of instruments, and an improved assessment of tradeoffs (including tradeoffs among pollutants) - all ingredients in more efficient policy.

3. THE MODEL

Fuel Demand. The input demand model and the empirical estimation closely follows work pioneered by Fuss (1977) and Pindyck (1979), and is reported here only briefly.

We employ a short term cost function:

C = C(Y, K, [P.sub.L], [P.sub.M], [P.sub.E] [[P.sub.E1], [P.sub.E2], . . ., [P.sub.En]]), (1)

where C is the cost of producing output Y, capital is K, and [P.sub.L], [P.sub.M] are prices of labor and material respectively, and [P.sub.Ei] is the price of energy source i. The function [P.sub.E] aggregates prices of individual energy sources (electricity and fuels) into an energy price (see below). Some restrictive assumptions are reflected in (1). The model assumes weak separability among aggregate inputs: energy, capital, labor and material. Furthermore, we assume that the energy aggregate is homothetic and exhibits constant returns to scale. This structure, which follows Fuss and Pindyck, serves to limit the range of functional forms and parameters involved when focusing on inter-fuel substitutability.

Separability implies that the marginal rates of substitution between fuels depend only on the use of fuels. Homotheticity and constant returns to scale further simplifies the fuel submodel: these two assumptions imply that fuel shares (in the energy aggregate) depend on relative fuel prices only - and that energy does not become cheaper or more expensive with the scale of consumption. This is what allows us to estimate parameters of a very simple energy submodel. While these assumptions allow for estimation of a demand system with a limited set of parameters, we emphasize that the assumptions themselves are not tested.

We use a translog second-order approximation to the cost function (1). From the first-order conditions for cost-minimization (given output and capital), we obtain the following factor-share equations:

[Mathematical Expression Omitted] (2)

where i, j = E, L, M; ln[P.sub.j] is the logarithm of price [P.sub.j]; and [Q.sub.k] is Y and K. The model for n fuel shares in the energy submodel is simpler:

[Mathematical Expression Omitted] (3)

Estimation of the demand system is first done for the energy submodel (3). With the parameters estimated, an instrumental variable for the energy price [P.sub.E] is created, and the aggregate model is estimated.(9) The estimated parameters yield fuel demand elasticities as follows:

For the energy submodel:

[Mathematical Expression Omitted], [Mathematical Expression Omitted] (4)

where [Mathematical Expression Omitted] are coefficients estimated in (3) and [Mathematical Expression Omitted] are predicted fuel shares (at means).

For the full model, with aggregate energy adjusting, we have:

[Mathematical Expression Omitted], [Mathematical Expression Omitted]. (5)

The Emission Model. Emission models usually are drawn from technically oriented literature, such as the U.S. EPA's emission modelling handbook, called AP-42 (US EPA, 1986).

For air pollution, these models typically use a function:

[Mathematical Expression Omitted] (6)

where [E.sup.p] is emissions of pollutant p (say, kilograms of particles per day, a flow), [Mathematical Expression Omitted] is the emission factor for pollutant p, energy source i, [X.sub.i] is the quantity of fuel consumed and t is a vector of technical characteristics for the equipment (say, a furnace), firing conditions, emission control devices, and the fuel (sulfur content, ash content, etc.). We assume that the vector t is unaffected by fuel price changes, so that the elasticity of emissions to a fuel price change is:

[Mathematical Expression Omitted] (7)

Where [[Epsilon].sub.ij] is the applicable demand elasticity of fuel i with respect to the price of fuel j: equation (4) when aggregate energy is assumed constant, otherwise equation (5).

From a practical perspective, the data and modelling approach require several restrictive assumptions: (i) regarding estimation of the demand model: that non-price determinants of fuel demand in the estimation period are constant or uncorrelated with relative prices; (ii) that price changes in the envisaged policy scenarios do not change emission factors for individual fuels.

Estimating Technique and Data. We follow Fuss (1977) and Pindyck (1979) and aggregate over plants in a region and estimate demand functions for the entire manufacturing sector. Aggregating over plants and subsectors is motivated largely by economy - it provides directly a model of the responsiveness of the sector as a whole, which is relevant for the policy question at hand.(10)

Each data set is a census of manufacturing firms and contains detailed plant level cost data, covering all plants with more than 10 workers (Chile) and 20 workers (Indonesia).(11) The Indonesian data cover the period of 1975-1989 and the Chilean data cover 1979-1986. The data used in the pooled cross-section time-series analysis are aggregated at the regional level, utilizing the panel structure by introducing regional fixed effects.(12) The numbers of observations were 240 for Indonesia (16 regions times 15 years) and 78 for Chile (13 regions times 6).

Fuel consumption data is reported in detail in both data sets, with information on both quantity and expenditure, to allow construction of fuel-specific price series (unit values) and thus facilitating estimation of inter-fuel substitution in the energy sub-model. We aggregate energy sources into three categories: electricity, heavy fuels and light fuels (Table 1). These categories are defined by three considerations: (i) assumed economics of substitution: electricity, heavy fuels, light fuels allow reasonably homogenous inputs in each group (see Table 1), thus representing relevant choices for the firm; (ii) modelling objective: the three categories also differ sufficiently in terms of pollution coefficients; (iii) demand system estimation: three categories yield few enough parameters - and high enough fuel cost shares, to successfully estimate a demand system.(13) For electricity, many firms in developing countries have substantial primary or back-up capacity for own electricity generation (see Lee et al. 1992). Our treatment of such generation is to net out own generation from electricity consumption, so that this activity is counted only in terms of the fuel used in generation.

The composition of the three categories of energy sources is shown in Table 1. Heavy fuels (heavy fuel oil, wood and coal), are relatively inexpensive per heat unit, but may require more of other inputs to deliver a unit of output than is required by, say, electricity. Electricity is relatively expensive per heat unit, but has other qualities, such as convenience and flexibility. These different qualities of the various energy sources give rise to imperfect substitution, and to our econometric estimation of substitutability.(14)

[TABULAR DATA FOR TABLE 1 OMITTED]

Table 2 presents fuel consumption data. The first two columns report cost shares. In the three next columns, we use tons of oil equivalents (toe, a measure of heating value, applied to the user) as a quantity measure. Indonesia has a lower average price of energy ($237/toe, as opposed to Chile's $260) and a higher energy intensity (238 toe/million$ output, as opposed to Chile's 139). Energy price increases - often recommended by external donors in petroleum rich countries - have long been a contentious issue in Indonesia (see, for instance, Pitt, 1985, Hughes, 1987). In Chile, electricity and heavy fuels are the two most important fuels in expenditure terms, and heavy fuels are most important in energy terms. Indonesia is different, with light fuels having the highest share both in energy terms and in expenditure terms. The industry's expenditure on energy is 3.6 percent in Chile and 5.6 percent in Indonesia.

In terms of emissions, we focus on total suspended particles (TSP), the measure of greatest relevance from a health perspective (see, for instance, Ostro, 1994 and Ostro et al., 1996) and sulfur oxides ([SO.sub.x]), which have a wider range of consequences, such as damage to fisheries, crops, ecosystems, materials and health. Calculations for other pollutants are performed, but are not reported in detail, though some figures are given in the concluding section of the paper and in two Annex Tables A3 and A4. For both types of emissions, [TABULAR DATA FOR TABLE 2 OMITTED] emission factors are higher for electricity than for heavy fuels and lowest for light fuels (Table 3). The calculations are based on standard models for emissions, using information about the fuels and the subsectors of manufacturing.(15) These calculations assume electricity is at the margin produced 50 percent through coal-fired power plants and 50 percent by clean technologies (say, hydro).(16) With these assumptions, electricity is more polluting than coal on a toe basis, since about two-thirds of the heat value is lost in the conversion to electricity. In our following presentations, we perform sensitivity analysis, alternatively viewing electricity as non-polluting. Electricity could be viewed as nonpolluting either because the sector is clean (emission-controlled - or hydroelectric at the margin) or because the perspective is strictly to address emission from the users, in which case electricity is non-polluting.

[TABULAR DATA FOR TABLE 3 OMITTED]

In Chile, heavy fuels are responsible for half of the emissions of particulates and around two thirds of sulfur oxides. In Indonesia, very little heavy fuels are used, representing only 25 percent of energy, as opposed to 68 in Chile. Historically, light fuels such as diesel have been heavily subsidized (prices are marginally higher than those for heavy fuels, as opposed to 50 percent higher in Chile), and natural gas and liquid gas have been available. Due to the lighter composition of energy, weighted emission factors (per toe) are lower than for Chile (4.1 kg TSP/toe, as opposed to 10.9 for Chile). Due to higher energy intensity, however, Indonesian manufacturing emits about the same amount of pollution per unit of output as Chilean manufacturing (less for TSP, more for [SO.sub.x]).

In addition to fuel data, the estimation of the cost share equations in the full model requires real output, real capital stocks, and unit values and cost shares for labor and materials. For Indonesia, due to lack of data, only the parameters of the energy submodel were estimated - yielding estimates of interfuel substitutability assuming aggregate energy constant. For Chile, it was possible to construct the quantity and price measures necessary to estimate the aggregate model as well.

4. ESTIMATION RESULTS: CHILE

Demand Elasticities: Energy Submodel. We will first estimate the energy submodel, which describes the extent to which fuels can be substituted for each other while aggregate energy is constant. Since the own price elasticity of energy is non-positive, this gives a conservative estimate of the extent to which emissions fall with an increase in each fuel price (see equations 5).

Table 4. Demand Elasticities, Energy Submodel: Chilean Manufacturing

                           Electricity      Price of       Price of
                              Tariff      Heavy Fuels    Light Fuels

Electricity Demand           -0.49(*)       0.35(**)       0.13(*)
                             (0.08)        (0.06)         (0.07)
Demand for Heavy Fuels        0.29(**)     -0.52(**)       0.22(**)
                             (0.05)        (0.07)         (0.05)
Demand for Light Fuels        0.21(*)       0.42(**)      -0.63(**)
                             (0.11)        (0.10)         (0.13)

Note: Fixed-effects model. Elasticities reported at means. Standard
errors in parenthesis.

* Significant at 5%;

** Significant at 1%. Parameter estimates for model without fixed
effects are shown in Annex Table A1.

The estimated elasticities (Table 4) are significantly different from zero and with the expected signs: negative for own price elasticities and positive for cross price elasticities. Own-price elasticities are around -0.5, cross-price elasticities between 0.1 and 0.4. These elasticities reflect that the manufacturing sector has significant flexibility to respond to price changes, even when the choice is limited to substitution among fuels.

These data can be used to calculate emission elasticities assuming aggregate energy constant, shown in Table 5. To illustrate, one could increase the price of heavy fuels to reduce particulates and/or sulphur emission, but the effect would be slight if electricity is polluting - since much of the response is substitution towards electricity. If electricity is not polluting, however, the elasticities of sulfur and TSP emission are about -0.5. This indicates that a 20 percent price increase for heavy fuels would reduce total emissions of sulfur and particles by 10 percent - an impressive contribution to any air pollution control program.

Table 5. Emission Elasticities, Energy Submodel: Chilean
Manufacturing

                             Electricity     Price of      Price of
                                 Tariff    Heavy Fuels   Light Fuels

Assuming Electricity Polluting:

Particulates (TSP)               -0.10        -0.08          0.17
S[O.sub.x]                        0.05        -0.22          0.16

Assuming Electricity Non-Polluting:

Particulates (TSP)                0.29        -0.51          0.21
S[O.sub.x]                        0.29        -0.46          0.17

Demand Elasticities: The Full Model. In the energy submodel, the "tradeoff" between energy sources is accentuated, since the use of one fuel can be reduced only if the use of others is increased. This is not so in the full model, in which other inputs can substitute for aggregate energy in response to a price rise. In fact, in the full model, a proportional reduction in all pollutants can be achieved by a proportional increase in all fuel prices. Table 6 shows price elasticities for aggregate inputs: energy, labor, and materials in a short term, cost function model. Own price elasticities are negative, and all are significantly different for zero at one percent. Of particular interest is the own price elasticity of energy, which is -0.94.

Table 6. Demand Elasticities For Aggregate Inputs, Chilean
Manufacturing

                        Aggregate Energy     Wage     Material Input
                              Price                         Price

Energy Demand                -0.94          0.16             0.78
                             (0.002)       (0.007)           0.007
Labor Demand                  0.06         -1.06             1.01
                             (0.003)       (0.02)           (0.21)

Material Demand               0.05          0.19            -0.25
                             (0.0005)      (0.04)           (0.04)

Note: Fixed effects model. Standard errors in parenthesis. All
coefficients significant at 1%.

When demand for aggregate energy adjusts, both own and cross price elasticities of fuels are shifted downwards, to be smaller (Table 7). Comparing the elasticities with those in Table 4, own-price elasticities have fallen (to become "more negative") and positive cross price elasticities have all fallen from positive to around zero.

Table 7.
Chilean Manufacturing: Demand Elasticities: Aggregate Energy
Adjusting

                         Electricity      Price of         Price of
                            Tariff       Heavy Fuels     Light Fuels

Electricity Demand           -0.82          -0.05           -0.08
Heavy Fuel Demand            -0.04          -0.91           -0.01
Light Fuel Demand            -0.12           0.02           -0.84

Emission elasticities - energy adjusting - are given in Table 8. As expected, the figures reflect a stronger tendency for price increases to reduce emissions. Assuming that electricity is polluting, emission elasticities with respect to electricity tariffs and heavy fuels' prices now range between -0.3 and -0.8. When electricity is not polluting, electricity tariffs have approximately no effect on emissions, and the price of heavy fuels is a more powerful instrument, with emission elasticities of -0.8.

Table 8. Emission Elasticities, Chilean Manufacturing: Full Model

                         Electricity      Price of         Price of
                            Tariff       Heavy Fuels     Light Fuels

Assuming Electricity Polluting:

Particulates (TSP)           -0.43          -0.43           -0.04
S[O.sub.x]                   -0.28          -0.58           -0.05

Assuming Electricity Non-Polluting:

Particulates (TSP)            0.05          -0.82            0.03
S[O.sub.x]                    0.01          -0.77            0.03

5. ESTIMATION RESULTS: INDONESIA

Demand Elasticities: Energy Submodel. Table 9 shows the estimated price elasticities of demand, assuming aggregate energy constant. As in Chile, own price elasticities are negative, but they are larger in absolute value: from -0.4 to -1.4. Cross price elasticities are positive, or insignificantly different from zero (between heavy fuels and electricity).

Table 9.
Price Elasticities, Aggregate Energy Held Constant: Indonesian
Manufacturing

                         Electricity      Price of         Price of
                            Tariff       Heavy Fuels     Light Fuels

Electricity Demand          -0.88            0.02            0.85
                            (0.09)          (0.07)          (0.09)
Heavy Fuels                  0.03           -1.37            1.34
                            (0.08)          (0.12)          (0.12)
Light Fuels                  0.19            0.25           -0.44
                            (0.02)          (0.02)          (0.03)

Note: Fixed effects model. Standard errors in parenthesis. All
coefficients are significantly different from zero at 1% level,
except the two cross price elasticities between electricity and
heavy fuels. Parameter estimates for model without fixed effects are
shown in Annex Table A2.

Emission Elasticities: Energy Submodel. Table 10 shows emission elasticities, using equation 7. Due to the higher own price elasticities (in absolute value) for heavy fuels in Indonesia than in Chile, emissions respond quite strongly to higher prices for heavy fuels, even with aggregate energy constant. When electricity is not polluting, a one percent price increase for heavy fuels reduces emissions 1.2 percent for particulates and 0.7 percent for sulfur oxides.

For Indonesia, estimating the full model is not feasible due to data availability. It is only possible to estimate the energy submodel, which provides conservative estimates of the responsiveness of emissions to energy price increases. With this constraint in mind, it is noteworthy that emissions from Indonesian manufacturing are so sensitive to fuel pricing, despite the already high share of light fuels.

Table 10.
Emission Elasticities, Energy Submodel: Indonesian Manufacturing

                         Electricity      Price of        Price of
                            Tariff       Heavy Fuels     Light Fuels

Assuming Electricity Polluting:

Particulates (TSP)           -0.40          -0.63            1.03
S[O.sub.x]                   -0.12          -0.56            0.68

Electricity not Polluting:

Particulates (TSP)            0.04          -1.24            1.2
S[Ox.sub.]                    0.09          -0.72            0.63

6. SUMMARY AND CONCLUSIONS

Emission reductions can be stimulated by changes in fuel prices, because of the effects that inter-fuel and input substitution have on the various pollutants. A priori, emission elasticities depend on how flexible firms are in their input use, how different the energy sources are in terms of the critical pollutants, and whether present consumption leaves much room for substitution towards cleaner fuels or energy conservation. These factors determine whether fuel substitution has the potential to reduce emissions significantly. Whether such a potential should be exploited by "dedicated" indirect instruments (such as presumptive fuel taxes), or by more general and flexible instruments (such as emission taxes), depends on other factors, such as the costs of monitoring and enforcement.

This paper provides a modeling framework to assess this potential and reports empirical estimates. Using data from Chile and Indonesia, we find that there is room for substitution towards cleaner input combinations, both towards cleaner fuels and also away from energy, towards labor, capital and materials (the latter is shown for Chile only). Emission elasticities with respect to heavy fuel prices - conservatively reported by a restricted cost function approach - are between -0.4 and -0.8 for TSP and sulfur oxides in Chile. For Indonesia, these emission elasticities are more conservatively reported (since they are estimated from a model of polluters that are more constrained in their responses), but nevertherless fall in the range between -0.4 and -1.2. This implies that if one increases the price of heavy fuels, then one can expect emission reductions for TSP and [SO.sub.x] that are in a range of 40 to 120 percent of the price increase.

One should be careful to assess the relative damage caused by different pollutants. For both case study countries, if the priority objective is to reduce small dust particle emissions (because of associated health effects, see Ostro, 1994, for Indonesia; and Ostro, Eskeland, Sanchez and Aranda, 1996, for Chile), increased prices for heavy fuels can deliver such reductions. Emphasis on sulfur oxides - more noted for damages to materials, plants and ecosystems, would give the same policy recommendation. For other pollutants, such as the ozone precursors (nitrogen oxides and volatile organic compounds), the suggested policy would be somewhat different, since light fuels are relatively more important (Annex Tables A3 and A4). Price increases for light fuels can be part of a strategy to reduce these pollutants. In Indonesia, because of an abnormally high share of light fuels in energy consumption, price increases for heavy fuels would increase emissions of VOC and CO.

If fuel taxes are to be used, it is important what other instruments are used or can be used. If electricity is not polluting at the margin (for instance if the power industry is effectively controlled) then both input prices and output tariffs in the electricity industry should be shielded from the price increases for heavy fuels. The reason is that substitution towards electricity in that case can contribute to the emission reductions. If, on the other hand, electricity producers are thermally based and uncontrolled, then price increases for heavy fuels will be more effective in delivering emission reduction if they apply to power producers as well - preferably on their input side similar to the way they apply to other users of fossil fuels.

Indirect instruments such as fuel taxes are blunt compared to first best instruments such as emission taxes, but the strategy can be refined as institutional capacity develops. As one example, illustrated here briefly by alternative assumptions for power plants, taxes on fuels could be introduced as presumptive emission taxes, to be partly refunded for users who can document lower emission factors than presumed. This point is illustrated when we show how price increases for heavy fuels should be accompanied by tariff increases for electricity if and only if power plants are polluting.

Finally, while the figures have shown responsiveness to emissions, much work remains before one can establish an optimal tax structure in practice (See Sandmo, 1975). Our estimates are more useful in a discussion of the recently popular proposals for green tax reform. We do not include any discussion here about single or double dividends from green tax reform, but have provided parameters indicating where a green dividend can be found.

The relative attractiveness of presumptive fuel taxes versus other instruments depends in part on other costs of selective or general energy price increases. Also, it will depend on governments' ability to implement alternative instruments (such as emission taxes backed by emission monitoring) and on the effectiveness of technical emission control devices (which reduce emission factors). First-best instruments, such as emission taxes, are desirable from a theoretical perspective because of their ability to elicit low-cost controls. This is particularly important when polluters are many and possess private information. However, such instruments may require too much in terms of monitoring emissions, and enforcement. Thus, for many countries, and in particular developing countries, the case for using indirect instruments can be persuasive: one can apply fuel taxes for demand management, perhaps accompanied by emission standards or other incentives for cleaner technologies. As institutional capacity develops, such as system can be developed, approaching the refinement of taxes based on emission monitoring. This paper has shown that inter-fuel substitution and energy conservation would be prominent among firms' responses even if such sophisticated instruments were to be used.

ANNEX

Table A1.
Demand Elasticities, Energy Submodel: Chilean Manufacturing

                         Electricity      Price of         Price of
                           Tariffs       Heavy Fuels     Light Fuels

Electricity Demand        -0.87            0.37            0.50
                          (0.008)(**)     (0.07)(**)      (0.08)(**)
Demand for Heavy Fuels     0.30           -0.49            0.19
                          (0.06)(**)      (0.12)(**)      (0.09)(**)
Demand for Light Fuels     0.78            0.35           -1.13
                          (0.12(**)       (0.17)(*)       (0.18)(**)

Note: Estimated without fixed effects.

* Significant at 5%;

** Significant at 1%. Standard errors in parenthesis.
Table A2.
Demand Elasticities, Energy Submodel: Indonesian Manufacturing

                         Electricity      Price of         Price of
                           Tariffs       Heavy Fuels     Light Fuels

Electricity Demand          -0.095           0.14            0.80
                            (0.10)          (0.08)          (0.12)
Demand for Heavy Fuels       0.17           -1.52            1.35
                            (0.09)          (0.14)          (0.15)
Demand for Light Fuels       0.18            0.25           -0.43
                            (0.03)          (0.03)          (0.04)

Note: Estimated without fixed effects. Standard errors in
parenthesis. All coefficients significantly different from zero at 1
percent.
Table A3. Emission Elasticities, Additional Pollutants:Chilean
Manufacturing, Full Model

                        Electricity      Price of          Price of
                           Tariff       Heavy Fuels      Light Fuels

Assuming Electricity Polluting

NOx                        -0.14           -0.68             -0.1
VOC                        -0.13           -0.39             -0.42
CO                         -0.14           -0.35             -0.44

Electricity not Polluting

NOx                         0.02           -0.70             -0.05
VOC                        -0.01           -0.33             -0.39
CO                         -0.01           -0.30             -0.42
Table A4. Emission Elasticities, Additional Pollutants:Indonesian
Manufacturing, Fuel Submodel

                        Electricity      Price of          Price of
                           Tariff       Heavy Fuels      Light Fuels

Assuming Electricity Polluting

NOx                         -0.1           -0.37            -0.47
VOC                          0.15           0.17            -0.32
CO                           0.12           0.15            -0.27

Electricity not Polluting

NOx                          0.12          -0.48             0.37
VOC                          0.18           0.17            -0.35
CO                           0.18           0.16            -0.34

1. An efficient outcome (though with different distributional implications) could also result if the polluter were subsidized to cut its emissions, but this could create incentives to become a member of this subsidized class.

2. See Eskeland (1994) and Eskeland and Devarajan (1996) on how standards should be accompanied by taxes on inputs and outputs. On the use of indirect instruments, see Eskeland and Jimenez (1992) Balcer (1980) Wijkander (1985) Sandmo (1976). For major stack sources, such as power plants, continuous emission monitoring is now feasible, though not in wide-spread use. Generally, pollution control agencies obtain measures of emission rates only, and at most with infrequent, pre-announced visits (See Russell, 1990). See Opschoor and Voos (1989) for an OECD review of economic policy instruments.

3. Our use of the word energy is conventional: From the user's perspective, energy is an aggregate of electricity and inputs used in combustion processes, such as fossil fuels and biomass. We use the term "energy conservation" when aggregate energy use per unit of output is reduced. For simplicity, we include electricity when we use the term "fuel substitution."

4. For each type of fuel throughput, the emission factor is determined by technical parameters of firing conditions, equipment, fuel, etc. Standards to control emissions typically will apply either to the determinants of emission factors (fuel standards, equipment standards) or to the emission factors themselves. An example of a study which uses the link between fuel consumption and air pollution externalities for the U.S. is Viscusi et al. (1994). It compares energy taxes that would internalize external effects with present tax levels, but does not estimate how emissions would change, as we do in the present study.

5. With per capita incomes growing at 6-7 percent over the last decade, Chile and Indonesia have per capita incomes of $3520 and $880, respectively, using official exchange rates, or $8890 and $4600, using purchasing power parity. Air pollution problems are described in Ostro (1994) for Indonesia, and World Bank (1994) and Ostro et al. (1996) for Chile.

6. Bacon (1992) discusses evidence from econometric as well as engineering approaches to assessment of inter-fuel substitutability and energy conservation. The empirical literature in the seventies generated methodological advances, as well as empirical estimates. Important landmarks, both critical to the method applied here, are flexible functional forms, such as the trans-log function (Christensen, Jorgenson and Lau, 1971), and the use of separability and homotheticity in the energy aggregate (Fuss, 1977), allowing the estimation of systems with more inputs. Fuss (1977) demonstrates the methodology applied in the present study.

7. The most interesting modelling is found in Jorgenson and Wilcoxen (1993), where emissions of carbon dioxide are linked to energy use in an estimated computable general equilibrium model. The CGE model disaggregates energy into coal, petroleum and electricity, but does not report on the role of substitution among these. Most other CGE models do not disaggregate energy use, and many use calibrated rather than estimated parameters. Boyd, Krutilla and Viscusi (1994) assumed fixed output coefficients, Stephan, van Nieuwkoop and Wiedmer (1992) assume Leontief and Cobb-Douglas technology. Other studies are Whalley and Wigle (1991), Glomsrod, Johnsen and Vennemo (1992), and Goulder (1995). An ambitious study is Duchin and Lange (1994) using an input-output model (thus focusing on technology) to ask whether an environmentally sustainable path for the world is feasible. Another study, Alfsen et al. (1990) analyzes emission consequences (C[O.sub.2] and S[O.sub.x]) of transition from oil and coal to natural gas in Europe, without estimating substitutability. Our methodology and results could serve as building blocks for such models in the future.

8. A fuel may have different emission factors for different types of users. In the short-run, when equipment is assumed to be unresponsive to fuel price changes, a fuel's emission factor for a population of users will be the individual factors weighted by their share in the total fuel demand change. Assuming uniform demand elasticities across users for a given fuel, the factors will be weighted by their shares in total consumption. It is probably conservative, if anything, to assume that emission reductions will be proportional to fuel demand: Fuel costs will likely be a greater share of costs for users with older, "dirtier" equipment, so there may be a systematic tendency that users with older equipment reduce their demand more in response to fuel price increases. To our knowledge, studies allowing comparison of demand elasticities across individual users of a fuel do not exist, so the assumption that aggregate demand elasticities apply to individual users seems reasonable.

9. We use the seemingly unrelated regressions (SUR, in SAS), imposing the appropriate parameter constraints. Standard errors in the elasticities are reported at means as in Pindyck (1979), p.62: [Mathematical Expression Omitted]. An appendix that provides details on the econometric model and estimation can be obtained from the authors.

10. Since differences in technology between industries may lead to aggregation bias, building aggregate models on more disaggregated results would be an important area for future research. For models of selected sub-sectors, see Guo and Tybout (1994) for Chile, and Pitt (1985) for Indonesia.

11. Indonesia: Biro Pusat Statistik, Census of Medium and Large Scale Enterprises. Chile: Instituto Nacional de Estadestica, Manufacturing Census.

12. An alternative strategy would have been to use random effects. The choice between these alternatives depends on "the context of the data, the manner in which they are gathered, and the environment from which they came" (Hsiao, p 43). The random effects model would be appropriate if one had a sample of regions and wanted to make unconditional statements about relations for the underlying population of regions. In our case, we are interested in making conditional statements about fuel demand relations for the regions in our sample (which is the population). Furthermore, random effects estimates would lead to biased estimates if the effects correlate with the explanatory variables. The region-specific effects are likely correlated with our unit costs of fuels, given that these in part reflect differences among regions in access to and transportation costs for fuels. Finally, a random effects model, even if it were applicable, would only improve efficiency. In our case, the fixed effects estimates have quite high t-statistics, so the increased efficiency would not influence our substantive conclusions.

13. An earlier version of this paper presented estimates with six fuel categories: Much fewer parameters were estimated with precision. Emission coefficients for heavy fuels, light fuels and electricity are weighted averages of their components.

14. Kaufmann (1994b) explores the heterogeneity issue in a study focusing on differences between fuels in marginal product per heat unit.

15. The pollution coefficients were compiled for individual fuels, presented here as weighted averages. We are grateful to Christopher S. Weaver, of Engine, Fuel and Emissions, California for this analysis.

16. In a given year in Chile, hydro accounts for 60-80 percent, but thermal power tends to play a greater role when demand drives marginal output changes.

REFERENCES

Alfsen, K. H., Lorents Lorentsen and Karine Nyborg (1990). "Environmental Effects of a Transition from Oil and Coal to Natural Gas in Europe." In Recent Modelling Approaches in Applied Energy Economics, Bjerkholt, O., O. Olsen and J. Vislie (editors). Chapman & Hall.

Bacon, R., (1992). "Measuring the Possibilities of Interfuel Substitution." Working Paper Series No. 1031, Policy Research Department, The World Bank, November.

Balcer, Yves, (1980). "Taxation of Externalities, Direct versus Indirect." Journal of Public Economics 13: 121-29.

Bhatia, Ramesh (1987). "Energy Demand Analysis in Developing Countries: A Review." The Energy Journal, Vol. 8, Special L.D.C. Issue.

Bjerkholt, Olav, O. Olson and J. Vislie (1990). Recent Modeling Approaches in Applied Energy Economics, New York and London: Chapman & Hall.

Boyd, R., K. Krutilla and W. Kip Viscusi (1995). "Energy Taxation as a Policy Instrument to Reduce C[O.sub.2] emissions: A Net Benefit Analysis." Journal of Environmental Economics and Management 29: 1-24.

Christensen, L. R., D. W. Jorgenson, and L. J. Lau (1971). "Conjugate Duality and the Transcendental Logarithmic Function." Econometrica 39(July): 206-207.

Duchin, Faye and Glenn-Marie Lange (1994). The Future of the Environment, Ecological Economics and Technological Change, Oxford University Press.

Eskeland, G. S. (1994). "A Presumptive Pigovian Tax: Complementing Regulation to Mimic an Emissions Fee." World Bank Economic Review 8(3): 374-394.

Eskeland, G. S. and S. Devarajan (1996). Taxing Bads by Taxing Goods: Pollution Control with Presumptive Charges. The World Bank, Directions in Development Series, Washington, D.C.

Eskeland, G. S., and Emmanuel Jimenez (1992). "Policy Instruments for Pollution Control in Developing Countries." World Bank Research Observer 7(2): 145-169.

Eskeland, G. S. and T. Feyzioglu (1997). "Is demand for polluting goods manageable? An econometric model of car ownership and use in Mexico." Journal of Development Economics 53(2): 423-445.

Eskeland, G. S. and T. Feyzioglu (1997b). "Rationing can backfire: The 'Day without a Car' in Mexico City." World Bank Economic Review 11(3): 383-408.

Eskeland, G. S. (1997). "Air pollution requires multipollutant analysis: The case of Santiago, Chile." American Journal of Agricultural Economics 79(5), December.

Fuss, M. A. (1977). "The Demand for Energy in Canadian Manufacturing." Journal of Econometrics 5: 89-116.

Glomsrod, S.T., Johnsen and H. Vennemo (1992). "Stabilization of C[O.sub.2]: A CGE Assessment." Scandinavian Journal of Economics 94(1): 53-69.

Goulder, L. H. (1995). "Effects of Carbon Taxes in an Economy with Prior Tax Distortions: An Intertemporal General Equilibrium Analysis." Journal of Environmental Economics and Management 29(3): 271-97.

Guo, C. and James R. Tybout (1994). "Relative Prices and Manufacturing Fuel Use Patterns: Plant-Level Evidence from Chile." Policy Research Working Paper No. 1297, Policy Research Department, The World Bank.

Hsiao, Cheng (1986). Analysis of Panel Data. New York, NY: Cambridge University Press.

Hughes, Gordon (1987). "The Incidence of Fuel Taxes: A Comparative Study of Three Countries." In The Theory of Taxation for Developing Countries, David Newbery and Nicholas Stern (editors), Oxford University Press: 533-559.

Jaffe, Adam B. and Robert N. Stavins (1994). "Energy-Efficiency Investments and Public Policy." The Energy Journal 15(2): 43-65.

Jorgenson, D. W. and Peter J. Wilcoxen (1993). "Reducing US Carbon Dioxide Emissions: An Econometric General Equilibrium Assessment." Resource and Energy Economics 15: 7-25.

Kaufmann, Robert K. (1994a). "The Effect of Expected Energy prices on Energy Demand: Implications for Energy Conservation and Carbon Taxes." Resource and Energy Economics 16: 167-188.

Kaufmann, Robert K. (1994b). "The Relation Between Marginal Product and Price in US Energy Markets: Implications for Climate Change Policy." Energy Economics 16(2): 145-158.

Kopp, Raymond, (1992). "Economic Incentives and Point Source Emissions: Choice of Modeling Platform." Working Paper No. 920, Policy Research Department, The World Bank, June.

Lee, Kyu Sik and Alex Anas (1992). "Cost of Deficient Infrastructure: The Case of Nigerian Manufacturing." Urban Studies 29(7): 1071-92.

Moss, Diana L. and James R. Tybout (1994). "The scope for fuel substitution in manufacturing Industries: A Case Study of Chile and Colombia." The World Bank Economic Review 8(1): 49-74.

Newbery, David and Nicholas Stern (1987). The Theory of Taxation in Developing Countries. New York and London: Oxford University Press.

Opschoor, J.P. and Hans Vos (1989). The application of economic instruments for environmental protection in OECD member countries. Paris: Organization for Economic Cooperation and Development.

Ostro, Bart (1994). "Estimating Health Effects of Air Pollution: A Methodology with an Application to Jakarta." Working Paper No. 1301, Policy Research Department, The World Bank.

Ostro, B., J. M. Sanchez, C. Aranda and G. S. Eskeland (1996). "Air Pollution and Mortality: Results from a Study of Santiago, Chile." Journal of Exposure Analysis and Environmental Epidemiology 6(1): 97-114.

Pindyck, R. (1979). The Structure of World Energy Demand. Cambridge, MA: MIT Press.

Pitt, Mark M. (1985). "Estimating Industrial Energy Demand with Firm-level Data: The Case of Indonesia." The Energy Journal 6(2) 25-39.

Russell, Clifford S. (1990). "Monitoring and enforcement." In Public Policies for Environmental Protection, Paul Portney (editor). Resources for the Future, Washington, D.C.

Sandmo, Agnar (1976). "Direct versus Indirect Pigovian Taxation." European Economic Review 7: 337-49.

Stephan, Gunter, Renger van Nieuwkoop, and Thomas Wiedmer (1992). "Social Incidence and Economic Costs of Carbon Limits: A Computable General Equilibrium Analysis for Switzerland." Environmental and Resource Economics 2: 569-591.

Ten Kate, Adriaan (1993). "Industrial Development and the Environment in Mexico." Working Paper Series No. 1125, Policy Research Department, The World Bank, April.

United States Environmental Protection Agency (1986). AP-42. "Compilation of Air Pollutant Emission Factors." Supplement A, October, Volume 1, Stationary Point and Area Sources." Research Triangle Park, N.C., USA.

Viscusi, K., W. Magat, A. Carlin and M. Dreyfus (1994) "Environmentally responsible energy pricing." The Energy Journal 15(2): 23-42.

Weaver, Christopher, S. and M. J. Reale (1994). "Development of Composite Fuel-specific Emission Factors for Air Pollutants in Java, Indonesia and Santiago, Chile." Sacramento, CA: Engine, Fuel and Emissions Engineering, Inc., mimeographed.

Whalley, J. and R. Wigle (1991). "Cutting C[O.sub.2] Emissions: The Effects of Alternative Policy Approaches." The Energy Journal 12(1): 109-24.

Wijkander, Hans. (1985). "Correcting Externalities through Taxes on/Subsidies to Related Goods." Journal of Public Economics 28: 111-25.