econometric modelling of international carbon tax regimes

~UTTERWORTH

0140-9883(95)00009-7

Energy Economics, Vol. 17, No. 2, pp. 133-146, 1995 Copyright © 1995 Elsevier Science Ltd

Printed in Great Britain. All rights reserved 0140-9883/95 $10.00 + 0.00

Econometric modelling of international carbon tax regimes

Clare Smith, Stephen Hall and Nick Mabey

An econometric model of fossil fuel demand has been estimated for eight OECD countries, relating coal, oil and gas demands to GDP and prices. In addition a model of endogenous technical progress has been estimated, aiming to include both price induced innovation in energy and structural change in the economy as long-term determinants of energy consumption. A number of possible international carbon/energy tax agreements are simulated, showing the impacts on carbon dioxide emissions and comparing the two models. In the long run the magnitude of taxes required to stabilize or reduce emissions would be large, and there are important differences between countries in price responses, which must be taken into account in international modelling and policy formulation. Keywords: Energy demand; Cointegration and error correction; Carbon taxes

International policies to reduce carbon dioxide emissions may have major macroeconomic impacts, which EGEM, a macroeconomic/energy model, is de- signed to simulate. In this paper, an energy demand model is used to compare the impacts of carbon and energy taxes on CO 2 emissions in eight OECD countries.

E G E M consists of an econometric model of energy demand, integrated into an international macroeconomic model (GEM: Global Econometric Model). GEM has been developed and maintained jointly by the London Business School and the National Institute for Economic and Social Re- search (NIESR). Energy sector equations were estimated for total fossil fuel energy consumption, which is then divided into three shares for coal, gas and oil, as functions of energy prices, GDP, a time trend and other determinants. In each case cointegration tech-

The authors are with the London Business School, Sussex Place, Regents Park, London NW1 4SA, UK.

Financial support from the ESRC Global Environmental Change Programme (grant no. Y320 25 3010) is gratefully acknowledged. This paper was initially presented at a conference on The Economics of Global Environmental Change, at the University of Birmingham, 9-11 May 1994.

Final manuscript received 19 December 1994.

niques were used to establish the long-term equilibrium relationships. Error correction models (ECMs) were then estimated, using the residual from the long-run relationship in a dynamic form, with the coefficient on the residuals representing the rate of adjustment towards the cointegrating equilibrium. This specification has the advantage of long-term stability together with rich dynamic structure.

An alternative formulation of the fossil fuel consumption equation was also estimated, including an endogenous technological change term in place of the deterministic trend. This model allows the effects of price induced technical innovation and structural economic change to be included, allowing the impacts of policies on the decline in energy intensity in the long term to be assessed. It is shown that the long-run effects of a carbon tax in stimulat- ing innovation in energy efficiency significantly increases the impact of the tax over the reduction due to direct demand elasticity.

The complete version of E G E M consisting of augmented G EM and the energy model contains the major effects and feedbacks of the energy sector on the rest of the economy. The strong empirical foun- dation of E G E M makes it ideal for investigating the economic effects of carbon taxes and similar policy instruments. Impacts on GDP, trade, inflation and

133

Econometric modelling of international carbon tax regimes: C Smith et al

other macroeconomic indicators will be forecast as well as CO 2 emissions up to 2030. Quantifying the differences in price response and in macroeconomic impact between countries allows assessment of incentive compatibility in multilateral negotiations. One of the main aims of the project is to look at the effectiveness of international agreements given these differences and to formulate sustainable policy agreements which take heterogeneity into account.

Description of the energy model of EGEM

For each country studied, the analysis is divided into two parts: aggregate fossil fuel consumption, and a system of fuel shares. Two possible models are developed for aggregate energy consumption: the first includes an exogenous trend in energy intensity while the second attempts to endogenize this trend. Quar- terly data from 1978-90 were used in estimation. Full details of the econometric analysis including results and significance statistics are available in Boone et al [6] and Hall et al [11].

Aggregate energy analysis using an exogenous trend

A two-state procedure was followed, involving a long-run cointegrating equation and a dynamic relationship in the form of an error correction model (ECM). The Granger representation theorem (Engle and Granger [9] ) states that if a set of I(1) variables are cointegrated (so that the resulting residual is stationary) then there is a valid error correction representation. A range of diagnostics were used to test the order of integration of the variables and it was concluded that logs of price and energy intensity could be considered I(1).

Initially, the Johansen procedure (Johansen [15]) was used to identify cointegration vectors between logs of fossil fuel intensity (demand per unit GDP) and a weighted average real fuel price. Generally four lags were included in the VAR, corresponding to the quarterly data, although some countries required up to eight to give a good result. A time trend was also included as a proxy for technical progress and structural change, by regressing the cointegrating vector on a time variable using OLS. The cointegrating vector together with the time trend gives a long run relationship between energy demand, GDP and price (1).

In (C , /Y t) = a + b. ln(RP t) + c.Z t + ~t (1)

where C = fossil fuel consumption, Y = GDP, RP = weighted average real fuel price, T = time and is a residual.

A dynamic relationship was then estimated in an ECM. Change in fossil fuel consumption is re- gressed against the residuals from (1), lagged once (the error correction term), together with lags of differences of the other variables as shown in Equa- tion (2). Generally up to four lags in each were included at the outset and retained according to significance:

A I n ( C t ) = a + ~'£:t-1 -If- E "Yi A l n ( c ) t - i i = 1

+ ~, ~iAln(RPt-i) i = 0

+ ~ 4~iAln(Yt_i) (2) i=o

Including the residual ~t allows past forecasting errors to be taken into account in forecasting the next change in fossil fuel consumption. The coefficient on this error correction term indicates the rate at which consumption adjusts towards the long-run equilibrium.

In some countries two or more significant cointegrating vectors were found, in which case residuals from both were included in the ECM, and the most significant retained. In three cases (Japan, Canada and the Netherlands) one of these embodied a positive relationship between price and demand, ie increasing demand increases prices. The lagged residual from these relationships, as expected, turned out to be non-significant in the ECM, indicating that causality from prices reduces energy consumption through the negative relationship.

The results from the long-run analysis, summarized in Table 1, show a negative price elasticity was found in each case although the values are fairly low. All countries show strong declining time trends in energy intensity.

Estimating an endogenous technical progress model

The time trend in the above model is taken to represent technical progress, analogous to the autonomous energy efficiency improvement (AEEI)

Table 1 F o s s i l f u e l i n t e n s i t y e l a s t i c i t i e s w i t h r e s p e c t to p r i c e a n d t i m e t r e n d s

Country P r i c e T i m e (% i ra)

C a n a d a - 0 .101 - 2 .17 F r a n c e - 0 .147 - 5 .67 G e r m a n y - 0 .063 - 3 .12

I t a l y - 0 .130 - 2 .19 J a p a n - 0 .133 - 4 .06 N e t h e r l a n d s - 0 .085 - 1.61 U K - 0 .045 - 2 .58

U S A - 0 .159 - 3 .09

1 3 4 Energy Economics 1995 Volume 17 Number 2


used in other models, a term coined by Manne and Richels [16]. Technological improvement and increasing energy productivity provide the possibility of reducing energy use in the long term, without increasing costs. Expressing this rate as an exogenous factor severely limits the model's ability to represent how policy or macroeconomic changes may affect technological development. Studies to mea- sure the AEEI empirically have been relatively few and the results inconclusive (Hogan and Jorgenson [14] ).

Decline in energy intensity may be due to several factors: structural change in the economy; the development of non-fossil fuel sources; price-induced innovation and non-price policy interventions, in addition to autonomous technical progress. As the time trend is substantial, it is important to refine the model to endogenise some of the determinants of this trend.

To estimate this endogenized trend, a model analogous to the error correction representation above was supplemented by an endogenous trend term, including determinants such as non-fossil fuel supply, manufacturing component of GDP, trade, and investment. Price is also included as high prices may stimulate innovation. In addition there is a stochastic trend, which is taken to represent the autonomous improvement in technology. The model was estimated using the Kalman filter, as described by Cuthbertson et al [7]) expressed as Equations (3a)-(3c):

Measurement equation

Aln(C)t = ot + T t + f l l l n ( C / Y ) t _ l

+ fl21n(RP)t-i + ]~.,%kAln(Xi,t_k) i,k

(3a)

where C = energy consumption, Y-- GDP, RP = weighted average real fuel price, AIn(X) -- lags of differences in consumption, GDP and price, and T is a trend (below).

In most cases the dependent variable used was total energy intensity (as opposed to the fossil fuel intensity in the initial model), including nuclear and hydro expressed as its fossil fuel replacement value. Although (by definition) increases in non-fossil fuel must be accompanied by decreases in fossil fuel for a given energy demand, this relationship was not always adequately represented by including the change in non-fossil fuel consumption in the endo- gonous trend. Using total energy consumption allows fossil fuel consumption to be calculated by subtracting exogenous forecasts of non-fossil fuel production.

It can be seen that this model subsumes the conventional elasticity plus trend model above, with the additional price term implying long-term irreversible improvements in energy efficiency as a result of higher prices. In a simple demand elasticity model, if the price were to be raised and later reduced, consumption would return to its original level; with endogenous technical change, this is not the case. The model has been estimated for the eight countries- results are summarized in Table 2, together with implied elasticities calculated by simu- lation as the percentage change in energy consumption caused by a 1% tax. As shown, this value depends upon the length of time the tax has been in place, as with endogenous technical change the elasticity continues to increase as long as the price remains high. This could be likened to the conventional short- and long-term elasticities, although the functional form is in fact quite different.

The main Z variable in the model was industrial production as a proportion of GDP, which was found to be significant and have the expected sign (energy consumption increasing with industrial production) for all countries except France and Germany; in the UK manufacturing was used instead. For the USA business investment was found to have a small negative correlation. However for the purpose of this paper, these variables have not been altered in simulations.

Transition equations

Tt = Tt-1 + g t - 1 + T I n ( R P ) t - 1 + ~ , r l i Z i i

gt = g t - 1

(3b)

(3c)

where g = stochastic trend and Z = determinants of endogenous trend, as discussed above.

Table 2 P r i c e c o e f f i c i e n t s i n endogenous technical change model

Country Price coefficient Implied elasticity Direct Trend 5 years 35 years

C a n a d a - 0 . 1 7 1 - 0 . 0 0 2 5 3 - 0 . 2 6 8 - 0 . 3 9 3

F r a n c e - 0 . 0 8 5 - 0 . 0 0 1 9 9 - 0 . 1 1 5 - 0 . 5 0 7

G e r m a n y - 0 . 2 4 8 - 0 . 0 0 2 3 1 - 0 . 3 0 3 - 0 . 5 7 0

I t a l y - 0 . 0 4 4 - 0 . 0 0 0 3 9 - 0 . 0 5 0 - 0 . 1 0 7

J a p a n - 0 . 0 1 6 - 0 . 0 0 2 3 8 - 0 . 0 9 5 - 0 . 6 4 5

N L - 0 . 0 8 1 - 0 . 0 0 2 4 9 - 0 . 1 2 7 - 0 . 4 3 1

U K - 0 . 0 4 5 - 0 . 0 0 5 1 5 - 0 . 1 2 6 - 0 . 7 1 3

U S A - 0 . 1 8 1 - 0 . 0 0 2 4 6 - 0 . 2 2 4 - 0 . 4 6 0

Energy Economics 1995 Volume 17 Number 2 135


Fuel shares

Consumption of coal, gas and oil were defined as shares of fossil fuel demand by their primary energy values. These are expected to respond to changes in their relative prices, and to changes in total energy demand. Additional determinants include changes in GDP and non-fossil fuel supplies. Fuel markets are complex: the three fuels are not wholly substi- tutable, and the fuel mix will be influenced by factors such as industrial mix, technological change, regulatory changes corresponding to policy objec- tives such as security of supply and environmental control, and also discrete events such as price shocks and strikes. Each of these will be of differing importance in each country, and for each fuel.

As for aggregate analysis, a two part procedure was used to estimate the equations, comprising a long run cointegrating relationship and a dynamic error correction model. This model is a modified almost ideal demand system (AIDS), originally due to Deaton and Muellbauer [8].

Long-run analysis proceeded by estimating functions for each fuel share, whereby shares sl, s2, s3 (for coal, gas and oil respectively) at time t are given by:

3

si, t = a i + ~, bijln(pjt/Plt) + ciln(C t) j=2

+ ~ dikln(Zkt) (4) k = l

where Pit, J = 1, 2, 3 are the prices 1 of each fuel, C t is total fossil fuel demand at time t, and the Zkt are a set of n exogenous variables of interest. The Z variables include such factors as GDP, consumption of nuclear energy, consumption of hydroelectricity, and in some cases dummy variables representing various one-off shocks such as strikes and events due to government policy, such as investment pro- grammes in power stations, as shown in Table 3.

The price variables are expressed relative to the price of coal as a restriction to ensure homogeneity in the system ie a rise in the price of oil has the same effect on shares as an equivalent fall in the prices of coal and gas, apart from the effect on the aggregate fuel consumption.

The Equations (4) are estimated separately using ordinary least squares (OLS). Each equation is estimated using the same set of variables. This is re-

1 Fuel prices are in domestic currency per tonne of oil equivalent, while demands are in tonnes of oil equivalent.

quired as the sum of the shares ~,Sit = 1, and so there is adding up across the share equations ie:

E a i = 1, ~. ,bi j = O, ~ . ,c i = O, ~_,dik = 0 (5) i i i i

Note that the system is singular ie only two of the equations can be estimated independently. Equa- tions for gas and oil were estimated, while coal's share was then obtained via (5). As the three shares by definition add up to one, this procedure has no effect on the coefficients obtained or their significance, as mathematically the calculation is identical whichever fuel is left out of estimation.

Estimation of Equation (4) is carried out with the aim of finding stable long-run relationships ie where both sides cointegrate together. Therefore it is desirable to obtain stationary residuals, and this was tested for using a number of diagnostics. It is clearly highly desirable a priori to obtain negative own price coefficients, and the best relationships were sought which embodied these. In some cases this presented difficulty because of the distortions in the energy markets under consideration, which was countered by inclusion of dummy variables as discussed above, for instance in the UK dummies for the miners' strike of 1983-85 explained the large drop in coal's share and increase in oil.

In the long-run relationships estimated, the oil share is found to have a significant negative own- price coefficient everywhere, usually with a value of 0.1-0.2. For gas the coefficients are lower, but all significant with the exception of Canada, where the gas price was not significant for any fuel share, and Japan, where the own-price coefficient is not significant for gas. Cross-price coefficients generally are positive and smaller than own price. Coal is the least price-responsive, generally with coefficients under 0.1, insignificant in the case of France, the UK and the USA, and negative with signs in some cases. Total consumption was significant in all countries but with no consistent pattern, although in five out of the eight countries the share of oil increases with increasing consumption and has the largest coefficients; gas is usually negative and has lower coefficients. GDP was significant in Italy, the UK and the USA. A time variable was used in Canada and the UK, and an S-shaped ogive in Japan, in each case correlated with the increase in gas use.

Residuals from the long-run relationships are fit- ted into an ECM framework to produce a dynamic model, analogous to the ECM in the aggregate case, but consisting of three equations, one for the change in each share.

136 Energy Economics 1995 Volume 17 Number 2

Econometr ic modelling o f international carbon tax regimes: C Smith et al

Table 3 Summary of fuel shares equations, long-run relationships"

Country Fuel Price and consumption coefficients Other explanatory share variables

i n ( p 2 / P t ) l n ( p s / P l ) I n ( C )

C a n a d a Coal - - 0.049 - 0.056 T i m e G a s - - 0.008 b - 0.274 D u m m i e s in 1978 and Oil - - - 0.057 0.330 1989

F r a n c e Coal 0.036 b 0.029 b 0.067 b D u m m y in 1986-88 G a s - 0.271 0.169 - 0.366 Oil 0.235 - 0.198 0.299

G e r m a n y Coal 0.058 0.071 - 0.278 Nuclear , hydro G a s - 0.062 0.093 - 0.057 D u m m y in 1981 Oil 0.005 b - 0.164 0.334

I taly Coal - 0.048 0.063 - 0.187 Y G a s - 0.072 0.036 b - 0.105 D u m m i e s in 1980-81, Oil 0.121 - 0 . 0 9 9 0.292 1982, 1984-86

J a p a n Coal - 0 . 0 1 3 0.046 0.069 Ogive f rom 1978-90 G as - 0.122 b 0.152 - 0.063 Oil 0.135 - 0.198 - 0.006 b

Ne the r l ands Coal 0.102 0.009 b - 0.011 b D u m m i e s in 1983-85 and G as - 0.153 0.060 0.062 1988 Oil 0.051 - 0.069 - 0.051

U K Coal - 0.016 b 0.222 0.251 Y, t ime G a s - 0.062 - 0.061 - 0.025 b D u m m i e s in 1984 and Oil 0.078 b - 0.161 - 0.226 1983-85

U S A Coal 0.126 b 0.032 - 0.328 Y, hydro G as - 0.014 0.039 - 0.039 D u m m y in 1986 Oil 0.002 b - 0.071 0.367

a D u e to add ing up, each g roup of th ree m u s t s um to zero; bdenotes coeff icient not statistically significant; Pi = price of fuel i; (1 = coal, 2 = gas, 3 = oil); C = total fossil fuel expendi ture .

However this is not a single equation but a system, which obeys adding up constraints and exhibits singularity. Hence it is not possible to estimate error correction equations for each share independently, nor is it possible to identify the dynamics or the adjustment coefficients of the system (Anderson and Blundell [1]; Barr and Cuthbertson [4]). Adding up implies that

Asl t + A s 2 t + As3 t = 0 (6a)

and

resl t + res2 t + res3 t = 0 (6b)

for all t, where resl is the residual from the long term cointegrating equation in s l etc.

The resulting estimated system must obey (6a), which implies that all equations contain the same set of exogenous variables on the right-hand side, with coefficients that sum to zero. Furthermore (6b) implies that the sum of the adjustment coefficients y~:

(the coefficient on res i in equation As~) down each column must be the same for each column:

3 3 3

i = l i = l i = 1

(7)

The most general specification for the residuals is

therefore to choose two, eg resl and res2, to appear in each of the equations for estimation. (This is equivalent to choosing three, as res3 = - ( r e s l + res2), but avoids having a linearly dependent set of independent variables.) The matrix (Yii) of adjustment coefficients obtained is therefore (3 × 2).

It would also be possil~le to choose just one residual in each equation, eg resl in the equation for Asl and so on. This is a more restrictive approach, yielding a matrix of coefficients that is diagonal with equal elements 3/11 -~- Y 2 2 = "Y33, all negative ie re- quiring that the shares adjust towards equilibrium each at the same rate. However, we have chosen the more general method, with resl and res2 in each equation, allowing adjustment at different rates.

For stability in the estimated dynamic system, we require that the eigenvalues of the 2 × 2 matrix (y#) (i, j = 1, 2) be negative. This is obtained if and only if

'~11 + Y22 < 0 (8a)

and

Y11'~22 - - ")/21")/12 )~ 0 (8b)

A similar argument applies in the case of the lagged dependent variables As( i ) t_ i. In this case we have


Econometric modelling o f international carbon tax regimes: C Smith et al

chosen to restrict the dynamics of each share to lags of itself eg A s l t _ i only are included in the equation for AS1 t and so on, which implies each will have the same coefficient to ensure adding up.

The resulting system of three equations is:

Asy,t = ot 1 + yy,lreslt_ 1 + 3,j,2res2t_ 1 m r n

+ E ,Asj,,_i + E/32j,,arp2,_i i = 1 i = 0

n ,/'/1

+ /33i,iArp3,_ i + ~.. ~_. ~j,k,iAZk,t-i i = 0 k = l i = 1

(9)

where each equation has the same dynamic structure and the same set of AZk . The system is estimated jointly using a non-linear least squares algo- rithm. All coefficients on the residuals, the relative prices and the independent variables add to zero down the columns. The coefficients on the lags of the respective dependent variables are the same for each share.

Fuel prices

Prices for the three fuels, net of tax, were compared with movements in the world oil price, expressed in domestic currency. A relationship for each fuel price was estimated (10):

eL f , t = ot + f l .WOL t + •TiZi , t ( 1 0 )

where e l f , t = log of net price of fuel f ; WOL = log of world price of oil; other variables included in Z may-include exchange rate, domestic price level (GDP deflator) and GDP, all in logs. Dummy variables may also be included. Stationarity of residuals was tested as an indicator of long-term stability.

For the price of oil in the majority of countries, the estimated equations show the required charac- teristics. The price of oil in each country depends strongly on the world price, with other factors ac- counting for short-term fluctuations and the smoothing behaviour of oil companies, contracts, domestic sources of oil etc. As expected the coefficients for the world oil price, /3, are positive and fairly close to one, and this is the most important explanatory variable. The use of domestic prices means the domestic price level is also likely to be significant and positive. Exchange rates and GDP are also significant in some cases.

For gas and coal, other country specific factors, particularly the GDP deflator, tend to be more important than for the oil price equation although world oil price is still a major factor. The coefficient /3 would be expected to be less significant and lower

than for oil but must be positive. However in some cases no relationship was found with the oil price: in Canada for both gas and coal, and in the USA for gas. This is thought to be because North American gas (and to a lesser extent coal) produced domestically is not closely linked to world markets, but prices are determined internally. By contrast, in Europe contracts linking gas prices to that of oil, and competition in electricity generation markets, provide stronger links with the oil price. Also in some cases the relationships include world price lagged by one year, indicating inertia in the response to world oil price fluctuations. In many cases, the relationships are very similar for gas and coal, as their prices are generally determined by similar factors.

Dummy variables were used to account for particular discrepancies. In particular in 1986, the collapse of the world oil price was not fully passed on to the consumer, and the average price paid after 1986 continued to show a significantly increased differen- tial over the world price. Gas and coal, especially where domestically produced, are susceptible to a range of policy factors affecting their prices.

Consumption levels and depletion are not taken into account, as they were not found to be significant in the estimation, nor considered likely to become so within 35 years although gas in particular may become supply constrained in the very long term. Future prices use a forecast of world oil price in GEM based on world export prices. The model leads to prices generally increasing at slightly more than the domestic inflation rate, although in some cases the coal price falls in real terms.

Energy use and carbon taxes

The energy model developed can be used to look at the impacts of different taxation strategies on fuel mix and carbon mix and carbon dioxide emissions. The fuel shares model provides a means of looking at interfuel substitution, and its importance compared with overall energy conservation. Carbon and energy taxes can be compared, along with other tax regimes such as applying the same percentage rates of tax in each country.

A carbon tax would be expected to be more effec- tive in reducing emissions than an energy tax, as it more accurately internalizes the externality. To compare carbon and energy taxes, two scenarios were run, corresponding to the EC proposal of US$3/bbl in 1995 rising to US$10/bbl by 2002 (in real 1990 prices), but in the first scenario wholly raised on energy and the second wholly on the carbon content of the fuel.



Many factors determine the overall elasticity of energy. Income levels are likely to play a part, together with the structure of the economy. Actual price levels vary widely between countries due to differences in existing taxes, which in the case of a carbon or energy tax implies that the percentage tax rate is higher in low-cost countries (USA, Canada) than those with high prices (Germany, Japan). The level of government intervention in such areas as efficiency regulation also varies widely. In a country where energy use is currently high and inefficient, there may be many low-cost opportunities for reductions, while countries that currently have lower energy intensity may find it harder to make further improvements.

Part of the difference between countries' price responses may be explained by the differences in prices and existing taxation in different countries (Hoeller and Coppel [13] ). In order to assess the magnitude of this effect, an ad valorem tax scenario was constructed in which the same percentage rate of tax is levied in each country, at a level giving the same total revenue overall, in the proportions on each fuel as for a 50 /50 carbon/energy tax in the UK. This translates to 20.8% on oil, 28.47% on gas and 83.95% on coal.

Table 4 compares tax rates in the UK (compara- ble with most of Europe) and the USA, to illustrate the differences between scenarios and countries.

Table 4 Prices and taxes on fuels in the three scenarios

Price Tax (%) £($) / toe E tax C tax Ad val

UK coal 116 56.9 61.6 84.48 gas 269 24.5 15.2 28.62 oil 428 15.4 13.8 21.03

US coal 57 157.9 186.0 84.48 gas 267 33.7 22.8 28.62 oil 274 32.8 32.1 21.03

The results from these three scenarios are compared in Figure 1, using the error correction model. These values represent the percentage reduction in CO 2 emissions in the year 2030 as compared to the base case value.

The levels of abatement are divided into conserva- t i o n - the amount caused by the absolute reduction in energy consumpt ion- and substitution, caused by the shift in fuel mix from coal to gas. It can be seen that there are large differences between countries, both in the amount reduced and the role of interfuel substitution. As expected, for carbon or energy taxes, low prices in the USA and Canada lead to these countries having the most abatement, with Japan and Germany having the least.

Canada, the Netherlands and the UK are the three major producers of natural gas among the eight countries, and they all show a significant reduction from substitution of gas for coal or oil,

8

o 7 o

6

5 > O 4

o 3 4--J ¢.A

2 GA

~" 1 cQ O 0 (D

-1

Figure 1

e c v e c v e c v e c v e c v e c v e c v e c v e c v

CN FR GE IT JP NL UK US TOT

Country & tax scheme

/ C o n s e r v a t i o n ~ S u b s t i t u t i o n

Impact of three tax strategies on carbon abatement



although this is subject to the perfectly elastic supply assumed in the price equation. France and Italy also show a high level of substitution, in this case away from coal. Germany, Japan and the USA, however, actually show negative substitution, indicating a shift away from gas and into oil, without a great reduction in coal. In the case of Germany this is probably due to the lack of price response in the coal industry due to government subsidy. In Japan, coal is relatively minor at present and gas is expensively imported as LNG: the system is highly constraified by their lack of resources and government intervention, again reducing the response to price.

Substitution between fuels is not always in the direction expected due to differences in their price levels, and in responses to price. For instance, for the USA forecast in 2000, the prices of oil, gas and coal are US$205, US$186 and US$54/toe respectively (in US$1990). The combined carbon/energy tax level in 2000, once the carbon contents of the fuel have been applied, works out at US$33, US$23 and US$40/toe for the three fuels respectively, re- suiting in additional tax rates of 32% on oil, 30% on gas and a massive 135% on coal. One would expect coal to lose some of its market share to both oil and gas. However, the analysis demonstrates that the price elasticity of gas is less than that of oil, with respect to either's price relative to coal. Hence oil takes over most of the share lost by'coal, while due to the overall decrease in fossil fuel consumption, the consumption of gas actually falls.

As compared to an energy tax, a carbon tax does not necessarily imply a gleater reduction in carbon emissions although it will be more efficient in terms of cost per tonne carbon removed. In the case of an agreement consisting of an international tax in addition to each country agreeing to some stabilization target, it can be seen that some countries would favour a carbon tax while others would benefit from an energy tax on the basis of carbon abatement. However this omits any consideration of the revenues raised from the tax, which will be different for each scenario.

The third ad valorem tax scenario was included in order to explain the extent to which differences in prices between countries determines the response. It can be seen that the response is rather more. uniform as would be expected, with Canada and the USA reduced to a similar level to Europe, and Germany and Japan increasing. This implies that large differences between these countries seen in the carbon/energy tax scenarios is more due to differences in price than behaviour or technical limitation.

Due to differences in fuel mix, consumption level

and price, the tax burden imposed by each scenario will vary according to country. Table 5 shows the tax revenue per unit of GDP for each country in 1995; these rates will increase by a factor of 3.3 between 1995 and 2002, as the tax rates increase, and decline as energy intensity declines thereafter. The total tax revenue for 1995 and 2030 is also shown, in constant US$1990 prices.

For the energy tax scenario, the rates reflect the fossil fuel intensity of each country. Those countries with a lower fossil fuel carbon intensity (ie those using more gas and less coal) will show a decreased revenue with a carbon tax, such as Canada, the Netherlands and the UK. The ad valorem tax will be lowest where current fuel prices are low, such as the USA.

Total revenue, while broadly similar for the three scenarios, changes considerably over time as each tax has a different effect on fuel mix and consumption. Thus it is difficult to compare directly the carbon abatement per dollar of tax from the three scenarios. The magnitude of the revenue reinforces the importance of the 64 billion dollar question of how these are recycled.

The above results, however, do not take into account the endogenized trend in energy intensity. The second model, estimated included a term allowing price to increase the rate of technical progress, would be expected to forecast larger reductions in emissions in the long run. Results from the two models are compared in Figures 2 and 3, for abatement in 2005 and 2030. A base-case forecast is compared with a scenario corresponding to the tax of US$3/bbl in 1995 rising to US$10/bbl in 2002, split 50/50 between energy and carbon.

Comparing the two models, it can be seen that endogenizing the trend does in most cases imply a greater potential level of carbon abatement. The models are broadly similar in 2005 but the difference becomes greater in the longer term. As the tax level reaches its maximum in 2002, for the error

Table 5 E n e r g y / c a r b o n tax revenue as a percentage of GDP and total for the eight countries (1995)

Country Energy Carbon Ad valorem tax tax tax

Canada 0.74 0.61 0.58 France 0.18 0.16 0.27 Germany 0.30 0.30 0.61 Italy 0.30 0.28 0.41 Japan 0.14 0.13 0.22 Netherlands 0.66 0.58 0.67 U K 0.53 0.42 0.66 USA 0.50 0.57 0.42 Total tax revenue for the eight countries (US$ million 1990) 1995 64 120 65 104 70 058



0 0

~S

~J o

(1)

~3 ,.Q

O

20

15

10

. L, EEEE, GE IT JP NL UK US TOT

C o u n t r y

iF CN FR

ECM ~ E n d o g

Figure 2 Comparison of error correction and endogenous technical progress models forecasts for CO 2 abatement in 2005

correction model there is little difference between the abatement level in 2005 and 2030, apart from some longer term adjustment to the higher price level. For the endogenous model however, the

intervening 25 years at the high price level brings about continued reductions in energy intensity.

Figures 4-6 show the effect of the tax in more detail on aggregated emissions in Europe, North

0 CO 0 0,/

QS

o

[/3

E O %

20

15

I 0

c~ CN FR GE IT JP NL UK US TOT O

C o u n t r y

10

ECM ~ E n d o g

Figure 3 Comparison of error correction and endogenous technical progress models forecasts for CO 2 abatement in 2030



950

9 0 0

r,_) -~ 8 5 0

o ~ 800

if/ •

7 5 0

o2 0 700 0

6 5 0

6 0 0

- 0 . 9 5

J

CO /,oe

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 . 7 5 ' 86 ' 90 ' 95 2 0 0 0 ' 1 5 ' 2 0 ' 2 5 ' 30

-0 .9

-0 .85

-0.8

' 0 5 ' 1 0

Y e a r

Figure 4 C O 2 emissions lands, U K )

B a s e c a s e - - T a x

and intensity in E C 5 ( F r a n c e , Germany, Italy, Nether-

o

CO

o o

o2 O o

America and Japan, using the endogenous technical change model. In each case emissions continue to r i s e - a tax of this size alone cannot stabilize or reduce emissions in the long run. The abatement relative to the base case is greatest in North Amer- ica, followed by Europe, and quite small in Japan. The relationship between energy consumption and CO 2 emissions is illustrated by the second line on

each graph, representing the emission per toe of fossil fuel consumed, which depends on the fuel mix. In Europe, there is a significant decline in this ratio, brought about by the increased use of natural gas. In North America and Japan the figure increases, implying substitution away from gas.

These changes can be clarified by looking at the changes in consumption of each fuel, as shown in

r_)

1.8

1 .75

1 7

~- 1 .65

o o 1.6

i~.55 [,,,.,

C,/ ~ 1 5 0

1 .45

1.4

1 .35

Figure 5 Canada)

~ , - , ~ _ ~[0 .95

[ - ~ C02/toe

I [ I I I I I r I 1 1 1 I I I I I [ ] 1 1 ] [ 1 1 ] ] I ] I T [ ~ [ ~

' 86 ' 90 ' 95 2 0 0 0 ' 05 ' 10

Y e a r

0.9

0.85

-0.8

0 ,75 '15 ' 20 '25 '30


C O 2 emissions and intensity in North America (USA and

0

\ L)

0

0 (9

~.~

cg o r.p



500

450

4OO

m o w ~ 3 5 0

300 o

/

~ U ~ / L O e

250 /

2 0 0 . . . . , , , , , , , , . . . . . . , , , , , , , . . . . , , , rq-r . . . . . . , , , ' 8 6 ' 9 0 ' 9 5 ' 1 5 ' 2 0 ' 2 5 ' 3 0

0.95

0.9

-0.85

- 0 . 8

2 0 0 0 ' 0 5 ' 1 0

Y e a r

0.75

0

\ 4.9

d 0

o3

o (9

4.9

q9

O3 0 0

- - B a s e c a s e - - Tax

Figure 6 CO 2 emissions and intensity in Japan

Figures 7-9. In Europe, as one would expect, coal and oil consumption decrease from their base case levels, with coal showing the greatest reduction, while gas initially increases, returning to its base case consumption by 2030. This shift in fuel mix explains the decreased CO2 intensity per unit of energy. However, in America, the absolute changes are large, and not in the direction expected. Although consumption of all three fuels declines and coal initially

decreases most, gas also shows a large reduction while oil initially increases followed by a smaller decrease, leading to an overall increase in CO 2 intensity per tonne of oil equivalent. In Japan the changes are smaller, and coal and gas are reduced while oil increases. As discussed above, this is due to differences in the prices and cross-price elasticities of the fuels.

A flat rate tax theoretically should be an efficient

500

4 5 0

o 400

o ~ 350

i ~ 3o0 250

200

1 5 0 ' 8 6

O i l

I ] l [ l [ l l l l l l l l l l l ] ~ [ l l l l l l l l l l l l l I l ~ [ l l ~

' 9 0 ' 9 5 2 0 0 0 ' 0 5 ' 1 0 ' 1 5 ' 2 0 ' 2 5 ' 3 0

Y e a r

B a s e c a s e . . . . . T a x

Figure 7 Fossil fuel consumption E C 5 ( F r a n c e , Germany, Italy, Nether- lands, U K )



9 0 0

850

8 0 0

0 '750

d 0

r!. o

r,.)

7 0 0

-£ 6 5 0

~ 6 0 0

G a

f

Coal 5 5 0

500

450

4 0 0 J l l l l l l l l l r l r r l l l l l l l r l l l r r l l l l r l l l ~ w j , ~ 1 1 1 1 1

' 86 ' 90 '95 2 0 0 0 ' 05 ' 10 '15 '20 ' 25 '30

Y e a r


Figure $ Fossil fuel consumption in North America

means of reducing emissions at least cost (Hoel [12]) as compared to setting an equivalent uniform target. 2 Consumers in the long term implement those emission reduction measures costing less per tonne of carbon than the tax rate, so internationally the

2A tax does have other advantages: if seen as correcting an externality it can be set at the marginal damage cost and so achieve the optimum abatement level for welfare maximization without the need to set a target. This may be seen as a drawback as the actual level of abatement will not be discovered until the tax has had its impact.

cost will be minimized. However the heterogeneity of the results above does not appear to tally with this theoretical framework, as the tax is only optimal in the case of perfect international markets (Golombek et al [10]). The differences between countries are in part due to physical differences (resource base, infrastructure, industrial and fuel mix) and in part to others factors such as government policy, consumer behaviour and institutional structure. An efficient international agreement would compensate for the physical differences and

350

300

o 250

o~ ~" 2 0 0

~ 15o

o 100

50

0 '86

/

• C O "

r r ; J l l J J J l J ~ l r ~ l ~ r r I ~ l q l r l l l l j p l l l i F i f r r

' 90 '95 2 0 0 0 '05 '10 '15 '20 '25 '30 Y e a r

- - B a s e c a s e - - T a x

Figure 9 Fossil fuel consumption in Japan



the variations due to consumer preferences, in that the emission reductions would be made where at lowest welfare loss. The differences between countries is to a major extent due to differences in other policies and to institutional barriers- in the structure of energy utilities, taxes and subsidies, efficiency regulation- which may continue to change and are not constrained by any international agreement. This analysis shows that the large differences that currently exist in fuel prices play a major part, which although partly for geographical reasons (resource availability, costs of providing roads paid for by fuel tax) is mainly due to differences in policy. The efficiency gain from a flat rate tax will thus be undermined as it is rather like trying to build a level playing field on top of a range of mountains. Har- monization of energy prices (at a higher rate than at present) would provide a second best solution, as this would reduce market distortions, but physical and consumer differences would be ignored; however it would be impractical politically as it would require domestic policy to be overridden. Setting individual country targets and/or different tax rates in each country may be the most efficient solution that is practically feasible.

Conclusions

Policy makers need to be informed about the macroeconomic implications of emission reduction and how these depend on the policy proposed, in addition to the effectiveness of the policy in meeting CO2 emission targets. The means of revenue recycling has been seen to be of great importance (Barker et al [3] ), and in any non-global agreement there is a possibility of carbon leakage - an increase in carbon emissions from those countries outside the agreement, brought about by changes in trade and/or world oil prices due to the abatement policies as examined by Smith [18]. There is an urgent need for models with a strong empirical base that can help answer these questions without making broad cross- country generalizations. At present, there are a number of mainly general or partial equilibrium models in use, as reviewed by Boero et al [5] and Weyant [19]. These often use estimated values for key parameters taken from other studies or from calibration to the existing (non-equilibrium) economy. While there are advantages to the general equilibrium approach, there is a lack of convergence in results, and many important interactions are being systematically overlooked.

Econometric modelling has the advantage of mea- suring the relationships between economic variables

empirically, which may include many interactions that have not been accounted for in a simple theoretical framework. In a second best world, a model that relies on first best theory may introduce large errors. Energy markets are often highly distorted and price responses vary between both countries and fuels. It has been shown that carbon taxes may have counterintuitive results due to complexities in fuel markets, even in some cases leading to an increase in CO 2 per unit of energy. There are large differences in the responses made in the countries studied which appear difficult to reconcile with the assump- tion behind economic optimality of a tax.

Atkinson and Manning [2] give an excellent review of energy elasticities in the literature, showing wide variations in values obtained according to the functional form, fuel, sector, estimation period and whether cross-country or time series data is used. In this study, values for overall energy elasticity have been found that are considerable lower than the commonly accepted long-run value of around -0.5, chiefly due to the cointegrating technique used, and the specification in terms of fossil fuel intensity. In the past, the main impact of the oil price shocks has been to reduce GDP in oil importing nations, due principally to trade balance effects. This leads to a decrease in energy consumption, which has been wrongly associated with the change in prices. In effect, what these models implicitly assume is that energy demand is only reduced at expense of GDP. As a carbon tax with revenue recycled within the economy will not impose such a large GDP shock, the impact on energy demand will be smaller than external price increases. Using a model formulated in terms of energy intensity as a function of price produces a much lower elasticity as has been found in other studies such as Neuburger [17].

The endogenous technical progress model estimated in this study has a novel functional form that looks at how energy intensity declines as a result of high prices, leading to irreversible improvements in energy efficiency. The long-term outcome of raising energy prices is thus decomposed into a conventional demand response and technological progress. The former comprises both short- and long-run effects as determined by the elasticity and the dynamics of the equation, while innovation is charac- terized by a continuing decrease in energy intensity while prices are high. The model in effect endoge- nizes the autonomous energy efficiency improvement or AEEI used in many models. Any long-term emissions scenario is crucially dependent upon the value of the AEEI, as it specifies directly the relationship between energy use and output. In modelling work its value is generally taken at some fairly



arbitrary value such as 1% pa varying for different world regions. Whether or not the value chosen accurately reflects existing trends, specifying the value exogenously excludes the possibility of technological development being affected by policy.

The double dividend of recycling revenue from a carbon tax to reduce both emissions and distorting labour taxes is now seen as making an increase in GDP possible. Taking energy saving technological change into account may increase GDP further, as if prices act as an incentive for innovation, this provides lasting change in energy productivity which would not be reversed by future price reductions.

Long-run energy elasticities as low as -0 .15 as found here make taxes a rather ineffective means of bringing about the large reductions in energy intensity required for atmospheric CO 2 stabilization. In the longer run, the induced technological innovation will have an increasingly important role, but requires very long planning horizons. Interfuel substitution away from coal and into gas has an important role to play in the short term in some countries, but its effect is limited.

One possible scenario is the recycling of a proportion of carbon tax revenues into energy saving investments, in the form of public sector investment or grants/subsidies to the private sector. Institutional changes and areas such as building regulations, energy labelling and transport policy will also play a major part. The economic advantage would be in overcoming market failures, such as the problems lower income households have in financing energy related investments, in particular when living in rented or public sector housing, and correcting for externalities.

If the major macroeconomic impacts of a carbon tax are seen in terms of changing the tax base of the economy towards a less distorting system, but other additional policies are required to achieve CO2 targets, the perspective of the problem is changed. It is evident that tax distortions will vary between countries, as will the response to a tax and the costs of non-price policies. E G E M will assess the variation in impacts across countries, which will aid in in- forming international policy negotiations and max- imizing the gains from co-operative agreements.

References 1 Anderson, G and Blundell, R 'Consumer non-dura-

bles in the UK: a dynamic demand system Economic Journal 1984 94 (Supplement) 35-44

2 Atkinson, J and Manning, N 'A survey of internatio-

nal energy elasticities' in Barker, T and Ekins, P (eds) Global Warming and Energy Elasticities Routledge, London (1995)

3 Barker, T Baylis, S and Madsen, P The Carbon Tax: E c o n o m i c and Policy Issues E n e r g y - Environment-Economy Modelling Discussion Paper No 3, Department of Applied Economics, University of Cambridge (1992)

4 Barr, D G and Cuthbertson, K 'Neoclassical consumer demand theory and the demand for money' Economic Journal 1991 101 855-876

5 Boero, G Clarke, R and Winters, L A The Macroeconomic Consequences of Controlling Green- houses Gases: A Suroey DOE Environmental Economics Research Series, HMSO, London (1991)

6 Boone, L Hall, S Kemball-Cook, D and Smith, C 'Endogenous technical progress in fossil fuel demand' in Barker, T and Ekins, P (eds) Global Warming and Energy Elasticities Routledge, London (1995)

7 Cuthbertson, K Hall, S G and Taylor, M P T Applied Econometric Techniques, Philip Allan, London (1992)

8 Deaton, A S and Muellbauer, J Economics and Con- sumer Behaviour Cambridge University Press, Cam- bridge (1980)

9 Engle, R F and Granger, C W J 'Co-integration and error correction: representation, estimation and testing' Econometrica 1987 55 251-276

10 Golombek, R Hagem, C and Hoel, M The Design of a Carbon Tax in an Incomplete International Climate Agreement Nota di Lavoro 22.93, Fondazione ENI Enrico Mattei, Milan (1993)

11 Hall, S Kemball-Cook, D and Smith, C Inter-fuel Substitution and CO 2 Emissions in Nine Countries Dis- cussion Paper 22-93, Centre for Economic Forecast- ing, London Business School (1993)

12 Hoel, M Stabilizing CO 2 Emissions in Europe: Indioid- ual Stabilization Versus Harmonization of Carbon Taxes Nota di Lavoro 8.93, Fondazione ENI Enrico Mattei, Milan (1993)

13 Hoeller, P and Coppel, J Energy Taxation and Price Distortions in Fossil Fuel Markets: Some Implications for Climate Change Policy Economics Department Working Paper No 110, OECD, Paris (1992)

14 Hogan, W W and Jorgenson, D W 'Productivity trends and reducing CO 2 emissions' Energy Journal 1991 12 (1) 67-86

15 Johansen, S 'Estimation and hypothesis testing of cointegration vectors Econometrica 1991 59 1551-1580

16 Manne, A and Richels, R G Buying Greenhouse Insur- ance: The Economic Costs of CO 2 Emissions Limits MIT Press, Cambridge, MA (1992)

17 Neuburger, H 'Energy use in an era of rapidly changing oil price: how OPEC did not save the world from the greenhouse effect' Enoironment and Planning A 1992 24 1039-1050

18 Smith, C Modelling Carbon Leakage in Incomplete CO z Abatement Treaties Discussion Paper No DP 26-94, Centre for Economic Forecasting, London Business School (1994)

19 Weyant, J P 'Costs of reducing global carbon emissions' Journal of Economic Perspectives 1993 7 27-46


econometric modelling of international carbon tax regimes

Documents