INTERNATIONAL COMPARISONS OF THE RESIDENTIAL DEMAND FOR ENERGY: A PRELIMINARY ANALYSIS* by Robert S. Pindyck Massachusetts Institute of Technology September, 1976 Working Paper #MIT EL 76-023WP * This work was supported by the National Science Foundation under Grant #GSF SIA75-00739, and is part of a larger project to develop analytical models of the world oil market. The author is indebted to Jacqueline Carson, Ralph Chang, John Donnelly, Daniel DuBoff, Ken Flamm, Ross Heide, Kevin Lloyd, and Eric Rosenfeld for their excellent research assistance in many aspects of this work, to the Computer Research Center of the National Bureau of Economic Research for assistance in some of the computational work, and to Mel Fuss, James Griffen, Edwin Kuh, and Leonard Waverman for comments and suggestions. INTERNATIONAL COMPARISONS OF THE RESIDENTIAL DEMAND FOR ENERGY: A PRELIMINARY ANALYSIS I. Introduction This paper reports on some initial results from an econometric study of the world demand for energy. The long-run objectives of this study are to estimate the determinants of total energy demand and interfuel substitution in the residential and industrial sectors of about twelve industrialized countries. Here we concentrate on the residential sector, and examine some very preliminary estimates of inter-country differences in the structure of demand. In modelling the residential demand for energy we assume that consumers make two decisions in purchasing fuels. First, they decide what fraction of their total budgets will be spent on energy, as opposed to such other consump2 tion categories as food, clothing, etc. Next, with the amount of money to be spent on energy taken as given, consumers decide which fuels to purchase, i.e. the fractions of energy expenditures allottedto oil, natural gas, coal, and electricity. Thus we assume (and we will empirically test this assumption) that consumers' utility functions are separable between energy and other commodities, i.e. that expenditure shares on fuels may depend on total energy expenditures, but are independent of the expenditure shares for other consumption categories. 3 1 This study is itself part of a larger project to develop analytic models of the world oil market. 2 For now we do not treat energy as a derived demand determined by the stock of energy-using appliances (in fact durable goods are a separate consumption category). In the dynamic models to be discussed later the effects of changes in appliance stocks will be included implicitly. 3 We thus have a "utility tree" along the lines described by Strotz [60], and the marginal rate of substitution between any two variables in the class of energy expenditures is independent of the expenditure on any other consumption category. 2 Estimating the demand for energy requires a model for the breakdown of total consumption expenditures. A number of such models have been constructed by others, some of them additively consistent (in terms of shares adding to one) and some inconsistent. 4 Typical model choices have included the additive logrithmic model, the linear expenditure system, and the additive quadratic model. 5 Usually these models have been estimated using time series data for single countries, but in some cases cross-country comparisons have been made using pooled time series-cross section data for a number of countries.6 We will extend this work by estimating both static and dynamic versions of the indirect translog utility function with pooled data. The advantage of the translog function is that it is a general approximation to any utility function, and therefore it does not a priori impose constraints of homotheticity and additivity. We will thus be able to obtain unrestricted estimates of own- price, cross-price, and income elasticities; this is important as there is reason to think, for example, that the income elasticity for energy demand differs from unity. In addition, we will be able to test whether homotheticity, additivity, stationarity, and for that matter, utility maximization are reasonable assumptions. We will also estimate alternative models for the breakdown of consumption expenditures. In particular, we will repeat Houthakker's 1965 study [31] using more recent data and our own consumption categories. This will permit us to explore some basic issues in the pooling of heterogeneous data, and to look for differences between short-run and long-run elasticities that might suggest priori specifications for dynamic models. 4 For an overview of models of consumer behavior see Brown and Deaton[ll] and Phlips [52]. 5 See Houthakker [30], Pollak and Wales [55], Houthakker and Taylor [32], Phlips[52], and Theil [63]. 6 See Houthakker [31], and Goldberger and Gamaletsos [23]. 3 In estimating models of consumption expenditures we wish to explore the extent to which higher energy prices might reduce the total consumption of energy. Although energy prices did not increase substantially during the period covered by our data (pre-1974), there is enough cross-sectional variation in prices to allow us to obtain price elasticity estimates, as well as estimates of income elasticities. In addition, by estimating models using country dummy variables and/or alternative groupings of countries we hope to determine the extent to which elasticities vary across countries. In estimating the demands for individual fuels we will also test a number of alternative model structures. We will again use our pooled data to estimate both static and dynamic versions of the indirect translog utility function. In doing so we can obtain unrestricted estimates of own-price, cross-price, and total expenditure elasticities, and test for homotheticity, additivity, and stationarity. Also, by estimating a translog model that includes both non- energy consumption expenditures and fuel expenditures we can test for separability. We will also use the multinomial logit model to break energy expenditures down into fuel shares. One advantage of the logit model is that it is relatively easy to estimate; as long as the share data represents aggregated samples of individual decisions (i.e. average shares for a large number of consumers) rather than individual decisions, ordinary or generalized least squares can be used. Another advantage of the logit model is that it allows considerable flexibility for working a dynamic structure into the specification. disadvantages, however. share data. The logit model also has Estimates become inefficient when there are zeros in the In addition, all cross-elasticities for a given own price are equal; 4 as Hausman and Wood have shown [26], they are the sum of the price elasticity for total expenditure minus the own price elasticity weighted by the share. This fact that cross elasticities are determined by total and own elasticities is restrictive, but much less so than the restrictions inherent in the linear expenditure system, additive quadratic model, and other "consistency" models. In estimating demand models for individual fuels we will explore the extent to which fuel shares shift in response to price changes and changes in total energy expenditures over both the short- and long-run. Again, there should be enough variation in prices through the combined use of time-series and crosssection data to obtain reasonable elasticity estimates, and determine the extent to which elasticities vary across countries. As one might expect, in work like this we are continually bound by data limitations. For many countries there is no good data available for some or all of the variables of interest to us. For other countries data exists, but obtaining that data can be an extremely time consuming and laborious task, so that choices had to be made as to which data were to be collected. These data limitations were one of the factors that helped define and delimit the modelling approaches used here. In particular, it necessitated restricting our detailed analysis of demand to a small set of countries. 7 Even for these countries, however, the quality of the data varies, and compromises had to occasionally be made. 8 The data used in this study is described briefly in this report; a much more detailed description is provided in a separate report entitled "A User's Guide to the MIT World Energy Demand Data Base." 9 7A much less detailed model of the demand for petroleum products is being constructed for a number of "secondary" countries for which only partial data is available; the results of this work will be described in a forthcoming paper. 8 For example, one of our consumption expenditure categories is "food, alcohol, and tobacco." For some countries a price index is available only for food, and this index was used since food is by far the largest component of the category. Working Paper No. MITEL76-01lWP, MIT World Oil Project, May 1976. 5 We stress that the results presented in this paper are preliminary and in many ways incomplete. Only some of our alternative model specifications have been estimated at this point. This report, however, does serve to provide initial elasticity estimates, explores problems in the use of pooled international data, and provides a framework for continuing modelling work. Hopefully it will also provide a source of feedback in the form of comments, criticisms, and suggestions that can help correct deficiencies in the present approach. In the next section we outline alternative specifications of alternative models of residential energy demand, and discuss the characteristics of each specification. Section 3 discusses some methodological issues in the estimation of energy demand models using pooled data. These issues include the use of purchasing power parities to make international comparisons, the question of accounting for thermal efficiencies in the use of energy consumption data, the formulation of an aggregate price index for energy, and the use of alternative estimation methods. Section 4 describes some of the characteristics and limita- tions of our data, and Section 5 includes the statistical results. 6 2. Alternative Specifications for Models of Residential Energy Demand As explained above, all of our models of residential energy demand involve a two-stage approach where first consumption expenditures are broken down into energy and other consumption categories, and second energy expenditures are broken down into expenditures on fuels. We begin here by reviewing the proper- ties of the indirect translog utility function with a time trend and discuss its application to both stages of the residential model. Next we describe some alternative dynamic specifications of the indirect translog utility function. As we will see, these specifications will permit us to explicitly include stock adjustment or habit formation effects. We then discuss the multinomial logit model and its application to the estimation of fuel shares. Finally we discuss alternative model specifications, including simple models. 2.1 Use of the Indirect Translog Utility Function The indirect translog utility function is a second-order approximation to any indirect utility function. The indirect translog function with time-varying preferences, introduced by Jorgenson and Lau [37], has the form:10 log V = + ZEailog(Pi/M ) + att + Izi log(Pi/M)log(Pj/M ij + EZi t log(Pi /M)' t + 2 Stt ) ~ ~~~~~~~~(1) 2 When the indirect translog function is used to model expenditure shares for energy and non-energy consumption categories, P i is the price index for consumption category i and M is total consumption expenditures. 10 When this function is The indirect translog utility function without time was introduced by Christensen, Jorgenson, and Lau [15]. The homothetic form of the indirect translog function was used by Christensen and Manser [14] to study consumer preferences for food, and the non-homothetic form was used by Jorgenson [35] to study a three-category breakdown of consumer expenditures in the United States. Berndt, Darrough, and Diewert [70] demonstrated empirically that the translog specification is more robust than other generalized functional forms such as the generalized Leontief or generalized Cobb-Douglass utility functions. 7 used to model fuel shares, Pi is the price of fuel i, and M is total expenditures on energy. The indirect translog function implies the budget share equations: si - PSX M aj + iBjilog(Pi/M)+ ++*I=l j 5a + ilog- t ..... j~~~~t i/M) + n (2) m ·- where X. is the quantity consumed of category i (or fuel ), t is a time trend (equal to zero at the beginning of the estimation period), and i Oi '- 'Oki' 8Mtk Note that the parameters ao, kt t, and Ott in equation (1) do not affect the utility-maximizing quantities consumed, and therefore cannot be identified. In addition note that the budget constraint implies that S = 1, so that only (n-l) of the share equations need be estimated to determine all of the parameters. The budget share equations are homogeneous of degree zero in the parameters, and therefore a parameter normalization is required for estimation. normalization = Za = -1. We use the A number of parameter restrictions are also required if the share equations are indeed based on utility maximization. particular, the parameters equations. and t must be the same in each of the n In share Since there are (n+l) parameters involved, and (n-l) equations are estimated, this implies a total of (n+l)(n-l) restrictions. Also, we assume that that log V is twice differentiable in its arguments, so the Hessian of log V must be symmetric. Bi =i , This implies the following i j, i,j = , ..., n (n-l)(n-2) symmetry restrictions: (3) 8 There are an additional (n-l) restrictions resulting from the fact that the parameters of the n th equation are determined from the parameters of the first (n-l) equations and the definitions of of parameter restrictions is Mi and Mt. Thus, the total number n(n-l). There are other restrictions that might be imposed on the indirect translog function, and tests can be performed to determine whether such restrictions are supported by the data. We will test some of these restrictions in this work, so we list them here. The indirect translog function is stationary if preferences do not 11 change with time. Stationarity implies that the parameters equal to zero, jail,. . ., n. it are all it 12 In estimating the consumption breakdown model, we might wish to test for groupwise separability between energy and the other consumption categories. Letting P1 and S1 be the price index and expenditure share for energy, and P2' P3 . ' Pn and S2, S3 , . .. , S be prices and shares for the other categories, separability would imply that the underlying indirect utility function can be written as logV = F(logVl(P2 /M,P3 /M, . . ., t), P1 /M, t) (4) If the underlying indirect utility function is groupwise separable, then the following restrictions must hold:13 A dynamic translog function, in which long-run elasticities differ from short-run elasticities, may still be stationary as long as the elasticities themselves do not depend on the particular time in which prices or income change. This is discussed further later. 12 13 Note that stationarity is equivalent to explicit neutrality. An indirect utility function is explicitly neutral if it can be written as log V = log V (P1/M, P 2 /M...,Pn/M) + F(t) See Jorgenson and Lau [37] for a derivation of these restrictions. 9 B12 where 1 : 813 2 is a constant. l 3, in . Pl (5) Even if the underlying indirect utility function is groupwise separable, the translog approximation need not be. Explicit group- wise separability ensures that the translog approximation is also groupwise separable. This requires the additional restriction that p1 = 0. We will also estimate share equations based on homothetic indirect utility functions.1 4 Under homotheticity the budget shares S. are independent of total J expenditures M. This implies that the income elasticities of demand for every commodity are the same and equal to unity. The underlying indirect utility function is homothetic if = where a is a constant. oa. j = 1, . .. , n (6) Explicit homotheticity will ensure that the translog approximation is also homothetic, and this requires the additional assumption that and a = 0. If the indirect utility function is explicitly homothetic t = 0, then it is also homogeneous. Finally, it is straightforward to test for explicit additivity, since a necessary and sufficient condition for explicit additivity in the commodities is that the indirect translog function is explicitly groupwise seperable 15 in any pair of commodities from the remaining commodity.5 constraints 1 are that ij = 0, i Thus the parameter j. is homothetic if it can be written as logV = F(logH(Pl/M, where H is homogeneous of degree 1. . . , Pn/Mt)t) 15An indirect translog utility function is explicitly additive if it can be written in the form log V = log Vl(P1 /Mt) + ... + log Vn(Pn/M,t). 10 We can test restrictions using a simple chi-square test. The appropriate test statistic is A -2 where '1 I and J6I log A = ^~~ n (logQ r - logl u) (7) are the determinants of the estimated error covariance matrices for the restricted and unrestricted models respectively. This statistic is distributed as chi-square with degrees of freedom equal to the number of parameter restrictions being tested. It is important to remember that there are only certain ranges of inputs over which the indirect translog utility function is a meaningful approximation to the underlying utility function. Consider, for example, the marginal utility of income (or of total expenditure), V agV - . M alogM = X = 3V/M: VP (ai+8Ealog4 -- M i Sij 4 +~ a.t)(8 ii ) (8)M The a. sum to -1 by the normalization, while the 3.. can be positive or nega1 1J tive. If some ij are positive, then as M becomes zero X can become negative, ij and if some 8ij are negative, then as M becomes increasingly large X can become negative. Thus, there are ranges of input space for which the translog ap- proximation may not be meaningful. It is important therefore to check esti- mated translog models by determining whether the marginal utility of income is positive over the range of historical (and forecasted) input data. After estimating models for the breakdown of consumption expenditures it might be useful to compute risch's welfare indicator for purposes of cross-country comparisons, e.g. to determine the relative effects on different countries of higher energy prices. This indicator is simply the income elasticity of the marginal utility of income, i.e. nXM= logX/alogM. For 11 a utility function that is well-behaved over the entire input space (which the translog is not)nkM would range from a large negative number (when M j (i .i-1 is zero) to zero (as M approaches infinity). Taking the log of equation (8) 16 and differentiating, we have for the indirect translog function: SE 8. ~~~~ - 1 - n>M =n~~~ =~~~ i i i i Note that as M + 0, nqM +- l, (9) C i -i M + Z.t log M + ZiS ij ij i itt so that a very small M is clearly out of the The same is true for M + o. "meaningful" range. We also need formulas for the calculation of income and price elasticities. logX /logM, The income elasticity of demand for good j, %M = is found by multiplying the share equation (2) by M/P. and differentiating: (10) 1+/+ 1+ - i i ji/3 iMi aM ilO)+ + a + jM E..t3logPi + Note that this is also the formula for the expenditure elasticity alog(PjX.)/alogM. alogX/alogPjis The own price elasticity njj J//S i- «iM a + 8Milo- + j = -1+ (ll) t i and the cross price elasticities nji = logX/logPi are j/ji -6 ;i = /S - Milog aM + S (12) + m.t i 16The same equation also applies to the direct translog utility function. 12 These elasticity formulas can be applied to each stage of our two-stage demand model, but they cannot be used to determine the total effect of a change in price or income on the demand for a particular fuel. If the price of oil changes, there will be a change in total expenditures on energy, and this will also affect the demand for oil. d log X/d log P is given by r] jj and X J i ~J J where P The total own price elasticity = P. fx j ) ME PE -j I X. aP. DhE J (13) aP.J JJ PE are the price and quantity of fuel j, ME is expenditures on J energy, and PE is the price index for energy. Thus, to determine the total elasticity we need an expression for the price of energy in terms of the prices of individual fuels. not determine P E Since fuels are not perfect substitutes we can- as a simple weighted average of the fuel prices. Instead we view PE as the cost of producing heat from fuel inputs, and use a translog cost function with constant returns to model this "production" process: log PE = Yo + Z Yi log P Pilog Pj + Z Z Yij log i (14) ij This is an energy price aggregator, and can be determined up to a scalar by estimating the share equations1 7 Si Y + Y log P p* Given equation (14) for the price of energy, we have aPE EaP' ap . 17 PE S Sj (15) We will discuss the energy price aggregator in more detail later. 13 * We can thus compute TjI from the fact that ax. X. DP.= P i where Tn ji (16) 3 is the partial own price elasticity for fuel j given by equation (11), ax. X. (17) __TI DME where Tj ME i }7 is the expenditure elasticity for fuel j given by equation(10), and 3E aPE X (1 + nEE) (18) where XE is the total quantity of energy consumed and nEE is the partial own price elasticity of energy consumption. (17), and (18) into Now substituting (15), (16), (13), we have * nj njj + njME(l+nEE (19) We can similarly compute the total cross price elasticity nji from *i '.Pax. '" i LaP ax. aE ap aiI aPE aPi] Ti + SinjME(1+nEE) (20) and the total income elasticity njM from * _ 'jM Note that since M X aXj E aM aM ME/aM = (ME/M)nEM, where (21) EM is the income elasticity of energy expenditures, we obtain * TI jM =TI J ME "71 EM (22) 14 2.2 Dynamic Versions of the Indirect Translog Utility Function A problem with the translog demand models described above is that they do not explain differences between short-run and long-run elasticities. Even when the time trend is included in the indirect utility function, the model is not really dynamic - tastes can change slowly over time, but there is no dynamic (lagged) response in demand to a sudden change in price. Thus adjustments (while possibly non-stationary) were assumed to occur instantaneously. There are two basic approaches that can be used to introduce dynamic adjustments into the translog utility function. The first approach involves specifying the translog approximation to the utility function (direct or indirect) to include lagged quantities, prices, or shares. The advantage of this approach is that "adding up" is always preserved in the resulting share equations without the introduction of additional parameter constraints. The disadvantage is that the translog approximation makes the dynamic specification somewhat arbitrary. As an example of this approach, we could write the indirect translog utility function as: log V = 0 + E'1log(Pi/M) + Z log(P i/M)lg(P J/M) (23) + Edilog(Pi/M)Di, t-l i i g(i/ ~ where Di t 1 is a lagged term in price, quantity, or share that is considered 18 an exogenous input to the determination of current share. 18 Logical choices for We are assuming that consumers determine their budget shares via utility maximization, i.e. they maximize utility at each instant ignoring the future, rather than maximizing the sum over time of utilities. The Di t 1 (together with current prices and income) static of time discounted simply represent the current state of the world. As shown by Hoel [28], even in a static model dynamic utility maximization can result in a different marginal utility of income. 15 Diitt-1 would be the quantity Xi t 1 or the share Si't-19 i t 19 to equation (23) yields the share equations: Si ad + d D = j + M Applying Roy's identity p l. igRysiett2 t-l + Ea1 ilog(P /M) EdiDit-1l + Milog(Pi/M) ,j,..., where aM and OMi are defined as before. (24) Note that unless all of the di are zero, the homothetic form of equation (24) - for which the ai nonlinear in the parameters. are zero - is As a result, estimation of (24) can be costly, even under the assumption of homotheticity. The shares in (24) will always add to one, however, even if lagged shares are used as the D tl' and - assuming that the errors are not serially correlated - the parameter estimates will be invariant to the choice of share that is dropped. A second approach that can be used is to introduce the dynamic adjustment directly into the share equations. This has the advantage of facilitating the use of simple and intuitively pleasing adjustment mechanisms. It has the disadvantage that "adding up" will not be preserved unless additional (and possibly highly restrictive) parameter restrictions are introduced. We consider two examples of this approach. is assumed to adjust to a desired level: Xi t = Xit 1 + 6i(X-Xi t_ where X i,t In the first, each quantity 1) (25) is the desired quantity of commodity i as determined from static utility maximization, and 6i is an adjustment parameter. equations P Sj Jil S SJ + (1- t/M( t ) [j Jil=-1jPj/t j,t-1 t-l/t 19 This yields the share (26) The form of equation (23) using lagged quantity Xi t1 was suggested by Manser [44,45], who applied it to the estimation oftfood demand. 20See [57]. The identity is: PiXi M = i alogV/a1ogPi alogV/alogM 16 or, using the indirect translog function for St S i.ii + (1-6j)Sj i J=t a ++ Ea Ilog(P 1 /M) + iMi t ci J t'l[Pj,t-l/Mt- 1 ] (27) M The parameters of the share equations (27) are estimated subject to the constraints * Ea S A are the same in each equation, and 1. k /6_ k- Note that the shares St need not add to one. Adding up can be imposed by estimating only n-I of the share equations, and determining the parameters th of the nth equation from ESt It = 1, but the estimated parameters will depend on the particular equation that is not estimated. Despite this deficiency, however, the specification of equation (27) permits the introduction of dynamic adjustments in a simple and appealing manner. ~~ Alternatively, we can assume that the shares adjust to the desired shares as follows:2 1 Si,t Si,t- + S 5 + ij(t-sjt-l) (28) Adding up requires that the sum of all changes in shares be zero: i (Si i it- ,t) ==' t- 1) si,t-z) (29) so that iij (Sjt Since the = ,t Cip- t-630 0 t and Sj,t-l1 sum to one, this equation implies the necessary condition that all of the columns of the matrix (ij sum to the same arbitrary constant, i.e. ~~~.· 21 This approach was suggested by Leonard Waverman in the context of dynamic adjustments in the translog production function. 17 i' 6= c' (31) is a vector of l's (ones), 6 is the matrix (6t), where 22 constant.2 2 and c is an arbitrary Note that if the number of shares is greater than two, there are alternative constraints on the ij that can be imposed to satisfy (31). Note ij also that (31) implies that 6 cannot be diagonal unless the adjustment coefficients for every share are the same, so that the adjustment of the th it h commodity share would generally not depend only on that share, but would depend on other shares as well. 2.3 Multinominal Logit Models for Fuel Choice Multinominal logit models have already been used to study the breakdown of energy consumption into demands for fuels in the United States23 and 24 Canada.2 4 Although the logit model is not based on assumptions of utility maximization, it has properties that make it appealing for this work. The model is consistent in terms of shares adding to one, and shares respond to price changes in a way that is intuitively appealing; as the share of, say, natural gas becomes small, it requires increasingly large price changes to make it still smaller. Finally the logit model is easy to estimate and permits us to easily introduce a variety of alternative dynamic specifications. 22 2 For a discussion of adding up conditions for more general lag structures, see Wall [65] and Berndt and Savin [9 ]. 23See Baughman and Joskow [6 ] 24 See Fuss and Waverman [68]. 18 We can write the logit model for the four fuel breakdown (oil, gas, 25 coal and electricity) as follows:2 5 Qi = efi(Xa) QT Qe4fi(Xo) T' ef x ) (32) e where Qi is the quantity (in tcals) of fuel i, QT EQi, and the fi are functions of a vector of attributes X and vector of parameters . Given this model, the relative shares of any two fuels can be represented as log(Qi/Q) = log(Si/S) = fi(x) - f((33)). Note that only three equations are estimated, since the parameters of the fourth equation are determined from the adding up constraint. In estimating fuel shares we include as attributes the relative price of each fuel. The relative oil price, for example, is the ratio of the real price of oil to the real price of energy, the latter being measured by the translog price aggregator described earlier. Other attributes may include per capita income, average temperature, and lagged quantity variables that allow shares to adjust dynamically to changes in price. Functional forms for the fi are somewhat arbitrary, but in the simplest model they might be 25 In effect we are assuming that consumer preferences are represented by a choice index which for the ith fuel, has the form fi(x)+ei(x), where i is an error term. Then the probability that a consumer would choose fuel i is Pi = Prob[fi(xo)+ i(x) > fj(x)+ cj(x)] for i # J. If the error terms (x) are independently and identically distributed with the Weibull distribution Prob[ei(x) < ]= e -e then the probability that fuel 1I will be chosen is given by equation (32). For f!lrthiert1srcusNLon see MFa(Iddeti ( 1 .CtC And MCFilddn 1191, (COx 161. Theil [61, and Chapter 8 of Pindyck and Rubinfeld []. For an interesting application to aggregate demand analysis, see Park [51]. 19 = Pi/PE' linear functions of the relative.fuel prices P where PE is the aggregate price of energy, as well as income Y and temperature T: fi(xB) = ai + biP (34) + CiY + diT This yields the three estimating equations (c-c)Y i (ai-a i4 44) + biPi i i - bP 4 4 4 + (cgc4 log(S /S) + (i (dd)T, 4) i=1,2,3 (35) Note that these three equations must be estimated simultaneously, with b4 constrained to be the same in each equation. The simplest means by which the preference functions can be made dynamic (e.g. to account for habit formation or stock adjustments) is to include the lagged share: + biPi + ciSit fi(xB) = a 1.(36) The three estimating equations are then log(S /S) = (ai-a4 )+ biPi -bp bt, + ic j 1 - Note that two lagged shares appear in each equation. i=1,2,3. (37) The three equations must again be estimated simultaneously, with both b 4 and c4 constrained to be the same in each equation. In Canada and the United States very little coal is used in the residential sector, and coal shares are generally less than one percent. In these countries, the residential fuel choice is effectively between three fuels, and we therefore use a three-fuel breakdown for Canada and the U.S., and a four-fuel breakdown elsewhere. 2.4 Other Models of Residential Energy Demand When working with pooled cross-section time series data it is often difficult to separately identify short-term and long-term effects, and determine the relative contributions to each from the cross-section versus time series variation in the data. This can be particularly true in the case 20 of the translog function, where nonlinear estimation is involved. It may therefore be useful to also estimate simple log-log demand equations for expenditures on each consumption category and each fuel. Houthakker in his 1965 study [3i] separated short-run and long-run elasticities by running separate regressions across countries and across time. It will be useful to repeat Houthakker's study for our own consumption date, and also apply the approach to the estimation of fuel demands.26 In addition, we will specify a dynamic version of Houthakker's basic model that should enable us to isolate the cross-section versus time series contributions to lag adjustments in demand. The basic demand equation for commodity i logqijt where j = ai + logYjt + Yilog Pi + t is the country index and t on commodity i is the time index, q at constant prices, y constant prices per capita, and Pi + iJt (38) is per capita expenditure is total consumer expenditure at is the relative price of commodity i. These equations can initially be estimated using simple weighted least squares, e.g. dividing each observation by the population of the country in the year concerned. Short-run and long-run effects can be identified by estimating "within country" and "between country" regressions. regression 26 The "within country" (short-run) is Houthakker's consumption breakdown had only five categories: food, clothing, rent, durables, and "other." We use our own consumption breakdown for which energy is a separate category. Goldberger and Gamaletsos [3] also reestimated Houthakker's log-log demand functions, but using pre- 1961 data and the same categories that Houthakker used. What is more interesting is their estimation of a linear expenditure system using the same data, and the comparison of the two demand systems. The approach was also used by Gregory and Griffen [21] to identify international differences and intertemporal change in industry structure. 21 log j log - (log = Yjt - log yj) + i(1og PiJt - log Pij) (39) and the "between country" (long-run)regression is log qij = ilog Yj + Y 1 °g o Pi + i Here the bar represents averaging over time. (40) Note that the "within country" regression is pooled, while the "between country" is purely cross sectional, and that deviations from means (over time) are used in the "within country" regressions in order to eliminate long-run effects. Equation (39) can also be run for each country separately, in order to determine how elasticities vary Alternatively, a pooled regression can be performed, with a across countries. multiplicative district dummy variable introduced to one coefficient at a time, e.g.: = 10ijDj(log Yjt - log yj) + Yi(log Pijt - log - log q log qit (41) Pij Equations (39) and (40) can also be estimated in first-differenced form. This crudely reduces trend effects, and also eliminates problems associated with the arbitrary choice of purchasing power parities. version is: Alog qi - Alog q = .~~~~~~~. ( log Yjt - Alog yj) + The "within country" i(Alog Pijt - Alog ij) (42) and the "between country" version is: Alog qij = 8iAlog Y (43) PiJ + YiAlog We can also specify a dynamic version of (39). We assume that demand qit depends not only on price and income in period t, but also on a "state variable" it: log qit If the demand is be negative. = ai + blog y + cl°g Pit+ dil°g sit (44) for durables, then sit will represent a stock, and di should If the demand is for a non-durable commodity, then sit will 22 represent a "habit" level, and di should be positive. The dynamics of sit can be expressed as = Alog sit (45) wil°g Si,t- log qit To obtain the demand equation where wi is effectively a depreciation rate. rewrite equation (44) as 1 sit = d-[log q ai clog bilog y (46) it which can be substituted into equation (45): Alog sit wi og qit di [log qi,t-1 ai - bil°g Yt1- ] c log (47) Now first-difference equation (46): Alog it =d [Alog qit - biAl°g yt - (48) ciAlog Pi1 Substituting this into (47) and rewriting, we have the estimating equation log q = it 0 + + 11o g, 1_1 12 Alog y '2 + a31g 3 lo Yt ~t-1 - Al g 4 Alo + t a g c 5 lo ~it-1' (49) This equation can also be estimated within countries and between countries. If this results in differences in the estimated value of a, it would indicate that adjustment response is not constant over time. 3. Methodological Issues in the Estimation of Residential Demand Models There are a number of problems that make estimation of the models described in the previous section less than straightforward. This is due in part to the fact that pooled international data are being used to obtain estimates, and in part to the nature of the models themselves. First, the comparison of expenditures or prices in different countries requires valuing different currencies in terms 23 of a common unit. Although using purchasing power parities for this purpose is probably more desirable than using official exchange rates, the choice of a particular index is not always clear. Second, a choice must be made whether to value energy quantities in gross or net terms, i.e. whether to adjust for thermal efficiencies of different fuels. Next, an energy price index must be obtained that accounts for fuel choice differences across countries. Finally, there are a number of econometric issues associated with the estimation of our models. 3.1 We examine these problems in this section. Use of Purchasing Power Parities Since all of the price and expenditure data for each country in our sample aremeasured in terms of the local currency of that country, a method is needed to convert these numbers into common units. One method which has been used by 27 a number of researchers is to simply use official exchange rates.27 This can be misleading, however, since official rates can differ considerably from equilibrium exchange rates, and tariffs, quotas, subsidies, and other controls can result in price structures that differ considerably from relative international prices. Alternatively, one could attempt to identify "free market" exchange rates between individual countries over time periods thought to reflect equilibrium conditions, e.g. during which trade balances were near zero. Even under free trade, however, equilibrium exchange rates only reflect the price equalization of internationally traded goods, which for most countries repre- 27 This approach has been used by Adams and Griffen [1 ], and Goldberger and Gamaletsos [23]. 24 28 sent a small subset of all market goods.28 A better approach is to use purchasing power parities (PPP's) to convert national currencies to some base currency. Purchasing power parities can be obtained explicitly by making binary comparisons between a base country (e.g. 299 the U.S.) and various other countries, using a fixed set of quantity weights. The problem, of course, is that two sets of price index numbers (Laspeyre and Paasche) can be obtained depending on whether base country or other country weights are used. In this work we use a Fisher "ideal" index (a geometric mean of these two index numbers) as a single index of relative purchasing power. 30 We use purchasing power parities by consumption category calculated by the German Statistical Office (Statistiches Bundesamt [59]) by means of detailed price comparisons. 31 These are binary index numbers with Germany the base country, and we use Germany as a "bridge" to convert to the U.S. as base country.3 2 28 As Chenery and Syrquin [12] point out, the relative prices of non-traded goods can be expected to increase with real per capita income, so that the use of official (or "free market") exchange rates leads to an underestimate of the purchasing power of the currencies of lower income countries. 29 Purchasing power parities can also be obtained implicitly by dividing a nominal national currency estimate of national product (or one of its components) by a base currency estimate of the same national product. This procedure was used recently by Lluch and Powell []. For a general discussion of explicit purchasing power parities, see Balassa [ ] and Allen [3 ]. 30 The use of a Fisher "ideal" index is suggested on theoretical grounds by Samuelson [68] and on empirical grounds by Kloek and Theil [36]. 31 Binary purchasing power parities were more recently calculated by Kravis et al. [39], but for only a subset of the countries in our sample. 32 Note that although binary PPP's permit us to make a transitive international ordering of purchasing powers, this ordering is not invariant with respect to the choice of "bridge" country. Kravis et. al. [39] also calculated multilateral PPP's by means of a regression model that estimates the purchasing power parity for a single category of expenditure as a function of all other international price ratios. Again, some of the countries in our sample are not included in the Kravis study. 25 The resulting parities apply to a base year, but we must construct intertemporal indices to deflate our time series. We do this using implicit price indices for each consumption category in each country, thus constructing an implicit ratio of relative intertemporal purchasing power in terms of a base year numeraire (normalized so that 1970 is our base year). The resulting base year purchasing 33 power parities for each consumption category are shown in Table 1. 3.2 "Gross" versus "Net" Energy Consumption As pointed out by Adams and Miovic [2], alternative fuels are not equiva- lent on a calorific basis as a result of the differing thermal efficiencies of energy consuming equipment. Since more efficient fuels are substituted for less efficient fuels over time, the measurement of an overall "energy elasticity" (i.e. the percent change in energy use associated with a 1 percent change in GNP) 34 will yield a larger number if thermal efficiencies are taken into account.3 4 This has led some individuals to suggest the use of "net" energy consumption (adjusted for thermal efficiencies) rather than gross energy consumption in the estimation of demand models. Nordhaus In recent studies of energy demand, for example, ['8, Adams and Griffen [£ ], and Fuss and Waverman [68], made the assumption that within each sector fuels are perfect substitutes, so that (given equal levels of non-fuel cost) interfuel competition is determined by relative net prices of fuels. Net consumption and net price are given by 33 For those countries also covered by Kravis et. al., our numbers are at all times within 10 percent of the 1970 Kravis numbers. 34 Adams and Miovic estimate an overall energy elasticity for the U.S. and several European countries of about 0.8 when gross energy quantities are used, and about 1.0 when energy quantities are adjusted for thermal efficiencies. 26 Table 1 - Base Year (1970) Purchasing Power Parities Consumption Category Transportation ConsumptionComnctn Country Consumption ~~- . Belgium - Apparel Food Durables Energy , 36.37 46.73 ~ ~~~~~ ~~~~~ . Canad .866 .828 France 4.82 4.26 Italy 467.2 364.5 , 32.40 ~ 1.022 . 2.21 5.90 , .576 5.78 4.00 5.16 5.76 3.13 606.5 400.0 523.1 343.7 2.94 4.77 7.77 . .. .818 473.3 ., 2.59 42.48 27.91 ~ ~~~~~~~~~~~~~~~~~~~~~~~~~ ~ , .689 . 2.14 i iim . 45.30 ~ .792 m m,.i Norway Other . 42.05 Netherlands T and Communication . 1.47 , i I . . 3.01 2.17 5.65 4.06 L~~. . 4.06 i , 2.60 i . ,. .K. .282 .208 .314 .235 .280 .281 .220 U.S.A. 1.0 1.0 1.0 1.0 1.0 1.0 1.0 3.13 2.26 3.76 2.84 2.84 3.40 2.44 est Germany For Belgium, Canada, Netherlands, and Norway, no "durables" PPP exists. A PPP for "other household" was used in going from these countries to Germany, and the "durables" PPP was used in bridging from Germany to the U.S. ** For Belgium and the Netherlands, the PPP relative to Germany refers to "electricity, gas, and water." 27 QNii qij and PN = flJQi j iJij (50) P (51) Pij = ij/ij where nij is the efficiency of fuel i in sector J. We see two problems with this approach. First, it is difficult to obtain reliable estimates of thermal efficiencies. Identification problems make econometric estimates infeasible unless unduly restrictive structural assumptions are imposed, and engineering estimates differ considerably from 35 source to source.35 As an example of this problem, we show in Table 2 engineering estimates of thermal efficiencies cited or used in four different studies. Note that these estimates differ considerably from study to study. A second and more fundamental problem is that fuels are not perfect substitutes (particularly in the short run), and there are non-thermal efficiencies (which we could label "economic") that also affect consumer demand. Fuel choice is also based on convenience, controllability, cleanliness, capital costs, etc., and the effects of these "efficiencies" (as well as thermal ones) will hopefully be manifested in the estimated parameters of our demand 35 Adams and Miovic [ ] attempted to measure thermal efficiences by assuming that fuel inputs are a constant proportion of aggregate economic output, and that there is no substitution between fuel inputs and labor and capital. Their production function was thus Y = min(aF, f(L,K)) where fuel input F is given by F = of fuel i (Fi ). EnihiFi, where h is the calorific content Since the hi's are known, they can estimate the n s up to a scalar multiple. The assumptions are extremely restrictive, however, and their results differ considerably from engineering estimates that they cite. For an engineering discussion of thermal efficiencies, as well as a set of estimates, see Hottel and Howard [29]. 28 Table 2 - Alternative Engineering Estimates of Thermal Efficiencies Citation or Use of Estimate Adams and Miovic Use of Estimatel Fuel ~ue~ Residential Industrial Adams and Griffen Residential Industrial Nordhaus Residential Industrial Fuss and Waverman Residential Industrial Gas .65-.72 .39 .60 .65 .70 .85 .75 .85 Solid .05-.60 .33 .50 .45 .20 .70 .50 .87 .65 .59 .60 .80 .65 .87 .80 .95 .99 1.00 1.00 Liquid .65 .40 Electricity .80 .80 models. -- It thus does not seem particularly relevant to measure fuel consump- tion in efficiency-adjusted thermal units, any more than it would be to measure food consumption in net calorific terms.3 6 We therefore choose to measure all of our energy quantities in "gross" rather than "et" terms. We assume that both thermal and non-thermal efficiencies have effects on interfuel competition, and that these effects will be picked up in the way that estimated fuel expenditure shares change as relative prices and income change. 3.3 A Price Index for Energy Estimation of our consumption breakdown models requires a price index for energy, and since price series for individual fuels are available, it would be preferable to use this data rather than an implicit index constructed from nominal and real energy expenditure series. 36 Since fuels are not perfect This is discussed further by Turvey and Nobay [6q]. 29 substitutes, however, a price index that truly reflects the unit cost of energy will not equal a simple weighted average of fuel prices. A typical 37 approach is to construct an approximate Divisia index as a means of aggregation. An alternative approach is to specify (and estimate) an aggregator function that relates the aggregate price index to the component prices. Any unit cost function could be used to represent the aggregate price of energy, but a logical choice is 38 the translog cost function. As an incidental advantage, the translog cost function (or "aggregator") provides us with an instrumental variable for estimation purposes. The translog cost function (which is equivalent to a homothetic and stationary indirect utility function with unit total expenditure) is given by log E = + YilogP + logP logP. Y (5 Assuming cost-minimizing behavior, the fuel share equations are then Si = Yi+ YijlogPj , i=l,...,n (5 The first (n - 1) share equations are estimated subject to the restrictions Zyi= 1 Yi = Yji and Yij = 0. The estimated parameters Yi and Yij are then A substituted in equation (52) to yield the estimated price index PE A Note that the energy price index PE is determined only up to an unknown E scalar multiple YO. The procedure is to pick one country (say the U.S.) as a base country, and then solve equation (52) for yo so that 37 38 See Jorgenson and Griliches [36] the price of energy and Hulten []. This is appealing as an unrestrictive representation of unit cost. Also, as Diewert [8 ] has shown, the Divisia index is "exact" for the translog aggregator function, i.e. it retrieves the actual values of the function. 30 in the base country is equal to 1 in some base year (say 1970). Relative price indices are thereby determined for all of the other countries. A problem remains regarding the number of fuels to be included in equations (52) and (53). Although four fuels are included in our demand model, very little coal is consumed in the residential sectors of the U.S. and Canada. This suggests that equations (52) and (53) should apply to a three-fuel aggregation (oil, gas, and electricity) for the U.S. and Canada, and a four-fuel aggregation for the remaining countries. Should this approach be used--as opposed to a four-fuel aggregation for all countries--a method is needed to "bridge" the U.S.-Canadian aggregator with the aggregator for the remaining countries. We use the follow- ing bridging method: (1) Equation (53) is estimated for four fuels for all countries except the U. S. and Canada. The unidentified parameter y0 in equation (52) is chosen so that the price of energy is 1 in Belgium in 1970. This permits the calculation of the price of energy for all countries except the U. S. and Canada relative to Belgium in 1970. (2) Equation (53) is estimated for three fuels (leaving out coal) for all countries. The parameter in equation (52) is chosen so that the price of energy is equal to 1 in the U. S. in 1970. The relative price of energy in Belgium in 1970 to that in the U. S. in 1970 is then computed. (3) Equation (53) is estimated for three fuels for the U. S. and Canada only. yO is chosen so that the price of energy is 1970. Now using the Belgium-to-U.S. conversion ratio determined in 1 in the U.S. in step (2), the price indices calculated in step (1) are converted to a U. S. 1970 base. 31 We said before that it is not clear whether a bridging approach such as that described above or a simple four-fuel aggregation should be used to construct the energy price index. using both methods. We therefore estimate energy price indices If the resulting indices are nearly the same, this would indicate that the relative size of the coal shares in the U. S. and Canada do not distort the fit of a fuel choice model that includes four fuels for all countries; this would indicate that coal should be included for all countries in the fuel demand models. If the results are significantly different, then coal should not be included in the U. S. and Canada demand models, and the bridging method should be used to calculate the energy price index. 3.4 Identifying Inter-Country Differences in Elasticities One of our objectives in estimating energy demand models is to determine the extent to which elasticities vary across countries, and the possible reasons for such variation. To identify regional variations in elasticities,'we must specify alternative ways of allowing for regional parameter variation when our models are estimated with pooled data. At the one extreme, we might assume that the parameters of our models are the same for all countries. Estimating the translog share equations (2) by simply pooling all of the data would restrict the parameters aj, 8ji, and 8t, to be the same in each country. it, The resulting elasticities could still vary across countries since relative prices and total expenditures are different in different countries, but such variation would be slight. At the other extreme we could estimate our models for each country separately; in the translog case the a's, in~ ~~~~~ country. gcsetea',i trn th and sI jt's could be different for every While this specification is least restrictive, it is likely to be infeasible due to insufficient data. 32 There are two compromise approaches that could be followed. One is to estimate models by pooling subsets of countries, so that parameters can differ across subsets but are the same within each subset. This might involve, for example, pooling the U. S. and Canada, pooling France, U.K., Italy and Germany, and pooling the Netherlands, Belgium and Norway. Re-estimation using alternative groupings could then be used to determine the validity of constraining parameters to be the same across countries in a subset. A second approach is to pool all of the countries, but to introduce regional dummy variables that would allow a subset of a model's parameters to vary across countries. coefficients a In the translog case, we might assume that the of the first-order terms in the Taylor series approximation can vary across countries, while the coefficients for each country. Ea Si D = ji and jt are the same This would involve estimating the following share equations: + ilog(Pi/M) +Sj +- .t , j o = 1,..., (n - 1) (54) k MkD + Zi &log(Pi/M)+ · Mtt where D are country dummy variables (Dk =1 kk Note that the usual restrictions on the k = for country k, and 0 otherwise). ji and jt apply, but ajk = -1 for each country k. jJk Alternatively we might assume that the coefficients Bji of the secondorder terms can vary across countries, while the a s for each country. and Bjt's are the same The share equations are then aj +U.ikDklog(Pi/M)+ jt.t j = ki J n a + Sik Dlog(Pi/M)+ Mtt ki Mk k m)t+ 1 , j = 1,...,(n - 1) (55) 33 Note that the restrictions on the jik 's are now that jik ijk for each country k, and aMik is the same in each share equation for every country k. Finally, note that variables whose variation is largely regional (as opposed to time-wise) can be introduced in additional to the regional dummy variables. In the translog case, for example, we might assume that a is a function of temperature T (which has both regional and time-wise variation), aj with a = a (56) + bT varying across countries, so that the share equations are Ea=kDk + bT + Eajilog(Pi/M)+ jtt kk i i(57) k S. ZaMkDk + EbT + Ealog(P/M) + 8Mt i k with aMk = aik = i -1 for each country k. We will estimate demand models by pooling data for alternative subsets of country, and by using regional dummy variables as described above. Hope- fully this would enable us to identify sources of inter-country elasticity variation. 3.5 Estimation Methods The choice of estimation methods involves a trade-off between the richness of the stochastic specification (and hopefully a resulting gain in efficiency) and computational expense. This trade-off is particularly severe given that all of our models involve systems of equations (even though for some models, e.g. the log-log models, the systems are not consistent, i.e. "adding up" does not hold). Ideally one would like to estimate a stochastic specification for which the error terms are heteroscedastic and autocorrelated both across time and 34 across countries within each equation, and are correlated across equations in the system. Estimating such a specification (which amounts to full generalized least squares), however, would be unreasonably costly even if the individual equations were linear in the parameters. If individual equations are nonlinear in the parameters (as is the case with our non-homothetic translog model), the estimation might be computationally impossible. We must therefore settle for a more restrictive specification that would hopefully capture the more important characteristics of the error terms. When estimating translog models (which can be nonlinear in the parameters and/or have cross-equation parameter constraints that are nonlinear), we ignore error term heteroscedasticity and autocorrelation within equations, and account only for error correlations across equations.399 In particular, we use iterative nonlinear Zellner estimation, which (under the assumption of no heteroscedasticity or autocorrelation within equations) is equivalent to full-information maximumlikelihood estimation.4 0 However we limit the number of iterations on the error covariance matrix to five; this reduces computational expense while still capturing at least 90% of the added efficiency that results from accounting for crossequation error correlations. Our logit models and logarithmic models are all linear in the parameters, and therefore our stochastic specification can be somewhat richer here. When 39 Accounting for within-equation heteroscedasticity and autocorrelation is certainly possible even if the equation is nonlinear in the parameters. One might use an algorithm that repeatedly linearized each equation and iteratively computed an error covariance matrix and estimated the linear equation for each linearization (see, for example, Eisner and Pindyck []). There is no guarantee, however, that final convergence would ever occur, and if it did the process would be extremely expensive. 40 See Zellner [67] and Gallant [22]. Oberhofer and Kmenta [49] prove that iterative Zellner estimation (iterating to convergence on the crossequation error covariances) is equivalent to full-information maximum likelihood. 35 estimating the equations of these models, we also account for within-equation heteroscedasticity. This is done using the following procedure. equation in the system is estimated using ordinary least squares. First each The resulting regression residuals, which we can label ukt, are then used to obtain consistent 2 estimates of the regional (country) error variances aok: .2 = IT (58) (kt) T-m-t where T is the number of annual observations for country k and m is the number of independent variables in the equation. Different estimates of the ak s will of course be obtained for each equation in the system. We then transform the data by dividing each observation by the appropriate estimated error term standard deviation e, and then re-estimate the entire system of k' 41 equations using iterative Zellner estimation.4 1 All of our estimation work has been carried out at the Computer Research Center of the National Bureau of Economic Research, using the GREMLIN experi42 42 mental nonlinear estimation package on the TROLL econometric software system. This package permits one to perform iterative nonlinear Zellner estimation conveniently and with reasonable computational expense. 4 MacAvoy and Pindyck [j~] used a similar approach to single-equation estimation that also accounted for time-wise autocorrelation. 42 For details on the estimation algorithm and its use, see Belsley [ ]. For a discussion of alternative nonlinear estimation algorithms, see Berndt,-Hall, Hall, and Hausman [8 ], Chow [3], and Gallant [22]. 36 4. Characteristics of the Data Unfortunately much of the data for this study could not be obtained from In some standard sources such as the OECD or the U.N. Statistical Office. cases the needed data (such as retail fuel prices) are not collected by these sources, and in other cases the data have been collected, but have been aggregated or categorized in ways limiting their usefulness for this study. As a result it was often necessary or desirable to go to the national statistical yearbooks of individual countries to obtain data. Nine countries are included in our sample: Belgium, Canada, France, Italy, the Netherlands, Norway, U.K., U.S., and West Germany. The data collected for these countries are described briefly below. 4 3 Consumption Expenditures. categories: These are broken down into six food (including alcohol and tobacco), clothing, durable goods, transportation and communication, energy, and "other". This last category includes housing expenditures (actual and imputed rental payments), expenditures on health services, and any other consumption expenditures. Data were obtained from the OECD's National Accounts, the national accounts publications of the EEC Statistical Office, the U.N. Yearbook of National Accounts, and national statistical yearbooks. The data are measured in current local currency units. ,,, 43 The data used here are part of a larger international energy data base assembled for use in this and several related studies. For a more detailed description of that data base, see [66 ]. Other researchers wishing to replicate or extend this study, or perform studies of their own, can access the data directly through the TROLL computer system of the NBER. 37 Price Indices for Consumption Expenditures. A retail price index (1970=100) was collected for each of the categories of consumption expenditures listed above. Although for some countries retail price indices were available directly, we constructed implicit price indices for all countries from consumption expenditure series in current and constant monetary units. Although price indices for energy are available, we use the energy price aggregator function described earlier in estimating our consumption breakdown model. 4 4 Data were obtained from the OECD's National Accounts, and the national statistical yearbooks of individual countries. Fuel Expenditures. Data were collected for total residential consumption expenditures on petroleum products (largely light fuel oil), natural gas, coal, and electricity. These data were generally obtained from the Statistical Office of the EEC's Energy Statistics and national accounts publications, or from national statistical yearbooks. In a few cases figures were obtained by multiplying the retail price of the fuel by the physical quantity of the fuel consumed; physical quantity data were obtained from the OECD's Energy Statistics tape. The data are measured in current local currency units. Fuel Prices. For each fuel, the data are countrywide averages of the average retail price. 4 In the cases of natural gas and The energy price aggregator requires data on fuel prices. For some countries our data on fuel prices does not go back as far as our consumption expenditure data. In these cases the estimated energy price aggregator was regressed against the implicit energy price index, so that data for the index, could be used to extend the aggregator backwards. 38 electricity, for countries with tariffs the price level chosen was the average price facing an average size household.4 5 When the price of natural gas differs from that of manufactured gas, an average of the prices weighted by the relative amounts consumed was calculated. Data were obtained from the Statistical Office of the EEC's Energy Statistics and Studien und Erhabungen, and from national statistical yearbooks of individual countries. All prices are measured in local currency units per tcal, but have been converted to 1970 U.S. dollars per tcal for estimation purposes. Fuel Quantities. for fuels. Some of our models use physical quantity data All quantities are implicitly derived from data on fuel expenditures and fuel prices. Other Variables. Units are tcals. Data were also collected for net disposable income, population, and temperature. The income data represent total net disposable income of all households, although for some countries only total private income data (personal income plus income going to non-profit institutions) were available. Income data were obtained from the OECD's National Accounts and the U.N.'s Yearbook of National Accounts, and have been converted to 1970 U.S. dollars. Data for the total population of each country came from the U.N. Demographic Yearbook, and are measured in millions of people. 45 Some researchers, e.g. Halvorsen [2], have used the marginal price of electricity as a measure of price. The marginal price alone is inappropriate, however, as has been demonstrated by Taylor [61]. The correct procedure is to use the average price at a normalized and constant rate of consumption, or to incorporate both average and marginal prices. We use only the average price, since that is the only data that is available. 39 Finally, our temperature data represent the average temperature over the five winter months (November - March) averaged over the principle city or cities of each country. The source is the U.S. Weather Bureau's Monthly Climatic Data for the World, and the units of measurement are degrees Fahrenheit. In some cases data for one or two variables was not available over as long a time period as was the case for the other variables used in the models. For example, for Italy the price of electricity was available only for 1963 to 1973, while other fuel prices and all fuel expenditures were available beginning in 1960. In constructing our fuel choice model, more efficient parameter estimates can be obtained by constructing an instrumental variable for the Italian price of electricity and using that to extend the price series back to 1960, rather than dropping the first three years of data for all of the other variables. 4 Instrumetnal variables were used in this way to fill in missing data points in a number of instances. In Table 3 we show the time bounds for all of the data described above; years in parentheses represent data points constructed via instrumental variables. Note that these time bounds do not represent those used in model estimation. Because of the need for overlapping bounds, only a subset of this data can be used for estimation work. 46 Letting PE1 represent the price of electricity in Italy, this variable I was regressed against a set of variables Z1 ... Zn that would include the prices of electricity in other countries and the prices of other fuels in Italy. This regression is used to generate values of PE , for 1960-1962. In using this data to perform the final regressions for uel shares, efficient estimates are obtained if weighted least squares is used. If 1 is the coefficient of PEI, is the error term for the fuel share equation, and is the error term for the instrumental variable regression used to construct kI, the error term in the final regression will be (v + £) for 1960-62 and after 1962. The resulting gain in efficiency is probably small, however, so we do not use weighted least squares. We thus obtain consistent, but slightly inefficient, parameter estimates. For a general discussion of approaches to the problem of missing data, see Pindyck and Rubinfeld [3], pps. 194-202. 40 0 I, CY) N- 0 Uj 0 0 N- U) U, 0 rU) U 0 N- 0 N- U) Ln Ln U, CV, ce) cn r- o r- I I L() LO) LO) Lf) r- r- I NU) o r- Lv, Lv, 0 U, U) CY) I 0 O U,) U, Lv, 0 0 r U,) Lf) _ N- L-1 Ur) 0 U, o Nq I I r- U, U,) r- U,) U,) U, U, NI NI 0 cl) U,) O N r-. r- I L1 U) I U, U) 0 rI U) U, O r- CY) I I Lf) U) 00 CY) CY) N- 0 Lf, 0 U,) Cv) CY) 0 Lr) 0 11) - I 0 I o' OrLr)1 - N- r- I 0 I C L) U) N-I 'IO V) I_ I I I %ID I %Dun) ur) r- Cv , rO U CY) (le) - r- I, IU) Un) CY) CY tr) U, Lr Lr l- '-I r-It Lv, N- 0U,) Cf, N- NIt Le) U, CY) ce~) Cv,) I I I r- rl_ r- r- r- r- Lv, 0I, U, Cr) 0- 0U,) CY) CY) 0- r- Lv, CY) M rI r- rI r- L) U) CY) CY) 0lU) 0- I, r- I NI Lv) U, 0 r- CY) M V) 0I 0I 0 %OI-C-C - Cv, 0- N- I I 0 U,) U,) LO Lv m clv r- . U, 0 00- N N- l 0 U,) U) Uv) 0- N- 0 0 0- I, U,) 0 I r- H- 0U, U,) 0 VCo Ul V) 0 0I, 0-co U,) cv, I, 0 r- N- U, CY) 0- 0 N- U, U) Mv U,) 0r- r-.I co Lr U,) HII r- -i rI U,) Lt] U) N- 0- 0 0 I I Lr2II U U,) 0 0U, Lf, 0 Loi, V, 0 Co U,I: OD 0 NI U, 0 0I Co ?, °iX O rI U,) 0) r- I Co U, 0 0H- U, 0 V, Lv, 0- r- Co Lv) Lv) CY) U,) Ur U, 0 r- r- 0 U,) I, U) Lr) H- Lf, U,) i 0I, Lr 0 0- o NI U) Lf) Lv, U,) 0) 0 H- U, I, I, O 0 0 CO Co L OD U, Ui U, Ln 0 -j 0 Lv 0 0Lv, 0 o0 I'D ID o or- N- I U,) I 0 o U,) %-C IC) ? o 0 o lZ oI u) U,I I O Lrl Co I U) r- 0 uLf 0o %D I 0 r- 0) r- IC ~o 0 0- fl o 0 10 o1.1J'1C uNC1 :N %O D I- Co H I-C r-- I I 0 U, Ln I-0 0 r- o 1 I'll4 _ I I0 0 N1 I I o0 I-C - U,) U,) Cv, r- N- IC4 0f Lr) r0 iv, 01 O CY) I Lir) I C) r- IOC Co I-O I' -- r- Lr) 0 0 Ln I L? 0-I UI U, Lv H 00 O V, CY rU) CY N- U) 'IC CY) r- Lr r- NI 0 U, CY Lrv NI 0 O V U) Cl) Lv, rI r, U) Lr U) C'-, NLv, 0 Lr) r- Lv, CY) r- Ch V) m 0- 0- 0' U,) cvn 0- 0l- Ln N 0CY) 00U,) 0 0U,) H- 0I, 0rU) Ul cv r, Cf) UI 0 -. U,) V) 0 0- 0 Lv, U,) U,) Hr- HN- I I H- 0) '%C U, o o I I U, U,) 0-4 I, 0 0- U,) 0 NOIQldlSNOD Lv, 0 U) 0- 0 Lv, 0 0I, 0 U, U,) HI H r-l II u) U, - I 0 U, 0 10rr-I I U, Lv, I I 10 0- l- o U, -q o 0 00 NIC %. o- %O l 0I 0) U, H rI 00 U,) N- IH Co 00 - 0 - N1 Cv, r- I I 0 0'- Lv, I, 0 I-C NU, U, U, U, 4- I U') Ln 01 cv, 0U, I Ln, rI cv Lv, a) .1i 0 of -H10 0 p 0 4A 0 C-.) Ca Er-l 0) rxz I-, IC 4-~ -H -e C-, 0-~ z H xn SfDINI aoIldc I'TrJT "'fn.T.T rp fl J7 0 a) H- cv, 0-_ UlI Un Cr -Ii CD co Lv, C -4 0 U, cJ 4-i -C) .- 4 .H 0 C,) SHDIUC c sd rl UI I-i 1.4 .C SNOIf.JJSNODX Cm 14 4-i HC- Lv, i i cd I-4 E Lv, 0I 4-.J 0 c- r- r-I Lv, H 014 CY) H N-f C CE N-1 I Ln U,) 41 Our basic translog models requre consumption expenditure shares and prices. It is useful to examine some of the share and price data before turning to the estimation results. Table 4. Data for 1962 and 1970 are shown in Note that the energy price index is not shown, since we compute a price aggregator from the fuel prices. 5. Statistical Results In this section we present the results of the estimation of the models set forth in Section 2. We begin with the estimation of the translog energy price aggregator, and present relative energy price indices for our nine countries. Next we present estimation results - and the implied demand elasticities - for Finally, we describe the results of the static and dynamic translog models. estimating the logit and log-log models. 5.1 The Price of Energy A price index for energy was obtained using two methods. First, a translog aggregator was estimated assuming a choice among four fuels in all nine countries. The estimated parameters for this aggregator are shown in column 1 of Table 5 (standard errors are in parentheses). tricity, (2) oil, (3) gas, (4) coal. Fuels are indexed in the order (1) elecNote that the parameter Yo is determined so that the price of energy is 1.0 in the U.S. in 1970. Second, the "bridging" method described in Section 3.3 was used. This involves first estimating a four-fuel aggregator for all countries exeept the U.S. and Canada; here Yo is chosen so that the price of energy is 1.0 in Belgium in 1970. The estimated parameters are shown in column 2 of Table 5. Next a three-fuel aggregator (leaving out coal) is estimated for all nine countries in order to find the price of eergy in Belgium in 1970 relative to that in the U.S. in 1970. 42 I - _~ N :n H. 0O -It .0 ) - N N O %D rD IF N Lr, NS '.0 .) '. U' cc H H O H H H * V) i )l ,I U) c,~ ND cc 00 AD t-Q '0 U.) -a ao 0% -I0 CO ,-4 Ha O f.cn 4m O '.0 u0n oM (V. Ln -t (', N H lV, 0 N 0 0 -t N a N H .* C--A 0% Ln. Sr 4t 00 C'4 --T '0 Ce cO H- -t oc -It Ln. 0H- 0% Ln 00 o0 u~ 0 p° Cv) O o0o -* r- v.D Cc C) cc 0 M o o O N C O U.) rC) o') 'o H OD o .)* 10 CYN N- o o 0 N H) cc 0 O PH on o U) x a om 00 c^l 0 I.-O o0H Cc cc D 0 A 0 H* oN 0H H o r- cq H rH o00 C) co O o H H H o a) A U) 0 C) cs C 0 C o -r -t 0 oO r- -Iu co o'Q Ci~ H H H H N N a OD ON 0o 0 -t IC0% O0 0C) *1.c0 CY.) Ln. H U O 4(I N CD F.- C) 01 O C1 o~ O r- cc o~ ooou,- H 00 H co eo 0s H O H N CY) H '.0%4 0oDo 0% N 00 H (V.) H H C C) H ccD oo C) 0 u% 0o C) H 0LV, 4t If) 0.,'0 0 oo co tn cc co H '.0 N oo 00 UX H 0% - OC) 0 C -0 U) N .) Q 00 0 oV, o C4 cc 00 N roo CO rn O -~ r- rD Ol 'o O 00 0; H--,I 0 '.0 00 r- H 10 -o 0 -t' - N N N N O> 0 N 0a H C) r- r- N 0N C'. CY) --T cc CV, * rC) as I-. -s H H C) H ONO H H r-N O O CO CO O U.) 0 C) U, 0 00 CV, o0 C) O0 H - 0) r- H- N H- N 0%*% C% 0 9 4 HH co PI H 0O N cc uL) N 0%A H C) N H H 00 - C)4 CY) 0 C (V.) *~ 0% (V O H Ir) 0 C) C) C) C) H NCO CY O H H (V. O) 0o C) Hs N H Ce) H N %D C) H N C1 o CV *~ * H H CO H N H CY~) N4 CU -IC w .iJ ..q oo cc CY0 ?--I I H N 0 N 0ae o o *~ c H C) H 00 0% HI ONO O H O O -t C) O 0% C) H 0 0 U) Ot. O (V00 01% o CO r-. N U.) C Ln u, CO -It Cl If) co 0o 00 0 1. N cc 0De~ O oo C) oC Cf) 04. H 0S C) cO o o' U) 00 cc C) H H 0%r co - o N OD C) - 0% L.) H ,H 00 C) 0% I_H OH -I CO cc CO O -n*~ H H * p -N u~ c1o r< 0 *Y~ N ' 4 ,C) H ' .. o H H CY) ccl C)H co H H Ns '0 14 N N It Lr 00 · 1 .~. N oh 00 - - - On r- a H H - H< o£. %o -4r (cV. o r-la 00 Cn. -,1 -t '.0 H Ct %~t~ N cn H -It -It 0 0o H - '0 U,- 0} - N 0% U) o 0 I H N r0 | -0 r_ to u~ C) m %D _O -t o~ o4 I I 0% H1 co H cn %O '0 %D J I II NE N *~C N0 .-I zU) *I l 0D H CD . C- l N -0 a% %D .H . a) U CN 1-i co N . '. 0 H C) C0% H- N..0 %D H H O H m H C.0 0% H r F.- H N '.0 0 HO ,.- '.o 0% H C) f. N1 C) 0% ao r-. H4 OH r. CD - 5-I U) ~4J s-i 4J a) U u o 0 o 43 En ,-4 '0 P-.-I N i o o o o o0 0 o Un -It -I oo ,l_ -I cn H N N 0 r-_ N 00 cn* *4 o 0 o o w C 4 10 un Y CJ 0 CN I cf) c' L) en a) 4-I) H oo 0o * ai U I'D E-1 *ri N z .r- .- o-It* X4) 0 ki r- 0Cq L() -o I c0 o4 N r_ -;T. 0 o C_ 0 H0 0 0 0 #- 0.' HC', M' m. H M 0 0 : 4 P4 0 :j ( o p4 $i0 i CU o00 (ii . a) N C/ H o o c- O . D 4 : rX *- r4 z U to t N- H o o o - Lf) U) O * TI C; - r- 0 10 r-l o 0 O CS 1 0 o O O c; 0 *- N- ON' '. NT N- 0 o% Nt cn N-_ 00 I 4) - ;> H 4 H- rH ,r- ',T en -n H 0- C. Ih I '.0 0 C', H ' Lr) -T Cn N r- ;> U-) C, r -< r- L( 00 H 0 00 o0. CI I I I >- ;,. N 0 . 00 0 0 H4 U- 0 c; 1 0 N- oo00 Hq ,-O rH CN ON 00O -H H 0 N4 N 0.' -It 0.' 0 L 0 0 Ln 'D 00 VL 0 o- 0 0 (S '.O 0 LO) H 0 . 0 0 * Ln 0 0 * I O Un n I Ln. O , N 0 C; - CN 0 0 0 * c; 0 0 N'.0 0r 0c cn ,H4 '.0 O 0 C; H e' 0 ' V) U) O 0I N cn O 4N 0 0 * I 0 CN HN 00 0 0 Ul) NT C '.0 0cI CY) 0 "v %-. 0. 0 0 0 H 0 O o o o o C, -I 'D cI ; I ;> 0 0 0 00 00 0 -%e 0 Nt H ,- r) Nt o - uL') .0 o Vu IT 0 H- % %0 H-q '.0 N- c; ;o 1 I ;.. * 0- o0 C -H N * o - 0 H- 0 0 O' 0 *- 1- rZ 4i o O O SI* H I o 0o 0 04 H C-* ':r- 0 %. o e\ CN ~o o o L) H eci -I 0S fLO * t - I 00 0 o~~~~~~~ c; 0 Nl U-) N 0 0 H fN- o o H 0 o 0 r4 d 0O o* o -cn C) N 0 0 0 <4 ,-% 0o i/1 C0 0; Lt) o N OI N < '. N- 'IO '.0 o 0 O0 o I CN 0 0U -It un *~o o* r- N Hn N c L U 04 O c; c; I 0 U) L1 C r-. 0 0 - CN 0 * * 0 N U-) N- Lu Lf) N Hd 0 - 0 * I T ,' Ln CN '.0 Nt -I T ,-4 -; On 0 CY) T cn cn I U%O -a O 0 -I O H cV) 0 t N '.0 0 rH Nt N - NT Ln -i -' cV) C, O 0I -C -IT c Hd M' H00 - 00 I I 0.' O' LO O co HCe 0oII ;; .a I -, T -- 0 0 ,9 1:34 - 44 This relative price was found to be 1.5239 (parameter estimates are shown in column 3). Finally a three-fuel aggregator was estimated for the U.S. and Canada (see column 4). Now using the Belgium-to-U.S. conversion price, the price indices can be converted to a U.S. 1970 base. The two price indices are shown for all nine countries in Table 6. Note that for all countries these two indices are quite close to each other, indicating that the relative size of coal shares in the U.S. and Canada do not distort the fit of a four-fuel aggregator (and thus a four-way choice model) that is estimated over all nine countries. We therefore use the four-fuel price index in estimating our consumption breakdown models. 5.2 Static Translog Models We begin with the breakdown of consumption expenditures. were estimated, and the results are shown in Tables 7a and 7b. Eight models All of these models are homothetic; the computational expense of estimating a non-homothetic translog system with six categories of expenditures proved to be inordinate. In the first four models all of the parameters are assumed to be the same for every country. Model (1) allows for time-varying preferences, so that the only restriction imposed is that of homotheticity. additional restriction of stationarity. Model (2) imposes the A likelihood ratio test was performed for this restriction, and the test statistic is 112.3. This is well above the critical 5% level of 11.07 (5 degrees of freedom), so that the hypothesis of stationarity cannot be accepted, conditional on model (1). Model (3) imposes the restriction of explicit groupwise seperability between energy (category 5) and the other categories of consumption. The test statistic for this model 45 mo -I cq m oo Oa -- C1 oo aW un 0 C, - , - , ,-,-, oN 1 %D cr fl c" 0.c ,, -Itr-i w0" ,'4 O0 r'. - ,- , ,- -,T, I r--- r- o I I '4 Ca J.JH cs .,( U n o'o o 0 o m o 0 o H "0 H H H 00 -, 0 N I'D 4 I In c - O D - - H I I (ON 0Nv un %D" n HHH 0 I V-i to O 0 0 oo W_ I I I I I r- r-q r-I r-I r-I r- ,H -t u'n ..t ,. HI I '-4 un n m w O w unCOC r- un H I I I CH C, I I I m -4 H M'MHH DIn H I III T m m oun W a r- 0 i*.r- %.I U)I'l*n. -, . -,*' -, m* m 4 * , , ,~ , ~~,H ,~ -T n -I -T m m -- n xD -. ,-i , -,-I o o I 0 o.,.,1 r. u' , o,4u'l oo o,, oN ao 0o flr C rsN <@oCnO r-4 I HHHH<H<H H HHH H-<H C1l 4-J O) N o N r-OCOOS e a c)N < v CN 0 U) :j L - r -1 r- -I 0 0 t) o o u) koo ul r /,-I 0 Co " H H r- " N H c-o , ---"OX Z oo oC7- oo ,_ . ur'lL C.. .I- c-4 - OI H Ln r-q m~ rl-I mn0 oCLr O O %D O b % n H 0 -H <1~ H 0 - -e- 4.C I "0 0 I: r" a) ,C bo Or e _.o "0 rCq -I 0 ri C" 00 C" II rO OaN HO 1. O Ln O 0 %D a H %O L rH 10 D o C- %' o Oo L-I ) Nt -I HH H n o o o-,, Trn CY) -z o0 ci I -) HH HH HH I .- , o ,m oo .- ,o , It o o ,- ' U,-1 o-H, ,,D 0 oI ,, ,- 1 o o oo r - - O -I ,- I-.,o c', 0o o o -r- ,- ,1 - oO CY) ,-, CY) OD - ON o -I , . O I I I HHH 0a O 0o ~ ~ ~oo n<Ho ~ N--0crcOaHOO'0 , 0 0| ,a crLOcrN-N-OOC)H ££ ocuiC< -r- < n co- I-,-- M C4 4% HH -IoC. -i C ,.I cV3CY)" N , M * *CO*C*.- . * . .t-* I I IH<H H HHH H HH HHH I (_ un H H " "s r-i r- Ha r- r " o CN oo %o-T , r- O o H-HO 0C 9 0 0 I I I H 4 H4 H (,4 oC cmN OO mOmc cOH-i r r r, Lm o LI oo 01- o cun 00 Ln -1 (N oq4 V- I vo IHHHHHHHoo X cH H-o HH W I I V) . m 14 oci) I II I (a/ I~ :3 0 u. H~oooo~oooo O ON -- M 0000 t 00 000 4 N Ln -I V- I I , * H* *HH *oH H* H* H rHH IHHH ~I0o HHHH * * 00 Nm Mn u Ln 00 O 0 O O uLn un N -O - MD Ds COo oM n COn - c aMn0 0 H 0 H0 OHHH 0H 00 CO r C Lf COcNIO T r. '4O-ArC4 %Dw IHHH<HHHHHHH co CON)-OCOOOm C - m -I n M ' M n %I'D %O%OD 'D %C)% %DLn %D urt Ln (-N I 0 I l~(3HO~ ~ %1 , ~i~,~~~~~~,- ,-~H ,-~ lH~ HA ~i ir- lllll~~~l~ I 00 44 r HU, ,YmmC 44 0 o0 o 0 o Ln r- I ONil Ln Ln on o r- o . C-0 Ln csT %D o II IHH HH r, r-%O~~L ~ (N L UnH H HH -I HHm C- m w I .oTO oN · ,-oo o o u}~o 0,-h LO cN H 0'n 0a Lm o0 CO0 0 r r--r- -o < V-' OU :3 Fx.4 O %D r I I I I I I I o ooooooo I I I I I I -H aI a af M cr r o ooj W %, pq-V I I I H HHHHH 4H HH H - -O0 ' :2 un t U I I I "' r 0 0 -I C .. L, U) r- Owa0 0 ,-iCN , n n n n ,D ,-,- - %I0 ,- ,- , -,- -1-,- -, aw rO O 4 oI r4 -0 r o- a o- r Ca a- oJ - a- r- I I I I an %O o- c un Ln I I , oo oH0 O-r- C; c ON0 HI "c O rn o oC; o NCmOO O CN CO n n n I' %OD xQD 1 1. 1.0 ID %D £ . rDN- P r, t0NmC HH I C -.ln I' I o r-1 HH 4 - <r - O r -q o a c _{ -4 H H H N H 46 Table 7a - Parameter Estimates: Consumption , 3. Homothetic, I 2. Homothetic, Parameter Ol 1 a2 3 a(4 a, 5 a1.6 t11 12 13 '14 15 B16 21 ~22 ~'23 P'24 P'25 ~'26 '31 ~'32 33 34 335 36 41 42 43 44 ,45 '46 51 s52 53 54 5 55 5s6 61 62 63 f64 65 66 B1T ~2T B3T B4T 135T II6T 1. stationary omothetic -0.1132 -0.1274 -0.4449 -0.0541 -0.0202 -0.2402 -0.0767 -0.0112 -0.0314 -0.0568 -0.0076 (.0043) -0.0908 -0.1240 -0.3182 -0.0340 -0.0319 -0.4011 0.0722 (.0095) (.0178) (.0083) (.0028) (.0053) (.0014) (.0080) (.0073) (.0029) 0. 0139 -0.0140 -0.0332 0.0119 -0.0408 0.0139 -0.0032 0.0334 -0.0032 0.0012 -0.0420 -0.0132 0.0334 0.1377 -0.1650 0.0079 -0.0073 -0.0112 -0.0024 +0.0257 -0.0348 0.0021 -0.0018 -0.0314 0.0257 0.0798 -0.0740 0.0129 -0.0130 -0.0568 -0.0348 -0.0740 0.1657 0.0225 -0.0226 0.0076 0.0021 0.0129 0.0225 -0.0452 +0.0001 -0.0073 -0.0018 -0.0130 -0.0226 -0.0001 0.0446 +0.0017 +0.0004 +0.0096 +0.0016 -0.0008 -0.0125 (.0030) (.0051) (.0026) (.0086) (.0258) (.0172) (.0052) 0.0000 -0.0332 -0.0032 -0.1650 0.0325 0.0238 0.1451 (.0180) (.0048) (.0023) (.0033) (.0028) (.0002) (.0006) 0 0 0 (.0011) (.0005) (.0002) , ,,,, .- -0.1083 -0.1253 (.0111) -0.4311 (.0041) -0.0381 (.0014) -0.0453 -0.2519 (.0056) 0. 0760 (.0015) 0.0126 (.0085) -0.0233 (.0073) -0.0392 (.0029) (0.0043) (0.0096) (0.0173) (0.0082) (0.0029) -0.1018 -0.1109 -0.3022 -0.0752 -0.0401 -0.3697 (.0018) (.0029) (.0057) (.0034) (.0008) (0.0049) (0.0049) (0.0076) (0.0074) 0 -0.0261 0.0126 (.0028) -0.0013 (.0059) 0.0235 (.0026) -0.0020 (.0009) (0.0029) (0.0049) (0.0026) 0 -0.0328 -0.0233 0.0235 (.0288) 0.0745 (.0172) -0.1626 (.0051) (0.0243) (0.0171) 0 0.0879 -0.0392 -0.0020 -0.1624 (.0176) 0.0423 (.0045) (0.0186) 0 0.1613 0. 0119 0.0012 0.0079 0.0238 -0.0283 -0.0067 -0.0408 -0.0420 0.0000 0.1451 -0.0067 -0.0556 non-stationary separability of energy Homothetic, stationary, separability of energy 0 0 0 0 (.0024) 0 0 -0.0261 -0.0328 0.0879 0.1613 0 -0.1903 0.0015 0.0004 0 0. 0091 0. 0010 0 0 0.0003 -0.0123 (0.0002) (0.0006) (0.0011) (0.0005) (0.0002) ~~~~~~~~~~~l Consumption are: 1 1 - Apparel, 2 - Durables, 3 - Food, 4 Transport * categories are: Consumption categories Communication, 5 - Energy, 6 - Other. 47 Table 7b - Parameter Estimates: Consumption 7 5 $4 6 Homothetic, stationary, country dummy variable p4 Homothetic, stationary, country dummy variables, seperability of energy 8 Homothetic, stationary, country dummy variables, additivity. Homothetic, non-stationary, country dummy variables i i~~~~~~~~~~~ n acxD2 DI D32 alc1 D 2 1D aD3 aD 4 aD 5 1D a2D 1D7 a21 1D 22 24 :tD7 r D 9 ;vD a3D aa 3 D2 D3 3 c3 D rvD2 a4D 5 a 4D 4 D8 a3D 7 a D9 a4D2 a5 D D4 4 OL4 avD aD5 a D7 D 8 6 CLS -0.0986 -0.1113 -0.1146 -0.1027 -0.1141 -0.1775 (.0047) (.0039) (.0048) (.0051) (.0028) (.0134) -0.0882 (.0046) -0.0852 -0.1266 -0.0899 -0.0592 -0.0587 -0.0286 -0.1358 0.0807 -0.0610 -0.1473 -0.1041 -0.0329 -0.3081 -0.4380 -0.4995 -0.3471 -0.6594 -0.2721 -0.2488 -0.3854 -0.1189 -0.1619 -0.1093 -0.1064 -0.0558 -0.0882 -0.0564 -0.0362 -0.1150 -0.0507 -0.0282 -0.0381 -0.0325 -0.0546 -0.0291 -0.0504 -0.0357 -0.0405 (.0011) (.0051) (.0049) (.0052) (.0046) (.0051) (.0027) (.0193) (.0041) (.0016) (.0046) (.0141) (.0152) (.0130) (.0147) (.0064) (.0380) (.0116) (.0041) (.0133) (.0068) (.0068) (.0063) (.0069) (.0034) (.0210) (.0059) (.0020) (.0065) (.0038) (.0029) (.0036) (.0037) (.0019) (.0100) (.0034) (.0008)! (.0036) -0.1094 -0.1153 -0.1262 -0.1123 -0.1198 -0.1806 -0.0992 -0.0864 -0.1366 -0.0944 -0.0635 -0.0632 -0.0337 -0.1376 0.0627 -0.0647 -0.1476 -0.1089 -0.3993 -0.3240 -0.4474 -0.5129 -03519 -0.6362 -0.2870 -0.2541 -0.3989 -0.0858 -0.1388 -0.0807 -0.0761 -0.0416 -0.0697 -0.0255 -0.0303 -0.0847 -0.0510 -0.0316 -0.0335 -0.0300 -0.0522 -0.0309 -0.0477 -0.0371 -0.0371 (0.0032) (0.0037) (0.0036) (0.0042) (0.0023) (0.0117) (0.0035) (0.0011) (0.0043) (0.0037) (0.0048) (0.0036) (0.0041) (0.0021) (0.0183) (0.0029) (0.0015) (0.0035) (0.0118) (0.0150) (0.0109) (0.0129) (0.0054) (0.0361) (0.0092) (0.0043) (0.0111) (0.0056) (0.0073) (0.0054) (0.0062) (0.0029) (0.0221) (0.0045) (0.0022) (0.0053) (0.0006) (0.0006) (0.0007) (0.0005) (0.0008) (0.0008) (0.0006) (0.0005) (0.0005) -0.0868 -0.0839 -0.1050 -0.0926 -0.1233 -0.1346 -0.0904 -0.0850 -0.1242 -0.1192 -0.0850 -0.0879 -0.0616 -0.1484 -0.0782 -0.0822 -0.1478 -0.1336 -0.3138 -0.2351 -0.3641 -0.4206 -0.3279 -0.2818 -0.2299 -0.2504 -0.3206 -0.0941 -0.1455 -0.0902 -0.0868 -0.0463 -0.1089 -0.0331 -0.0303 -0.0949 -0.0510 -0.0316 -0.0335 -0.0300 -0.0522 -0.0309 -0.0477 -0.0371 -0.0371 (.0019) (.0020) (.0019) (.0021) (.0017) (.0025) (.0018) (.0016) (.0017) (.0018) (.0019) (.0018) (.0020) (.0016) (.0024) (.0016) (.0015) (.0016) (.0066) (.0072) (.0066) (.0075) (.0059) (.0090) (.0063) (.0056) (.0061) (.0023) (.0025) (.0023) (.0026) (.0021) (.0031) (.0021) (.0020) (.0021) (.0006) (.0006) (.0006) (.0007) (.0005) (.0008) (.0006) (.0005) (.0005) -0.1013 -0.1135 -0.1180 -0.1046 -0.1199 -0.1567 -0.0953 -0.0936 -0.1292 -0.1000 -0.0697 -0.0685 -0.0421 -0.1329 -0.0593 -0.0611 -0.1356 -0.1147 -0.4226 -0.3509 -0.4672 -0.5273 -0.4085 -0.4007 -0.3444 -0.3389 -0.4165 -0.0806 -0.1244 -0.0791 -0.0714 -0.0325 -0.0835 -0.0163 -0.0114 -0.0812 -0.0548 -0.0345 -0.0399 -0.0372 -0.0550 -0.0412 -0.0518 -0.0355 -0.0442 (.0059) (.0047) (.0057) (.0060) (.0043) (.0052) (.0067) (.0034) (.0062) (.0049) (.0054) (.0046) (.0049) (.0033) (.0052) (.0049) (.0034) (.0046) (.0096) (.0093) (.0088) (.0096) (.0057) (.0109) (.0091) (.0053) (.0091) (.0065) (.0068) (.0059) (.0064) (.0041) (.0071) (.0063) (.0043) (.0060) (.0061) (.0042) (.0055) (.0053) (.0040) (.0036) (.0065) (.0033) (.0054) 48 Table 7b, cont. Parameter 5 6 7 8 mI -0.2590 (x6 D -0.2414 -0.2304 -0.2926 -0.1265 -0.4719 -0.4468 -0.2284 D3 (63 (I D4 (6 D 4 (X D (.6 D (x D (~6 6 6D7 !hD8 (I6 lS1z I' Q DI 2 p1 0.0407 I r ^ -u. UU4/ 0.0381 -0.0067 -0.0117 -0.0558 -0.0047 0.0331 -0.0769 0.0091 -0.0004 f314 t32l 13 5 16 1321 322 B3 3 324 325 -0.0769 0.1769 0.0107 -0.0120 -0.1368 -0.0067 0.0091 332 133 t34 1335 336 1341 342 -0.2338 . (.0072) ", "" k. uuoo) (.0081) (.0057) (.0033) (.0051) (.0088) (.0052) (.0024) (.0278) (.0138) (.0054) 0.0107 (.0113) 134 5 0.0245 (.0037) 1354 0.0245 1355 0.0004 -0.0009 -0.0558 1356s is6 136 0.0399 362 -0.1368 -0.0232 -0.0009 0.1769 33 364 0 (0.0029) 0.0324 O.0056 (0.0083) (0.0060) 0 0 0 0 0 -0.0681 -0.0052 0.0298 -0.0659 (0.0049) (0.0086) 0 0 0 0 0.0077 (0.0056) 0 0 -0.0659 0.2037 -0.0206 (0.0283) (0.0155) 0 0.0056 0.0077 -0.0143 352 1353 (0.0071) -0.1496 1343 -0.0232 -0.0117 -0.0004 -0.0120 0.0353 -0.0052 0.0336 0.0324 134 4 1346 351 -0.3352 -0.4189 -0.3193 -0.3084 -0.3018 -0.3656 -0.5166 -0.4495 -0.2897 -0.4446 0.0399 0.0382 f336 f3 1 -0.0206 -0.0025 0 0.0099 (.0028) 0 0 0 0 0 (0.0135) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0336 0 0 0 0 -0.1496 0 0. 0099 0 -0.0681 §T f2 T 3T 0 0 0 0 0 0 0 0 0 0 0 0 0 4T ~i5T 0 0 0 365 364 6T - -0.2601 -0.3268 -0.2490 -0.2349 -0.2968 -0.1453 -0.4759 -0.3313 - 0 0 0 0.1743 0 0 0- -0.2408 -0.3071 -0.2273 -0.2174 -0.2513 -0.2585 -0.4310 -0.3851 -0.2142 0.0351 0.0050 0.0133 (.0079) (.0040) (.0076) 0.0017 (.0055) -0.0081 -0.0470 0.0050 (.0043) 0.0057 (.0022) -0.0063 -0.0026 -0.0065 0.0048 0.0133 -0.0063 0.0335 -0.0172 (.0071) (.0057) (.0026) 0.0158 (.0053) (.0173) (.0105) -0.0391 0.0017 -0.0026 -0.0172 0.0204 (.0118) 0.0074 -0.0097 -0.0081 -0.0065 0.0158 0.0074 -0.0027 -0.0060 -0.0470 (.0035) (.0051) 0.0048 -0.0391 -0.0097 -0.0060 0.0970 0.0005 0.0007 0.0054 (.0002) (.0002) (.0003) -0.0012 -0.0001 -0.0053 (.0002) (.0002) - 49 (against model (1)) is 112.8, which would indicate that the assumption that energy is separable from other commodities is not supported. Model (4) imposes the restriction of explicit additivity (together with stationarity), and since the model is homothetic, this means that all of the secondorder coefficients are zero. The test statistic for this model (against model (2)) is 442.3, and this is also well above the critical 5% level (25.0 with 15 degrees of freedom), so that additivity cannot be accepted, conditional on model (2). In the last four models country dummy variables are introduced so that the first-order coefficients (ai) can differ across countries. Model (5) is stationary, so that the hypothesis that the a i's are the same across countries is tested by computing the likelihood ratio statistic for model (2) against model (5). The value of the test statistic is 1345.5, which is well above the critical 5% level of 64 (48 degrees of freedom), so that we must assume that the ai's indeed vary across countries. differ considerably across countries.) (Note that the a i's in model (5) In model (6) we allow the a.i's to vary across countries, but we add the restriction that energy is explicitly groupwise separable from the other categories. The test statistic for this model (against model (5)) is 2.92, which is below the critical 5% level of 11.07 (5 freedom), allowing us to accept the assumption of energy separability. egrees of We view this test of energy separability as more definitive than the last one, since it is based on a model for which the ai's vary across countries, which we have seen is essential. Model (7) adds the restriction of explicit additivity. The test statistic for this model (against model (5)) is 180.65, which is above the 47 5% level, so that we do not accept additivity. 47 Finally, model (8) includes no We also test the hypothesis that the ai's are the same across countries conditional on the restriction of additivity (model (7) against model (4)). The test statistic is 1607.6, which is well above the 5% level, again supporting retention of the country dummy variables. 50 restrictions other than homotheticity. It differs from model (5) in that pre- ferences are allowed to vary over time. The results of these tests indicate that a non-stationary model based on a non-additive indirect utility function with first-order coefficients that very across countries is needed to estimate price elasticities of consumption expenditures.48 We now retain models (5) and (8), and present their implied price elasticities in Tables 8a and 8b. Note that the own price elasticity for energy is close to -1 in both models, which is what would be implied by energy separability. Own price elasticities vary between -.5 and -1.8 in model (5), but only between -.9 and -1.4 in model (8). The rather large elasticity for food in model (5) is somewhat surprising, but this probably does not reflect a true price response, and is instead the result of food prices rising slightly with food shares dropping considerably as consumers spend larger incomes on other goods. This is supported by the lower estimate of nFF in model (8) where preferences can change through time. Cross-price elasticities are presented only for energy in order to save space. Most of these are near zero except for food and transporta- tion. The negative cross-price elasticities for energy and transportation are surprising, but again may represent something other than a true price effect; as energy became cheaper during the 1960's, the infrastructure grew that could make possible expansion of the transportation and communication shares. We turn next to the breakdown of energy expenditures. estimated, and the results are shown in Table 9. homothetic. Ten models were Of these models, ine are Again, we had considerable difficulty estimating non-homothetic models, and the one that we were successful in estimating - model (5) - is somewhat suspect, since although convergence of the nonlinear estimation procedure was reached, the resulting parameter estimates are rather large. 48 We must stress, however, that these tests are all conditional on the assumption of homotheticity, and we have not tested the homotheticity restriction since we have been unable to estimate the non-homothetic model. 51 To -o CNi C',l %O I Lrl , oo o I HrH H I I * I. f-... I*I 1 1 N U4 o t.- H H , 0o I* I I *4 1I * -I --- c ,- I t.1 . H I 1 II * 1 ) L * * II 0O- %.D - oc n o o, I 0 I -CN Os co . * -a* I I n .' c'o * I I I 1 O I i II H H II H 1 HH r %O I H I ,0 Ln U) 1 1 I * oH H ,1t- OO D N , % r- 0c H HH cnH o I M n I H-I , . oo 0 cn ' m I I I I I H C..4~ 00 · -I eC-. Lrn- H ,-c, Io ,--i ..- ION -. H L) 4J (U o0I 0 ,.-I r-q cu 0 z oJIJ II 0 ,-I ~4 -H 0 n~ CN I U)q H P4 l. I I -o ,.. o' HH H Iloo o o o -I ~. ,' oo HI HI HH I I I I HH II HH I ,- I mQ ,-I -Iq I I I I I I I oo oo 1 Oc. HH · HH 'o r- C I1 I1 I I oo I I I I - oo *1a *%ts - HH HH I I I I I I I I I I I I 0I *0.II - a o00Io ' I* II II *-ao ,Ia o co o T ONcc o,10 IO m '~oo IT ..i-I1Il -,1 M~ c,~ m. roo I I -ac-)H OoOC HH <.O HI7 II · -Im-I .c oOO c,,.. O1-o cn, Ln.-I o o o-L~ co ,,, -I oo l r- n co 1l @. cn -I l un CI -I2 LI I0I I '-. o~ U) I I ' I r-M~ ¢ U0 O Lm I -C H .1- U) w I oqI 1-4 ,4,4 I coI cn U) a) 0 H I I r-. Il 1-I XI I r-II) ,-q -4 w hII U) 0 H 1r-4 N HH I .--. 11 bo p 9-I I 1 1 I Ml H-IH D.o' O 00o 0 0 I oo Cc oo -. co II I I I I '0 II H U) 0 FzJ mH 0O'. HHI I HH I I I I I* I 1 . 1 HH o0 o I *4 I * 1 1 Ln H L U oIcs oo 0 c I *4 I 1 -a I I 1 N 9:) c'l oO CcX riI 4 co U) Io I %O rV- I c7 a% 'Ioa 7 H I %O T- C-100 crco C\00 ON0'ON '.0N0a'.0%0 H H~-- H H4 r H I H coo cc co " co D %0 o. o'D o CO oD %O0 CNJ00 H co % %O a-lla. H HIr- r-I -I -H -H H- I 9¢ p rq PF 5= r-7 97 E, FH 9=' w F F ~~~~_r %= -¢ 5= ~ S 5=~ 4H H N H 5:- 1: 52 0: oo 0 -1-1 HH, HH I I I H Hr- I C'r < I O~0O C! I I - oo c J oCi -O H e00 O I I I ~- I -H I m OO--o.O0,cm4\ Co<D -4 r- I I 0 HH I I I I cn 04 -, 0 0o H I H IH cn o OC I oo r- H o r-oo 0o oL r- >~ C" c4 c1 I CN c; ooc oo c; I : I C I C; Io C~JC~4 HH CNl lt~ I I I U I I-4iT -I I f 00 C ,-4-I 00 0 0 0 0 00 .O * *~ . * * I . I I I CN o* Oo o o 0i V'% HH- 0 o0 00o ooQ o I I 00I *;V-4 * ?-.1 *00 * 00 * * ooo *0 * oo~~~ oo oo 0*0I 1-4 w 4. r10 r r -t H* co * 1 i-a0 4t) 1 1 1 4 00 OM MC4 JH r-*r Ln Lr I II II II 1 00 c c oH 00 oH 00 Co C~CoC HH 4 C LI Ln I cn ,-4O om cao oo o0D coo0 c 40c0 c1 N c, I I I u'r N I'D '\ ,- HII ) *; * 0sW -II 'Ii 0 z o2 *-I 0 l-i 0 H. H. 1 c' D H'.0O HH . HH OH-r-0 O -. r. 00r o 0o HI I II II cN ,-- mo H HHH -I-4 OMa ,-i ci II III I II IIHH r- % O C 4J Cd C- II II * * * 1 tt i.N I I 1,0 1~0 N C l -l C) co U1 cC r,H- , Ced4 H EH c00C C*N1Cf) 00 Co 00 O- O HH- -,Ct Sf1 ON ON c 0 C14 z O H --I c * . ' 00 %O ONcc oO OO 1. Lt 0 o0 I I e14 g-i -OO 0 HH~1 O0 LLr) ') cot 0 00 HH1CxCq HO i uI --m c~~~c~~ cC; c c~ ;c c;co - C'O' He Ico I o'< 02 0 i-4 I-4 0 02 -4 0 . H p C, ON0 H w 4I4 w 0 1.1 E4 I II HH 00 ,-I I I00 I) Io In ,-I I Cl I cIC4nC H I I I I U) H- r- 'OrC") O -H HH i I ~0 II H I H I I CYC OP-I H I H " 00 I I q I I cN c 0 0 00 ' H m -- NO oH O MOC)N C 4N H N I HH I4 I4I I00 III IHH II I I ,4 I IHHI IHH 0 ~1) 0s o0- ONC - cn ur - H0 -H I I Lt1L oC It 0 I OH I I 0 I I o L n H0H - C-N4 HH HH HH HH I I II II I II I o of 00 I I HH I I 00 II 008 1 C '- -I O -I -t' < r 6 II II 00C1 cr CS l I N-. ,r. lCq c4',-4 cx lr,.I 00 Ot oH I O 00 c N e14eN c III dd H.H-C) Cy cH Ho H ,a C-N I If1 0 Lr L 000000q I II II II II II II ll H HII 1-4 Nc Cs CS * ** * I 00 Io C; I * * I C) 00 i r *8 (o O l * * e- tC'4 0o O O I I I-iII -! e" .0 C'4 N HOt~r! I l-I 1l to p oo~~~ oo oo~ ooo I I -, C * *; I* I *; I*I *; * O O H 00O o 0 i 0 I I I 0I,-0 0 00 0I 0I 00 II -0 ¢'0aoo 0oco0 coo @oi a) c00 c c0 00 N C >0 C O c,, csd *9 oo 9o0 09o CN C'4 00o LLOo 00 i ICI 9 9 I-I !-I 00 o 00 OO OO *9 cHc csa; HeN HH C4 oo0 O 0000Or- I uI C; .D Lfl l0) C) 00 rII)C) *6 00o oo oo C',OO %e m"a r-I --.I 1.D%,. * 1 I I I I C OO *d 00o H U) 1-i 400'IO %co HHaN r- -I Cd co O% ", Cq I 0 ec ,4" 'IcO HHN H HH ,.-P. I v4--It-4 -ir- C,oO%O 'IOX HH, r.4--i I a0 oC oiC O O ('400 C", a0 % C,, 00 %O %~O %D00 C'D O O HH o,%H HH HH H6% N rv- rq 4 ,-4r' l r" r-i V- -.V4 r- i C", oO e.o % C', eo~00 H ,N Na" vrI r-q l -I r-i "4all 'IO ' 'dD ''aHH -- oO HH e, 00 r-a r-4 II 4c cd H- I %= ~ I 9n e roH H= w w r= 9 m r-, p w $= PLI 7-po E-4 ap 1:4 r-P, p w 9P w C?4 w 9:, :p 53 HcTON OHC'J " 01 'a 00 Oq0 0 I 000O0000 II I C)OOOH I ~ 0O'I ~ -~I 0rr- 0ONC4 . 00 gC)zt 00 o0 r4 0 It r.0-# cOHrn LL~~~~~~000 r_ IOP z '0 C) " .... ·..... -** 0 I I II ,. S -It u C O - -rI c0Jr C) -0 H 000 )1 "L qr-C . i I 1 i i i i 1 · i 1 i c000 uMco0 ·H 4o 000 4 o o oo (V 00 ·i 00 00C O0 %OI cxT 0 a-C) ) C' H -H I O I U tO M r -I co 000 000 n rOm 0c 0 0 0cn0 1 0H 'IO 4i.J a) IC4 .-I a) H co U ~ r C) C a) Qr4) 'C14 -tH %' -r%- *- -I000 - 000 C) s-' H 00 %n 0 CDC C7U H) II Ir- H I II 0 ;C;C i0 0 i i C Ii;C;C iii'.C) iii CNrILn 0-1t 0 C; d00m0 OLn00 00cq-q0 ON - C') C~00 p.. -I-0 C1 Ue 4..I r.l 4-) r' 00~'0 !....IL ~~~~- O-4 rV) e' J'C; C;.C; N- C. -* H ' -~I 0 0 tf 00 cn a% 00H III I I-I1I% co I · in00 ON00000000000000 I I II 'C-Jr C; 0COo 0 'J-c' ~ -4 It )S ) -4 N N O0 C' 0 -tr000 Hz 0 ~rC N1re'P 000 C )CDC ; qr- CDC III C') cq )C>0C I I .C)C) C) 54 Table 9 8 7 Homothetic, stationary, count.dum.vrbls Homoth.,stat., additive,count. dummy variables -0.3630 0.0283 -0.3308 -0.0368 -0.2560 -0.1142 -0.0624 0.0860 -0.0335 0.0304 -0.2632 0.0717 0.0384 -0.1214 -0.0522 -0.0030 -0.2310 -0.0303 -0.2611 -0.2192 -0.3262 -0.3631 -0.2450 -0.0501 -0.3231 -0.3101 -0.1085 -0.4062 -0.5458 -0.4146 -0.6386 -0.3777 -0.7836 -0.6115 -0.5450 -0.8277 0.3098 -0.1226 -0.0563 -0.1309 -0.1226 0.1874 -0.0259 -0.0389 -0.0563 -0.0259 0.0262 0.0560 -0.1309 -0.0389 0.0560 0.1138 -0.4537 -0.0157 -0.3461 -0.1585 -0.2586 -0.1743 -0.3011 0.0000 -0.1209 -0.1370 -0.3689 -0.1201 -0.1239 -0.3005 -0.1330 -0.0824 -0.3335 -0.1967 -0.1715 -0.1674 -0.2524 -0.2654 -0.1794 -0.0027 -0.2188 -0.2454 -0.0204 -0.2378 -0.4480 -0.2814 -0.2378 -0.2613 -0.6899 -0.3978 -0.4211 -0.6620 0 0 0 0 0 0 0 62T -0.9116 -0.3367 -0.8078 -0.5990 -0.7065 -0.4047 -0.5678 -0.3417 -0.5453 0.0293 -0.2522 0.0693 0.0379 -0.1247 -0.0343 -0.0150 -0.2240 -0.0331 -0.2329 -0.2016 -0.3022 -0.3342 -0.2220 -0.0372 -0.2987 -0.2884 -0.0823 0.1152 -0.2095 0.0407 -0.1047 0.0531 -0.5238 -0.1485 -0.1459 -0.3393 0.0791 -0.1125 -0.0429 -0.0763 -0.1125 0.1759 -0.0292 -0.0343 -0.0429 -0.0292 0.0270 0.0451 0.0763 -0.0343 0.0451 -0.0871 0.0104 -0.0013 B3T -0 0005 a4r 0.0086 alD2 cXID cx1 D12 aiD 3 ax JLD4 a1 D 5 cxlD 6 ( I D7 aiD e aid 9 c2D (x 2D 2 cxD 3 cx!D4 ca2D A,25 aID 2 3 aO2D a2D8 a2lD6 a2D a2D3 9 aa(3 D D4 3 2 CxD a D6 3 7 a3D4 a 3D? a3D e a 3 D9 a4 D 1 a4 D 2 a4D ca4 D 3 a4D 5 c 4D 6 a4D7 a 4 D8 cx~D9 all $12 813 31 4 a21 822 823 a24 53i 332 a33 83~ a1 a42 43 a4 4 _ (.0898) (.0574) (.0810) (.0920) (.0769) (.0481) (.0814) (.0666) (.0850) (.0770) (.0480) (.0726) (.0782) (.0689) (.0393) (.0626) (.0547) (.0737) (.0326) (.0209) (.0298) (.0334) (.0282) (.0184) (.0301) (.0239) (.0310) (.0423) (.0286) (.0137) (.0347) (.0120) (.0074) (.0016) (.0013) (.0006) (.0554) (.0307) (.0502) (.0557) (.0485) (.0240) (.0433) (.0369) (.0522) (.0731) (.0404) (.0700) (.0735) (.0667) (.0290) (.0501) (.0483) (.0706) (.0269) (.0146) (.0251) (.0272) (.0243) (.0111) (.0198) (.0179) (.0259) (.0317) (.0235) (.0100) (.0348) (.0117) (.0200) (.0210) (.0210) (.0170) (.0148) (.0162) (.0138) (.0138) (.0138) (.0145) (.0153) (.0153) (.0122) (.0108) (.0118) (.0053) (.0053) (.0053) (.0055) (.0058) (.0058) (.0047) (.0041) (.0045) (.0069) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 (.0190) (.0190) (.0190) 0 0 0 10 Homoth.addit., country dummy variables 9 Homothetic, Country,dummy variables Parameter . Cont. . -0.6452 -0.2191 -0.5375 -0.3440 -0.4382 (.0224) (.0233) (.0224) (.0224) (.0224) -0.3777 (.0242) -0.5104 -0.1855 -0.2884 -0.0735 -0.3015 -0.0567 -0.0624 -0.2410 -0.0656 -0.0130 -0.2720 -0.1411 -0.1434 -0.1376 -0.2242 -0.2382 -0.1531 0.0272 -0.1880 -0.2181 0.0042 -0.1380 -0.3419 -0.1816 -0.3554 -0.1677 -0.5839 -0.2885 -0.3243 -0.5746 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0120 -0.0040 -0.0018 0.0062 (.0230) (.0204) (.0195) (.0220) (.0229) (.0220) (.0220) (.0220) (.0238) (.0226) (.0200) (.0191) (.0082) (.0086) (.0082) (.0082) (.0082) (.0089) (.0084) (.0076) (.0071) (.0011) (.0011) (.0004) 55 In the first six models all of the coefficients are constrained to be the same across countries. is homotheticity. In model (1) the only additional restriction In model (2) we add the restriction of stationarity. Model (3) allows time-varying preferences, but adds the restriction of explicit additivity. Model (4) includes both stationarity and explicit additivity restrictions. Models (2) and (3) are tested against model (1), and model (4) is tested against model (1). The computed test statistics for models (2), (3), and (4) are 10.44, 45.36, and 44.64. These are all above the critical 5% levels, so that we cannot accept the hypotheses of stationarity or explicit additivity. Model (5) is non-homothetic, but the stationarity restriction is imposed. To test for homotheticity, we compare this model to model (2) (so that the test is conditional on stationarity and on the ai's being the same for each country). The calculated test statistic is 61.92, which is well above the critical 5% level of 9.5 (four degrees of freedom), so that we cannot accept the hypothesis of homotheticity based on this test. Nonetheless, because of the problems involved in estimating non-homothetic models, we continue to impose the restriction of homotheticity. Model (6) has the same specification as model (1), but it is estimated using only four years of data: 1960 (or the first available year, if later than 1960), 1964, 1968, and 1972 (or the last available year). We can expect that the use of data at four year intervals will result in a model that is more representative of the long-run. Then, by comparing elasticities derived from model (6) with those derived from model (1), we can determine whether model (1) and all of our other static models - are more representative of the long-run or the short-run. 56 Models (7)-(10) are homothetic, but country dummy variables are introduced so that the first-order coefficients (i's) can differ across countries. Model (7) has no additional restrictions, model (8) is stationary, model (9) is stationary and explicitly additive, and model (10) is additive but non-stationary. We first test the hypothesis that the ai's are the same across countries by comparing model (7) with model (1), model (8) with model (2), model (9) with (4), The calculated test statistics are 900.0, 860.4, and model(10) with model (3). 813.6, and 889.2 respectively, and these are all far above the critical 5% level of 46 (32 degrees of freedom). of country dummy variables. This provides strong support for the retention Furthermore, it leaves us with some doubt about (Unfortunately we were not able the validity of the test for homotheticity. to successfully estimate a non-homothetic model that included country dummy variables.) We next test stationarity and additivity by comparing models (8) and (7), models (10)and (7), models (9) and (8), and models () and (10). The calculated test statistics are 50.76, 55.44, 90.0, and 85.68 respectively. These are all well above the critical 5% levels, so that the hypotheses of stationarity and additivity cannot be accepted. Thus whether or not we permit the ai's to differ across countries, the data do not support the assumption of constant fuel shares for each country that are independent of price. It is interesting to examine the elasticities implied by some of the nonadditive models that have been estimated. Own price elasticities for models (1) and (6) and own price elasticities and income elasticities for model (5) are shown in Tables 10 and 11. Note that all of the own price elasticities are close to unity for oil, gas, and electricity, but are much larger for solid fuel. Removing the homotheticity restriction results in price elasticities that are more uniformly close to -1, and results in particularly large changes in the elasticities for oil. Given the results of the likelihood ratio tests, however, we could consider differences in the oil and gas price elasticities from -1 as noise. The solid 57 0J2 r ;: % C4 eO 00 ,I I I I 1 HH I 1 H I I C.11 H H H H HH I I I I I IO I m...I-0 I On It r- I cd 4-ia) r -I" 0000 I I I I I ONO Cl 0oNH0or- r·.(i I iK 4Cx HI HI HH I I HHo II II OH C)I r-I l l I HI C1 C Cl HI HI Cl -) Hd Uoo 4-i %o c0 o0 IH o I'o 0 c0 HC'C Oe~ a) 4 0 *I * . o, a, ,-4 ,-HI E1 o1 II HIHH 00o mn 4-0 . I I. . L~ 0 r 0 z ,-! N I C.I I I o C H z H0 1£ o7 H I Lt)O Co l OY - r--0 f-r IKI H4 HH osO Ur F-=C O -I O <' o- H C) , H I. o' . c4 0 0 c oI II I II Q) 0 -a) E- z I I I I 04 -i 'Cr4 U0 Lr r-. He c~ o ~ o~ HH cl ~- 00 o0 ~1 HH I I ci-,-4-) * -K HI HI P4 - oC4 00c 14 tL- I ~4- Ho OO IIl I I oda)r--! 4-) 0 c-cW 11 <i0 a) a-) 0 r.14.. H 02 r-H 4-i I I HC' 0 H I co I I I I I I Hi-I 0 - 0 mIOI I It E HH I I I I I I It~00C HH4 A c4o- o0 o- r- 1-I I-4 I IH: O r--,-C)ON ~l r-O r-0 -. _o I I '.O'-I r- ro~ 1 'h C') 0Go w-I ,-I Xf~ U 4-ie.I CJo Ho q4 ) rq 0 zH -H -W .1 1 H H.-I 1II IHHI 0II0 0 1 1 01 2 0a)4U I-i "~so)00 H e0 0-) >4 -H III IrII r.H HH % Cl rl 0 00 Cl l m C)i02,'C) oI %D c, %D I HH I 0 0 1-4 0 H-- o,Ilo I CC vJ OJ o) O C'2 I t_I Ioo a,I oo i::(14 C1 C I I csoO 3c EH HH H H O _O- HH )s COl_1= H: r-..I -I . r-I -- I -I, I00 I 0 H Il r-I r-. 0II HH 0,_ 1 . C\2 ooHH 4H H CCOCN~ H e-c HHoo C CHH Lf<Q FI Oi IIHH -q OO HH l - Ir! '. H E.i O 4i N 0r.; '-4 I 2 0 ,-4 I 0 t-4 :t lII 0 *-I, C CI co o. o0 c, c10 HH HI \." C* O O CI" C H 02t 0 I D HHo CY C O l 4-i co co C') O ('24 - l: l:: CV a 'C H 0a 00 H n a :- 4-i H H0 X 58 00 : ,HH * I K * * * 0 LC e~%O cn cn CN ,- Ln'-It,-4 ,-I I I I I " n ecn I I 1 I I I *C * KK4¢ I I O H 0o,ao M I * *,*, a * 1; I 1 r- 00 r-r-Ho o H rH. ... .. CqJ H r- H- I II Cm Lr- --T oo n < I'D 0 1, -t .,T " C"c 00oo00 e[~~i)ci H-r-I4 HHq c .-I ."C,4 , Ln Un o cn o Hr 0 vDor- CN C,4 O I II I I I I I C 1 0 .4-J ctS 4-) 1 C1 1 1 1 1 %OCn * * 1 0 00 : 0 [0 " Cq 4 %DLn r- 0 *. * -1 0z 0 ,cu §0 Iz 00 oo4r-- * * H 100' aI I I IIL O0 I I HH I- INr I HH. *4 * u r- cyO' C)cn e Kf co CUn a) U z 4-i CU H U3 I U .1-1 ,i-J Cd Q (2) c .rq rz U) H- IO I \0c0 * 1. O vo I II HHr .r1 I-J oC *qr- 0"~00 '- $ co I I p -4 14 U) '-H 1 I l *~ 9 ,-I ,-I I 0 I CY)c rl ,-I li ,-I I I I 00 eo.o 99 ,-I C1 4-i --I U C4 4 4-i -H- .,-I o - p 4i U -H 0 Ud U) * 00,-I 1 00 I 1-JH4 *~~~~~~~~~~ 1 H H- r- H 0oo rll r--i HH r-4 U) CU PL 00 O 1-4C II r-c H ,0 I II r-I 00 oH U IC * -H II I I,--I r00 r-0 o00 10-4 I · . 1jH 1010 r- r-i I * *I * I Lr *S O r- II co 00 l -II Co 4-t * 14 1- CL 00 * 1 00 0 11D * C14 C1 -S 00 bo 11 M It 00 I HH IIl C-400 pC HH II r-HHir %D 10 - 0 0 HsooH H l l c0l -', %,O r- r-I Ce') COC"1 -4' 00oo 1-bH L)00 C N"4 H - r-I 'I H HH li -400 14 00 0 CU 1-H"q H r-D 0 r-oo OH r- roH ,) 0: 11 CS H 0 o %O % 'O1 10 O .t 4U) U) :> c1 00 O ,- c1400 ID , cr o c o I'D aS oO ,- -I o400 m mE ,-- ,-I 4-i CU H- 0O r- 14 1= 04 04 e'J " In . t :t CU H rx- z .1 HH 11 r- U) -I r- 0; o (1 4) En 59 fuel elasticity is considerably greater than -1, and the electricity elasticity is stable across the models, so that we may reasonably take these o1;ticittep to be in the vicinity of -1.4 and -1.2 respectively. The income elasticities are, of course, all 1.0 in the homothetic model, but differ considerably from 1 (at least for three out of the four fuels) in the non-homothetic model so as to suggest re-estimating that model, perhaps under alternative restrictions. Some of the income elasticities (e.g. for solid fuel) are highly unstable, so that for the time being we have no grounds to assert that these elasticities differ from 1. The reader should compare the elasticities of models (1) and (6) in Table 10. Note that these elasticities are quite similar, and usually do not differ by more than 20%. This leads us to infer that our static translog models are probably more reflective of long-run behavior than short-run behavior. This is comforting, since most of our estimated own-price elasticities (for the consumption model as well as the fuel model) are greater than unity. Of course the final determina- tion of whether these elasticities are long-run or short-run must await the estimation of a dynamic model. In Table 12 we present partial price elasticities for model (8). contains country dummy variables that allow the This model 's to differ across countries, and although these parameters do not enter into the calculation of elasticities, permitting them to vary would yield better estimates of the Bij's. In Table 13 we present partial price elasticities for model (7), which is the same as model (8), except that it does not contain the restriction of stationarity. (Recall that using our likelihood test we rejected stationarity.) By comparing Tables 12 and 13 we can determine the effect of imposing the unwarranted stationarity restriction. 60 Note that the elasticities in Table 12 are generally much larger than those in Table 13, and vary considerably across countries in ways that in some cases cannot be explained on economic grounds. This variation across countries - and across time - is reduced considerably in Table 13. In view of these elasticities, and in view of our likelihood test results, we retain model (7) as our "preferred" model. Note that the own price elasticities differ considerably from those for model (1) (see Table 10), particularly for solid fuel and electricity. The oil price elasticities are surprising because they are so large, exceeding -2 for some countries. Comparing the estimated values of 22 for models (1) and (7), we see that the model (1) estimate is insignificant, while the model (7) estimate has a t-statistic of over 5, so that the data strongly support the larger elasticities. In fact the prices of fuel oil fell over the period 1962-1970 for most countries in our sample, and the size of the oil shares increased considerably - usually at the expense of the solid fuel shares. Part of this shift in shares was probably due to changing availabilities of fuelburning equipment, but (over the long term) much of the increase in oil shares could indeed be attributed directly to falling oil prices. Note that the cross-price elasticities between gas and electricity, and between solid fuel and electricity, are negative. It is doubtful that these fuels are truly complements in the residential sector; these elasticities are probably the result of the installation of new gas lines and electric power lines at the same time during the 1960's in many European countries, together with a diminishing use of coal. Note also that cross-price elasticities with gas (and the own price elasticity of gas) are very large for West Germany. This is the result of an extremely limited availability of gas for residential use in that country - the share of gas has been about .02. 61 r- I 0 -N- LCO- 00 D II II * * .1 - . .- . ln Lrsev oqcl co) H4 -q NH CN I I HHo I wI OH CoNO , I-I r-4 -4 4Ar4 oor-- t"N N," H-4 HH I I I cn I I I I I I I ·c -IC -* C H,-- T4 .Lr) - - H C1- Ci II CA 0 r-,-- 4IC-ICH H ~0 - cq 0 0 NO -4 A HH L N-- HH ' ' -c ' '' " tH ,-IN ,c' iC . C' , N-H o) L" HN I en ON NO N H.H ulf eq Cr-- t o. · · ID OO O HO CN '-o o 0 0 cI c O " cN cqC i ' -K . Ci NU00 t Ln N Ien N CMN HH ,-- O CN C 0C O I I' N I I a: oO c1 , c1 -- o 0o 00 ,-HH NIN H OI 0 C 0 C0)C ,,-I 00 No O O C II .- ' O O - H I'' OO HO H-I -I II -) O 0 z 0 00 ~ -N C CS Ii C -I0 N" I I ,-IH,rl ,-I oo cN IIt~l I 00 O r-0 Crr-iLn 0' M -I I I I o I) I I I I CN N N-~ (X II I I r H-r. %DO U en HO H-H - C'O . H O HH I un -I CNN- N H O * 00H *' * HH HH I * CD*H .H-I' * jI' I I %C,-ALn' Ln * . O t %O cn -i -~ c q -I - N 0c) HN I I c-I -ic-Ic 0 HO I ,- H -tU~ HH icc HH HH Lu oo r- r I I LnCo 0 o o0, O D O ~HIHI II nN I I II N-- -i H- CNN I H I oo N N - C) I I I %C %O HH % I'DO%O N O: ,--r-I 1-I -- I I ,- y- -I HHf N 0 CO PI N uLn L HN U, Ci N c- I-_ H- -I . * CN T '0 -. Ln I I * m m' t * r- q O - - -I cN "N * N Hr-H -I N c· H i - CN · CNNCI CN e HO HH CNH 'c * ** *H ·* -cic ic -ic-K L Oq Cn O O ,- * H *c* · · cq 0 C4O0 %DCY 'IO 'IC'D 9-1 0 HH -cn r N Cn r N-N- OH 0 HOH i . Nc . -.. * * · 'IO'O HHa% l r I On D * IH * * N - N cn -C . . l ,- O %O N H - TN4cV') r- r- . nN un -I 0 CN NO o * '0 , t HH . . N-- H4H ON00 'C M Cn .-n m CN O O . . . . . I* 00 00 n a C4 r- II NONOo HH "H0H N oo ON I D O NNO 0O o N N N %O NC -- -- 0%C H- C-, t _- cv) _ - C) C c-I HH - 4C- HHir- C" (C) N H 00 H I I m H ,I I I ,O') r-1 00 0 I I CoH1 I ' -T CNH CNN I I 4i) Hwz . . LncN c r- L()H 4.44 u) 1 *I NCO NO cnN *· * *· I 0OO n -q O O O O * . . . I'I I. I H; 4-A 4 - K-K * -K * * I O 0 -I t C~ NO H O - * -*-! OH H I H Cd - %N H i H'CH H r4, Ni -H ° D N-* - LI'~ Q 0 o'P t r- rO O Ln I I N O N~ONON NO N~O -'-i0-t oo-0 £OD-OD ,%HO- 0£O O@0 .1' en Cl, * ' *T 5=: Cv - C7: 62 lY c oC, 0 L D "C r~-D %O%D 0% ,D 00( II ;3 I Il II4 I I4 I II I cn W) *n U)T * *II :0 0 r- o H LnH r I I I,I., r- o o 0%MC %O 0 r- o ,4 * r-* M D. -I u'% 00 0 oHo c H 4 c8c~ c~ I en t H H HUAO ' OH 00 . 00 .. I"" HH Lun o0'o *C .1cC C C 00 ICK * - * ... ... C e4J Dco 0 C~r 4H - -I 0 0 00 00I c; c; c8c 00 00 O00 HO O c00 cAO *8 I 6I Co II II 00 HC4 N II HH HH 00 o o o0o HH~ -o 00 00 00 o --I oIIro II II 00 II C L r-% I.<1 ,D C1 HH I I C *CIC COH HHc; HH HH H *.M CU % C0% 0C %' O L I4 OH 4C %-0 HH ;c;0 H I HH 00 II II IIll 0 00 Cn ., n COLA4 4I HH II 0 ?O HO cn r- COHC l I H- O O 1' 0 I0I C o o H,- r LAna cc; HH; c; HH 4 4 0 00 c0IIl 004 Ln00 COO 00 OH OCO 0 * * OH %LA 00 %?O o rCe %?'DC-iH H- c;; 0 c 0 0 0 O - 0 n 00 * 8c; N CA N o~4Co a00 c J0 0 -It O .-0 c8c; ICC * C; o o II c; O0 C4l H- - 1Cl 1 1 I I0 I0 I * Cs O H*r C Un n P- %C 00C oo;C00CC o0o 1 1 I I *- M o o4 o3o 0 z -W '44 4-i H - HA n I I I I I oIo I o '0o 00 0 0O OH I I L CD CN4H HH 00 00 00 *I*IHH*I** CO CM o ' ?H 1%--.? I. C %?'J H HO C4C4 HH0 0 0 * 0* 0 I* 68 68 68 66 C;C00 II o LA M r- 1-4 OH1 I r-- c CO1 I - O -H %? LuAO COCD cH O HH t-i HH H H C4 C'I C' L C.IL -N~ I I H H- I HH 4c II ? 0I 0I -4 r- i I CON I i I 00 or 00OHI O0 I II rT4 COC HO O cC4 LA O-Nc %0 N r -% O 00 O C4H P ,-N c . HO cn'c0 HH 'ON · D c H -tc HOH H-- H-t H cON C* * H00 00 I1 I1 HO 0000000000 o O q q oIIo 44 0 I Iv; I I 4;c I4I c;I c; I -tH ; 0'-:i c; c; I ; I 00H I'd II II88 r~ 8 C;8 8 88 0 'T Ir-I ~ ~t O~ L., o W O,-;-Il · '1' ,-,-A~ OH ** *I -H * * ICC ** 0'0 oH 0000 o00 c; 00o 00 rNLA iC00'Nfl HH 00 a N 00 D I I Io II I 00** CON 00 0 I 0C I .4 -H ,-.I ul LACC;oO C co H<eH 40lK sc*r-H.. L.n I .O ?HO N I I I I I I 0c" ? H- c4cC'C cnc Ifc;c * O**H * H*I 1 p-4 I I c4LE 00N. cOH ,-I r-I 0c c 00 * C; COC 0"O C'J% c C JL 0 a% co4 C400 co0 C4 0 0 cN 0 CO N 0 e0 40 C 00 C 0 'OO ON 'O'DO %O 'O 0'0 0'O 0'ooN 0'ON O O 'O %O 'o %O 'O O %O O 'O O 'O O 'OO '%O ON0 0'0 0 0'ON 0'0a O 0O 0'0 0 0'0'00 ON'Ol HH~ID. H,-I H r-I HH4 rqIH HHf HHr41 'hO O HH HH - -f94rHH 0 0 0' HH~1, -I C~4I C C I~ 00 0 c; O00 co i- 1c;c II I I CN40 W 04 HH WHr-I 4 -4 N q C" C" o0 C CN00 0' 0' '.O O HH4 H-I I'DIO HHfrq CA 00 I I aI !- 4-i Wc i-4 ,t & C r') ( = r . _ - C7 o4 1= c) .. -4 N 9= ,-~~ - N ° gr J= ~ ~ ~ ~ * 9= ~') ('~ 4 : ( ¢~ = ..1- - * 5= ,.1.~~~~~~~~ 4. 63 I II Un I ,_ 4 _; 1 C4 Lnr-ir l- oo I CNI C CI co r- o o -H O r- 0 U3 0 HOI H ~-~o r, ro0 0 ~--~- I I I KK * Cn H O r- 00 00 0. 0 cN I 1Ci C1C * * o'c 0c 'r- r-. o r3 , rI r rA ~r--. i . co %D c4 I --, I r- r-_ Coo r-ocr- C.-, I . r- -4 * c0 o o HH HH00o a r- oq o -.-CuI :T Ln me', * C D U'nLn Inr-r-4 -T H-H C, r-i ILn v-I un-t co I I 00 r-4 -0 HrIcq n - o ~I I I I ' .C -'O o o Oq '4 - r- *~ * o CI I I - --iL O r- 0- 00 I o N N O Ln *~ 8 00 008 uo c'Qc I -- -1 00 00 I I o o oo 00 o c 00Ir 0 0 I I oo1 ~-oo j go C; (J I I 0 0, 0 o I 00 00 C' c'17 '1) ~..oo oI I I c ·-I 0 Hc I HO 00CC . I r- 0 z HO oo (O O' r-.r'- -. II. I Ln O r-4 -4 I I C'-)0 ( N 4 C) - 00 II I o cO o'001 * *- * -* -ic I I I I o,- o oo 'o£I OI xOD .H o O o 0oo o ur 0 0 z I I CL0 HCo - " co 00 COO 00 I I 00i CHOC) I I 4I-J .i Or- C*i-I I I I I I II 'H -- CY) - C0 I 0C r- r-II I cn a%,-o Hmo o3o Olo Lfe' -oH 0 II I II II J cd u3 'lc °°q T°, oc d.. I r--O o'-o Ldc~ c~4tc; r ~o, o3 o~ 03Ch . d. 0 m 00o O c-i o -0 oo O Ol -- C14¢,, CY)oq C,1¢,, ,-- I go O C * r-. gg I , I II II I I C* ' . Lo .c) o. .,I) .N 00 00 J~~c; gg6 : co coo .J. I I '00 00 00 00 00 00 ~4 c~'u3 ~,1o 0~0~ -IC 14 :- 0 -4 UA ,-o HeN H-i I o o - 0 4-i I I -H I oN 00 I I K .c . . I I I I M- N I .I . * ciC I I L HH< H H II eNe l (N o ¢,4oq es N uLoc eNeN r- -o HH~ CQN H - CV . oeq IIll KK C ICC -H HO o c; r-C; c; ~C) 00 00 a00 0000 00 c c) I HcoH nH aH'r-H .H '-0r ,1 u 1 i.1' 0' c, o"1 * 00 O0 OH 00ul) 00O , -. v ~ ooa C1 C1 001o 0o4II 0I4 0-I 0-I 0 0 00 I I HH 00o I I CN r-I I I U) ,0 Os Hr- 0 fr4 c r O HH I I H coo ,H3o o uL) III H HII H 0I I 0 r-o C oo oC0O II io o u o~o o -(v,-O~c;~Q ¢,1 o c~c; .rc; eN~C~) 0 r- 00 v3 - r-i 'N O ~0 n N r- r- I II I ~ ~Hgo;e eH r1 H -g N*-g NNN o 0 '-4 C' o C) oH e-o D Ha 4 Co H- 'qC4oo '0i 'cDD O cs $:7 OH ( m F ,i 0 O H -t.t : t F= c oO Cq o NC0 D C'IO HH o - = 1 HH ~~~~~~~ -4 H C100 , ClC0!0 eNCO %D D H oH Y 4 .1 F' 1= - HO .t-4 N Cq 0 ¢, % eNCD DO H aH - I C-i 004 9= F C1 0 CN co CNCX % I'D V. HH ci 04 C1 00 %DO C 0 %O% H H CHH aa aH H HH c. N 0) 0) et r7 S= cn Z _ ,0)~ C14 00 %D a H C' 00 %0% % H HN ~ ~~~~~~~~~~~~~~~o -t 9 -I 64 These price elasticities are "partial" elasticities, in the sense that they do not account for the effect of fuel price changes on total energy expenditures. In Table 14 we present total price elasticities for our two Recall from preferred models - fuel model (7) and consumption model (8). Table 8b that the price elasticity of energy, as calculated from our consumption breakdown model, is close to 1. This explains only the total fuel price elasticities are nearly the same as the partial elasticities. 5.3 Dynamic Translog Models (to be added) 5.4 Logit Models of Fuel Choice We estimate a number of static and dynamic logit models to describe the dependence of fuel shares on prices, income, and temperature. The "decision functions" in these models are linear or logarithmic functions of relative fuel prices .Pi (the price of fuel i divided by the price of energy), per capita income Y, temperature T, and, in the case of the dynamic models, lagged shares. Recall from equation (37) that this leads to a set of three equations that must be estimated simultaneously, since certain coefficients are constrained to be 49 the same across equations. We therefore use iterative Zellner estimation to estimate all of the models. 49 Note that even without cross-equation coefficient constraints, simultaneous equation estimation is desirable in that insofar as errors are correlated across equations, it yields more efficient parameter estimates. 65 Our static logit models are of the general form a log(S 4) = where ai4k =ik-a4k i~k I ai4kDk + biPi -b4P k=l ci 4 + Ci4Y + di4T, i=1,2,3. ci-c4 , and d=di-d4, 4 ik-4k' ci4'ci_'4' i4 4k (59) as in equation (351. The Dk are country dummy variables (countries are ordered alphabetically), and the fuels are ordered (1) liquid, (2) solid, (3) gas, and (4) electricity. Parameter estimates for four of the more promising static models are shown in Table 15 (t-statistics are in parentheses, and the R2 for each equation is shown). Model (1) is linear in prices and per capita income. the intercept dummy variables are highly significant. Note that most of The price coefficients, on the other hand, are insignificant for solid fuel and electricity, and significant but with the wrong sign for natural gas. The income variables are all significant; as expected, increasing incomes lead to greater use of oil and electricity and less use of solid fuel (we have no a priori expectation regarding the income effect on gas). Temperature variables are added to model (2), but these are all insignificant, and they provide no improvement in the price coefficients, so they are dropped from the remaining models. Model (3) is the same as model (1), except that price dummy variables areintroduced for solid fuel in Canada and the U.S. (CNSD, USSD), and for gas in Norway and West Germany (NRGD, WGGD). There is virtually no solid fuel used in the residential sectors of Canada and the U.S., and little or no gas used in the residential sectors of Norway and West Germany. This is not because prices are too high, but because in Canada and the U.S. other fuels are readily available that are cleaner and more convenient, and in Norway and West Germany the extremely limited supplies of gas are not made available to residential consumers. Note that the price dummy variables are indeed highly significant. 66 Table 15 - Parameter Estimates for Static Logit Models 2 1 Linear in prices, income Parameter al 0.3153 (0.56) 3.5200 (7.55) -0.0818 (-0.20) 0.9231 (3.12) -3.2645(-14.94) -0.8303 (-4.86) -0.3529 (-2.44) -0.4702 (-3.29) 0.1918 (3.02) -0.6383 (-4.12) -2.4808(-18.61) -0.3018 (-5.57) 0.8470 (5.29) -1.0693 (-6.13) -0.1109-(-1.44) 0.0375 (0.09) -1.8056 (-6.44) -6.1332(-27.96) -0.5859 (-2.30) -0.6565 (-2.74) -0.8341 (-4.78) 0.6726 (2.71) -9.0474(-48.38) -0.2165 (-1.73) -0.6366 (-4.84) -2.5981(-22.63) -3.2036(-64.17) -0.000 (-3.92) a 24 1 a341 a142 a242 a342 a 43 a243 a3m3 a144 a244 a 344 al45 a2 4 a 3 45 al 46 a 24 6 a 346 a 14 7 a2 4 7 a3 4 7 al 48 a 2 48 a 488 a 33 m a 14 9 a2 49 a 3 49 bl b2 -3.985Xl10- 5 (-0.89) 5.841x10- 5 (11.46) 1.990X10- 6 (0.27) b3 b4 CNSD USSD NRGD WGGD 0.0002 (1.71) -0.0014(-12.00) -0.00013(-2.79) C14 C24 c34 dl4 d 24 d3 4 I Eqn(1)R Eqn(2)RA' Eqn(3)R 4I Linear in Prices, income, temperature 0.6937 (0.73) 3.2409 (3.88) -0.3132 (-0.73) 0.8080 (2.16) -3.2060 (-10.72) -0.7675 (-4.42) -0.3581 (-2.05) -0.5571 (-3.15) 0.1167 (1.51) 0.6105 (-2.38) -2.6112 (-11.42) -0.4263 (-4.61) 0.8150 (5.08) -1.1026 (6.29) -0.1303 (-1.69) -0.0999 (-0.22) -1.7827 (-5.24) -6.0818 (-27.72) -0.6553 (-2.44) -0.7667 (-3.08) -0.9432 (-5.13) 0.6398 (52.46) -9.1359 (-46.23) -0.3062 (-2.28) -0.6672 (-4.16) -2.5580 (-18.39) -3.1633 (-57.23) -0.0004 (-4.02) -3.335X10-5(-0.72) 5.678X10-5(11.09) 5.670X10-6(0.75) 0.000173(1.58) -0.00141 (-11.93) -0.00015 (-3.17) -0.00385 (-0.19) 0.01185 (0.66) 0.01138 (1.64) .7844 .7868 .9938 .9961 .9938 II 4 3 Linear in prices, income; dummy variables Linear in logs of prices and income dummy variables 0.1270 (0.220) 15.3070 (3.73) 2.5185 (5.140) 5.3262 (1.37) -0.5301 (-1.22) -0.1918 (-0.05) 1.4187 (4.53) 1.3114 (4.08) 1.6660 (1.42) 42.4883 (3.92) -0.3672 (-1.90) -0.4862 (-2.17) -0.2941 (-1.98) -0.3534 (-2.47) -0.4241 (-3.20) -0.4367 (-3.33) 0.2643 ( 4.06) 0.2179 (3.63) -0.6262 ((-3.97) -0.6724 (-4.36) -2.3730 (-19.25) -2.3873(-19.62) -0.2805 (-5.26) -0.2950 (-5.03) 0.9534 ( 5.78) 0.8988 ( 5.68) -0.9705 (-5.95) -0.9788 (-6.11) -0.0339 (-0.43) -0.1001 (-1.36) 0.7692 ( 1.79) 0.4054 (0.91) -1.5148 (-5.26) -1.7199 (-4.79) -5.8846 (-23.72) -12.9588(-10.76) -0.1635 (-0.60) -0.2765 (-0.97) -0.3621 (-1.48) -0.6554 (-2.46) -0.0889 (-0.35) -0.1535 (-0.52) 1.0376 ( 4.00) 1.0249 (3.95) -4.5508 (-2.85) 32.9861 (2.30) 0.0869 ( 0.62) -0.0058 (-0.04) -0.5879 (-4.37) -0.6306 (-4.83) -2.5431 (-24.08) -2.5653(-24.78) -4.1449 (-16.28) -10.1876 (-4.70) -0.00047 (-5.17) -2.0431 (-5.88) -2.033X10- 5 (-0.50) -0.3358 (-1.10) 7.414X10-6(0.33) -0.0343 (-0.23) -9.118x10-6(-1.15) -0.0380 (-0.13) -0.00059 (-4.13) -5.0851 (-4.22) -0.00063 (-2.75) -4.7594 (-2.93) 5.333X10-5(2.36) 0.8221 (5.37) 0.00057 (3.80) 0.9395 (3.23) 0.00021 (1.86) 0.00013 (1.17) -0.00118(-10.12) -0.000119(-10.41) -0.00016(-3.38) -0.00013 (-2.86) .7856 .9949 .9957 .7759 .9949 .9962 .9962 I .- 67 The price coefficients are improved in that three of them have the correct sign, and the fourth, while positive, is insignificant. Model (4) is the same as model (3) except that it is logarithmic in prices. This results in no improvement in the estimated coefficients. We therefore retain model (3) as our "preferred" static model. Our dynamic logit models are based on the assumption that the choice of fuels this period depends on the relative shares last period, as well as this period's prices and income. The dependence on past shares is intended to incorporate both habit formation and stock adjustment effects. It leads to equations of the form 9 log(SiS4= ai4 kDk + biPi - b 4 P4 + ci4 Y + iSitl-4S4,t-l (60) i=1,2,3 Note that this is not a Koyck adjustment model. The coefficients i can be greater than 1 (although we would expect them to be positive), and in general a change in price will not lead to geometrically declining changes in shares over time. We present estimation results for three dynamic models in Table 16. The first model is identical to static model (3), except that it contains lagged share terms. The second model is identical to static model (4), but with lagged share terms. The third model is the same as the first, but all of the intercept coefficients ai4k are constrained to be the same across countries. Model (1) is our "preferred" dynamic model. Note that many of the inter- cept dummy variables are significant, so that inter-country differences must be 68 Table 16 - Parameter Estimates for Dynamic Logit Models 1 Parameter 2 Linear in prices, income 3 Linear in logs of prices Linear in prices, income; no intercept dummy variables i a 1 41 a2 4 1 a3 4 1 a1 4 2 a2 4 2 a 3 42 a 1 43 a2 4 3 a 3 43 a 1 44 a2 4 4 a3 4 4 a 14 5 a2 4 a 3 4 5 a 1 46 a 2 46 a3 4 6 a1 47 a2 4 7 a 34 7 a1 48 a2 4 8 a3 4 8 a1 4 a2 4 9 a3 4 9 bi b2 b3 b4 CNSD USSD NRGD WGGD c14 C24 C34 1 2 3 4 Eqn(1)R2 Eqn(2)R2 Eqn(3)R2 -0.9339 1.1268 -1.0223 0.1902 2.8482 0.2822 -0.0600 -0.2086 0.0600 -0.1934 -1.2759 -0.1720 0.0935 -0.6289 0.0416 0.8205 -0.1717 -4.2357 0.0412 0.2821 0.5868 0.0673 1.7224 0.2219 0.0543 -1.1047 -2.8778 -0.00018 (-2.09) 3. 3997 ( 1.86) (-2.60) ( 0.65) (2.31) (1.44) (-0.52) (-1.66) ( 0.79) (-1.37) (-5.65) (-1.65) ( 0.60) (-3.54) -0.9392 (-0.57) ( 2.18) (-0.55) (-13.92) (0.18) (1.21) (2.25) ( 0.29) 5 3.612x10 -3.967xlO- 5 -1.317xlO 5 -0.00061 -0.00090 (-1.10) ( 1.56) (-0.31) (-4.70) (-10.29) (-2.88) ( 0.93) (-1.94) (-1.91) (-4.30) (-4.20) 9.740x10 5 0.00068 6.993xlO 5 -0.00092 -4.552xlO 5 5.1800 1.6636 3.9691 1.9629 .8785 (4.77) (6.01) (0.80) (-8.01) (-1.04) (11.14) (3.23) (6.29) (6.23) (1.03) (-0.25) (0.02) (-0.40) (11.25) (0.44) (-1.21) 0. 0586 -0.1237 46.5229 0.0941 0.1357 -0.2573 -0.0169 -0.2713 -1.2910 -0.2430 0.0086 -0.6728 -0.0579 0.2468 -0.5274 -12.2412 -0.2719 0.0162 0.3140 -0.1421 51.1490 0.0805 -0.1848 -1.1784 -10.7211 -0.6379 0.2183 -0.1956 -0.0849 -5.4310 -6.7050 .9963 .9971 .9968 _ i~~~ i -0.3446 0.0295 -2.1596 (-1.36) ( 0.10) (-6.92) (-1.94) (-5.53) (-2.29) (0.05) (-3.84) (-0.84) (0.61) (-1.44) (-12.00) (-1.12) (0.06) (1.18) (-0.59) (3.76) (0.56) (-1.08) (-4.92) (-6.00) (-2.40) (0.71) (-1.59) (-0.34) (-4.37) (-4.38) (7.50) 1.1992 1.636x10-5 -0.00093 5.0881 1.8127 4.1353 1.7554 .8820 .9962 i (-2.05) (-0.22) 0. 9452 -8.872x10 i 5 (5.07) (0.19) (-8.10) (-1.96) (10.22) (3.27) (5.78) (5.65) -7.062x10- 5 (-2.09) 4.969x10- 6 ( 0.22) 5.200x10-5 ( 3.02) 1. 859x10 -6 ( 0.66) -0.00020 (-11.48) -0.00107 (-50.76) -0.00013 (-6.88) 9.049x10S ( 0.69) -7.703x10- 5 (-1.07) -0.00047 (-5.05) -0.00017 (-1.92) 4.9224 (15.70) 4.3937 (14.55) 12.2985 (12.49) 2.2369 (13.51) .8354 .9934 .9335 69 accounted for in estimating elasticities. The price dummy variables for solid fuel in Canada and the U.S. and for gas in Norway and West Germany are highly significant, and are retained. Three of the four price coefficients are significant and have the correct sign; the fourth (for solid fuel) has the wrong sign but is insignificant. This is not very surprising - solid fuel shares are low (and falling) in some countries with low solid fuel prices because cleaner and more convenient fuels have become more readily available. Finally, note that all of the lagged share coefficients are positive and highly significant, so that dynamic adjustments seen to play an important role in the determination of fuel shares. Model (2) is logarithmic in prices, and the result is a loss of statistical significance for two of the four price coefficients, with no noticeable improvement in any of the other coefficients. Model (3) is linear in prices, but contains no intercept dummy variables. The result is that three of the four price coefficients have the wrong sing. We therefore use model (1) to calculate dynamic elasticities, and to forecast changes in shares. In Table 17 we present price and income fuel share elasticities for our preferred" static model for two different years, and in Table 18 we present short-run and long-run price and income fuel share elasticities for our "preferred" 50 dynamic model.50 (We must stress that the solid fuel elasticities for Canada and the U.S., and the natural gas elasticities for Norway and West Germany are of dubious meaning since the respective fuel shares are so close to zero.) 50It is straightforward to calculate income and price elasticities of shares for the static logit model. From equation (35) we can obtain the income elasticity for a model with n shares as follows: dSk/Sk - dSn/S Since n = (ck-cn)dY dSk = 0, we have that n-l I dSk k=l n-l t = k=l Sk S n dSn + Sk(ck-cn )dY} = -dS , or, (cont.)+ 70 %\IC00 Cq ,-i -I I r-4 00' H Ln CD "00 -i I-4 c r_ o q I I I I u o r O I' I' 1 I I O-4H -Mc H- O <4 c/3 -iI I I O0 " cq oo * r- r-H *. HH oo * I *-W nH mJ ·L . i I I 10 .jctO u . I · I " o 'oO 4-0-t 0 II H I I Hei 3 O 4-I z0 I I I 0coc I 'I0r-q -I r-4 co H· \0 HFI*41I e * uIH . CoX o o -H l cn r- (-I c 00o O-' *-~ *oO * e rI. .r-I I -J I .r4 4q0 I En -4 ¢0 -H 4-J o r- I p w 40 I + H H o'-~ c~4CN 00 I 1 C' I'" z U] C- r-4 rI 0 .-I H oe -4 rI rI I I o 1p I E4- .4 H .H ~C3 -H -D 0n Cd C~) I C- CqCI * I HO H-I ,-* -* - . . 0 · ' · oo H r-. Ii qCY I C4* H un co ·e s · I rI ILO 0 cn I · o O, 0aJ O q0 uI H %1 C14-I 02 -H Cnn - 0 00 I HH OH II I. -I.I I I I' OH O O I1 '00 Ln I'I coo r-4,-4 I- *4 0 foo 0O or0 00o IH 1-. "o0 "o 0 HH aN cO 'IO H ON aN c'Co HHO aN O 0 4 0 " C) 0 0 4c 0w 0 -i I %OHr rI o · ·t Ho rI cw H0 O II 00 D - - O O I 0-4 00 0 C 0o r-i w * ...C 0 I w w U cd 4- I *1-4 oq u 0 0 raI u oo c 00c ,', I I -H 4I -H .r4 cn o Ln cq · c *' ' -I 4-4 il CO C O 'Dor-r4r4 rq .Or--4 c io w_ O O CHo ao oro .- I4 -H u H4 0- -H -' H 02 02 0n do U H0 .I0 ~o HH --4 ,-4 0 ,-I co 0n bc v Hl ,-Iet -- 71 rU . oo Lr o :3 cn Cq O~* oo0< o*- C" oo r H I' 0'oi C'oI I ,I 0 I C4 N CN% I l_ IC Io I I :3: ILC') , (I'FO %0r- H ,I 00 ,-4I I CoI - I' I' 1-O m H' 4I I' Y CSl * -It. or- H'fr r- - n r- In .-HI .-4 rc- oo I Ln -'O r- %O . %O Lt * .V I. I* C I ,o c,, I ,-I 0 u .H 0w cl I I w I Hq oI I I' OCN HI I I I ,-.{ I HI CJ 04 I In I z C-) w cU 0 4 ci -H 0 1-1 a OC 00co 0 0 (4 4 c4 ,4 0( - I I I I co an 00o' I I r.) N 0 0 0(") 9 " i~O I u P.4 I I.- O 4 I '-I Cn M H I' I' C I H riHC 0 u H 4~ r- I I r-O0 oI I esHI 0,,.0 lz 0 4- 1-4 OH li ' "4 .1- (t -H "r4 "4J (N-1H( 'q I' 0 I 0 C-) 4. Ie, * - CU O N I 0H Ur oO ,ICS uI * H O ooo OOL(r H0a I I I I I* I* *n14 " 4.J H ,'H 4I OC I ( N o NI 1 1 1 1 O O o4u' (NLIr H I .H u'n H-l Ir1. -H *0 O' a 0c 'O' O -Io*O* C4* %D C0 I ' I' ,I ,- 0 I I -e C. PLU I .co -I * t* O'.o . HDk r4 I I Hq H- CU H'40 N %D 01 ON a I I IIo I I ,- r- o oo' Hc (N ,I H-iN II o D I'ON II I -I It 00oo cU II' ( C-) I a'N Ia 00i- o* H 1- Or- P,I o Oq H cn I l I p p 00 In w GI ". -H 0x n O C ~0 ' Lfn M Ln ~o.a1- ~C HO'4 I , H-HO - O (N O M I I O- U L' l C: C mH 4 1 ' I I I 4JJ 00Q 11 U1 i En 1. um 1. cn 1. II11 0-4 II U)0 al En F4 * U2 u ".- -H -. 4 0 CU 0 0' -H *1. - (1 C0 bO 02 O . . U 0 H 0 _ CU 72 Note that with the exception of liquid fuel, the own price elasticities (These are share elasticities, but for in the static model are quite small. very small changes in price, they will approximately equal the quantity elasThe larger elasticities for liquid fuel are supportive of the ticities.) results of our static translog model with country dummy variables. Nonetheless it is surprising that these elasticities are so much smaller than those obtained from the static translog model - and in fact are unreasonably small for solid fuel and electricity, and have the wrong sign for gas. We assume that this is a result of the fact that the logit model is more restrictive than the translog model, and carries restrictions that do not yield a valid representation of fuel demands. The income elasticities for the static model are less disconcerting. It is reasonable that solid fuel has a negative income elasticity (note that unless all income elasticities are zero, at least one must be negative), since coal is indeed the "inferior" fuel for residential use. As European per capita incomes have increased, more and more homes have had oil, gas, and electric n-l 50(cont.) dSn/Sn = cndY(l-Sn ) - d CkSk k=l After some manipulation, this reduces to dS /Si n-l ' (ci-cn) niY = dY/Y - I (ck-cn)SkY k=l To obtain the own price elasticities of shares, note that dSk/S - dSn/S Then using the fact that S nii dSi/Si dPi/Pi i i = bkdPk dSk = 0 and ISk =1, we can obtain: bi(l-Si)Pi , i=l, ..., n-l biSiPi , i=n We do not derive analytical expressions for long-term and short-term dynamic elasticities. Instead, these elasticities are calculated by simulating the dynamic model. 73 heating equipment installed, and of these latter three fuels, oil has usually been preferred. Price and income elasticities calculated for the dynamic model are extremely large in the long run in many cases. If all of the price elasti- cities had the correct sign, we could interpret this as indicating that in the long run fuels are nearly perfect substitutes, so that if the relative price of a particular fuel increased from its "equilibrium" level, the share of that fuel would become close to zero after enough time had elapsed. This is indeed reasonable, particularly if the "long run" is long enough (we use ten years for the long run in claculating elasticities). However some of the price elasticities have the wrong sign, even though three of the four estimated price coefficients have the correct sing. In addition, the signs of both the price and income elasticities change from country to country, and in some cases the sign of the long run income elasticity is the opposite of that for the shortrun elasticity. In addition, if fuels were perfect substitutes, long-run income elasticities would be close to zero, which they certainly are not. It appears that the dynamic model is unstable, with changes in fuel shares cascading so that equilibrium adjustment to a price or income change is never reached. The results of simulating both the static and dynamic models are shown in Table 19. Actual values of prices and incomes are used up to 1973, and extra- polated relative prices and incomes are used after 1973, with the extrapolations done using the 1960-1973 growth rates for each country. The 1975 and 1980 simulated share values, then, show the projected effects of a continued change in relative prices implied by each model; they do not represent our "forecast" of future shares since there is good reason to expect that relative prices will not continue to move as they have in the past. 74 Table 19: Simulated and Actual Fuel Shares (1) = Simulation of Static Model (2) = Simulation of Dynamic Model (3) = Actual Values BELGIUM Liquid Solid Electricity Gas ; (1) 1962 1966 .09 .16 1970 1975 1980 .20 .26 .33 (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .09 .09 .14 .23 .48 .97 .14 .22 .53 .42 .31 .15 .06 .54 .45 .27 .08 .56 .43 .31 .16 .17 .20 .21 .20 .16 .17 .15 .18 .18 .21 .24 .29 .39 .41 .21 .24 .30 .19 .25 .28 .30 .02 - .00 .20 .14 .01 CANADA Liquid Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) 1962 1966 1970 1975 .37 .36 .41 .44 .4 .42 .67 .97 .3.38 .36 .35 - .02 .01 .00 .00 .02 .01 .00 .00 .03 .01 .00 - .17 .17 .16 .14 .15 .15 .09 .01 .16 .19 .16 - .44 .46 .43 .42 .43 .41 .23 .02 .43 .44 .48 - 1980 .48 .97 .00 .00 - .12 .01 - .40 .02 - - FRANCE Liquid 1962 1966 1970 1975 1980 Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .09 .14 .15 .20 .26 .08 .13 .16 .18 .05 .08 .12 .20 - .41 .32 .24 .14 .06 .43 .33 .19 .07 .01 .44 .31 .20 - .24 .26 .28 .30 .29 .24 .25 .31 .49 .86 .24 .26 .27 - .26 .29 .33 .36 .38 .25 .29 .33 .25 .08 .24 .30 .33 - ITALY Liquid 1962 1966 1970 1975 1980 Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .11 .13 .14 .12 .12 .10 .13 .15 .12 .09 .08 .16 .24 - .17 .13 .09 .06 .03 .18 .13 .09 .06 .03 .20 .12 .07 - .27 .27 .27 .27 .26 .27 .28 .25 .20 .13 .26 .27 .25 - .44 .46 .50 .54 .58 .44 .46 .52 .63 .75 .45 .46 .43 - 75 Table 19- Cont. NETHERLANDS Solid Liauid I 1962 1966 1968 1975 1980 _ _ Gas ._ - . l Electricity - (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .30 .35 .29 .42 .46 .27 .28 .31 .86 .97 .24 .34 .39 - .26 .18 .17 .06 .03 .26 .19 .14 .02 .34 .18 .20 .19 .12 .19 .21 .18 .16 .21 .22 .04 .01 .18 .19 .26 .28 .33 .34 .35 .27 .31 .33 .08 .02 .24 .28 .29 .00 .21 NORWAY Liquid Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) 1962 1966 1970 1975 .09 .15 .17 .00 .11 .14 .15 .01 .13 .14 .14 - .23 .16 .13 .14 .22 .15 .15 .18 .21 .17 .15 - .01 .00 .00 .00 .01 .00 .00 .00 .01 .00 .00 - .67 .68 .70 .85 .66 .71 .70 .80 .66 .69 .70 - 1980 .00 .00 - .10 .17 - .00 .00 - .89 .82 - USA Liquid 1962 1966 1970 1975 1980 Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .34 .34 .31 .28 .25 .32 .24 .11 .07 .05 .32 .34 .30 - .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 .00 - .25 .24 .24 .24 .23 .27 .29 .32 .23 .12 .26 .25 .24 - .40 .42 .45 .48 .52 .41 .46 .57 .70 .83 .41 .41 .45 - U.K. Liquid Solid Gas Electricity F . _ (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) 1962 1966 .06 .09 .38 .32 .29 .25 .20 .41 .31 .26 .17 .04 .39 .32 .24 .20 .21 .22 .24 .26 .17 .23 .28 .47 .86 .20 .36 .38 .39 .46 .50 .27 .10 .05 .04 .07 .06 .06 .18 1970 1975 1980 .13 .08 .07 .04 .36 .42 .44 1$i.. .01 mm I I . I _ _ . I l .25 . .38 .38 .32 .08 $ II W. GERMANY Liquid 19t2 1966 1970 1975 1980 Solid Gas Electricity (1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3) .[ .26 .25 .29 .33 . .22 .30 .83 .98 ..23 .23 .29 - .12 .08 .05 .03 .01 .11 .13 .09 .06 - .02 .02 .02 .02 .02 .02 .02 .02 .01 .00 . .08 .05 .02 .00 .o .64 .68 .66 .64 .7 .68 .63 .14 .01 .63 .66 .63 - .02 .02 - 76 Elasticity Country Estimate US -.12(S.R.),-.50(L.R.) .16(S.R.),-.63(L.R.) -.28 Source Energy own price US US US (a) (b) (c) -. 40 (d) US -.50(S.R.),-1.70(L.R.) (e) Canada Norway -. 33 to -. 56 (f) -.30 -.35(S.R.),-.78(L.R.) -.63(S.R.),-1.30(L.R.) -.42(S.R.),-1.30(L.R.) -.38(S.R.),-.42(L.R.) (g) W. Germany Italy Netherlands U.K. 6 countries* (e) (e) (e) (e) -.71 pooled 20 OECD countries-pooled (e) (h) -0.42 Energy income Fuels own price (partial) U.S. U.S. .10(S.R.),.60(L.R.) .20(S.R.),.80(L.R.) (a) (b) U.S. Canada .27 .83 to 1.26 (c) (f) Norway 1.08 (g) 6 countries* 1.09 (e) pooled 20 OECD countries-pooled 151 (h) Canada Norway 20 OECD coun- oCa cP r-4 Fuels own price (total) gas & oil: -0.96 electricity: -0.34 electricity: -.22 to -.60 oil: -.33 (f) (f) (g) (h) tries-pooled gas: Canada coal: -0.81 gas & oil: 1.24 electricity: 1.88 (h) (f) (f) Norway electricity: (g) U.S. U.S. electricity: -L0 to -1.2 gas: -. 15(S.R.), -. 01l(L.R.) oil: -. 8(S.R.),-l.l(L.R.) (i) (a) (a) elec.: -.19(S.R.),-1.00(L.R.) (a) U.S. gas: oil: -1.34 -1.89 (b) (b) electricity. -1.13 (b) gas: (j) U.S. -1.05 (h) 0 to 1.4 -1.28 to -1.77 electricity: -0.40 (j) U.S. gas: -. 91 oil: -.91 elec.: -. 84 (Q) (Q) (Q) Canada gas: (k) 20 OECD coun- oil: tries-pooled gas: -1.11 (h) coal: -. 98 (h) -.20(S.R.),-1.3(L.R.) -. 52 (h) l_ SURCES: (a) Joskow and Baughman [69] (b) Baughman and Joskow [6] (c) Nelson (c], Jraan [47] cn r 1S (e) Nordhaus [48] (f) Fuss and Waverman [68] (g) Rdseth and Str~m [56] IMN AA--- -- A1^_ Aw ri i (i) (i) Halvorsen [25] Liew [41] (k) Berndt and Watkins [10] - __ I t 0\ _ TT _T ._ _ _- r , 77 The historical simulation performance of the static model is quite good, with share errors less than 10% in most cases. Note that the modPl 98). projects solid fuel shares to drop to below 10% for most countries by This is the result not of an extrapolated rise in the relative price of solid fuel, but of an extrapolated rise in incomes. The dynamic model shows, for each country, the share of one particular fuel becoming close to 1. Again, this could be interpreted as near-perfect substitutability (in Belgium, for example, the 1980 estrapolated price of liquid fuel falls relative to the prices of the other fuels). in the dynamic model. However, it is more likely the result of a basic instability Clearly further work needs to be done in specifying and estimating dynamic logit models for fuel shares. 78 6. Summar and Conclusions The results presented in this paper include a wide range of implied elasticities for total energy demand, and for the demands for individual fuels. For example, the elasticities for our "preferred" static translog fuel choice model are considerably larger (about twice as large) as those for our "preferred" static logit fuel choice model. As we stressed in the beginning of this paper, however, we view these results as very preliminary, and we see their main value in suggesting where further work could be best directed. The results for both the translog models and logit models must, at this stage, be considered inconclusive. Translog models must be estimated in their non-homothetic form (perhaps using a different computer algorithm), so that we can test conclusively whether the assumption of homotheticity can be maintained. This is essential, since all of our other tests (stationarity, additivity, etc.) were based on this assumption of homotheticity. In addition, the imposition of the homotheticity restrictions greatly reduces the power of the translog model. Under homotheticity explicit seperability (additivity) of any category implies that its own price elasticity is -1 and its cross elasticities are zero. The calculated own price elasticity of energy in our consumption breakdown model is -1.01, but a different number might have resulted had the model been non-homothetic. We also need to better identify the difference between short-run and longrun elasticities. One way to do this will be to estimate static models across time and across countries separately. However, it is also necessary to obtain better specifications and estimations of dynamic models. 79 In some ways the variations in estimated elasticities between our different models is not surprising. A survey of the existing literature indicates a very wide range of published and unpublished energy elasticity estimates. Observe in Table 20 that long-run price elasticity estimates for the United States range from -.33 to -.89 for oil, -1.01 to -1.77 for gas, and -0.4 to -1.21 for electricity. 51 Own price elasticity estimates, for total 52 residential energy consumption in the U.S. range from -.28 to -.63.52 A similar range of elasticity estimates exists for Canada and the European countries.5 3 Unfortunately we are unable, with our results at this point, to narrow these "confidence intervals" on demand elasticities. Hopefully, such a narrow- ing will come about with further work. 51- 51SeeHalvorsen [25], Baughman and Joskow and Cope [27]. 52 See Baughman and Joskow [6] [6] , Liew 41], and Hirst, Lin, , Nelson [44], and Jorgenson [35]. 53SeeAdams and Griffen [ 1], Berndt and Watkins R6dseth and Strom [56], and Nordhaus [48]. 10], Fuss and Waverman [68], 80 REFERENCES 1. Adams, F.G. and J.M. Griffin, "Energy and Fuel Substitution Elasticities: Results from an International Cross Section Study," unpublished, October, 1974. 2. Adams, F.G. and P. Miovic, "On Relative Fuel Efficiency and the Output Elasticity of Energy Consumption in Western Europe," Journal of Industrial Economics, November, 1968. 3. Allen, R.G.D., Index Numbers in Theory and Practice, Aldine Publishing Co., Chicago, 1975. 4. Balassa, B., "The Purchasing-Power Parity Doctrine: A Reappraisal," Journal of Political Economy, June, 1964. 5. Barten, A.P., "Maximum Likelihood Estimation of a Complete System of Demand Equations," European Economic Review, Fall, 1969, pp. 7-73. 6. Baughman, M.L. and P.L. Joskow, "Interfuel Substitution in the Consumption of Energy in the United States," M.I.T. Energy Laboratory Working Paper, May, 1974. 7. Belsley, D.A., "Estimation of Systems of Simultaneous Equations, and Computational Specifications of GREMLIN," Annals of Economic and Social Measurement, October, 1974. 8. Berndt, E.R., B.H. Hall, R.E. Hall, and J.A. Hausman, "Estimation and Inference in Nonlinear Structural Models," Annals of Economic and Social Measurement, October, 1974. 9. Berndt, E.R. and N.E. Savin, "Estimation and Hypothesis Testing in Singular Equation Systems with Autoregressive Disturbances," Econometrica, September, 1975. 10. Berndt, E.R. and G.C. Watkins, "A Residential and Commercial Natural Gas Demand Model for Ontario and British Columbia," in The Economics of Oil and Gas Self-Sufficiency in Canada, Conference Proceedings, Department of Economics, University of Calgary, October, 1975. 11. Brown, A. and H. Deaton, "Surveys in Applied Economics: Models of Consumer Behavior," Economic Journal, December, 1972. 12. Chenery, H., and M. Syrquin, Patterns of Development, 1950-1970, Oxford University Press, London, 1975. 13. Chow, G., "On the Computation of Full-Information Maximum Likelihood Estimates for Nonlinear Equation Systems," Review of Economics and Statistics, February, 1973. 81 14. Christensen, L.R. and M.E. Manser, "Estimating U.S. Consumer Preferences for Meat, 1947-71," unpublished, February, 1974. 15. Christensen, L.R., D.W. Jorgenson, and L.J. Lau, "Transcendental Logarithmic Utility Functions," American Economic Review, June, 1975. 16. Cox, D.R., Analysis of Binary Data, Methuen & Co., London, 1970. 17. Deaton, A.S., "The Measurement of Income and Price Elasticities, European Economic Review, Vol. 6, 1975, pp. 261-273. 18. Diewert, W.E., "Homogeneous Weak Separability and Exact Index Numbers," Technical Report No. 122, Institute for Mathematical Studies in the Social Sciences, Stanford University, January, 1974. 19. Domencich, T., and D. McFadden, Urban Travel Demand: A Behavioral Analysis, North-Holland, Amsterdam, 1975. 20. Eisner, M., and R.S. Pindyck, "A Generalized Approach to Estimation as Implemented in the TROLL/1 System," Annals of Economic and Social Measurement, January, 1973. 21. Frisch, R., "A Complete Scheme for Computing All Direct and Cross Demand Elasticities in a Model with Many Sectors," Econometrica, April, 1959. 22. Gallant, W.R., "Seemingly Unrelated Nonlinear Regressions," Journal of Econometrics, February, 1975. 23. Goldberger, A.S. and T. Gamaletsos, "A Cross-Country Comparison of Consumer Expenditure Patterns," European Economic Review, Vol. 1, pp. 357-400. 24. Gregory, P., and J.M. Griffen, "Secular and Cross-Section Industrialization Patterns: Some Further Evidence on the Kuznets-Chenery Controversy," Review of Economics and Statistics, August, 1974. 25. Halvorsen, R., "Residential Demand for Electric Energy," Review of Economics and Statistics, February, 1975. 26. Hausman, J. and D. Wood, "Using the Share Ratio (Logit) Specification for Estimation of Price Elasticities in Demand Models," 1975. 27. Hirst, E., W. Lin, and J. Cope, "An Engineering-Economic Model of Residential Energy Use," Oak Ridge National Laboratory, Technical Report #TM-5470, July, 1976. 28. Hoel, M., "A Note on the Estimation of the Elasticity of the Marginal Utility of Consumption," European Economic Review, Vol. 6, 1975, pp. 411-415. 82 29. Hottel, H.C., and J.B. Howard, New Energy Technology: Some Facts and Assessments, MIT Press, Cambridge, Mass., 1971. 30. Houthakker, H.S., "Additive Preferences," Econometrica, Vol. 28, No. 2, 1960. 31. Houthakker, H.S., "New Evidence on Demand Elasticities," Econometrica, Vol. 33, No. 2, April, 1965. 32. Houthakker, H.S. and L.D. Taylor, Consumer Demand in the United States 1929-1970, Harvard University Press, Cambridge, Mass., 1970. 33. Hudson, E.A., and D.W. Jorgenson, "U.S. Energy Policy and Economic Growth, 1975-2000," The Bell Journal of Economics and Management Science, Autumn, 1974. 34. Hulten, "Divisia Index Numbers," Econometrica, November, 1973. 35. Jorgenson, D.W., "Consumer Demand for Energy," Harvard Institute of Economic Research, Discussion Paper 386, November, 1974. 36. Jorgenson, D.W. and Z. Griliches, "Issues In Growth Accounting: A Reply to Edward F. Denison," Survey of Current Business, May, 1972, pp. 65-94. 37. Jorgenson, D.W., and L.J. Lau, "The Structure of Consumer Preferences," Annals of Economic and Social Measurement, Vol. 4, Winter, 1975, pp. 49-101. 38. Kloek, T. and H. Theil, "International Comparisons of Prices and Quantities Consumed," Econometrica, July, 1965. 39. Kravis, I.B., Z. enessey, A. Heston, and R. Summers, A System of International Comparisons of Gross Product and Purchasing Power, Johns Hopkins University Press, Baltimore, 1975. 40. Lau, L.J., "Duality and the Structure of Utility Functions," Journal of Economic Theory, Vl. 1, December, 1969. 41. Liew, C.K., "Measuring the Substitutibility of Energy Consumption," unpublished, December, 1974. 42. Lluch, C., and H. Powell, "International Comparisons of Expenditure Patterns," European Economic Review, Vol. 5, 1975, pp. 275-303. 43. MacAvoy, P.W. and R.S. Pindyck, The Economics of the Natural Gas Shortage, 1960-1980, North-Holland Publishing Co., Amsterdam, 1975. 44. Manser, M.E., "'TheTranslog Utility Function with Changing Tastes," Discussion Paper No. 333, S.U.N.Y. at Buffalo, Economic Research Group, December, 1974. 83 45. Manser, M.E. "Elasticities of Demand for Food: An Analysis Using Non-Additive Utility Functions Allowing for Habit Formation," Discussion paper #357, S.U.N.Y. at Buffalo, Economic Research Group, August, 1975. 46. McFadden, D., "Conditional Logit Analysis of Qualitative Choice Behavior," in P. Zarembka (editor), Frontiers in Econometrics, Academic Press, New York, 1973. 47. Nelson, J.P., "The Demand for Space Heating Energy," Review of Economics and Statistics, November, 1975. 48. Nordhaus, W.D., "The Demand for Energy: An International Perspective," Unpublished, September, 1975. 49. Oberhofer, W., and J. Kmenta, "A General Procedure for Obtaining Maximum Likelihood Estimates in Generalized Regression Models," Econometrica, May, 1974. 50. Organization for Economic Cooperation and Development, Energy Prospects to 1985, Paris, 1974. 51. Park, R.E., "Prospects for Cable in the 100 Largest Television Markets," Bell Journal of Economics and Management Science, Spring, 1972. 52. Philips, L., Applied Consumption Analysis, North-Holland, Amsterdam, 1974. 53. Pindyck, R.S., and D.L. Rubinfeld, Econometric Models and Economic Forecasts, McGraw-Hill, New York, 1976. 54. Pollak, R.A., "Habit Formation and Dynamic Demand Functions," Journal of Political Economy, Vol. 78, 1970. 55. Pllak, R.A. and P.J. Wales, "Estimation of the Linear Expenditure System," Econometrics, Vol. 37, 1969. 56. Rdseth, A. and S. StrSm, "The Demand for Energy in Norwegian Households with Special Emphasis on the Demand for Electricity," University of Oslo, Institute of Economics, Research Memorandum, April, 1976. 57. Roy, R., De l'Utilite-Contribution a la Theorie des Choix, Hermann, Paris, 1942. 58. Samuelson, P.A., "Analytical Notes on International Real Income Measures," Economic Journal, September, 1974. 59. Statistiches Bundesamt, "Internationaler Vergleich der Preise fr die Lebenshaltung," Preise Lohne Wirtschaftsrechsnungen, Reihe 10, various years. 60. Strotz, R.H., "The Empirical Implications of a Utility Tree," Econometrica, Vol. 25, 1957. 84 61. Taylor, L.D., "The Demand for Electricity: A Survey," The Bell Journal of Economics and Management, Spring, 1975. 62. Theil, H., "A Multinomial Extension of the Linear Logit Model," Inter--national Economic Review, Vol. 10, 1969, pp. 251-259. 63. Theil, H., Theory and Measurement of Consumer Demand, North-Holland Publishing Co., Amsterdam, 1975. 64. Turvey, R. and A.R. Nobay, "On Measuring Energy Consumption," Economic Journal, December, 1965. 65. Wall, K.D., "Inter-Equation Constraints and the Specification of Dynamic Structure," Working Paper No. 119, Computer Research Center, National Bureau of Economic Research, January, 1976. 66. World Oil Project, MIT, "A User's Guide to the MIT World Energy Demand Data Base," MIT Energy Laboratory Working Paper #76-011WP, May, 1976. 67. Zellner, A., "An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias," Journal of the American Statistical Association, June, 1962. 68. Fuss, M. and L. Waverman, "The Demand for Energy in Canada," Working Paper, Institute for Policy Analysis, University of Toronto, 1975. 69. Joskow, P.L., and M.L. Baughman, "The Future of the U.S. Nuclear Energy Industry," Bell Journal of Economics, Spring 1976. 70. Berndt, E.R., M.N. Darrough, and W.E. Diewert, "Flexible Functional Forms and Expenditure Distributions: An Application to Canadian Consumer Demand Functions," Discussion Paper 76-06, University of British Columbia, January 1976.