Consumer’s Surplus Without Apology Robert D. Willig The American Economic Review, Vol. 66, No. 4 (Sep., 1976), 589-597. Stable URL: btp//links,jstor.org/sici?sict=0002-8282% 28 197609%2966%3A4% 3CS89%3ACSWAIE2.0,CO%3B2-0 The American Economic Review is currently published by American Economic Association, ‘Your use of the ISTOR archive indicates your acceptance of JSTOR’s Terms and Conditions of Use, available at hhup:/www.jstororg/about/terms.hml. JSTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at hup:/ www jstor.org/journalsaea.himl Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the sereen or printed page of such transmission, STOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals, For more information regarding JSTOR, please contact jstor-info@ umich edu, hupswww jstor.org/ Mon Feb 23 13:09:11 2008Consumet’s Surplus Without Apology By Ronen’ ‘The purpose of this paper is to settle the controversy surrounding consumer's surplust and, by so doing, to validate its use as a tool of welfare economics. I will show that observed consumer's surplus can be rigorously utilized to estimate the unobservable compensating and equiva- lent variations—the correct theoretical measures of the welfare impact of changes in prices and income on an individual. I derive precise upper and lower bounds on the percentage errors of approximating the compensating and equivalent vari- ations with consumer's surplus. These bounds can be explicitly calculated from observable demand data, and it is clear + Economie Research Department, ell Laboratories. ‘This paper fs drawn from doctoral research done at Stanford University under the guidance of Jim Rosse. | lam grateful fo his support and forthe standards of pro fesslonalexcllence he tied to teach me. I would seo lke to thank Megina Jack for considerable editorial "Throughout, the term consumer's surplus i used to refer to the area to the left of en individuals fixed Income (Marsalian) demand curve and between the relevant price horizontals. ‘The concept of consumer's surplus originated in 1844 Gee Jules Dupait) and has been controversial ever since. Alfred Marshal, who popularized the tol, stipulated that fori to be validly ‘sed the marginal ulity of money must be constant (Marshal p. 842 or David Katone, p. 152). However, Harold Hoteling wrote that consumer'ssurpluses give 4 meaningful measure of socal value. This breaks down ifthe variations under consideration ae too large a part ofthe total economy of the person." (p. 289) John Hicks to, stated only» gentle caution: In onder that the Marshallian measure of consumer's surplus shouldbe good measure, one thing alone is necdful~that the income effect should be smal” (p, 177). More recently, ‘though, Paul Samuelson (pp. 194-98) concluded that consumer's surplus isa worse than useless concept (be: fause it confuses), and TAD. Little (p, 180) agreed, calling itno more than a “theoretical toy." Nonetheless, theorists and. cost-benefit analysts have persisted in thee use of the tool. For Justification they resort (ace Eo]. Mishan, pp. 337-38 for example), with no formal theoretical support, to statements similar to. those ‘quoted above from Hotelling and Hicks, wD, Wiettc* that in most applications the error of approximation will be very small. In fact, the error will often be overshadowed by the errors involved in estimating the demand curve. The results in no way depend upon arguments about the constancy of the marginal utility of income. Consequently, this paper supplies spe- i¢ empirical criteria which can replace the apologetic caveats frequently em- ployed by those who presently apply con- sumer's surplus. Moreover, the results imply that consumer's surplus is usually a very good approximation to the ap- propriate welfare measures. To preview, below I establish the validity of these rules of thumb: For a single® price change, if, |;A/2m| <.0 1A /2m?| <.08, and if |A/m®| $9, then alal 2m Herc, A=consumer's surplus area under the demand curve and between the two prices (positive for a price increase and negative for a price decrease) C= compensating variation corre- sponding to the price change quivalent variation correspond ing to the price change consumer's base income respectively the largest and smallest values of the income + While I restvct attention to single price change here, analogous, but more complex formulae are derived for ‘multiple price changes in my papers (1973, b). S89590 THE AMERICAN ECONOMIC REVIEW elasticity of demand in the region under consideration. ‘Phe formulae place observable bounds on the percentage errors of approximating the C or E conceptual measures with ob- servable A. For example, if the consumer's measured income elasticity of demand is 0.8 and if the surplus area under the demand curve between the old and new prices is 5 percent of income, then the compensating variation is within 2 per cent of the measured consumer's surplus. ‘The ratio | A) /m'® can be interpreted as a measure of the proportional change in real income due to the price change. In most applications, the ratio will be very small. Measured’ income elasticities of demand tend to cluster closely about 1.0, with only rare outliers. Thus it can be expected that §]4|/2m°, the most impor- tant of the terms in (1) and (2), will usu- ally be small enough to permit conscious and unapologetic substitution of A for C or E in studies of individual welfare. Should | A /2m? be large, A would not be close to C and E. For such rare case formulae are provided below in Section IV which enable the estimation of C and E from the observable 4, 7, m°, and A. 1. The Compensating and Equivalent Variations In this section, I present definitions of conceptual tools to measure the costs or benefits of price changes to an individual consumer. While these theoretical mea- sures are not directly observable, the anal- ysis that follows in succeeding sections will Show that they can be empirically esti- ‘mated with consumer's surplus. ‘Throughout [ will be assuming that the consumer behaves as though he were Ur the ratio can be interpreted using the words of Hotelling. quoted in fo, 1 as the relative size of the vatlaton * Formulae (1) and (2) refeet the cautions (ef. 1) of both Hotling and Hicks, SEPTEMBER 1976 choosing his consumption bundle X=X', x4, “to maximize an increasing, strictly quasi-concave ordinal utility fune- tion U(X) subject to the budget con- straint 5p,.X*=m. The resulting demand functions, denoted X‘(p, m), are assumed, to be differentiable. The indirect utility function, defined by 1p, my U[X%(p, m), X2(p,m),- Xp, m)| relates the price and income parameters to the maximum level of utility the con- sumer can achieve under the resulting budget constraint. Clearly, by nonsati- ation, [(p, m) is monotone increasing in income m, and decreasing in prices ‘The indirect utility function can be used to make statements about individual welfare. Let the base, initial situation be characterized by prices p° and income 1", while an alternative situation can be sum- marized by p’, m’. The economic well-being of the consumer in the different situations can be compared by means of the ordinal ranking of the numbers ((p°, m®) and. Ap", m’). ‘Another way to effect this welfare test is to compare the income change m'—m® to the smallest income adjustment needed to make the consumer indifferent to the change in prices from p° to p’. If m’—m® is larger, then welfare is greater in the new situation, and inversely This test level of income adjustment is called the compensating income va denoted by C below. Symbolically, (3) Kp, m®) = 100", m? + ©) ‘The welfare test above YU, m') 2p, mi!) asm! — mC follows immediately from (3) by non- satiation. Thus the compensating vari- ation is an individual's cost-benefit con- cept which makes price changes per- fectly commensurable with changes inVOL. 66 NO. 4 Similarly, the equivalent variation in income (B) can be defined? by (3) Up, mw = B) = I, m®) In words, ~E is the income change which has the same welfare impact on the con- sumer in the base situation as have the changes in prices from p* to p’. It reduces the impacts of different price changes down to the single dimension of income As such, the equivalent variation concept can be used to rank the consumer's levels of well-being under various sets of prices. With the definitions ((p°, m°-2)=I(9', mi!) and Kp", m®—E")=I(p", m!), these Welfare tests, too, follow from nonsati- ation: (6) Up", m®) 2109", m") as BME BY Up", m8) 21pm!) as mn — EE ml ‘The welfare tests (4) and (6) show that the compensating and equivalent va ations are cost-benefit concepts which can be used to evaluate the impact of micro- economic policy on an individual.* These concepts derive practical importance from the fact that they can be estimated from observable consumer's surplus. H. Consumer's Surplus ‘The compensating and equivalent v ations can he most incisively studied and related to consumer's surplus by means of the income compensation function.’ This is denoted by up] p°, m®) and is defined to be the least income required by the con- sumer when he faces prices p to achieve the same utility level he could enjoy (by “The definitions (3) and (8) correspond to those of Hicks, p. 177, and Samuelson, p. 199, "They als can serve as building blocks for methodol es to make socal wllare judgments. The Compensa- tion Prindple isa wellknown example (ee Tibor Saitovsiy) "This theoretical tool was introduced by Lionel MeKenzie, and delinitively stadied by Leonid Hurwcz and Hivof Urawa, WILLIG: CONSUMER'S SURPLUS $91 maximizing behavior) under the param- eters p®, m°, Thus, by definition, Up, mC |p, my] = Ip°, mo) ially, we have ® H(p°| p,m) = me Now, we can see that the compensating and equivalent variations can be expressed. or redefined in terms of the income com- pensation function. From (3), m°+C: u(?"| p?, m®), or combining with (8), (9) C= wlp"|p®, my — mpl 9%, my Similarly, from (5), m°—E=n(p"| p’, m®), (10) B= n(p'|p", m°) — wl] 9", m?) ‘These relationships serve as the bridge to consumer's surplus. It is well known® that aul |p%, m®) aj ay Xp, w(p|p%, m*)) ‘This system of partial differential equa- ions, together with the boundary condi: tion (8), is the heart of analytical welfare economics.’ The compensating and equiva- lent variations, or any measure of indi- vidual welfare that accepts the individ- ual’s own consumption preferences, can be calculated from the complete demand fune- tions via (11) and (8) Restricting attention to changes in a single price, p,, let p?=(p%, p%,..., 2) and p'=(pi, Ph ~~» pi). Use the Funda- mental Theorem of Calculus and (11) to rewrite (9) and (10) as * See Murvice and Uzawa fora state-of-the art det- vation, Heuristialy, (11) say thatthe first-order in- coine change, dy, tequied to compensate for the pice Increase, dpi jist the augmentation needed to buy the old consumption bundle, XP, (7% my atthe new ices pd Poss os rather than at the old prices "The irelevance to this calculation of the concomitant ‘substitution effects is the result of the envelope theorer. "This point of ew was taken by Herbert MobringS02 THE AMERICAN ECONOMIC REVIEW (12) c= fx, H(Pas Bose 5 Bal Phy By «+ Pas YD (3) B= fx Ont BPs Boas Pal PUG Pas «= Baym) ADL ‘These formulae express the compen- sating and equivalent variations as areas under demand curves, between the old and new price horizontals. The demand curves are not Marshallian in that the income parameters are not constant. Instead, they are Hicksian compensated demand curves, because the income parameters include compensation which varies with the price to keep the consumer at a con- stant level of utility. The only distinction between C and £ in (12) and (13) is the level of utility the compensation is de- signed to reach. Referring to Figure 1, C is the arca PYpibe under the demand curve compen- sated to U(p°, m®). This curve crosses the Marshallian curve X'(p, m®) at pf, since u(p"| p®, m)=m®, With pi> pi, if Xt is noninferior (@X'/Am >0) this compensated curve lies above the Marshallian one for bi> Pi, since wl pr, p3, Pa] p®, m®) =m" whenever p> pi. Similarly, Bis the arca Piflaf under the demand curve compen- sated to I(p’, m®). This Hicksian curve crosses the Marshallian one at pt, and lies below it for p.< pj. The area usually called consumer's surplus is p%pjae, defined by the observable Marshallian demand curve. Denoting this area by A, we have, then, C2A>E, for noninferior X! (the in- equalities reverse for X? inferior). Of course, it also follows immediately that if there is no income effect (@X'/am=0), =E. ‘These qualitative results may be useful for some cost-benefit analyses. For exam- ple, suppose a policy would raise both an individual’s income and the price of a non- inferior good. If the observable con- SEPTEMBER 1976 x4 p,u(pip%mo)) xp, (plpym0)) Ficurr 1 sumer’s surplus area A were greater than the income boost, it could be inferred from the inequality that C also would be greater. ‘Then, from the welfare test (4), an analyst could conclude that the policy would be injurious to the consumer. However, usually more information than this is needed about C and £. What is required is a methodology to estimate the welfare measures from observable data, In the next section I show how Cand E can be explicitly calculated from ob- servables when the income clasticity of demand is constant. IIL. Constant Income Elasticity Constant income elasticity of demand for X* means that AXP) m) am XC, m) Then, we have the simple differential equation dX*/X!=n(dm/m) which can be integrated from X*(p, m®) to yield X¥p,m) = Xp, m8) [sy ‘The entire income compensation func- tion can be derived by substituting this expression into (11) and solving the resultVOL. 66 NO. 4 ing differential equation with boundary condition (8). We have, suppressing un- changing arguments, ‘on m[£] Taw Ro = XO, [4] senda = (YX, mp ‘Then, integration between p? and pi, remembering that y(p%) = m®, yields bea) = [wo ty aay a > fx, m dps for #1, and for Zh oowen Hence, after rearranging we have these explicit expressions for the income com- pensation function In wpe) ~ tn (15) (Pt |p, m') afi (S) froma] ne (16) w(t m') = went fe Xun do] ‘These give the welfare measure w in terms of the potentially observable con- stant income elasticity of demand and the consumer's surplus area under the Mar- shallian demand curve. Let us denote this, area by 1 an =f, Pou middp From (as) we see that if =0, u(pi| pt, m®)=m°+-A. However, from (16) WILLIG: CONSUMER'S SURPLUS 593 wwe see that if preferences are homothetic, the consequent unitary » does not imply any equialities among C, £, and A. Below, for expositional convenience, I ignore the case = 1 Recalling the definitions of C and B, (9) and (10), and loosely applying to (15) this Taylor approximation, ‘ (tomer gti y 1-9 (where « means “approximately equal to”), wwe get cmaate, C-A_ 9A rr a ee ‘This was the striking result on the per- centage error of approximating C with A which was previewed in the introduction, The next section will establish this for- mula rigorously for nonconstant income elasticity of demand. IV. Estimation Results Assume that in the region of price- income space under consideration, and ‘yare upper and lower bounds, respectively, On (AXP, m)/m)(m/X"(p, m)), with neither equal to 1." It follows from the Mean Value Theorem that ay (*)< Son X"(p,m) ~ <¢ Let us consider the welfare impact of a ) form; > m price increase from p? to pl. (pi P%, me") > uC pt] f°, m®) for pr, can set m= y(px) and m= uC p2)=m? in (18) 2 This region i (pm: uma a), OS Dew ph FeT m= ym (Lau pp, mw), Oy <1 and 3G, >0}- 1 ither org can be arbitrarily close to504 THE AMERICAN ECONOMIC REVIEW [ee } Rearranging, and substituting from (11) yields siete] oo OS KO, mys bbs = ay = ays SX p, mon) Integrating these relationships with re- spect to pi between pt and pi (as in (14)) preserves the inequalities. Rearrangement of the resulting relationships yields these bounds: la ay mfrza-9S]* Sa(p"| 8 m9) 0 A and 140-2 >0 For the case of a price decrease from A to pi, since (pil pt, m")0, LEA) A/m">0, and if a, and ¥ are sufficiently close in value."” “OF course, in the limit, as approaches 4, (19) reduces to the constant elasticity formula (15). Moreover, we shall see that if the absolute values of 7A /2m? and $4 /2m! are small, then (20) ‘and (21) reduce to elegant rules of thumb. Table 1 displays the numerical values of the following coefficients for selected choices of and a: The most plausible cause of the negation of these conditions is (@X'/Om) m/A')—e, However, regions in which is identically sero can be ignored, since there Voth sand are unchanging. To handle’ the case in hich V0 and aN am0 near the boundary of the televant region, hounds on can be derived from bounds ‘aN dn Because these are enerally more gross than (09), the best approach i to tke this tack only inthe vicinity of the singularity, use (19) on the rest of the path of integration, and splice the sets of incqualtes together. The formulae for such procedarescan be found in my’ 197Sa, papers, An explcle solution for w when Xam sindependent of misao reported thereVOL. 66 NO. 4 WILLIG: CONSUMER'S SURPLUS $95 Tame so 88:8 8 Bee TS Ie CS eR aa =D 2 re 2 2 = * 2 = 2.2 3 2 2 2 = a = a a 2 2 a 3 2 22 22 8 8 2 z te me 2 22 2 2 eos ee oss 1085 hos ca aC Cama a ioe tap ose og a io cos tastes ote oT tas a tee Cees agra “ap “Age ee ee alae ais 000 ie ass fe ca Sakr “owe i 4e er eee aha oxo. ees ae es ae came ae 2: = OE * Fach group of three numbers includes, from the top, 8/2, (CHa) Ahad The entry * indicates that (4 (Ue) <0. tol /o, and (C= (tae) ‘The latter two expressions encompass the forms of the bounds in (20) and (21), when a is interpreted as |A]/m?.!¥ It can For example, the value of the lower bound in (20) wT05 .088, Thiscan be found in aera when y=2and jn e ‘Table 1 as the value of ((—(—n)9)""™—1-+0}/a| when y=2 and o=.05.596 THE AMERICAN ECONOMIC REVIEW be readily seen from the table that for the ranges of parameter values studied,!* when |2/2| is small (say less than .05), na/2 is close enough (within .005) to the actual bounds for most practical purposes. ‘This numerical observation corroborates the loose application to (15) of the Taylor Series expansion in Section II. More im- portantly, it establishes the rules of thumb previewed in (1) and (2). Addition of (1) and (2) yields a check on the numerical proximity of C and E: when | 7A /2m| <.05, | 4A /2m®| <.05, and A/m'| <9, alAl c- es Bale m lal So, the analysis hinges on the magni- tudes of and A/m®. As discussed in the introduction, in most practical applica- tions |14/2m*| and |A/m?| are likely to be small enough for the rules of thumb to apply. If not, equations (19)-(21) and Table 1 will be useful. Even if the calcu- lated error bounds are too large to be ignored, the compensating and equivalent variations may still be usefully estimated from the data via the formulae. V. Individual Welfare and Consumer's Surplus With the approximation results in hand, let us return to the question of how to make statements about individual welfare, based on observable data. Remember from (4) that 1p’, m’)Z1(p%, m®) as m’—m"2C. With the empirical ’ information that Hp°, m®), Up, m!) < 1p, m®), itm! —m>T ifm! =m 0. that [Ajne] <9. SEPTEMBER 1976 1f Cand T are close in value, (24) pro- vides a welfare test of considerable power." If |44/2m"| and pA /2m°| are small enough, both and C can be safely replaced in (24) by A. Otherwise, they can be calculated from 9, 9, A, and m. ‘To conclude, at the level of the individ- ual consumer, cost-benefit welfare analysis can be performed rigorously and un- apologetically by means of .consumer's surplus. Another welfare comparison (which may be useful foran analysis of toca welfare with a Bersonian socal welfare function) is made posible by the fact (sce HRurwice and Uzava) that u( p,m), viewed asa func Hion af # and m, isa proper indzect utility function ate |p,m) = Bm where Eis the equivalent variation aswocated with a hang from f° to p. Hence this particular ordinal in ‘iret wily function can be exactly expressed by areas ‘under compensated demand curves, a (3), oF can bie timate fom consumer's srplas via (19), Ql, or Q), REFERENCES K. J. Arrow and T. Scitovsky, Readings in Welfare Economics, vol. 12, Homewood 1969, J. Dupuit, “On the Measurement of the Utility of Public Works,” (1844), translated and reprinted in K. J. Arrow and T. Scitovsky, eds,, Readings in Welfare Economics, vol. 12, Homewood 1969, 255-83. J. Ri Hicks, A Revision of Demand Theory, London 1956. H. Horelling, “The General Welfare in Rela- tion to Problems of Taxation and of Rail- way and Utility Rates,” reprinted in K. J. Arrow and T. Scitovsky, eds., Readings in Welfare Economics, wol. 12, Homewood 1969. L. Hurwice and H. Uzawa, “On the Integra- bility of Demand Functions,” in J. S. Chip- man et al, eds., Preferences, Utility, and Demand, New York 1971, 114-48. D, Kataner, Static Demand Theory, New York 1970. LM. D. Little, 4 Critique of Weljare Eco- nomics, London 1937. A. Marshall, Principles of Economics, 9th ed., ‘New York 1961.VOL. 66 NO. 4 L. W. McKenzie, “Demand Theory without a Usilty Index,” Rev. Econ, Stud., June 1957, 24, 185-89. E. J. Mishan, Cost-Benefit Analysis: Am In- sroduction, New York 1971. H. Mohring, “Alternative Welfare Gain and Loss Measures,” Western Econ. J., Dec 1971, 9, 349-68, B.A, Samuelson, Foundations of Economic Analysis, Cambridge 1947. T. Scitovsky, “A Note on Welfare Propositions WILLIG: CONSUMER'S SURPLUS sor {in Bconomics,” reprinted in K. J. Arrow and T. Scitovsky, eds., Readings in Welfare Economics, vol. 12, Homewood 1969, R. Willig, (1973a) “Consumer's Surplus: A Rigorous Cookbook,” tech. rep. no. 98, Economics Series, Inst, for Mathemat, Stud. in the Soc. Sci., Stanford Univ. 1973. + (1973b) “Welfare Analysis of Policies Aflecting Prices and Products,” memo. no. 153, Center for Research in Econ, Growth, Stanford Univ. 1973.