working paper department of economics ON THE MISUSE OF THE PROFITS-SALES RATIO TO INFER MONOPOLY POWER Franklin M. Fisher Number 364 February 1985 massachusetts institute of technology 50 memorial drive Cambridge, mass. 02139 /on the misuse of the profits-sales ratio TO infer monopoly power^ Franklin M. Fisher Number 364 February 1985 Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/onmisuseofprofitOOfish On the Misuse of the Profits-Sales Ratio to Infer Monopoly Power Franklin M. Fisher Department of Economics, E52-359 Massachusetts Institute of Technology Cambridge, Massachusetts 02139 ABSTRACT A large literature tests the structure-conduct-performance paradigm by regressions which use the profits-sales ratio as the dependent variable. That ratio is used because, under constant returns, it is generally supposed that profits/sales equals the Lerner measure of monopoly power. Profits/sales ratios are also used to make inferences about monopoly power for individual firms. This paper shows, however, that unless proper account is taken of the valuation of capital and the treatment of capital costs, profits/sales will be a very substantially inaccurate surrogate for the Lerner measure. Such proper account requires Hotelling valuation of capital goods which real firms do not use. The full analytics of the problem are worked out and some theorems proved relating the error to the cost of capital, the growth rate of the firm, and the depreciation method used. Examples are used to show that the error is poten-^' tially enormous, and the error is shown in general to be related to the variables used on the right hand side of the regression studies referred to above. The conclusion is reached that such studies are totally worthless. That conclusion will doubtless spark a certain amount of controversy as did the conclusion of the parallel paper on the accounting rate of return on assets or equity (Fisher and McGowan, American Economic Review, March 1983). i SEP 3 1985 REC£3Ve3 ON THE MISUSE OF THE PROFITS SALES RATIO TO INFER MONOPOLY POWER Franklin M. Fisher Massachusetts Institute of Technology* 1. Introduction It is popularly supposed that, Lerner measure of monopoly power — — under constant returns, (price - marginal cost)/price is equal to the ratio of profits to sales. once that, one Moreover, shows This paper leaves the simplest one-period model, generally not the case. the this is the errors can be quite large and are likely to be systematic. The reason is not hard to find. Constant means returns constant returns to all factors including capital, so that unless proper account is taken of capital costs, the profits-sales ratio will simply misstate the Lerner measure. Assuming, city that the prices of capital goods are constant, costs involved for simpli- the capital will be depreciation and foregone interest opportunity cost of capital). (the costs The reported value of such depends on the way in which the firm values its capital stock. It should therefore generally ling come as no surprise that capital costs not be correctly measured unless the firm uses valuation in which capital goods are valued at the will Hotelpresent the remaining net revenue streams they generate using the firm's risk-adjusted discount rate (Harold Hotelling). Real value of firms do not do this. This capital-theoretic problem with the easy measurement monopoly power difficulty has a familiar ring to it. Precisely the occurs with the use of the accounting rate of of same return on capital value or on equity to infer monopoly profits. Indeed, of the profits-sales ratio to the Lerner measure relations the resemblance to the relations of bear a striking rate of return on capital to the economic rate of return studied in my earlier paper with John J. Those McGowan (Fisher and sought a refuge from our results in the who accounting the McGowan) use the of profits-sales ratio instead of the accounting rate of return have not found one (William Long and David Ravenscraft; Stephen Martin The analysis of monopoly power cannot be so cheaply 1983, 1984). accomplished. with the issues surrounding the accounting rate As the turn, existence profits-sales witz, of capital-measurement problems presents a slashing attack, re- with the LeiboMartin and 28-32) lays out some of what is involved. pp. of ratio has been noticed in the literature. particular, in (1983, 2 But, de- Martin's assertion that such problems are well known (Mar- spite tin 1983, p. 32), they do not appear to be well understood. The full analytics of the problem appear not to have been spelled out nor has the importance of the problem been realized. That importance literature able the substantial. very large using the profits-sales ratio as the dependent vari- is There is in regressions purporting to test various structure-conduct-performance paradigm. Martin, a propositions (See, of for example, 1983, Leonard Weiss, and David J. Ravenscraft.) Yet that variable dependent is subject to substantial capital-valuation problems studied here. from error Moreover, the contrary to facile assertion that one can assume such errors are distri- the buted independently of the right-hand side variables used in such for example, studies (See, Martin 1983, it is shown 32), p. below that such independence cannot be assumed. Of course, about the monopoly power of individual inferences such use is sometimes made. a independence make justify the use of the profits-sales measure to not would even the unwarranted assumption of Commission's Line-of -Business The staff of the firms. Yet Federal Trade program proposed to use the data generated by that program to identify targets of antitrust prosebased on profits-sales ratios (Long cution ants to pointed the Antitrust Division of the Justice to high monopoly power. cites 1983 known) £i. ^il) for profits-sales ratios as Thus, F. M. Scherer i and consult- Department convincing have proof of (one of those whom Martin the proposition that these problems are well states (Scherer p. 620) IBM's normal strategy was to set prices exceeding manufacturing costs by at least a factor of four. If this does not reflect monopoly power, what does? Apart from the fact that such a statement ignores IBM's tial non-manufacturing costs, so far as I glosses over the question of capital costs. a need for a can tell, Evidently, substanit at best there is systematic consideration of what is involved here. 2. Formal Model: Hotellina Valuation suppose the firm produces output, x, from capital To begin, and labor inputs, K and L, respectively, according to a produc- tion function: X (1) = F(K, L) assumed once continuously diff erentiable. L a flow, as is x. The time variable, t, has been omitted. firm faces an inverse demand curve, The where p is price. ble inputs) is w, The wage of labor Begin given by p = p(x), (which stands for all varia- and the price of a unit of capital is considering the simplest case in which by norma- The interest cost of capital is r. lized to be unity. invests Here, K is a stock and in a single project starting at time 0. the In that firm case, the firm maximizes its present value, given by: oo J (2) ) [pF(K, L) differentiation Denoting - by wL]e ^^ dt - K subscripts, . the conditions are: R'Fj^ = w (3) all t and oo (4) j R'Fj,e ^^ dt = 1 where subscripts denote partial differentiation and; first-order R' (5) is marginal = xp' (x) + p • Note that R', revenue. like the other variables, depends on t. Now, gives the first-order condition with (4) capital as of time It respect the moment at which the capital is bought. 0, is not hard to see that a similar condition will hold for any later time, with the right-hand side of (4) replaced by an appro- priate value for a unit of older capital. Indeed, firm values a unit of capital of age according to a the dule, ^0. to where V(0) V(0), = 1, suppose that sche= and it is natural to assume V(oo) Then consistent planning will require that V(.) satisfy: oo J (6) R-F^e""^^^"®^ dt = V(e) e so that V(e) of a is the present value as of marginal unit of capital. This, of course, valuation (or "economic valuation"), and which satisfies (6) by V* ( . ) Now that V{.) I shall denote the V(.) Obviously, V*(0) = 1, and I shall 0. assume that the firm does use Hotelling = is Hotelling reserving the unstarred symbol for , general valuation functions. assume that V*(oo) = of the future benefits V*(.) and differentiate (6) with valuation, respect to so 0, obtaining: R'Fj, (7) The right-hand = - V*' (0) side of (7) + rV*(0) is readily seen to be the instanta- consisting of depreciation plus the cost of capital, neous puted interest on the funds tied up in Since (6) and (7) unit of capital. a are equivalent, given (4) im- 4 and the fact that V*(0) = 1, the firm (or the analyst) can use either one. If then optimal behavior is characterized totally in (7) is used, — of values defined at time terms Martin emphasizes (Martin 1983, later marginal revenue products. on the left-hand side of sented (0) , requires forecasting In any event, (7) what is buy schedule V* If the firm and sell capital of different ages according ( . ) , repre- is the full marginal benefit of an additional unit of capital of age 0. at time could 30), but which is quite mis- p. since Hotelling valuation, V* leading, property of "myopia" which a then the right-hand side of long-run marginal cost of capital of age to the would be the true (7) at time 0. if Hotelling valuation It is therefore not surprising that, the prof its-to-sales ratio is used and constant returns assumed, does in fact equal the Lerner measure. To see this, assume that is homogeneous of degree one. Denoting profits/sales at F(K, t L) by M(t) : px - wL - [rV*(t) - V*'(t)]K = M(t) (8) px where again it must be remembered that all the variables on right-hand side of simplicity, of r). (9) (8) By depend on t (3) and wL + [rV*(t) - (7) (with the for and Euler's Theorem: V*'(t)]K = R' (LF^ Li so that exception, the + KF., JS. ) = R'x , p - R' M(t) (10) = - = 1/h This is the Lerner measure where h is the elasticity of demand. full exploitation of monopoly power on the part of assuming the firm. Note, however, that this assumes that Hotelling valuation is used. If (as is nearly always the case for real valuation schedule is used, other , any hold and firms) then (10) will not it will not follow that profits/sales equals the Lerner measure. To go more deeply into this, somewhat. tion Let G(t, 0) = R'Fj^ G(t, t) recalling be the marginal revenue product at Then, in the previous notation: time t of capital of age 9. (11) nota- it will aid to simplify both marginal revenue the marginal physical product of capital will generally depend on t. G(t, t) the stream of net benefits which flow from a marginal is again thus that unit of capital acquired at time 0. two arguments, I and write G(t, 0) as having because such a stream will generally depend both on the calendar date, t, at which the benefits are received and on the age of the capital involved, 0. We must now consider the nature of such dependence. In full generality, t is not it is one-hoss the marginal product of capital at time necessarily independent of the age of the — a special case shay — capital. then capital is essentially type with capital of different substitutes until one vintage evaporates. vintages of If the perfect More generally, there technical change embodied in the capital stock or be may may be physical depreciation. a there Hence the net benefit stream from particular unit of capital may depend on the capital's vintage, t - 6, and thus on its age, 9, given the date, t. however, is only depend as important for our purposes least At fact that benefit streams will the on capital age but also on calendar demand may change over time and so may wages. disembodied technical change. outputs different as In response, such not effects, generally time. Thus, There may also be the firm may choose and different employment levels at different This will make marginal revenue and the marginal product times. of capital vary over time. Only in the quite unrealistic case of a totally stationary environment will this not be so. consider Now, in and (6) match (7) marginal certainly . the generalization of Hotelling valuation as If depreciation plus imputed interest benefits on all types of (a G(t, 0) , then so that Hotelling valuation, V*(t, 0) mea- must generalize to: (7) = -V*(t, to condition sufficient for profits/sales to equal the Lerner sure under constant returns) (12) capital are 9) + rV*(t, 9) is: oo V*(t, 9) = (13) / G{t, u)e"^^"~®^ du 9 Note that the depreciation term in (12) takes no account of the fact that when capital is one year older it will be a year later. In effect, the firm consistently planning at t behaves as though it faced a schedule of capital prices, V*(t, 0), at which it can buy or sell capital of differing ages. feature of (12) and (13) is important The Hotelling that valuation in the model which leads to profits/sales equalling the Lerner measure, depends not merely on capital age, 9, but aJLso pp calendar time This has i.. the following implication. Even when such dependence does not occur, expect we do not firms to use Hotelling valuation and the depreciation sche- real dules which correspond to it. has stream a For example, benefit where the positive bulge some time after the investment is Hotelling valuation would require the firm to write up the made, value of existing capital by taking negative depreciation as that bulge gets closer. thing. Obviously, real firms do not do this sort of the present case, In however, the fact that Hotelling depends directly on calendar time would require valuation firms to adopt different depreciation schedules for otherwise identical capital of different vintages solely because wage rates or demand changes over time. practice, In principle, they obviously do not. studies will profits/sales use Hotelling firms ought to do this; in The hope that firms in actual depreciation make thus and equal the Lerner measure under constant returns is plainly forlorn. more word must be added before proceeding. One seen, the case of Hotelling valuation requires even profits to capital if the profits-sales ratio is to equal the sure. Because In r As we have be adjusted by subtracting the imputed practice, — this is seldom, interest Lerner done if ever, the risk-adjusted cost of capital accounting — on mea- directly. is not known. studies which make any adjustment for the opportunity most of cost capital do so by leaving it out of the computation of profits term in the value of capital divided by and putting the right-hand side of any regression. a sales If Hotelling on valuation used and if the capital-output ratio had no other effect on were the Lerner measure, then the coefficient of that variable In practice, as Leibowitz be r. (p. 8) would points out, that coeffi- cient is often found to be significantly negative. I shall comment later on a possible reason for such a For the present, ing. imputed find- simply consider the procedure of leaving interest out of costs and placing the value of capital divided by sales on with profits/sales as the dependent variable. Suppose that the resulting regression coefficient for that variable is b. ble This is equivalent to subtracting r times that varia- from the dependent variable and obtaining a regression coef- ficient on the right-hand side of b - r. Such a subtraction, however, amounts to evaluating profits/sales allowing for imputed interest at the correct interest rate, valuation of capital. r, but with the firm's own In what follows, I assume that to have been done and consider the consequences for estimating the Lerner measure as though it were done explicitly. Studies or opinions which make no such adjustment are obviously hopeless. 10 3. A Stationary Environment: The Most Favorable Case Suppose then that firms do not use Hotelling valuation. Is it nevertheless possible that profits/sales will equal the Lerner measure under constant returns. useful To examine this question, it is to go to the most favorable case, to ignore the issues just raised, and to assume that the firm exists in an essentially stationary environment dar firm, and time. one in which benefits do not depend on calen- (To make this consistent with net investment by must suppose that demand shifts in a very special way the firm invests to secure the same marginal revenue in periods.) In this case, we can replace G(t, 0) the by f(9). all Allowing for this change in notation: f(e) (14) = - v*'(e) + rv*(e) Now denote the amount of investment which the firm makes time u by I(u). one (For simplicity, type of capital; I at assume that the firm uses only generalization is easy.) Then, for any valuation schedule, V(.), the sum of total depreciation and total imputed interest at time t is given by: t ^- (15) / I(u) [-V (t-u) + rV(t-u)] du -oo On the other hand, total net returns to capital, the depreciation and imputed interest corresponding to Hotelling given by: t (16) ^ " I(iJ)f(t-u) J -oo 11 du valuation, are capital costs are subtracted in Since understatement overstates costs (B measure. (B Lerner measure, the A) > means profits, Only if B while an overstatement of such Lerner the A will profits/sales give = an profits/sales that means that profits/sales understates A) < such costs of calculating correct the result. 4. The E rror Formula expression the for measure Lerner instructive very is It in now consider to difference between the precise the profits/sales stationary-environment and the under case consideration. Let the true value of be L* L* = (p - R')Pf where p is price. Lerner the measure. Then Equivalently, px = R'x/(1 - L*) (17) Let measure. L the profits/sales ratio be Then, using (15), (16), and B - A shows estimate lated that supposed Lerner (17): - A) (1 - L*) R'x px This the = L - L* = (18) (B — the error in the use of profits/sales of the Lerner measure is very unlikely to as an uncorre- be This with variables which affect the Lerner measure. is because L* itself appears on the right-hand side of (18) Indeed, be side the only possibility of avoiding this result would if the appearance of x in the denominator of of (18) the numerator. the somehow cancelled out the appearance of This does not seem a very realistic 12 right-hand (1 - L*) in possibility, we shall now see that it is definitely not true in the and most interesting leading cases. we must explicitly account for see this, To physical the depreciation of capital and the time it takes for new machines to be installed and running. I do this by altering the description the technology slightly and assuming that one unit of capital of of age 9 is equivalent to a fixed number of units, it is natural to measure capital Then capital. a one-hoss shay — corresponds capital to no deterioration while the capital is in a constant for a (9) Consideration of good.) (The case of (7) 0^9^ and the life above, (14) use the of however, that the marginal revenue product of capital of age 9 must shows be — efficiency in each of which has the same marginal product. units, of new a (6), f(9), so a (9) is proportional to f (9) and we might renormalize by weighting capital of age 9 by f(9). as well From (16), B is then the capital stock in the new efficiency units. More precisely, suppose that, at time t, to capital of vintage u is L(t, u) the labor assigned and that the output produced thereby is: (19) x(t, u) = F(f(t - u)I(u), L(t, u)) Then (19) states that the effects of capital age can be as a capital-augmenting technical change. constant returns marginal revenue is not changing) and theorem on capital aggregation (see, 559-60) Since we are assuming all labor has the same , captured wage (and since it follows from a well-known for example, Fisher, that total output is given by the production function: 13 pp. = F(B, X (20) where is defined in B employed by the firm. measured capital L) (16) and L is the total amount of Note that (7) and labor imply that we have (14) so that the marginal revenue product of B is unity. Now consider changes over time with w and R' held Since F(., .) constant. is homogeneous of degree one and L will be chosen to have marginal physical product equal w/R', the ratio of L to B It follows that x/B will also be constant and will be constant. that R'x = mB for some constant m. Substituting in (18) yields B - A L - L* = (21) (1 - L*) mB The constant, R' , m, depends on w and on R', it is determined by the technology. firms with the same true Lerner measure, values profits/sales taken In particular, different L*, will have different and different firms with the same value of m of m, different have but, given w and values of L* This means that . error in cannot be the as a substitute for the Lerner measure be independent of the Lerner measure itself or of to will the variables assumed to influence that measure. It is case (and below) . the special particular helpful in interpreting the numerical examples given Suppose the technology is Cobb-Douglas, so that F(B, L) (22) Then illuminating to evaluate m in a = B^L-"-"^ fact that the marginal revenue product of B implies that R'x = B/a, so that m = 1/a and (21) becomes 14 is unity B ~ A L - L* (23) = a (1 - L*) B appearance of the Cobb-Douglas parameter in this way is The no accident. Returning to the more general case of any constant- returns production function, and tion, Euler's Theorem, profit-maximiza- fact that the marginal revenue product of B the is unity imply: B = R'x - wL (24) , so that 1/m = 1 (25) - wL/R'x , or one minus labor's share of the value of output when that value is in terms of marginal revenue rather than of price. Cobb-Douglas production this case, function, As in the will depend on the parameters but it should come as no surprise that the errors which come from the mismeasurement of capital bulk when (true) the of larger capital costs are a large share of total costs than they do when that share is small. In the case in which the aging of capital does not have capital-augmenting effect on its productivity, are less readily come by. It remains generally a simple formulae true, however, that the error depends on the Lerner measure and on the nature of the technology. case, the It is obvious that, as in the capital-augmenting error must be absolutely larger, more important capital costs are as I and a ceteris paribus the fraction of total costs. shall return to a discussion of the error formula also to the use of the Cobb-Douglas example. 15 For the below, pre- however, sent, want to put aside the leading but special case I capital-augmenting of general the changes due to capital age and return to examine the question case of whether to without it is possible that B = A and profits/sales Hotelling valuation, equals the Lerner measure. 5. The S im plest Case: Exponential Growth To examine the circumstances under which B = A, define t K* = (26) / I(u) du , -oo the total value before depreciation of the firm's capital stock. Then A/K* is average depreciation and imputed interest per dollar of capital put in place in the past, while B/K* is similar a average of net returns to capital. For profits/sales to equal the Lerner these two averages have to be the measure, obviously magnitudes happens in if each weight, applies equal to that would be the case if the two averages; used Hotelling valuation. firm I(u)/K*, This same. I now investigate the circum- the stances in which the two averages will be equal even though their components are not. I shall consider only the simplest and most favorable case, that of exponential growth in which: (27) I(u) = e^" this is at least a case in which one can imagine the stationarity assumption to hold with demand and all variables growing at same rate g (which does not make it a realistic 16 case, the however). As we shall now see, even in this unreasonably favorable case, it is generally not that profits/sales true equals the Lerner measure. To see this, define: so = e"5^(B - A) C(g) (28) that is equivalent to profits/sales = C(g) the we see that, in the exponen- Letting 9 = t-u, Lerner measure. equalling tial growth case: oo C(g) (29) = J - V* [rV*(e) ' (0) -rV(e) + V (0) ] e~^® de oo - f (9)]e 5® d9 [f (9) where (30) f(.) f (0) = -V + rV(9) (9) should be interpreted as that stream of benefits which would make V(.) (the valuation actually used by the firm) the appro- priate Hotelling valuation. Theorem 1. In the exponential case, Lerner measure if g = r. Proof : Consider (29) at g = profits/sales equals the ' r On the one hand, since the price . of a unit of new capital has been normalized to be unity, 17 oo (31) j f(e)e~^® de = V*(0) = 1 On the other hand. oo (32) / [-V(e) + rV(G)]e ^® de = - V(e)e"^®| = 1 V for the same reason. This accounting result parallels that for the relation here the cost of capital, economic rate of return. return on the margin.) r, said. ing this 95) (Of course, r except , is involved instead of the is the economic rate of When the growth rate is equal to the cost of capital, everything comes out all right. however, the rate of return and the economic rate of return in the exponential growth case (see Fisher and McGowan, p. that between In the present case, is just about the only positive thing that can be It is not even true that the weak result that the account- and economic rates of return are on the same side of the growth rate has a parallel here. This C(g) can most easily be seen by observing that (29) shows to be the present value of the difference between two bene- fit streams, f(.) and f(.), discounted at interest rate g. Since there signs is nothing to prevent that difference from changing an arbitrary number of times, C(g) deed an infinite number of) roots, 18 can easily have several (in- must be one. Fur- of which r ther, C(g) will generally not be monotonia in g. and are unrestricted except by f(.) between and f(.) firms reasons, by considering one way in this f(.) accelerate are which relation the is likely to be restricted. obvious For likely to use depreciation schedules depreciation. depreciation all it may be possible to get a bit fur- however, In practice, than this is (32), These propositions are exemplified below. that can be said. ther and (31) So long as f(.) If they are able so to which accelerate to make the value of their capital stock as never greater and sometimes less than would be the case under Hotelling depreciation, then Theorem definite result does emerge. a Suppose 2. that V(9) £ V* (6) It is: for all 9 2 0, with the strong inequality holding for some set of values of 9 of non-zero measure. for growth rates sufficiently close Then, zero, to profits/sales overstates the Lerner measure. Evaluate (29) at Proof. g = 0. Then: CO GO But V(oo) = C(0) (33) the first = = integral must [V J (9) -V*'(9)] integral is zero, V*(oo). be d9+ / [V* (9) since V(l) Under the given positive. r Hence, =1 -V(9)] d9 V*(l) = assumption, at profits/sales overstates the Lerner measure. g = 0, the B > . and second A and By continuity, the same result is true for growth rates sufficiently close to zero. 19 In the proof depends on observing that, essence, growth rate, the firm is taking depreciation equal to at thfe zero initial value of a single capital good, and this is the same using either valuation. It follows that all that matters is imputed interest, and accelerating depreciation must reduce that. Note that essentially the same proof leads to: Theorem 3. Suppose that one of two otherwise identical firms uses a depreciation always less schedule which results in a valuation capital than or equal than that of the other and non-trivial time periods. ficiently close to zero, less for Then, for identical growth rates suf- the firm using accelerated depreciation will have profits/sales greater than the other firm. Note that, at low rates of growth, accelerating depreciation does not increase the profits/sales ratio. Of course this A simi- because of the steady-state nature of what is going on. lar result is holds for the accounting rate of return (Fisher and McGowan, p. 86) As one might expect, rates of growth, lar, the opposite result holds at very high for high enough rates of growth, a firm accelerating depre- ciation relative to Hotelling valuation will have understating the Lerner measure. g goes to infinity, the values of To profits/sales see this, observe that, as the only things that matter in C(g) will depreciation installed capital stock. the In particu- although this is less interesting. and foregone interest on newly- But the value of new capital stock same in all valuation systems, so the sign of C(g) only on the difference in the depreciation being taken, 20 be is depends and C{g) will be negative under the stated assumption. identical otherwise two of relative Similarly, if one depreciation accelerates firms then at high enough rates of growth that to the other, firm will have a lower profits/sales ratio. is monotonic- It is tempting to conclude from this that C(g) ally decreasing in the growth rate in such but this need not be true. tion cases, that depreciation is higher, means Accelerated depreciation but it also means that fore- therefore, Even in such cases, gone interest is less. [f(0) al values of g at which C{g) =0. The Size and Behavior of the Error: Examples It large enough to know that such not is problems can exist, It is important to know whether they are likely to however. instructive very Consideration begin by working out a to set examples. of of the leading case discussed above in It is which the effects of capital aging are capital-augmenting shows that it useful to do this by evaluating times denote by use (23) Z) (B - A)/B (which I For a Cobb-Douglas technology, . shall is some- we can then to obtain an order of magnitude for the errors. Examples are fairly easy to generate. Each example requires choice of f(.) subject to the restriction that oo (34) be and whether they will tend to be correlated with variables used in regressions purporting to explain monopoly power. a - can change sign more than once, so that there can be sever- f(9)] 6. accelerated-deprecia- / f (0) e"^® de 21 = 1 This is done in various ways as described below. each of the examples below is worked out addition, In different three types of depreciation schedule. — straight line for :! which f (9) The first has: v'(e) = (35) for -1/T v(e) = (T - e)/T where T is the life of the capital good (the point at T, becomes and remains zero. The second type of depreciation is the continuous-time equi- valent of sum-of-the-years digits. It has T - e V (9) (36) • = ^~ - V(9) e(2T - =1 - TV2 d switch Define straight line when the latter method becomes faster. = 2/T and 9* = T - 1/d, . T^ The third type of depreciation is exponential with a to 9) X so that d = 1/{T -9*). Then this method has V(9) (37a) ^ 9 i for = - d e~^^ 9* V(9) = e"*^® and e -d9* V'(9) = - (37b) V(9) = (T - 9) d e""""^^ T - 9* for 9* ^ 9 ^ Each in T. of the tables giving the results for two parts. rate, r = .2; In the first half of each in the second half, 22 r = .15. Z table, = (B the - A)/B is interest These seem reasonable enough for illustrative purposes. in the rate differ r In practice, that should be used, preted as the cost of capital including for must be r inter- premium. a risk we should expect from the analysis above, As firms of course, Z = A)/B - (B is most often greater in absolute value for growth rates far from r than it is for growth rates close to common, when interest it would be absolutely I values 2 they mistake to conclude that those a the results In any event, small. then to the question of how to judge given in the tables are high or Z indicated, problems the for in As L-L* already error case. (In B-A 1-L* = the Dividing (23) by L* yields roughly similar results.) (38) whether other capital-augmenting cases will yield view of (21) and (25), a ) ( L* ) ( L* B Obviously, this means that a given value of a high, rule of thumb can be given using the rough a low. formula (23) for the Cobb-Douglas-capital-augmenting to are percentage points each. come of rates are low than when less case are given for growth rates from zero to 30% exponential steps of not of the Lerner measure are likely to be relatively although become a this suggests that the problems with profits/sales as an estimate severe although this is not Since growth rates as high as 15% are property. universal r (B - A)/B corresponds greater percentage error for low values of L* than for high ones, a fact I shall discuss below. Ravenscraft of -Business Cobb-Douglas (p. 31) gives a mean value for L from the Line- data of .0648, excluding imputed interest, estimates for the United States suggest an 23 while average value of measure actually exclusion given If one were to believe that profits/sales about .25. a on average and were to L* of imputed interest, about forget this would mean that the in the tables below should be multiplied by the values about 3.6 to obtain an order of magnitude for the errors as a fraction of the Lerner measure. Because of the exclusion of imputed this number is an underestimate. profits/sales Lerner point whole The that is of the It is therefore of some interest to judge the of the error without assuming that L and L* are orders of magnitude. same the Solving as a function of error close This is easy to do and leads to roughly together on the average. express however, paper, far from being a trustworthy estimate is measure. magnitude the p this of interest, for (23) Dividing L. L*, one can L, one by obtains: L - L* 1 = (39) a L where, again, Z - L ( L = (B - A)/B. Z ) ) ( 1 - aZ This gives the error as a fraction of the observable profits/sales ratio, L. the tables sufficiently given above below and for a For the values of approximately .25, however, small that the (underestimated) multiplier in Z is aZ of 3.6 remains roughly correct (as do the other multipliers given in footnote 8) ; somewhat greater values are required when is positive, and somewhat smaller ones when Z Z is negative. In sum, the Line-of-Business data suggest that the values in the tables below should be multiplied by at least 3.6 to be inter- preted as fractions of observed average profits/sales ratios. 24 If that the average profits/sales ratio is believes one true Lerner measure average (a dangerous assumption) same multiplier gives the error as a , also the then the fraction of the average true Lerner measure Table gives the values of 1 case in which f (9) They pattern. A)/B for a one-hoss the maximum value of f(6) is chosen to (B - A)/B exhibit the simplest are all positive for growth rates near zero decline monotonically passing through zero, , for low growth rates, we should also expect, the-years' digits value of - A)/A, (B absolutely lowest depreciation 1 and 2 and must, at above. As as they This is to be expected, given Theorems g = r. shay thereafter. Here the values of satisfy (34).) - is constant for 10 years and zero in all the examples, (As (B the use of sum-of- absolutely yields the highest with straight line depreciation yielding the values and exponential depreciation values somewhere in between the two. [TABLE is evident, It HERE] given a multiplier of 3.6, that the values in Table 1 for low rates of growth are quite large given to 1 be worth attention, and the magnitudes will increase progress to later tables. generated comfort with sum-of-the-years' digits depreciation, small errors soon disappear. Table (on the order of "only" 1 as we Certainly this is true for the values in the notion that straight line depreciation relatively enough but leads 20% or so) any to will happens to be one of the more favorable cases for the use of profits/sales. Table 2 shows that even what appears to minor change in the benefit profile, 25 f{.), be a relatively can make an enormous difference Table in the magnitude of the errors. table, like case. The that now the benefits from the capital good only presents the results for 1, difference is That a one-hoss shay begin two years after the expenditure for that good is the future. At low growth rates, the values of tripling - (B A)/B generated by straight-line for sum-of-the-years' digits. results the profits/sales about Even in Table 1, doubling and those at four percent correspond to lower bounds for errors in at the latter 's average value of from about 24% to Now, 80%. ranging this results in roughly depreciation and somewhat less than exponential growth, In the benefit profile of Table 1 is shifted two years other words, into made. they correspond to lower in Table 2, Even with a from about 76% to about 145%. much bounds lower multiplier than 3.6, these are large errors. [TABLE 2 HERE] The results in Table percentage but they illustrate errors, In that table, also correspond to (absolutely) large 3 10 years at a unit rate of decay of f(9) different phenomenon. has been chosen to decline exponentially for f (6) £ 10) after which a f (9) (B - f (9) = Ce for Again, the maximum value becomes zero. is chosen to satisfy is that the values of (in other words, The new feature in this table (34). A)/B are negative at low growth rates and increase, rather than decrease monotonically with g. [TABLE Table 3 profits/sales growth. As shows that it 3 HERE] is a mistake to conclude overestimates the Lerner measure at low Theorem 2 suggests, 26 that rates of if Hotelling depreciation is relative to the depreciation methods accelerated actually used instead of the other way round, then underestimation will result. this regard that Ravenscraft in (Note (p. 31) finds that the value of the profits/sales ratio in the Line-of -Business minimum data is substantially negative.) As one should expect from Theorem it circumstances, in these now the least rapid depreciation method, is yields that 3 growth. the (absolutely) largest errors at low should also come as no surprise that It function, straight which f(.)f corresponds to Table 3 by delaying its increases the shift moves Hotelling depreciation closer to the three examined. schedules tion positive Such a depreciamethods It makes the errors for some and those for others negative, of the of the errors for low rates of growth. values absolute rates shifting by two years this time reduces rather than start line, but while some of the errors naturally become absolutely small, not all of them do, and the size of the errors (or their sign) cannot be predicted with- out knowledge of the benefit profile, given in Table 4 This rate 4 digits and change sign at rates of growth in Table 3 {f(0) are too large (absolutely) because the 27 the r. - f(©)} (29)). of course, that the values of to generate them is extreme. than other is because we have now reached cases in which It may be objected, used The results for depreciation are not monotonic in can change sign more than once (cf. given are HERE] also illustrates a new phenomenon. sum-of-the-years' growth These results 4. [TABLE Table f(.). (B - A)/B example That example involves a very rapid exponential decline in benefits, so that benefits are close to zero for a good deal of the ten-year life of the capital goods while the firm is forced to depreciate the goods involved at a non-negligible rate over the entire ten years. There an are three reasons to be careful before offering First, objection. benefits after taxes, ciation method. 9 such the example can be though of as one with including the effects of the chosen depre- Second, extending the life of the capital good, as in Table 4 makes things better, not worse. Third, this as suggests, the general principles being examplified do not rest on the choice of particular examples. for Table 5 an exponential decay case identical with that of Table with a life of five years rather than ten. of presents the results, (B - A)/B The resulting values for low rates of growth are value than those in Table 3, but 3 smaller absolute in but they still translate into enor- mous errors in the use of profits/sales. [TABLE 5 HERE] is possible to continue with many more It examples. (In- deed, only the desire to keep benefit shapes, f(.), relatively and probably unrealistically ing even simple keeps one from illustratThe conclusion more complex behavior.) should now be clear, liable — Profits/sales is however. estimate of the Lerner measure. using it can vary from very small to they will be, as functions — to a be drawn totally unre- The errors involved f righteningly large. in Which what sign they will have, and how they will behave of the growth rate cannot be knowledge of the benefit profile, 28 f(.). determined without No conclusion as to the monopoly power of a firm or group of firms can be drawn from such a Further, all these results come from the unreasonably measure. favorable stationary-environment case of exponential growth. 7. Regression St udies and "Random" Errors This does not end the matter, however, for profits/sales is also used as a measure of monopoly power in regression studies in which it is the dependent variable. errors in In such studies, the dependent variable become part of the error term. led some practitioners create errors such them assume p. 32) uncorrelated with the has to assume that no problem because one might just For brevity, regression. Martin 1983, (e.g. This variables well as used the in I shall refer to this as the assumption a dangerous procedure. of "independence". This is, at best, The assumption that error terms are uncorrelated with regressors is often made. as here, it has no better foundation than the assertion Usually, that the investigator has been unable to think of a reason why it should not hold. Where errors are small, assumption may do little harm. large indeed, degression Where, as here, they can be very one is risking a great deal. studies that The very fact a the very large variance make one suspect (although it certainly does those that appear to recover systematic results in presence of errors which can obviously have should such an independence errors are systematically related to the not prove) variables used, even though one would wish to assume otherwise. In any event, even such a wishful-thinking assumption cannot be made in the present case. The error involved in the use 29 of profits/sales ratio is systematically related to the vari- the ables used in regression studies, and the independence assumption is therefore false. To (18)). This consider the error formula (23) see this, It contains a term in - L*) (1 (or (21) which does not cancel out. means that errors will be absolutely larger for firms of the Lerner measure than for high ones values low effect or with this (and will be even more pronounced in percentage terms) This . says that it will not generally be the case that the error can be taken to independent of the variables used to be explain the Lerner measure. More precisely, suppose that L* is supposed to be related to a set of variables, X, by: L* = (40) in X P +6 Assume that the usual notation for regression. ^ mean has zero and is distributed independently of X. We know that L - L* - L*)u. in fact distributed is in the form (1 independently of X; stochastic. Suppose that for convenience, indeed, For convenience, independently distributed. u is take X to be non- suppose further that Then b, u and the least-squares ^ are estimate obtained from regressing L on X has the property: E(b) (41) Only = E{(X'X)'^X'L} = t ^ (1 - E(u)) if u has mean zero can one assume that this creates no bias (and no inconsistency — the problem does not disappear for large sample sizes) No doubt there will be some willing to assume that E(u) = 0, 30 the lack of basis for such an assumption now but possibly make it are asserting that because a who Those exposed. lies large one might as well assume it to an unknown sign, be number has zero. There is no basis for believing that the benefit profiles characteristic happen of capital goods in the American be distributed relative to the to economy depreciation just methods used so as to make the average value of u = a(B - A)/B zero. fact (B - that The A)/B can be either positive or negative is not a sufficient basis for such a belief. the absence of such an unwarranted assumption, In (also unwarranted) assumption that u is even the distributed independently of the variables used in regression analyses will not rescue such studies. Assuming that - E(u) 1 0, > then such studies will have their coefficients biased towards zero if E{u) zero if E(u) < This sort of problem would only be avoided if 0. the dependent variable were log such Even made the in profits/sales because ables a - L) (1 choice of dependent variable literature as — will not — rescue commonly not studies a surrogate for the Lerner measure. it is clear that u itself is correlated with used from and away > in such studies. I using This is vari- the comment briefly on some of the effects involved. (a) who recognize interest into Capital Intensity. — account We saw above, As already observed, that the opportunity cost of investigators capital needs to be included frequently attempt to by using the capital/output ratio as a however, — imputed take regressor. that relatively capital-intensive 31 it firms While we cannot say will have relatively high (absolute) errors. what sign those errors are likely to have, it is again foolhardy to suppose that this means they can be assumed zero. there is some evidence here as to the sign of such In fact, errors. the If book value of capital were also the value (and Hotelling depreciation subtracted from coefficient of risk-adjusted has Retelling profits) , the the capital-intensity variable should equal the cost of capital (assuming that no direct effect on the Lerner measure) intensity capital Leibowitz . (p. 8) points out, however, that such coefficients are often found to be significantly negative. A distinct possibility is results from negative values of Depreciation Methods. (b) (B - A)/B, methods that 12 of with faster depreciaslower producing algebraically larger errors than ones at low enough rates of growth. however, as in Table 3. this It is obvious that the choice depreciation method influences the errors, tion that Gerald Salamon points large firms tend to use accelerated out, depreciation This will lead to an upward bias in the coefficient of methods. a firm size variable. Growth Rate. (c) As we have seen, the size of the error is influenced by the growth rate of the firm. Studies which do not include variable will have errors correlated with any independent variable correlated with growth rate. the growth rate as an explanatory Firm size or market share appear candidates here, since they will be influenced by past growth. Even tion above. the inclusion of the growth rate requires the that the environment is stationary in the sense assumpdescribed (Departures from stationarity can also be correlated with 32 included variables, but it is hard to say anything specific about this.) Risk. (d) The value of difference in the errors. used obviously makes r differs over firms because it is of capital including a risk premium. cost the Yet r At sizeable a low it seems likely that firms engaged in relatively activities will have relatively larger errors (absolutely) rates, will relatively risk-free firms. Some variables, growth risky than such as re- search and development expenditures, may be related to participation in risky activities; others, such as the variance of firm or the Hirschman-Herf indahl index may relate to the sizes risky nature of the industry involved. 8. Conclusion The conclusion to be drawn from all this must be can (p. 91) on the companion issue of the relations of accounting and economic rates of return. them sedly I do no better than to paraphrase the conclusion of Fisher and McGowan who clear. Economists (and others) believe that analysis of the profits-sales ratio much are deluding themselves. the will tell The literature which suppo- relates concentration and the Lerner measure does no such thing, and examination of the profits-sales ratio to draw conclu- sions about monopoly power is a totally misleading enterprise. 33 A ppendi x: A Note on the Use of Tobin's q — recent papers have used Tobin's q Several the ratio of the market value of a company's stock to the value of its capital — assets the dependent variable in regressions intended as test hypotheses about monopoly power. Michael A. Stephen Ross, (See Eric Lindenberg to and Salinger, and Michael Smirlock, Thomas Gilligan, and William Marshall.) Such a use has much to recommend it to the use of the accounting rate of return or relative profits/sales ratio; in particular, problems relating to the risk disappear as the variable itself includes the market's evaluation of the required risk premium. Nevertheless, the capital-theore- problems which eviscerate the meaning of studies tic using the other two variables also affect those using Tobin's q. In appears the first place, in the value of the capital assets the denominator of q typically does not include activities such conclusion tends This that are to bias q upwards in industries important and may lead such activities are correlated where erroneous the to the development value of intangible assets such as past research and activities. which with monopoly power More present directly paper, denominator torically even related to the problems considered in the assets in the the valuation of physical of q raises serious problems. taken accelerated depreciation will show value for their capital stock, ceteris paribus have used slower depreciation methods. an Firms that have his- , a lower book than firms that Further, the size of such effect depends on the rate of growth of the firms involved. Yet depreciation methods may be correlated with variables such as 34 firm size. is certainly true, It the capital-valuation problem affects only the that of as pointed out by Salinger Tobin's denominator q and not both the numerator and denominator does in the case of the accounting rate of return. as Hence a error in capital valuation relative to Hotelling percent 160) (p. it one valua- corresponds to a one percent error in the measurement of q. tion It is not fully clear whether that makes q more or less sensitive to such errors than are the accounting rate of return profits/sales ratio. enough as I to pose a the suspect is true, the sensi- the errors in valuation seem likely to be tivity is less, large Even if, and very serious problem and quite possibly to vitiate the use of the measure altogether. Not even Tobin's q fully escapes the problem that invest in capital goods for the future stream of benefits goods will generate. ment cost would help, firms those While valuation of assets at true replacefirms' book values usually do not provide an appropriate valuation for analytical purposes. 35 References Fisher, Franklin M., "The Existence of Aggregate Production Functions," Econometrica October 1969, 22, 553-77. , McGowan, and John J., "On the Misuse Accounting Rates of Return to Infer Monopoly Profits." of Ame- rican Eco nQmic Review, March 1983, 21, 82-97. Hotelling, "A General Mathematical Theory of Harold, Journal pf September 1925, 2fi, tion," the American Statistical Deprecia- Association , 340-53. Leibowitz, S. J., "The Measurement and Mismeasurement of Monopoly Power," mimeo, March 1984. Lindenberg, Eric and Ross, Stephen, "Tobin's 3 Ratio and Indus- trial Organization," Journal of gusiness , January 1981, 54 , 1-32. William F., Long, Lean, Curtis L., Wagner, David F., Ravenscraft, David J., and Ill, Benefits and Costs of the Federal Trade Commission's Line of Business Program, Volume Analysis Federal , Trade Commission, Bureau of I; SiafJE Economics, September 1982. William Long, F. Accounting and Rates Ravenscraft, of Return: David J., "The Comment." Misuse of Economic American Review. June 1984, JA, 494-500. Martin, Stephen, Market, Firm, and Economic Performance Brothers Center for the Study of Financial . Salomon Institutions, Graduate School of Business Administration, New York University, 1983. 36 Misuse "The ., of Accounting Comment," America n Economic Review David Ravenscraft, Line ^M Salamon, J., . Rates June 1984, JA, 501-06. "Structure-Profit Relationships Business and Industry Level," Review of St a ti st ic s Gerald ^ February 1983, ^, "Accounting Rate L., at the Economics (2i 22-31. of Return, and Tests of Economic Hypotheses: Error, Return: of Measurement The Case of Firm Size," mimeo., June 1983. Salinger, Michael A., "Tobin's a. Relationship," tration-Profits Summer 1984, Scherer, H, Unionization, and the ConcenFand Journal of Economics , 159-70. F. M., Review of F. M. Fisher, J. J. McGowan, and J. E. Greenwood, Folded, Spindled, and Mutilated: Economic Analy- sis and Sj. iIj. 3L.. Journal ol gconoinic Literaturer June lEMj. 1984, 21, 620-21. Smirlock, Michael, "Tobin's a and Gilligan, the Thomas, and Marshall, Structure-Performance American Economic Review , William, Relationship," December 1984, JAr 1051-60. 37 * am indebted to Paul L. I Joskow for helpful discussions but retain responsibility for error. 1. problem is not limited to constant returns and The profits-sales ratio. Any attempt to calculate the Lerner measure involve at least an implicit measurement of marginal must Such a measure, Leibowitz.) J. For simplicity, (This point is well made I assume constant that seems a questionable assumption when though in returns as does most of the empirical literature, throughout this paper, even cost. if direct, must take capital costs properly into account to produce a valid result. S. the studying the effect of firm size. 2. sure Some recent papers have used Tobin's q as a simple While that measure monopoly power. of seems mea- definitely superior to either the profits-sales ratio or the accounting rate of return, its use does not avoid the capital-theoretic difficul- ties here discussed. 3. I comment briefly on this in the Appendix. There are other problems with Scherer's statement. My normal strategy for many years has been to weigh 150 pounds. 4. For simplicity, I assume that the price of a new unit of capital is always constant at unity so that there are no gains or losses stemming from capital price changes. 5. the This can also be shown more laboriously by optimal assignment of labor in the Cobb-Douglas considering case given in which (19). 6. This valuations proviso essentially rules out the case are discontinuous and the valuation used by the 38 firm is below Hotelling valuation only at isolated points. For simplicity, the entire analysis and the examples have 7. been worked out in before-tax terms. applies if after-tax analysis same the interpreting f (.) and McGowan, pp. It is not hard to show that magnitudes as the ^£i£x-tax benefit stream. 95-97.) Of course, are used, Fisher (See in that case the examples would involve different pre-tax f(.) for different below depre- ciation methods. Since Ravenscraft (like others) does not find a 8. sion coefficient interpretable for his capital/sales variable that is such that easily it is not simple further to adjust the as r, figure to take account of imputed interest. ever, regres- It is 3.6 how- clear, an adjustment for any reasonable value of r would make the appropriate multiplier very much higher. Ravenscraft At the maximum value for profits/sales given by (.5371), the lower bound for the multiplier is approximately .22; the minimum so that the corresponding to the low end of the range is unbound- multiplier ed. value for profits/sales is negative, For comparative purposes, using Census of Manufactures data (Ravenscraft, p. 30) , the lower bound for the multiplier at the mean of L (.2188) is be roughly .89; is about these .34; at the minimum (.0096), at the maximum (.4232), it is about it Even 25.8. lower multipliers make the errors in the tables below loom very large. 9. value of 10. See r footnote 7. It should be noted, however, that a of .15 may be a bit high for an after-tax example. It may still be objected that a decay rate of unity 39 is very high, although how one is to know that without any empirical evidence on f ( . ) If lower decay rates are is not clear. used, depreciation becomes less accelerated relative to Hotelling three depreciation methods examined. In the limit, a zero decay rate corresponds to a one-hoss shay example as in Tables As we have seen, the 1 and 2. such an example generates positive, rather than negative errors at low rates of growth. Correspondingly, as the decay rate goes from unity down to zero, the errors at low rates of growth move upwards from negative values to positive ones. course there is some decay rate at which such errors are small (although that rate will not necessarily be the same for each the three depreciation methods used) . It would be to suppose that such a decay rate must be the however, Of of foolhardy, "realis- tic" one. 11. There is also the probability-zero case that u is corre- lated with the regressors in just such a way as to cancel out the effects of 12. A)/B and (1 - L*) . There is no need to discuss this. It could also result from a negative correlation of capital intensity stemming from a systematic between capital intensity and the shape of f(.). 40 (B - relation \m c C o CO O K E M O f-H r» 00 00 o r*- V r~ o 00 o vo o o r» ^ o i-H n CD CN o> in CN vo n (N m CN CM cs Ol i-( •-( i-< 1^ o QO re m n n o ^ (N r- o o 00 a^ i-H O <H n V o o in tc o o in vo o o o <N ^• •a- •-I I-H o o O O i-H o> in ^H in r-) ^ in in vc vc o o I I I vo in CO vo •«• eri o\ —i tH O ^ ^ o o o o ^ o o ^ m o o •-H in o o o o o I t^ o ^ CN r^ r- r» r» 00 I I o o o o O o .-( in rlO iH ^ o O o 1 1 in vo r- «)• Ol a u r» 4J C -r- re —1 n o> o o r» o 00 o o CO l~ \o o o o in in o o •a- O) ^ o o o o o ro o 00 r~ >-i o o 00 in o o o 00 in o o o 1 fN a ^^^ ^-^ < tfl o o o n vo o 00 O o VO •H <H o r~ rH o o 1 00 <H o> r~ vo 00 ^• fn lO (-« o o o CO in 1 1 1 in CO rsi o n o o 1 O tN <N CM o o vo o o 1 IN VO rs n T in •H tn ^T .^T o o oo o n OO w N CO 1-1 CQ ^«^ 0) iH ^ 1 14-1 (U o c re H o o to 01 3 l-l r-l re 0) re p- c 0) c o > o 1 Q. lu 01 o m i-i E QC re 40 >• t-i o m o\ in^» •-( iH fH <H o O O ^ 00 ov Ol o o 00 ^• CO o o o o o o r>- "» m in o o o o iH r>- <<r CN o o o o m n iH O O O O O O O at f-H in cs o o m o o 1 1 1 CNJ O o ^•invovoo%mom«*>ooo^ CT>o^<Nf'iinr~«cMinoin oo cj O o m o VO o o in c>» o o o m o> o o I I I I I o 00 C>4 •H o o O o r>-mr>4ovvomor--incMOCM Cs»C>»CNfH.-|.iH»-HOOOOO O r»- r>- o r^ iH •-( o a Ol in o n cs •« c •-< >-l re •<H o o vo vo o ^ o ^r <N o> o o n CO r~ o o in T VO o o m o o o in vo p~ r~- ^• e*i CN o o o o f-i M ^ O O o o o o o r^ o o «N o ^ o vo o CO o o fH (N iH <a> vo r-( •-( CO <H o CN 1 1 n r-i fS o o 1 oo o o m n o o .-1 CM vo CNl • • 1 o n ir> c o c in o\ CM o% o\ oi CM 00 m vo fN o m ON ro o> ,-t 1-1 i-l r^ m in f-H m o r~ o vo CM o c c X u en CO k' j-i -H oo •H vo oo n ^ m V m , iH CO o\ fs O o m "» o> ON vo ^^ rsi i-i i-t fH E n 3 0) — en >- O rOl 00 o o \o o n o o CM vo i~ O n vo (^ eo .-I i-l o in o n «o in in O o o o o o o o I I I 1 I I I ^ d m vo m ^ vo «N r» vo 1-^ CM o o ^ Ot o n o •H o\ n in n 1 1 1-1 o o o o 1 1 r^ vo 00 f-i m m o m o ^ o 1 1 O ^• vo CS CN CO CN O) OV <n o o m o 1 1 1 VO CN 00 CN o ro CN CN vo in iH CN CN in o o I I I VO r* en vo CN • II u o ^ bC TO vo ^ m o o vo (U cs x: w CM rH r-< ^ •» vo o a Ot H CN O o in CNI fS o o CO CN a\ vo ^• 00 o o iH o\ o\ vo fH O O 1 1 1 o CN «N CN l-i o 00 i-t o "» o M o I (U c o u to CN ^ ^ 00 ^ O oo O 00 VO o o ^ CN I < O" I a >1 CO s o Si cu td H 3 o c c o a. X > cd 0) a fH r» 00 n in n rCN n o •«• o\ CN . CO 03 m CN V ^ CN CO •H CO CN 1-1 o\ o\ rn iH VO in o\ o o a\ ^• o o o o O in ^ in O o vo in o iH I ^ o n 00 CN u n ^ U-J CO O U E re :: W en OJ Q) -H 00 -H >- Q 00 r~ «T • o n iH ^ ^• • o o CN o >» n o • • o vo n o m vo ^ ro • o o\ CT> VO CN • o o iH CN CM • O VO en vo i-t • o vo in fH i-l • O •-I o\ in o o • o o o o o • CO H VO o o m 1 n vo O o o a vo n o o 1 1 1 1 CN •H • n ov r-4 • • • CM II U "91 (30 lU •H. C re I- CN VO cn -H o> «r m m in in o m n ^1 r>- CN ^ CN VO TT CN CN CN rH 00 o o CN o vo o 00 o o ^ OV VO ^ O O n 1-1 00 CO CO in m o o o o o I I O CN CM fH CN o o I I VO CN 00 CM vo CM CO «c o c c CO E 3 ij cn ^j — o (0 oo -H 01 • W > O o o 1 1 1 1 n m vo in —1 vo a\ CN CO o\ r~ •-t o o m o o t n •» <9 o ^ m ro o r- l-H in »»• lu, r> r~ in CM vo o^ CN in CN • o o CN • o o .-i • o u> o^ CO \o CD (N »— 1— o CN fH . ««• o o O o o 1 1 1 vo • 1 • • r-i n CN o o CM fvl CN o o r^ m lo o in f-H i-( o n r» r— VD CO vo a> m n 04 •-( t-( I— I— CN 00 ro a\ CO vo t-i n CO o i-< fH r-t o\ o\ IN ^> n en ^• in •H lO O O o vc o »-H co ^r fN •J" 00 m 1 II u 00 00 01 •H C 10 -H vo CN in o\ vo CN •T o o n o 1 1 1 »— i-H ^• o o\ CN CN vo in vo vo r» V m CN in CN vo CN o\ »*> i-t i-H o o o 1 1 1 VO 00 O CN vo 00 O •-< <-l r-t r-( m r» o o ^• CN o o o o o o n o T CN o m CN 00 m •H <-i CN r~ •H o o o O o CN CN «»• CN VO CN CO CN o o o o o n r~ n o o ^ en CN CD rH in to iH iH fH o O O r~ rcn n CN 'T m m •a- r- "O- o O in CN O 4-1 CO CD 1 <u u (U c n o o CN O o O CN O **^ y-K tn 00 •H >H 0) < o n ^-^ .-1 n •H 0) rH ^ VM 4.1 o to H c c o a. « (U CO 0) ^3 W u > u , c 01 c n o m i~ ^ n fH o ^ O n CO n o 1 1 1 O vo in in o 1 a. o rH CN VO T iH fH m o m o m vo ^' CN VO r00 fH O o O o o 1 1 1 1 1 in CO cn vo in in 1 •a- fH «N in CO r- o o o ^• .H r- o O o ' n m 1 Ed <T> B 3 W in in cn 00 CO n 1- -H • • (0 cio T-i LU 0) PH Q CN VO CO CN CN CO o\ fH <Tl ^ ^f CN CO in iH CN CN fH • • • in o o o o O o o O m o o 1 1 1 1 1 1 1 1 • r»- CN o o o o 1 1 • o o o o o o o o o o o V CO m o o o CN «N • in CN • O o • o o * n o o o> • rH fH • O tvi II JJ ^00 (U c •H -H (0 rH W (30 r» vo vo vo in vo vo in • • iH VO P~ o o ^ o 1 1 1 o o CN O • vo ^ cn n o 1 • VO o 00 o ov n CN o n n CO CN in CN a\ fH rH • • • • n o o 1 1 CO o O O CO r~ CO ^ • • o o 1 tn •H O O • vo n ro o • o vo o ov in n n vo fH fH rH • • «• • o O O VO CN CO CN n 1 CN ^ CN o r^ CM 'T CO r-t 1— •-i m a. U- CD o j-i CO E = oo ra — 0) W Q >< m o CO <N 00 • o O 1 1 r- «N o\ rn «-< • • o o o o o • O fM 0\ r~ o o • r- ID *r ro V m o o • o o m o o o o o I 1 1 • 1 CM o o o f-1 • vo m o o o o n o o CO o o o CN m o o • 1 CM in i-H »— in "• (N 5 o '^ i-i o • in o 1 fN r» o r^ vo -< O II r>- l-t ts «N o o o o o V o o 1 1 1 i-H o o fM in -I • (T, in -< O CO <N o\ tN -H O o» r» O O CO o o o O o m o o in T T O o o o o o 1 1 1 1 1 • • • «o r~ in • CN CD I II i-i c o •H d to 1-t <u a o ^ i-H CM «N r^ iH o o I I r~. in vo ri i-( m o V o o o m CM o o o o o I 1 1 1 ,-( o i-t vo CM \o ^ r» o • • in r~ • r> \o o o o • f«- tiH o o • o\ in CM o o • ^ «H n o o • in ^« ro f o O o • O o cn 0) c o o o a. ea ^«» ^ < uX to C>l o o 00 o CO o CM CM CM ^ VO CM CD CM u to 1 •a- >*• o 1 (U r-i ^ IW O ts CO H ca ^ 3 •o fH to > <u :% JJ o c c o (U .-H O Cb u >> o ^ 0V fH CO ^• ^ 1-1 1-1 iH o o O ^ CO o o 1 1 1 1 in in C3 "fll o^ iH iH tN r* -H m CO rH CO r-i VO CM VO "B" n O O o o r* 00 CM o o <-i r~ iH O o VO CO o o o o m o o o o o o o o lO o o o o iH iH O O O O o o o in ov o o o iH r» CO iH O O 1 ' .H OJ Q to (U >-" IW CM o E 3 CO tn vj to OJ to >< 4J •H 00 •H a o o ^• o o o O O o o ro O CM O o "«• . iH tN O O VO o CM o o o 00 1-1 o o r*' VO m r* ^• o o o o o o o o o o CM OV o o o 1 1 o\ o\ l-t o CM in o o o VO o o 1 1 1 VO in CM n o o n o o 1 1-1 0V It u o 000) to -H ov CM CM O 1 00 o o CM "J" r*- ^1 ^T CO VO ^^ iH l-t ^ o o o o o 1 1 o VO PO 1-1 r» ^• CM o ov ^1 m CO VO o 00 o o o 1 o o 1 CO CO CN o o n VO 1-1 o o 1 CO VO o o o 1 o o o o o CM 1* o o o o VO o o o o o o m o o o 1 1 o CM • o tN CM T CM VO CM CO CM o ri m n •-H IT) t-i 1-^ •— O o vo c G C o c \D o o fN m o o o o o o o o o o 1 1 1 1 1 1 ^ 1 00 u-i en O 4J tn tj -^ y: 1 x: oc 0) ov On ON •H C CO -H w n on 1 • 1 • 1 • 1 • o o —t CN in o o >< -w ON o\ o o o ^ o o r~ "T vo o o M Q E « 3 01 vo t~ r^ o ON o o o> Ol vo • ON •-) vo vo •H • • o o <r 0\ in r-l ,-i i-t r-t o o o o O o I I I I I <*1 f-H Oy CT> in in O o o o o 1 1 1 in 00 m «o 00 CN 00 CM f— o VD .H 00 00 »r 00 • O o o Iin o o o • 4> o o V d o o o o I I I 00 VD CSl . O o Ol in i-i n o o 1-1 i-H vo <N • • O o vo tN r-i vc ^' o o CM m o n m o 00 o CO en LD o o .H VD n o o • 00 CN r~ (N tn r- ON Ol 00 o o o o r~ in ^T o o • r>- vo o ^H • o o f-H rN in in VD o ^ o o p» On O o o o m n O iH r^ CN .H CM vo CN 00 CN O CM m o o o p~ ^ o o n .-( i-H in VO o o r>- CO CO CN 00 ren vo vo ^" n CO O o o n o o vo rS ON CN CN •« CN O iH ON o «»• iH CD ID U >-^ 0) c CO ^.^ /-v < 1 m n « H ^O 00 vo o CO o O o 1-1 cs iH CN o. n (U a Nw^ .-H U-l H xj ra 0) —1 XI o o •H o 3 c 01 c o i-l o. 0] 0) ffl > X O ^ •H CN I-l m o vo ^ O o o ^ m ^ o 1 1 1 iH r^ C n c Q a. X >> 1 ) ON vw CO O tn XJ u B D TO 01 -^ OO •H W >• Q m •-I .-( r«- CO oo CN O o o o CJN p«- vo r~ p>. - CO "T rt •H O n ON o o 1 1 -H n o o r*' 1 vo n m o o 1 m in vo o o o o o I I I 1-1 in ON ^ "O- r~ d o o 1 vo o o o o o O o I 1 I I in CN O O o o o o o 1 1 r-l CN o ^ O o I O o O O ^• vo »H O O ON ON o o O o o o o n u JJ ^i- in in ^ o _f^ CN CN CN in ON »H • • oo •H re U 0) C -H -H jj &0 m r» vo vo ^ o O O o 1 1 1 1 o o CN o vo o r~ ON CO o m ON 00 in VD vo CN ^^ CN • • • o n -H ^ O o o 1 1 1 n i-t • 00 o • • o o o o oo o o o o o o o o • o CN ON ON •H • CT> • o o V ro o • O o VO CN CO CN m rin • • O S'SOk c Of+5 MIT LIBRflRlES 3 TDflQ QD3 Dt,2 Bbfl j^CLr C^^ ^-5 Oy\ \?0XLk~ c^^^r~