Digitized by the Internet Archive in 2011 with funding from Boston Library Consortium IVIember Libraries http://www.archive.org/details/voluntarydisclosOOfarr i^ working paper department r of economics Voluntary- Disclosure: Robustness of the Unraveling Result, and Comments on Its Importance Joseph Farrell Number 374 May 1985 massachusetts institute of technology 50 memorial drive Cambridge, mass. 02139 Voluntary Disclosure: Robustness of the Unraveling Result, and Comments on Its Importance Joseph Farrell Number 374 May 1985 TN 84-407.1 Voluntary Disclosure: Robustness of the Unraveling Result, and Comments on Its Importance* by Joseph Farrell November 1984 Telecommunications Research Laboratory GTE LABORATORIES INCORPORATED 40 Sylvan Road Waltham, Massachusetts 02254 *To be published in the Proceedings of the U. C. Santa Cruz Conference on Regulation and Antitrust, R. Grieson, editor, Lexington Books 1985. TN 34-407.1 1 1. INTRODUCTION Used car dealers have alv/ays had the Federal Trade Commission In 1976, in their cars. poor reputation for revealing faults a (FTC) proposed a strong rule requiring inspection by dealers of each of 52 systems or components, and revelation of the results, together with estimated repair cost, and other disclosures. The National Automobile was rule Dealers Association (NADA) citing unreasonable, estimates about ten times what the FTC believed. May 1980 the FTC tentatively proposed would not be required, buyers. of claimed that such inspection cost of $200, NADA's protests were effective, and in a different rule, under which inspection but defects known to the dealer must be disclosed to Having apparently defeated the compulsory-inspection rule, the National Independent Automobile Dealers Association (NIADA) new rule, about a NADA and turned on the claiming that difficulties in knowing what a dealer had known would make this rule tantamount to the previous proposal. However, in August 1981 the FTC adopted the second rule. The rule did not immediately decreed that FTC rules take effect, because in 1980 should be subject to veto by both houses of Congress (acting together) within an uninterrupted 90 day period. FTC Improvement Congressional Congress had (This was called the Because the Act required an uninterrupted 90 Act.) extended the recess review period until May 1982, when houses of Congress voted to veto the "known-defects" disclosure rule. dentally, the rule was the first to come before Congress under the days, a both (Inci- 1980 Act.) Thus the rule seemed dead, especially after a move to attach a revised version of its provisions to a FTC funding bill was defeated in committee three months later. However, the courts had been active meanv;hile in the area of legislative vetoes. ruled a In January 1982 "one-house veto" unconstitutional, in a the U. S. Circuit Court of Appeals in Viashington had (either house of Congress could veto agency actions) case involving the Federal Energy Regulatory Commission rM 34-407.1 2 the same court extended that finding to tv7o-house vetoes, In October, (FERC). provision of the veto and in particular the FTC Improvement Act, 1980 case brought by Consumers Union and Public Citizen against Congress. of 1983, the Supreme Court upheld these on the grounds rulings, in a In July that vetoes violate the constitutional provision that everything must be presented to the President before becoming law: somewhat illogical argument, since the agen- rulings themselves are not normally presented to the President. cies' Thus the a the known-defect rule appeared to have been reinstated. FTC then decided to take another look at the rule, However, and in July 1984 the five-member Commission voted 3-2 to abolish it, citing difficulty of enforceSo ended ment. tortuous history of one of the FTC's most controversial the rules In this paper, efficiency? A we How much difference does this make ask: conclusion tentative rather little difference. surprising - to me - is to that it The reader can judge whether or not this noticeably supported by the analysis below, which is the result economic makes view is of thinking about disclosure in general rather than the result of trying to analyze this case in particular. The principal existing model dently) to Grossman (1981) and Hilgrom (1981). tion 2. disclosure of voluntary is due (indepen- It is summarized below in Sec- The "unraveling result" claims that voluntary disclosure will lead to the same results as compulsory disclosure, and therefore it makes no real dif- ference whether disclosure is required or not. Plainly, for the passion brought something is missing in that model, to bear in the 1980-1982 lobbying battle testifies that, whatever the results may be, made it does matter. by Grossman A natural candidate and Hilgrom that fully informed about their products. a model in Farrell and Sobel (1983) sellers informed. Although \-ie it is for modification is common knowledge In section 3, to I the that assumption sellers are use a simplification of examine the effects of having not all reproduce the continuity result of Farrell and TN 34-407.1 3 we also Sobel, sense I see in this version that continuity may be misleading, in a discuss. In Section 4, requirements will actually increase In Section the 5, models I try just to evaluate amount I the of information acquired and allocative gains discussed lack welfare there clearly are such benefits. examples, the The result is ambiguous. disclosed. Although use the Farrell-Sobel model to examine whether disclosure I benefits from information. from information, However, by considering reasonable numerical show that the benefits of simple allocative efficiency may be sur- prisingly small. IN 34-407 . 4 THE BASIC MODEL: A BASIS FOR DEPARTURE. 2. For completeness, and to introduce notation, will I now describe the model discussed by Grossman and Milgrom. Each of a large number of sellers has available for sale one item, which known to the seller but not to buyers. concerning lies are assumed to be impossible, Buyers are concerned with expected value, and compete or adequately deterred. to buy, The seller may make any true statement and buyers will believe it: q, is initially Its value to each of many buyers, q, not value himself. he does that the price of an item certified to be of quality q will be q, so while the price of an incompletely certified item will be the expectation of In q. forming expectations, buyers know the prior which is represented by q, F(») on [0, a buyers also 1]: distribution of (overall) continuously dif ferentiable distribution function take into account any equilibrium effects (see below) For a full and careful treatment of the model, see Milgrom (1981). I simply give the unraveling result and Suppose that about q? seller refuses a Clearly, to a Here, heuristic justification. disclose q. What should buyers infer they should not infer that q is at the top of the range - then lower q's would follow that concealment strategy). (for if they did so, But then buyers' beliefs have to be such that if q were in fact at the top of the range, then the seller would rather reveal q. argument to the range remaining after the top q's drop out A were p. little Then, more if q formally, > p, a items are of quality q < p. suppose the price for we Ne.xt, a seller would rather label, . . apply the . and so on. wholly unlabeled so same item that all unlabeled Since p must equal the average quality of unla- beled items, this implies that none have quality strictly less than p: but the only way this can happen in equilibrium is for p to be zero. Thus we have: PROPOSITION 2.1 (GROSSMAN, MILGROM) 2.1 With the assumptions discussed above, buyers are as pessimistic as possible that is, if q is not revealed, or is only partly revealed, - that q has the result, lov;est value buyers infer compatible with whatever has been revealed. in equilibrium buyers know the value of q for each item, As a and the allo- cation is as if revelation were compulsory. In Sections 2.1 of 3 and 4, we will briefly discuss allowing for incompletely informed sellers. follow, which we will not do here, lies are in effect impossible. the impact Another on Proposition natural path to is to relax the very strong assumption that M o4-4G7 3. . ALLOWING FOR UNINFORMED SELLERS: A CONTINUITY RESULT AND A CONTRARY RESULT this In (1983) section, briefly we describe model the Farrell of This enables us to it by making the fraction G of informed sellers exogenous. 1: a of unlabeled items Sobel then we simplify which allows for costs of sellers becoming informed; show that the equilibrium price p(G) and continuous at G = is result (due to Farrell and Sobel) which seems to suggest that the Gross- man/Milgrom assumption (G = 1) is reasonable as an approximation. then show that the derivative p'(G) is argue that this suggests Initially, However, F(»). item, without a (negatively) seller infinite at G = can, by paying a cost this being observed by buyers. q, and a and , I particular seller's c have prior beliefs like buyers, c, learn the quality This information cost tributed according to the distribution function G(»); pendently of 1 certain non-robustness. sellers do not know q: they, a However, we c is is not observable c'^, such that rent between buying and not buying information. We a is dis- distributed indeto buyers. Equilibrium will now be characterized by two related variables: for unlabeled items, and a cutoff value c of his a price p c^-seller is indiffe- consider only "unlabeled" and fully labeled items, because partial revelation of q would reveal that the seller was informed, and thus Grossman - Milgrom would apply to him. Informed sellers who conceal q are riding on the coattails of uninformed (therefore, on average, of average quality) sellers. We have two equations for p and c*. First, the seller's ex-ante decision problem yields as an equilibrium condition: P = Second, - p c* + must /J be max(p, q) dF(q) the average (3.1) quality of unlabeled items. This includes the genuinely uninformed, and the informed who found q < p: average TN O4-407.1 7 p ^ [1 - G (c-)] /J G(c-)] - [1 q dF(q) + G(c-)/P q dF(q) + ^3_2) G(c-) F(p) Existence and properties of solutions to (3.1, For and Sobel. purposes of the this are 3.2) we section, discussed in Farrell simplify by assuming that G(») has a very special form: G(c) =GifO<c<l = 1 if c = (3.3) 1 Thus a fraction G of sellers have zero information costs be informed) while , never be informed. p[l - (1 - have large enough information costs that they will G) Then we have G + G F(p)] = a single equation determining p: G]q + G /P q dF(q) - [1 where q is the mean value, (and so will always (3.4) q dF(q) / Integrating by parts in (3.4) gives us H (p,G) E p (1 - G) + G /P F(q) dq - (1 - G) q - (3.5) The function H is dif f erentiable in p and G, and has positive partials in both variables. This implies that p is uniquely defined given G, and that p is continuous and monotone decreasing in G. ^= (q - q>OatG which becomes |S-= (1 /P F(q)dq P) + - G) + G F (3.6) l,p = = 0, 1 v;hile (3.7) (p) which becomes zero at G = However, we have , p = 0. Til S4-4C7 . 8 Thus, dp/dG = 3.1 - Summarizing, we have at G = 1. oo PROPOSITION: Defining p(G) by (3.4), the function p(») is uniquely defined, continuous However, at G = and monotone decreasing. 1 , p'(G) = -oo. The significance of the infinite derivative is the following. nuity result shows form of robustness of the unraveling result, in the sense a that if G is very close to 1 then p will be almost zero. Taylor expression of p(G) around G = p(G) = For values p(l) if G = close of G, = p(l) Plainly, - (3.9) (3.9) 1) gives a does if G = [1 for G - 1, however, 1] ought to take concerning whether we to deal v;ith cases in which G is "near" 1. 1 < we (3.9) different sense from (3.8) the uniform distribution on [0, G + G p] to 1. that p(G) = - oo for G < 1, any more than 1.) To close this section, we solve - close, This then gives us = not claim (3.8) claims that p(G) = p[l one-term 1: but not very should rely on the model with G = (Of course, a (3.8) (G - oo This is 1 another term in the Taylor series. P(G) The conti- a simple example explicitly. so that F G] i- + G -i-p^ (x) e x. Let F(«) be Then (3.4) becomes: (3.10) iW S4-407 . 9 Multiplying by 2 G and rearranging gives: G^ p^ + 2 G (1 whose solution p "^ p = G 0.3 P - ^-^791 so that a 0.23. [/(I G) p - - G (1 (1 - G)] G) = is > for instance, Thus, - - G) - if G = 0.91, 0.09 = 9% change in G, (3.11) then 0.21 „ ^, °-23 oTTi = from 1 to 0.91, produces a change from p = The lesson is that continuity results must be viewed with care. we ask about robustness, we may be concerned with more than continuity. to p = When TN S4-407.1 10 DO DISCLOSURE REQUIREMENTS INCREASE DISCLOSURE? 4. In this section, information more apply the Farrell-Sobel we emerges if would be this informed are more model to ask whether likely to obliged to be Given the amount of information acquired by sell- reveal their information. ers, the (19S3) true; however, the prospect of compulsory revelation We will show reduces the incentive to become informed, given the value of p. that the net results on information flow are ambiguous. In Farrell-Sobel model, the suppose that there is will be obliged to a described at the beginning of Section as probability z that a seller who becomes > reveal his information. the degree of enforcement of disclosure (We rules, can think perhaps, of z as 3, informed measuring although that inter- pretation might suggest that there vjould be more disclosure near p, and less near zero, than we will suppose.) Then equations (3.1) and (3.2) are modified as follows: (3.1) becomes p = -c* + zq + (1 - z)/ max(p, q) dF(q) (4.1) and (3.2) becomes p ^ 1 It is G(c*) + z G(c^)]q + - [1 - easy to G(c*) + see z that - (1 z) G(c*) 0(0^^) + (4.1) represents (1 - z) G(c'^) a /g q dF (q) ^^_2) F(p) downward-sloping curve in (c*, p) space, which shifts down and left with an increase in z. (4.2) also represents a downward sloping curve, but this curve moves up when z increases. c*.) (A larger z makes buyers less cynical about unlabeled items, given Thus we have two cases: 34--;07.1 [N 11 if (4.1) i) is steeper than then dp/dz (4.2), these effects tend to reduce and dc*/dz > information transmitted: the information is obtained by sellers,- and p rises, so less ment is more attractive. 0. < Both c* falls, so conceal- These effects work against the direct reve- lation-producing effect of the increase in z , overall the result so is ambiguous. if (4.2) ii) is steeper than both indirect effects, then dp/dz (4.1), as well as the < and dc*/dz > direct effect of z, 0. Then lead to more information being revealed. say whether Can we case (i) or answer that question, we examine form on [0, (1 - z) (ii) is (4.1) be to expected? In and (4.2) when F(») an attempt and G(») to are uni- Then (4.1) reduces to 1]. p^ c* = 2p + 1 - C^p^ + (1 - c* + zc*)(2p - (4.3) and (4.2) becomes (1 - z) - 1) = (4.4) The (absolute) slopes are given by (4.5: slope of (4.3) = 2p(l-z)-2 2-2p(l-z) slope of (4.4) = ^'.: "^^P^ ' ^P ^ ^ ^ 20^^(1 - z)p + 2 We can readily compare (4.5) is greater, (4.5) and ;/^ {1 - (4.6) z increases revelation, the incentive (4.6)' only when ^ z is close to 1: then which tells us we are in the ambiguous case (i). This ambiguity should not surprise us of ^ . c^ + zc^) to become : we know that the "direct effect" but the prospect of compulsory revelation reduces informed, and the less cynicism of buyers about unla- beled items makes it more tempting to conceal information. IN O4-407.1 12 5. WELFARE EFFECTS OF INFORMATION The alert reader may have noticed that, efficiency benefits of information. information is Farrell and Sobel (1983). go through. unknov.'n (as quality is the expected social Constructing models with little more complicated: a there are no The reason is that the social value measured by market price) of an item of value of an item of known quality. in the models above, a social value of see Burdett and Hortensen (1980) or However, the basic ideas described above will still In this section, we try to get a sense of the size of the alloca- tive efficiency benefits of information. Used cars differ in their probability of breakdown, their costs variation: of experiencing people's being far from a value service of a breakdown. time differs; station differs, Among the the and people differ in reasons probability for of the a latter breakdown and people vary in how well they cope with the consumer's nightmare of auto repair. The direct allocative efficiency effects of more information lie in get- ting the most reliable cars to those who most value reliability. To gain some notion of these benefits, we carry out an illustrative calculation. TN S4-407.1 13 5.1 PERCENTAGE SAVINGS FROM INFORMATION: DIRECT EFFECTS Suppose that cars differ in their probability p of having breakdown, a and suppose p is uniformly distributed (among used cars) on [.1, moreover that cost the drivers to c of such a between one hundred dollars and one hundred eighty dollars, suggested above. (One might expect those with the very for uniformly reasons the highest ^ Suppose .9]. breakdown varies < costs of breakdown to choose new cars, or to have their cars inspected: thus the upper tails will be truncated.) Then efficiency requires that the car with break- down probability p be assigned to the driver with cost c = 180 - (p - = 190 - 100 p. 100 .1) Such an arrangement will give total cost of breakdowns C* = Q . ^ o /? .1 - P(190 100 p) dp This "probability of breakdown" may be an aggregate figure, ferent breakdowns by seriousness, in the following sense. n different possible breakdov/ns, If buyer i. j has cost c and weighting dif- Suppose there are car has probability p. a of breakdown i, of breakdown then the cost of assigning this car to him is n p.c^ Z (A) Suppose that, for all buyers c^/c^ - 11 for some c., repair is this case, "/c k c j , k, and all i, (^) ' independent of twice k's, then the i. This says that, same will be (not likely to hold exactly, true if j's cost for for all other a certain repairs. In like most aggregation conditions, but :N S4-407.1 14 .8C* = 95 (.1)'] [(.9)' = 95 X - 'I' [(.9)^ (.1)^] - 33.3 X .728 .8 - = 76 - 24.26 = 51.74 Now suppose instead that the revelation system is imperfect, and all cars with p > indistinguishable to buyers. are .5 Thus those cars are matched randomly to the buyers in the lower half of the cost distribution. / p(190 - 100 p) dp + f't p = 95 [(.5)2 - (.1)2] .8C = + 60 = 95 X .24 [(.9)2 - - 7 • Now we have: 120 dp [(.5)2- (.1)2] (.5)2] 33.3 x .124 + 60 x .56 = 22.8 - 4.1 + 33.6 a reasonable central case) we can write (A) as c. p. Z i=l c. and write n p I . c . 1 '^i i=l as p (for this particular car). imizing I pc , Then the optimization problem reduces to min- where the sum is over cars or buyers (equivalently) TN 84-407. 15 = 52.3 This is only about 1.5% greater than the full information cost figure. If all cars are indistinguishable, we have random allocation, .SC" = 140 f^ and p dp = 70 X [.9^ .1^] - = 70 X .8 = 56 Thus the direct efficiency effects of information are rather modest: c" - 4.26 c* = 56 In other v/ords, alent of a 7 .076 as a fraction of no-information costs. even if information does not emerge at all, this is the equiv- 1/2% increase in breakdowns: serious, but not a terrible problem. We generalize this calculation by noting that E(pc) = cov (p, c) + E(p) E(c) and that, (5.1) while random allocation makes cov in this case makes cov (p, c) = - A'ar(p) (p, var(c). c) = 0, efficient allocation Hence the proportional sav- ings from efficient allocation are equal to % E(p) °c E(c) (5.2) the product of the "relative variations" o /E(p) and o /E(c). Note that (5.1) P is general, but (5.2) applies only when efficient allocation gives a correla*-• t:j 3-;--;o7.i 16 tion coefficient of (-1) between p and c. In general, perfect negative relationship between p and that this relationship is also linear, as but it is only in special cases c, required to obtain (5.2). We next derive a simple expression for ables are , a Thus, in information is less valuable than (5.2) suggests. general, b] efficiency requires uniformly distributed. where Suppose x (5.2) in the case where the vari- uniformly distributed on is [a, Then < a < b. a+b E(x) - b-a ^^^(^) = b?l 4 i''^^ = I2 ^^-^^' So the relative variation of x is 0^ ^1 (b-a) ^/3 a+b = E(x) ^ 2a a+b _1^^_ /3 If b = ma, where m > 1, this becomes » ^ . Thus, as one would expect, 1+m large dispersions - i.e., large values of m (for p and/or for c) give greater cost savings; but in no case can information save as much as a third of the costs. The percentage values of m are given below: 1.25 1.5 2 4 1.25 0.4 0.7 1.2 2.2 3 1.5 0.7 1.3 2.2 4 5.5 2 1.2 2.2 3.7 6.7 9.1 4 2.2 4 6.7 12.0 16 10 3 5.5 9.1 16 22 10 savings for some reasonable IN 84-407.1 17 The notable thing about this is that only for quite large values of m do the savings become "substantial" (say, in excess of Remember also that 10%). represents an extreme comparison: perfect information versus no informa- this saw we As tion. achieves quite a above, lot of if some of the potential the information savings emerges, from full it already information. Thus, these direct allocative efficiency benefits from information are surprisingly small. distributions have If the will be smaller. central tendency, the gains from information a For instance, suppose each distribution has uniformly distributed, v;ith (1 - being an atom at the mean. X) the same for the two distributions of its weight \ (of p and of c), Provided we still get a X is very simple formula for the proportional gain from information: ,2 X c p E(p) E(c) (With more general distributions, write dov;n the efficient allocation is provided all cars have positive value, body in the to efficient allocation. Thus the result is the fact each will be allocated to some- social value of knowing about repair whose cost is the same to every buyer is zero. to simple and manipulate.) The intuition behind this surprising (to many people) that, less a One's intuition tends start from the private incentive not to be the person landed with the bad car 5.2 INDIRECT EFFECTS V^hen ket, the this goods of different qualities are not duly distinguished by the marv;ill have used car market, feedback one v;ill effects on see this incentives to maintain quality. in two forms. First owners In of cars TM 34-407.1 18 intending to sell them used will not have enough incentive to maintain their cars. Second auto makers, who sell to new buyers most of whom (at least tra- ditionally) will sell the used car, will not have as much incentive as would be desirable to make the cars last longer than These effects may well be important, but I the tenure of first owners. do not know how to quantify them. IN £4-407 . 19 CONCLUSION 6. We tentatively view this episode as one in which market participants fought hard over rather doubtful distributional gains, and in which the direct efficiency stakes were smaller than one might think. of regulation suggest that questions close Interest-group theories to purely distributional will tend However, it is far from clear in this case that to produce a lot of conflict. all parties were behaving rationally. Economic does a poor theory participants. It may be Alternatively, favor, despite the fact the attitudes of market many market participants may have Sellers were almost uniformly against more been wrong in their prediction. enforcement. of predicting that theory's predictions of the outcomes under dif- ferent rules are mistaken. disclosure job equally uniformly in that prices measure value under all sets of rules, so Consumer representatives were that not all consumers would benefit from more market information. This raises an important methodological problem. entry and exit, and very low level of sunk costs, tate in writing down zero-profit equations industry with expected not rules, with each this free to entry and exit have firm would earn argument; views strong a perhaps, ease of few economists would hesi- for the used-car industry. Yet an ensuring zero profits would naturally be on the rules normal profit. it In view of the is that of the Evidently, important game: whatever something is wrong comparative advantages translate into rents which will be affected by changes in the rules. itself is not a very interesting observation: but the caveat against ral" jump to a false conclusion may be important. the This in a "natu- ilj 54-407 . 20 7. 1. Automotive News 2. Burdett, K. , , REFERENCES (1979-1984), passim and D. Hortensen, . (1980), "Testing for Ability in a Com- petitive Labor Market," Journal of Economic Theory. 3. Farrell, J., and J. Sobel, (1983), "Voluntary Disclosure," unpublished notes. 4. Grossman, S., (1981), "The Informational Role of Warranties and Private Disclosure about Product Quality," Journal of 5. Milgrom, P., (1981), "Good Nev;s Lav; and Bad News: and Applications," Bell Journal of Economics 12. 5bjSLv .. O and Economics 24. Representation Theorems Z-^-'"?^ ^i:v';;\V-;-/;,;,;:=j:;x.,t J, V,vr'?;7 >;.':.,,. K^"! ,:;,;;• ::-"?'«'?v Date Due Lib-26-67 MIT LIBRARIES DUPL 3 TDSD 00Sfl57M3 fl