Knowledge that Transforms

In his recent paper in this Review, Glenn Harrison (1989) argues that the conclusions of James Cox et al. (1982, 1983, 1985, 1988) in their studies of first-price private-value auctions are not well supported, because of shortcomings in the way their experimental investigations were designed, analyzed, and reported. Harrison argues that the expected cost of deviations from risk-neutral Nash equilibrium (RNNE) bidding in these auctions was quite small (less than $0.05 at the median), so that in terms of expected monetary payoffs (payoff space) many subjects had little to lose from deviating from the RNNE strategy. Harrison suggests that the significance of the differences Cox, Vernon Smith, and James Walker (hereafter CSW) report between subjects' bids and the RNNE bids (deviations in the message space) may therefore need to be reexamined. In discussing Harrison versus CSW we have three primary points to make.' First, in arguing that it is more natural to evaluate subject behavior in expected payoff space (Harrison, 1989 p. 749), we think Harrison has overstated his case. However, we agree with his more important point that looking at the cost of deviations is a useful diagnostic tool for determining when experimenters are likely to have lost control over subjects' incentives. Further, as we will show in Section I, this part of Harrison's critique applies with special force to CSW's studies of bidding. Second, a broader examination of the results of private-value auction experiments indicates that risk aversion cannot be the only factor and may well not be the most important factor behind bidding above the RNNE found so often in first-price privatevalue auctions. The most telling evidence here is bidding above the dominant bid price found in second-price auctions (Kagel et al., 1987; Kagel and Levin, 1990) and the risk-loving found under several treatment conditions in CSW's (1984) own multipleunit discriminative auctions (auctions in which the high bidders pay their bid price). These and other data inconsistent with risk-averse bidding are largely ignored in CSW (1988) but are nevertheless relevant to the substantive issue of risk aversion in private-value auctions. They are discussed in Section II. Third, there are data gathered in other investigations which provide strong support for the view that the deviations from RNNE bidding reported in first-price auctions are not the results of the low expected cost of such deviations. However, these data, unlike the higher-stakes payoff data that CSW offer in response to Harrison, are not consistent with CSW's subsidiary conclusions that the data can be well accounted for by a narrow class of risk-aversion parameters for the bidders, together with the assumption that all agents are playing a Nash equilibrium of the resulting game of incomplete information. A key difference between these experiments and CSW's is that if subjects do not respond to CSW's treatment condition (increasing the payoffs from experimental to U.S. dollars) their behavior will be consistent with CSW's theory. In contrast, * Department of Economics, University of Pittsburgh, Pittsburgh, PA 15260. We thank Jack Ochs and Emilie Roth for thoughtful discussions on earlier drafts of the paper, Jim Cox and Glenn Harrison for helpful comments on the initial draft of the paper, Susan Garvin for research assistance, and Ray Battalio, Carl Kogut, and Don Meyer for providing us with access to their data. Research support was provided by the Information Science and Technology and Economics divisions of the National Science Foundation, the Alfred P. Sloan Foundation, and the Russell Sage Foundation. The usual caveat applies with special force. IWe do not respond to specific comments that CSW (1992) make in response to our comment as, in order to avoid indefinite regress, the ground rules for this debate required us to comment on CSW's criticism of Harrison, after which they would be given the opportunity to respond to our comment.

Suppose that you must choose between lotteries S (safer) and R (riskier) shown in the top half of Figure 1. There are 100 possible states of the world and the consequence of S and R are indicated for each state. The states of the world with the consequence win are separate from the states with the consequence win $20,000. You may find S less appealing knowing that if you choose it and one of the states 71-100 occurs you would have received $20,000 had you picked R. Suppose that you must choose between lotteries S' and R' shown in the bottom half of Figure 1. Though the prospects (probability distribution of consequences) of S' and R' are identical to those of S and R, the states of the world with positive payoffs overlap. You may find S' more appealing knowing that if you choose it and one of the states 1-30 occurs you would have received $20,000 had you picked R', but S' pays an $8,000 consolation prize. A series of papers by Graham Loomes (1988a, b, 1989) and Chris Starmer and Robert Sugden (1989) provides evidence that in evaluating identical prospects the juxtaposition of consequences against different states of the world has a systematic effect on choice under as predicted by regret theory (David E. Bell, 1982; Peter C. Fishburn, 1982; Loomes and Sugden, 1982, 1987a). Regret effects violate expected utility theory and all other prospect-based theories of choice. All of the laboratory evidence of regret or juxtaposition effects, however, has been generated under the problem representation in Figure 1: the matrix of state-contingent consequences. While Loomes (1988b p. 468) argues that evidence of regret effects indicates that prospect-based theories ...may all be failing to capture an important element in decision making under uncertainty, this paper suggests that the experimental evidence of regret effects observed so far is specific to the matrix-problem representation. Nearly 400 subjects completed versions of two questionnaires in which choice problems were presented in different formats. The juxtaposition of consequences sways choices, but only under the matrix presentation. The first questionnaire indicates that regret effects are not observed when statecontingent consequences are described by ticket numbers, rather than a matrix. The second questionnaire indicates that if the matrix format is changed slightly to a simple proportional format subjects apparently compare prospects and ignore the juxtaposition of consequences. Regret theory predicts that choice depends on the juxtaposition of consequences, but regret effects should be invariant to different representations of state-contingent consequences.

The Choice of a Conservative Central Banker in a Multisector Economy

In Ordover, Saloner, and Salop (1990; hereafter OSS) we showed that a downstream duopolist may have an incentive to backward integrate in order to foreclose its downstream rival from a source of upstream supply. As a result of the vertical integration, the downstream rival's input price increases, giving the integrating firm a competitive advantage in the downstream market. Moreover, in equilibrium, the foreclosed downstream rival does not find it profitable to negate these effects by integrating itself. OSS considers a four-stage game involving two upstream firms, Ul and U2, and two downstream firms, Dl and D2. In the first stage, the downstream firms can bid to acquire Ul. If there is an acquisition, (say, Dl acquires Ul to form F1), upstream input prices are set in the second stage. In the third stage, knowing the input prices it faces, D2 can attempt to acquire U2. Finally, in the fourth stage, Dl and D2 compete 'a la Bertrand with differentiated products. In his comment, David Reiffen (1992) makes three distinct points. First, he notes that once Dl has acquired Ul, the equilibrium of the subsequent subgame depends critically on how upstream prices are set in the second stage. He argues that our results depend on the ability of Fl to commit to a high upstream price. This criticism previously has been made by Oliver Hart and Jean Tirole (1990). We show below that the results in OSS do not depend on the ability of Fl to commit. Instead, our main result stems from the fact that vertical integration changes the firm's incentives to engage in price-cutting in the input market. The notion that vertically integrated firms behave differently from unintegrated ones in supplying inputs to downstream rivals would strike a businessperson, if not an economist, as common sense.1 We show that there is theoretical merit to that common-sense view. Second, Reiffen argues that the game considered by OSS is similar to a game in which there is no vertical integration but, rather, where a nonintegrated Dl has a first-mover advantage in the downstream market. In such a game, Dl benefits from the ability to commit to the price that its Stackelberg leadership position gives it. Reiffen considers this to be additional evidence in support of the claim that OSS depends critically on Fl's ability to commit, not on vertical integration. We explain why the mechanism by which Dl is able to raise its profits in the sequential one-shot game considered by Reiffen is conceptually quite different from the mechanism by which Dl profits from vertical integration in OSS. Third, Reiffen argues that our results depend on there being only two upstream firms. This is correct in the symmetric Bertrand model analyzed. However, as we discuss below, our results do obtain as long as the upstream price is decreasing in the number of firms, as occurs in many oligopoly models, particularly when costs vary across firms. The next section summarizes the conceptual problems pertaining to upstream pricing. These problems are resolved in Section II by means of simple and natural price * Ordover: Department of Economics, New York University, 269 Mercer St., 7th Floor, New York, NY 10003; Saloner: Graduate School of Business, Stanford University, Stanford, CA 94305; Salop: Georgetown University Law Center, 600 New Jersey Avenue NW, Washington, DC 20001. We have benefited greatly from several discussions with Faruk Gul. Research support from the National Science Foundation (grant 8813943-IRI), the Sloan Foundation, and the C. V. Starr Center for Applied Economics at NYU is gratefully acknowledged. 'For example, the Japan-U.S. Strategic Impediments Initiative is predicated on the proposition that the loosely linked Japanese firms that form the various keiretsu do discriminate against nonaffiliated firms.