Bargaining versus posted price competition in customer markets

International Journal of Industrial Organization 21 (2003) 223–251 www.elsevier.com / locate / econbase

Bargaining versus posted price competition in customer markets Timothy N. Cason a , *, Daniel Friedman b , Garrett H. Milam c a

Department of Economics, Krannert School of Management, Purdue University, West Lafayette, IN 47907 -1310, USA b Department of Economics, 217 Social Sciences I, University of California, Santa Cruz, CA 95064, USA c Department of Economics, Ryerson University, Jorgenson Hall, Toronto, ON, Canada M5 B 2 K3 Received 8 February 2001; received in revised form 16 October 2001; accepted 10 May 2002

Abstract We compare posted price and bilateral bargaining (or ‘haggle’) market institutions in 12 pairs of laboratory market sessions. Each session runs for 50–75 periods in a customer market environment, where buyers incur a cost to switch sellers. Costs evolve following a random walk process. Coasian and New Institutionalist traditions provide competing conjectures on relative market performance. We find that efficiency is lower, sellers price higher, and prices are stickier under haggle than under posted offer.  2002 Elsevier Science B.V. All rights reserved. JEL classification: D4; L13 Keywords: Posted offers; Experiment; Sticky prices; Market power

1. Introduction The ancient practice of price negotiation, or haggling, has largely been replaced in industrialized societies by sellers posting prices. Posted prices save time, a *Corresponding author. Tel.: 11-765-494-1737; fax: 11-765-494-9658. E-mail addresses: [email protected] (T.N. Cason), [email protected] (D. Friedman), [email protected] (G.H. Milam). 0167-7187 / 02 / $ – see front matter  2002 Elsevier Science B.V. All rights reserved. PII: S0167-7187( 02 )00056-5

224

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

tremendous advantage in retail markets—imagine having to negotiate individual prices for every item you purchased at the supermarket! However, haggling has not entirely disappeared and, beyond time costs, it is not clear which institution better promotes efficiency. Nor is it clear how the two institutions influence the division of surplus. Such questions become increasingly important as electronic commerce transforms transaction costs, including time costs for negotiations, which can be conducted now by ‘shop bots’ (e.g. Eisenberg, 2000). Laboratory comparison of market institutions has a history going back at least to Plott and Smith (1978). The laboratory is especially appropriate for our questions because it allows control of key features such as time costs that confound inferences from field data. The present paper reports a laboratory experiment comparing the efficiency, the surplus distribution and the price dynamics of two market institutions, the familiar seller posted offer institution and a new bilateral bargaining institution called ‘haggle.’ In the haggle institution, as in the continuous double auction, each buyer and seller can continuously improve her own offer or accept a counterparty offer at any moment during a trading period of known duration. In haggle, but not in the continuous double auction, each offer is extended only to a single counterparty, not to all possible counterparties. In the posted offer institution, sellers post a single take-it-or-leave-it price each period, which buyers either accept or reject. The laboratory environment we use is a ‘customer’ market, a term coined by Okun (1981). Each buyer is attached to a particular seller and must pay a switch cost to transact with another seller. Switch costs are important in labor markets (where they include search, relocation and retraining) and in many intermediate and wholesale markets, and are present even in retail markets (think of the shopper’s extra time cost in an unfamiliar supermarket). Customer markets are the natural habitat for haggle, since starting negotiations with a new counterparty normally entails some setup cost. The posted offer institution is also common in customer markets and has already been studied in the laboratory (e.g. Cason and Friedman, 2002). No formal models are available for comparing market institutions in customer markets,1 but it is not hard to find cogent arguments with conflicting predictions. A strong oral tradition rooted in Coase (1960) holds that parties will find ways to exhaust mutual gains unless artificial barriers are imposed (De Meza, 1998). The continuous bargaining protocol in haggle imposes fewer barriers than the unilateral, one-shot, take-it-or-leave-it offer allowed in the posted offer institution, so haggle should be more efficient. Moreover, haggle permits first degree price discrimination and thus should enhance efficiency while allowing sellers to capture a greater share of the surplus. New institutionalists might make the opposite predictions, based on incomplete 1 Formal comparisons in a static setting (i.e. no ongoing attachments or persistent cost shocks) include Julien et al. (2000) and Camera and Delacroix (2001).

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

225

information considerations. The posted offer institution is a solution to the ‘fundamental problem of commerce,’ the tendency of buyers and sellers to underreveal their willingness to transact in order to extract more favorable prices (e.g. Myerson and Satterthwaite, 1983; Milgrom and Roberts, 1992; McKelvey and Page, 2000). Posted offer constrains these tendencies more than haggle and hence should facilitate more efficient exchange. According to this view, sellers’ commitment technology in posted offer should allow them to extract a larger share of the surplus than in haggle, where constraints and bargaining opportunities are more symmetric. The two market institutions are of special interest from an historical or evolutionary perspective. Haggle spans the gap between pre-market gift exchange and early organized marketplaces. Historically it has tended to be displaced by more complicated or structured forms of exchange like posted offer (North, 1991), but it still holds a major share in modern economies (e.g. procurement, car dealers, and wholesale transactions) as well as in traditional economies like Morocco (Geertz et al., 1979). In the field, transactors’ time costs are higher for haggle than for posted offer. Additionally, the expansion of retail from single proprietor ‘Mom and Pop’ outlets into chain stores leads to agency problems that favor posted prices. These factors account for the shift into posted offer in modern economies, particularly for standardized, small ticket items. The use of new information technology such as the Internet and automated ‘price bots’ may overcome these obstacles. Indeed haggling is reemerging in e-commerce with such sites as www.BargainandHaggle.com and www.Priceline.com, perhaps because sellers believe it will allow them to extract a larger share of the surplus. In the laboratory we can remove agency problems and equalize the time cost in order to test such conjectures and thus better understand the forces driving the evolution of market institutions.2 The customer market environment also provides a non-trivial setting to study price dynamics. Do transaction prices track competitive equilibrium prices when sellers’ production costs change, or are transaction prices sticky? Early work by Scitovsky (1952) attributes price stickiness to switch costs based on a modified kinked demand model. This and later models do not clearly predict which market institution will have stickier prices. Do sellers price discriminate between attached customers and new customers? Theoretical work by Klemperer (1987) suggests that if sellers are unable to discriminate between attached vs. unattached customers, they will compete vigorously early on and will raise prices once they achieve a base of attached customers. Taylor (1999) examines subscription markets, such as banking and long-distance telephone service, and shows in a variety of settings that firms will offer lower prices to new customers. Although 2

Kirchsteiger et al. (2001) employ a clever alternative laboratory research strategy to investigate the evolution of market institutions, by allowing traders to select their own information conditions and matching rules.

226

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

our laboratory environment is not closely adapted to either of these theoretical analyses, it does provide evidence regarding the joint dynamics of price and attachment, with and without the ability to price discriminate. Laboratory researchers have studied the posted offer institution extensively; see Holt (1995) for a survey of work from Plott and Smith (1978) and Ketcham et al. (1984) up through the early 1990s. These studies confirm that the posted offer institution does allow sellers to extract a greater share of the surplus than do more symmetric trading institutions such as the continuous double auction. Efficiency also is lower in posted offer, but both effects tend to decline over time in the repetitive stationary environments used in most of these studies. It is not clear whether these results extend to customer market environments with switch costs and with changing seller costs each period.3 Laboratory studies of pairwise bargaining (e.g. Roth and Murnighan, 1982) also have a long history; surveyed in Roth (1995). Almost no studies embed continuous bilateral bargaining in a multilateral market context, as we do in the haggle institution. One exception is Hong and Plott (1982). Motivated by a proposal that barge operators post prices at the Interstate Commerce Commission, this experiment compared bilateral (telephone) negotiations with posted offer trading. The same 33 subjects (22 sellers and 11 buyers) used both institutions in four sessions. Hong and Plott find that relative to negotiation, price posting leads to collusive behavior including higher prices, lower volume, and reduced efficiency. Our results are different, not surprisingly given the many environmental and institutional differences, such as the number of traders, demand shifts instead of variations in costs, no switch cost instead of positive switch cost, and free-form verbal instead of computerized price-only negotiation. A series of papers by Davis and Holt (1994, 1997, 1998) is also relevant, since they take a step toward introducing negotiation in a standard posted offer, multilateral market context. Their posted offer markets featured a single round of structured negotiation. Buyers could ‘request’ a discount, and sellers could offer a single, private discounted price. As in haggle, this allows sellers to price discriminate. In order to limit buyers’ incentive to cycle through sellers repeatedly, the authors introduce a small (5 or 10-cent) switching cost. Davis and Holt find that the opportunity to offer secret discounts increased the variance of market outcomes (Davis and Holt, 1994), did not substantially improve market performance in a nonstationary environment (Davis and Holt, 1997), and led to the breakdown of explicit cartel agreements (Davis and Holt, 1998). 3 Cason and Friedman (2001) study a posted offer market with switch costs, but not in a customer market environment. Guided by the static theoretical model of Burdett and Judd (1983), their sellers have identical constant costs and buyers see iid random samples of posted prices. As the sample size and cost varies, the observed prices vary rather smoothly between the extreme predictions of Diamond (1971) (monopoly pricing) and Bertrand (1883) (marginal cost pricing), roughly in line with price dispersion predictions of the theoretical model.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

227

The next section describes our experiment, a balanced panel design with twelve matched pairs of sessions for the two trading institutions. Each session has two or three runs of 25 trading periods. Sessions differ by switch cost (zero, low or high) and trader experience. The following section presents the results. Posted offer proves to be (economically and statistically) significantly more efficient than haggle. Decomposition of efficiency losses shows that, consistent with the new institutionalist view, the dominant source of lower efficiency is low volume; that is, traders in haggle more frequently fail to complete mutually beneficial transactions. Market power, measured as seller markup over the competitive equilibrium price, also varies across the two institutions. Contrary to the new institutionalist view, markups are significantly higher in haggle than in posted offer. The data show consistently higher markups for attached customers than new customers, consistent with the Taylor and Klemperer models, and this effect is stronger in haggle in low switch cost sessions. Prices are also sticky, more so in haggle than in posted offer markets. Finally, we find that buyers switch to new sellers less frequently in haggle. The last section summarizes the results, offers some interpretations and conjectures, and suggests new avenues of investigation.

2. The experiment The main treatment variable is the market institution. In the posted offer treatment each seller enters a single take-it-or-leave-it price, and each buyer purchases at most one indivisible unit from one seller at that seller’s posted price. In the haggle treatment each seller posts a ‘list’ price that serves as a starting point for negotiations with individual buyers. Each negotiating buyer–seller pair sends binding price offers in a negotiation window. Only the two parties to the negotiation see the window. Either party can send a better offer at any time, and either party can accept the other’s offer at any time until the period ends. The message space is restricted to prices in dollars and cents. Each seller begins at his list price and cannot withdraw or increase his offer price to any attached buyer within the trading period, nor can any buyer reduce her bid price. Additional details can be found in Appendix A, which contains an abridged version of the haggle experiment instructions. A complete version of the instructions is posted on the journal web page. To create a customer market, we introduce attachments and switch costs. Each buyer begins each period attached to some seller, whose posted or list price she observes costlessly. In order to see other sellers’ prices or initiate negotiations with a new seller she must sink a switch cost C > 0. If she does so she observes the offer or list prices for all sellers (who are not further identified). In the posted offer treatment she can then accept any of these prices; in the haggle treatment she can initiate negotiations with any one of the sellers. She enters the next period attached

228

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

to the seller she purchased from most recently. Thus the switch decision is a more inclusive version of search. Switch cost C is constant across buyers and periods, and is publicly announced before the first period. It is varied across sessions at three levels: zero (a baseline for comparison), low (20 cents) and high (50 cents, over half of the median surplus in a transaction). Customer markets feature long term but impermanent attachments. To discourage permanent attachments, the experiment introduces random production cost innovations. In each session the production costs follow a random walk with a mean zero additive innovation to marginal production cost each period. In particular, seller j’s cost in period t evolves randomly according to c jt 5 c jt 21 1 e jt , where the innovations e jt are the sum of two components, one common to all sellers drawn from U(215, 15) and one independent component drawn from U(25, 5). The larger correlated component keeps sellers’ costs closer together in later periods. To sharpen comparisons across institutions and across switch costs, we used the same sequence of random shocks e jt for all sessions with the same experience and cost process conditions. The realized cost sequences are shown in figures posted on the journal web page. Table 1 summarizes the timelines for each trading period for the two market institutions. In the posted offer treatment the period is over after each buyer has either rejected all available price(s) or else accepted one of the posted prices she sees (that of ‘her seller’ if she does not switch, or all prices if she chose to switch). In the haggle treatment the period is over when the 60-s negotiation period ends.4 Table 1 Timelines for the each market period of the posted offer and haggle trading institutions Posted offer

Haggle

Sellers learn own production cost c

Sellers learn own production cost c

Sellers post prices

Sellers post list prices

Buyers observe own seller’s price

Buyers observe own seller’s list price

Buyers may choose to switch (at cost C)

Buyers initiate negotiations with own seller; may choose to switch to negotiate with another seller (at cost C) at any time

Buyers accept or reject seller offers

Through bilateral negotiation sellers and buyers may reach transaction terms

Median transaction price announced to all

Median transaction price announced to all

Buyers forecast upcoming median price

Buyers forecast upcoming median price

4 We ran pilot sessions with periods lasting 180, 120, 90, 60 and 45 s. Except in the 45-s pilots we saw little impact on trade volume; after the first few periods most trades occur in the last 30 s. Therefore in the experiment we let the negotiation period run for 90 or 120 s in the first few periods and 60 s thereafter.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

229

Then the sellers and buyers view interim screens that summarize their own activity in that period and their own cumulative profit, as well as the median transaction price (across all sellers) for that period.5 Buyers forecast the next period’s median transaction price, at the same time that sellers are posting their prices for the period. The buyer with smallest absolute forecast error totaled over all periods earns a modest prize ($5.00). Each session is divided into runs of 25 periods, and buyers and sellers are randomly re-matched to new trading partners at the beginning of each run. Each session has three sellers who produce to demand (i.e. no inventories) at a uniform (constant) marginal cost each period with no fixed cost. Each seller’s capacity is 3 units, so the maximum quantity that sellers can supply in total is 9 units per period. Five buyers each demand only 1 unit per period.6 Buyer values are constant over each session at 300, 275, 250, 225, and 200 cents. At the end of each 25 period run, assignments of buyers to these values are rotated and attachments are reinitialized randomly. Fig. 1 displays the supply and demand for an example period. Table 2 presents the experimental design, which employs 24 sessions. The design is balanced, with equal numbers of sessions in each trading institution and switch cost treatment. Identical software was used at all sites. Experienced subjects participated in an earlier session listed in the table, at the same site and in the same trading institution. We were able to complete more periods in the experienced sessions because the experienced sessions required a significantly shorter instructional period. Subjects typically earned between $20 and $30 per 2-h session (including instructions). Half of the data (the posted offer treatment) were reported previously in Cason and Friedman (2002). This earlier paper focused on other research issues, such as the market impact of information differences and changes in random cost sequences.

3. Results This section begins with an overview of the data, and then compares relative market efficiency, market power, and price stickiness. Our design allows for sharp matched pair hypothesis tests for many of the comparisons because seller costs follow identical paths in the two market institutions for each treatment. In testing 5 Cason and Friedman (2002) report sessions in the posted offer treatment with no feedback and with full (all transaction prices) feedback. Based their results, as well and Davis and Holt secret price discounts work mentioned earlier, we decided that reporting only the median price last period represents the most useful middle ground for our institutional comparison. 6 If buyers try to accept more units than a specific seller has offered, then in the posted offer treatment the buyers who transacted most recently with that seller receive priority. Any remaining units are awarded first come first served, and in the haggle treatment all units are always awarded first come first served.

230

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Fig. 1. Example supply and demand (period 22, experienced).

for differences across market institutions, we also match experience, switch cost, and experimenter (UCSC versus [Purdue or USC]). In testing for differences across any other treatment, we match market institution and the other treatments. To economize on journal space we have posted on the journal website some supplementary tables reporting additional data analysis details. Table 2 Sessions and runs by experience, switching cost and institutions Inexperienced subjects

Experienced subjects

Switching cost

Switching cost

Institution

$0.00

$0.20

$0.50

$0.00

$0.20

$0.50

Posted offer Haggle

232 232

232 232

232 232

233 233

233 233

233 233

Notes: A total of 24 sessions are reported. N 3 M within a cell indicates N sessions, each with M 25-period runs (The experienced posted offer sessions actually ran for four runs (100 periods), but here we report only the first three runs in order to be comparable to the haggle sessions). One session in each cell was conducted at the University of California at Santa Cruz, one session in each posted offer cell was conducted at the University of Southern California and one session in each haggle cell was conducted at Purdue University.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

231

Figs. 2 and 3 show prices in a matched pair of final runs, one posted offer and the other haggle, with 50 cent switch cost and experienced players. Open circles represent unsold offers or list prices and filled circles represent completed transactions. A horizontal bar in each period shows the competitive equilibrium (CE) price. Transaction prices stay at a fairly constant level in both cases, remaining well above CE prices. We see a sharper contrast across institutions in Figs. 4 and 5, depicting another matched pair, this time middle runs with 20 cent switch costs. The posted prices in Fig. 4 track the CE prices fairly well as seller costs rise and subsequently fall. The haggle transaction prices in Fig. 5, however, remain well above the CE price in all but a few periods (32 through 35). At the same time, trading volume falls below the CE level, as we document later.

3.1. Efficiency Inefficiency, the loss of potential gains from trade, has two main sources in most markets. Volume (V) or undertrading losses occur when a buyer and seller who could profitably transact in competitive equilibrium fail to do so. Extra-Marginal (EM) losses occur when a buyer or seller transacts who would not do so in competitive equilibrium, thereby displacing a more profitable trade (e.g. Cason and Friedman, 1996).7 Customer markets open a third source of inefficiency, the switching costs incurred by buyers.8 Table 3 reports mean efficiency losses by source for each session and treatment. Efficiency and efficiency losses are listed as percentages of potential surplus, so in competitive equilibrium efficiency would be 100 and all three losses would be 0. Efficiency is lower than in many other laboratory markets, ranging from 78 percent in an experienced high switch cost haggle session to 95 percent in an experienced zero switch cost posted offer session. As in previous experiments (e.g. Ketcham et al., 1984), average efficiency increases with experience in the posted offer institution with zero switch costs; with positive switch costs we find slight increases in three of four session pairs. Notably, average efficiency does not increase with experience in haggle for any switch costs. More importantly, with the exception of the inexperienced zero switch cost sessions, efficiency is always lower in haggle than in posted offer, usually 5–8 percentage points lower. The trading institutions have no consistent difference in switch costs or EM losses, but 7 Assigning inefficiency is ambiguous when in the same trading period an extramarginal unit has displaced an inframarginal unit (EM inefficiency) and also another inframarginal unit has failed to transact (V inefficiency); in these cases (relatively rare in the data) we cannot identify which inframarginal unit to assign to EM and which to V. Here, we resolve the ambiguity by using the average value of the untraded inframarginal units for each source. 8 Due to the random variation in seller costs, for the market to be fully efficient buyers need to incur switch costs from time to time. These ‘efficient switches,’ when they occur, help reduce EM and V inefficiency.

232 T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Fig. 2. Posted and transaction prices for posted offer session USC503cx (experienced, $0.50 search).

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251 233

Fig. 3. List and transaction prices for haggle session PU503hx (experienced, $0.50 search).

234 T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Fig. 4. Posted and transaction prices for posted offer session UC204cx (experienced, $0.20 search).

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251 235

Fig. 5. List and transaction prices for haggle session UC204hx (experienced, $0.20 search).

236

Switch cost

Institution

Zero

Posted offer Haggle

Low

Posted offer Haggle

High

Posted offer Haggle

Inexperienced

Experienced

Mean efficiency (%)

Switch cost losses (%)

Extra-marginal losses (%)

Volume losses (%)

Mean efficiency (%)

Switch cost losses (%)

Extra-marginal losses (%)

Volume losses (%)

84.1 (78.3, 89.8) 88.7 (90.9, 87.1)

0 (0, 0) 0 (0, 0)

6.1 (3.4, 9.0) 4.3 (2.6, 6.6)

9.8 (18.3, 1.2) 7.0 (6.5, 6.3)

94.2 (93.6, 94.9) 86.9 (89.2, 84.6)

0 (0, 0) 0 (0, 0)

4.0 (4.0, 4.0) 3.3 (3.6, 3.0)

1.7 (2.4, 1.1) 9.7 (7.1, 12.3)

86.7 (86.4, 87.0) 82.0 (79.5, 84.6)

7.5 (7.0, 8.0) 5.0 (6.6, 3.3)

2.8 (3.0, 2.7) 4.8 (4.9, 4.8)

2.9 (3.6, 2.3) 8.2 (9.0, 7.3)

86.3 (87.6, 85.1) 82.0 (82.2, 81.7)

4.2 (5.0, 3.3) 3.7 (5.4, 2.0)

6.6 (6.2, 7.0) 6.0 (5.1, 6.9)

2.8 (1.1, 4.6) 8.3 (7.2, 9.3)

83.6 (86.6, 81.2) 80.6 (81.4, 80.0)

6.5 (5.3, 7.5) 6.1 (3.6, 8.2)

4.3 (2.8, 5.6) 4.2 (5.8, 2.8)

5.5 (5.3, 5.7) 9.0 (9.2, 8.9)

85.4 (88.5, 82.4) 79.1 (77.7, 80.6)

8.4 (6.6, 10.1) 6.2 (7.2, 5.3)

4.5 (4.8, 4.2) 6.1 (8.1, 4.2)

1.7 (0.1, 3.3) 8.5 (7.1, 10.0)

Note: Values in parentheses are the UCSC session mean followed by the USC or Purdue University session mean.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Table 3 Inefficiency decomposition

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

237

there is a difference in average V losses, most striking in experienced sessions: 1–3 percent in posted offer versus 8–10 percent in haggle. These differences in efficiency across trading institutions are statistically significant, even based on a very conservative non-parametric signed rank test on paired session means. To ensure statistical independence, we used only one observation per session (the overall mean) for each (in)efficiency measure. Overall efficiency is lower in haggle for 11 of the 12 pairs of sessions, a significant result at the 5-percent level (Wilcoxon signed-rank test statistic512; two-tailed Pvalue50.034). Extra-Marginal inefficiency and switch inefficiency are not significantly different by trading institution, with Wilcoxon test P-values of 0.97 and 0.25, respectively. The source of the efficiency difference is volume inefficiency, which is greater in haggle for 11 of the 12 pairs of sessions (Wilcoxon signed-rank test statistic512; two-tailed P-value50.034). This confirms that the lower efficiency in haggle is mainly due to trading volume losses.9,10 Table 4 provides a parametric decomposition of the efficiency losses, based on a tobit model with separate observations for each period. The tobit specification accounts for the restriction that efficiencies are bounded above by 1 and inefficiencies are bounded below by 0. To account for the fact that a session’s outcomes are not independent across periods, we employ a random-effects error structure, with each session-run as a random effect. The point estimates for the haggle dummy indicate that efficiency is 5 percentage points lower than with posted prices and volume losses are 5 percentage points higher, while other losses

9 For readers who prefer parametric regression analysis to these nonparametric matched-pairs tests, we also report regressions that employ one observation per session for each market performance measure, using dummy variables for all treatments as explanatory variables. The results usually concur with the matched-pairs tests, although the statistical significance of the treatment dummy variables is often limited, due in part to the small number of degrees of freedom and low power in these regressions. For example, we estimated (Efficiency Measure)5a 1 b(Institution dummy)1c(20-cent switch cost dummy)1d(50-cent switch cost dummy)1e(experienced dummy)1f(USC site dummy)1 g(Purdue site dummy). We estimated separate equations for the four efficiency measures in Table 3, all using a tobit model to account for the bounds of 0 and 1. Although only marginally insignificant or significant, consistent with the matched-pairs tests, the estimated b coefficients indicate that volume inefficiency is higher (t51.38) and total efficiency is lower (t51.76) in haggle than in posted offer. Also consistent with the matched-pairs tests, switch inefficiency (t50.25) and extra-marginal inefficiency (t51.02) are not significantly different across trading institutions. 10 This failure by subjects to execute mutually-beneficial transactions is not observed in the well-known bargaining experiment reported in Hoffman and Spitzer (1985). Unlike the present experiment, this earlier study involved face-to-face, verbal negotiations, and it did not fix a time limit on the bargaining. Our bargaining protocol is more similar to McKelvey and Page’s (1997) recent experiment. Like us, they observe frequent bargaining breakdowns. Their experiment includes both complete information and private (payoff) information treatments, and breakdowns occur significantly more frequently with private information. Our experiment features private information.

238

Variable

(1) (2) (3) (4) (5) (6) (7) (8)

Intercept 51 if switch cost5$0.20 51 if switch cost5$0.50 51 if experienced subjects 51 if haggle Institution Observations Log-likelihood Restricted log-likelihood

Efficiency

Volume inefficiency

EM inefficiency

Switch inefficiency

Coefficient

S.E.

Coefficient

S.E.

Coefficient

S.E.

Coefficient

S.E.

0.92** 20.06* 20.08** 0.02 20.05* 1482 373.14 331.01

0.022 0.026 0.030 0.028 0.020

0.06** 20.01 20.009 20.18 0.05* 1482 1213.74 1144.97

0.019 0.020 0.018 0.020 0.017

0.04** 0.009 0.005 0.007 0.008 1482 1880.77 1812.30

0.008 0.012 0.013 0.012 0.010

0.01** 0.05 0.07* 20.05 20.009 1482 1528.54 1515.99

0.030 0.030 0.030 0.008 0.007

Note: ** denotes significantly different from zero at 1 percent; * denotes significantly different from zero at 5 percent. Estimates are based on a random-effects error structure u it 5 ti 1 eit , where the term ti reflects the session-run specific random effect.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Table 4 Tobit regressions for efficiency and inefficiency

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

239

differ insignificantly. These regression results also indicate that overall efficiency is 6–8 percent lower in the treatments with positive switch costs, attributable at least in part to greater switch inefficiency. The point estimate indicates 2 percent greater efficiency in experienced sessions, but this experience effect is not statistically significant.

3.2. Market power Next we examine relative market power, measured as mean markup of transaction prices over the competitive equilibrium price. The results are reported in Table 5. In both institutions, average markup increases with the switch cost and (perhaps surprisingly) with experience; the sole exception is the experienced, high switch cost sessions. In every treatment, average markup is higher in haggle than in posted offer, usually by at least 50 percent. Fig. 6 illustrates the higher average transaction prices in haggle for the high switch cost treatment. This figure plots the time series of the difference in mean transaction prices (haggle2posted offer) for each period, for both the inexperienced (panel A) and experienced (panel B) sessions. Our experimental design permits this direct period-by-period comparison, because we used the same random cost sequence in each session. Of the 50 inexperienced periods for this high switch cost treatment, in only one period (period 38) did the mean posted offer transaction price exceed the mean haggle transaction price. In the experienced sessions for this treatment, transaction prices appear about equal in periods 31–60; but prices are systematically higher in haggle

Table 5 Mean markup (transaction price2CE price) Switch cost Zero

Posted offer Haggle

Low

Posted offer Haggle

High

Posted offer Haggle

Inexperienced

Experienced

6.6 (1.6, 10.7) 11.7 (0.6, 23.1)

8.3 (9.7, 7.0) 12.6 (11.4, 13.9)

11.0 (19.1, 3.4) 17.4 (13.7, 21.0)

18.0 (12.7, 23.8) 22.6 (23.6, 21.6)

14.6 (8.8, 20.6) 25.6 (25.5, 25.6)

13.2 (11.9, 14.5) 24.7 (21.9, 27.8)

Note: Values in parentheses are the UCSC session mean followed by the USC or Purdue University session mean.

240

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Fig. 6. Panel A: Difference in mean transaction prices (haggle2posted offer) for inexperienced, high switch cost sessions. Panel B: Difference in mean transaction prices (haggle2posted offer) for experienced, high switch cost sessions.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

241

in earlier and later periods.11 Another paired signed rank test (again with each pair of matched sessions contributing one statistically independent observation) confirms that markups are significantly higher in haggle (Wilcoxon sum of signed ranks58, n512, P-value50.012), controlling for switch cost, experimenter, and experience levels.12 Table 6 provides parametric decomposition of seller markup, with correction for autocorrelated errors to account for possible price stickiness as discussed below. The dependent variable is mean transaction price minus competitive equilibrium price. Both switch cost dummy variables are highly significant, indicating markups roughly 12 cents higher for positive switch cost treatments. The fact that the interaction terms are insignificant suggests that the impact of switch costs does not decline significantly later in the run. The significant haggle dummy confirms that markups are higher in this institution than in posted offer, by an average of 7 cents. These higher observed markups coinciding with greater volume losses under haggle raise the additional question of whether sellers see a net gain in profits due to these higher markups. We examined seller profits using the paired test of signed ranks, with each session pair contributing one observation, and found that they are significantly higher in haggle (sum of signed ranks59, n512, P50.016).13 Table 6 Price markup regressions (dependent variable5mean transaction price2CE price (in cents))

(1) (2) (3) (4) (5) (6) (7) (8)

Variable

Coefficient

S.E.

Intercept 51 if switch cost5$0.20 51 if switch cost5$0.20* periods left 51 if switch cost5$0.50 51 if switch cost5$0.50* periods left 51 if experienced Periods left 51 if haggle institution

0.295 12.64** 20.18 11.84** 20.16 1.57 0.17* 7.27**

3.2 4.1 0.1 4.2 0.1 2.2 0.07 2.2

Observations: 732 R 2 : 0.643 Note: ** denotes significantly different from zero at 1 percent; * denotes significantly different from zero at 5 percent (all two-tailed tests). Estimates are corrected for autocorrelation. 11

Because of a run of positive common cost shocks, seller costs increase very rapidly in periods 26–38 of the experienced sessions. We will later see that transaction prices are ‘stickier’ in haggle, and this may account for the exceptional price comparison in periods 31–60. 12 An OLS regression of markups on the treatment dummy variables—analogous to the efficiency regressions described in Footnote 9—provides a similar conclusion. Markups are higher in haggle, although only at marginal significance levels for this test (t51.66; two-tailed P-value50.115). 13 The OLS regression of seller profits on treatment dummy variables indicates (insignificantly) higher profits in haggle (t51.37; two-tailed P-value50.190).

242

Switch cost

Inexperienced Previously unattached

Low

Posted offer Haggle

High

Posted offer Haggle

7 9.5 9 25

Experienced Attached for one or more periods

(Attached2 unattached)

Previously unattached

Attached for one or more periods

(Attached2 unattached)

11 17

4 7.5

14 16

17 24

3 6

13 26

4 1

7 16

13 22

6 6

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Table 7 Median markup by attachment length

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

243

Are haggle markups higher because the institution allows sellers to price discriminate between attached buyers and new customers? Table 7 shows median markup for new vs. previously attached buyer–seller pairs in positive switch cost sessions.14 It shows that new customers on average receive a discount in all relevant treatments, consistent with Taylor (1999). In low switch cost sessions the discount is larger in haggle than in posted offer, with inexperienced (7.5 vs. 4 cents) as well as experienced subjects (6 vs. 3 cents). In high cost sessions it is not clear where the discount is larger. Do markups for attached customers increase the longer they are attached? We looked at the data in various ways and found little evidence for a systematic relationship in either institution.

3.3. Sticky prices Cason and Friedman (2002) argue that price stickiness is best measured by comparing the median change in period-to-period transaction prices to the corresponding change in competitive equilibrium prices. To maintain comparability across the two trading institutions, here we look at transaction prices for each continuing buyer–seller pair. Prices are sticky in both trading institutions, since the median CE price change is always 7–8 cents while the median transaction price change is 3–6 cents and is 4–5 in most treatments. This stickiness is lowest with zero switch cost, and stickiness increases with switch cost in all instances except for low switch cost in haggle. Median transaction price changes are generally a bit smaller (and thus prices stickier) with haggle than with posted offer. To sharpen the comparison, we classified an individual seller as ‘sticky’ if her absolute cost changes exceed her absolute price change in at least half of the periods. Overall, for the posted offer market 24 of 36 sellers were ‘sticky’ vs. 34 of 36 under haggle, a very significant difference according to Fisher’s exact test (P,0.01).15 Although transaction prices vary less from period-to-period in haggle than in posted offer, the list prices posted in haggle exhibit approximately the same variability as offer prices in the posted offer sessions. For example, sellers’ median absolute change in list prices for haggle and in posted prices for posted offer are both 5 cents. In posted offer 18 of 32 individual sellers have a median absolute 14 Since in haggle sellers negotiate individual transaction prices with each buyer, classifying transactions into ‘new’ and ‘attached’ categories is straightforward. In posted offer, by contrast, a particular offer price might be accepted both by new and attached customers, since the institution does not permit price discrimination. In these cases the price contributes observations to both the new and attached categories, with the number of observations determined by the number of accepting buyers of each type. 15 The breakdown by switch costs is 7 / 12, 7 / 12, and 10 / 12 sticky sellers for posted price and 10 / 12, 12 / 12, and 12 / 12 for haggle for the zero, 20 cent, and 50 cent switch cost cases, respectively.

244

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

Table 8 Switch frequencies (percent) Treatment Zero switch cost

Posted offer Haggle

Low switch cost

Posted offer Haggle

High switch cost

Posted offer Haggle

Inexperienced

Experienced

56 (59, 52) 60 (75, 42)

57 (44, 70) 78 (97, 59)

18 (16, 19) 12 (16, 6)

10 (12, 7) 8 (11, 4)

7 (6, 7) 6 (4, 8)

8 (6, 10) 6 (7, 5)

Note: Values in parentheses are the UCSC session mean followed by the USC or Purdue University session mean.

change in posted prices of 5 cents or less, and in haggle 20 of 32 individual sellers have a median absolute change in list prices of 5 cents or less.

3.4. Buyer behavior Table 8 indicates that, not surprisingly, switch rates are lower in higher switch cost sessions. More interestingly, switch rates seem lower for haggle when switch costs are positive—for both inexperienced and experienced sessions and for both low and high switch costs. The difference in switch rates is greatest when inexperienced subjects face low switch costs. More formally, we form matched pairs of all 25-period runs with positive switch costs in the two trading institutions and test whether the distribution of switch rate differences is centered on zero. The test rejects the null hypothesis that switch rates are the same in the two institutions in favor of the alternative that switch rates are lower for haggle (Wilcoxon sum of signed ranks539, n520, P50.011).16 This lower observed switch rate is consistent with the earlier result that seller market power is greater under haggle.

16

A more conservative test that treats each pair of sessions (rather than each pair of runs) as the unit of observation marginally fails to reject this null hypothesis of no difference across institutions (P50.117). This could be attributed in part to the small sample size of only eight pairs of sessions with positive switch costs. Similarly, a tobit regression of each sessions’ switch rate on the treatment dummy variables does not indicate a significant difference in switch rates across trading institutions. This regression estimates six coefficients using only 16 session observations, and only the switch cost dummy variable is significantly different from zero.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

245

4. Discussion Our laboratory results comparing haggle (bilateral bargaining) markets with posted offer markets can be summarized as follows: 1. Haggle is less efficient at extracting potential gains from trade, mainly because mutually beneficial trades are more often missed (‘V losses’), and not because of inefficient customer search or extramarginal trades. This result is consistent with the new institutionalist view and inconsistent with the Coasian view. Unlike most other market institutions studied in the laboratory, efficiency does not increase with experience in haggle. 2. Sellers obtain a larger share of realized gains with haggle than with posted offer. This is contrary to the new institutionalist view that giving sellers the commitment technology in posted offer should increase their share of the gains. It is consistent with the view that haggle encourages classic market power with its deadweight (V) losses. 3. New buyers receive lower prices, consistent with the work of Taylor and (indirectly) Klemperer. In low switch cost sessions, the effect is stronger in haggle, as one might expect since haggle but not posted offer allows direct price discrimination. 4. Prices are sticky in both institutions, and especially in haggle. Sellers’ chosen prices dampen the fluctuations in their production costs, and reflect only about half of the resulting fluctuations in competitive equilibrium price. 5. In sessions with positive switch costs, buyers switch sellers less frequently with haggle than with posted offer. One might ask whether the V losses in haggle are due to insufficient time for negotiation. We do not think so because these losses are not reduced either by longer negotiation periods in early trials and in pilot experiments, or (as we have seen in the data reported above) by subjects’ experience. On the other hand, the V losses could be due to ‘deadline effects’ in bargaining (Roth et al., 1988). As in previous bilateral bargaining experiments, our subjects frequently reached agreement within the last few seconds of the trading period, no matter how long it is and no matter how experienced the subjects. Perhaps sometimes traders hold out for their counterparty to concede in the last few seconds and that concession does not materialize. We leave to another paper the real time dynamic analysis of haggling. After thinking about our results we noticed a possible explanation for many of the performance differences based on a subtle informational difference between the two market institutions. A switching buyer in posted offer sees his actual alternative opportunities because sellers are committed to their posted prices. By contrast, a switching buyer in haggle only sees a weak signal of alternative opportunities because list prices are not committed, and so the switch option is less

246

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

valuable. Therefore in haggle buyers are less inclined to switch and sellers are better able to exercise market power via larger markups and lower trading volume. Note that these institutional switch value differences are also present in posted offer and haggle markets in the field. For example, consider a procurement agent contemplating switching parts suppliers for an assembly plant. This agent could estimate different values of switching depending on whether she must negotiate terms with a new supplier, or whether she can simply accept or reject nonnegotiable seller offers. More experiments are necessary to test this explanation and other possible theories. The Coasian versus New Institutionalist theme was secondary in the present paper, but some investigators might wish to pursue it further by simplifying the laboratory environment (e.g. draw private costs and values independently each period rather than random walks, and eliminate attachments) and deriving explicit testable models. A different empirical project would be to interpolate new market institutions with information and negotiation procedures between haggle and posted offer, and study these new institutions in the laboratory. For example, subjects could bargain by following a strict two or three round alternating offers protocol. The impact of differing market information could also be studied, for example, by allowing switching buyers to observe sellers’ transaction price history and / or their current production costs. Our current design features a competitive equilibrium that tends to favor the buyers in terms of the division of surplus. This asymmetry may be behind some of the divergence from the CE results that we find. One interesting extension would be to remove or reverse this asymmetry and see if key results—such as the higher prices in haggle compared to posted offer—continue to hold. Another useful extension would allow sellers to choose which trading institution they wish to employ. Our results suggest that sellers’ profits suffer when they adopt posted offer rules, but as we discussed in the introduction there exist agency rationales for adopting this less flexible pricing institution. It would be useful to include such agency complications in an experiment that addresses the endogenous development of market institutions. It would also be valuable to include new innovations such as bargaining bots and Priceline.com style haggling that offer the opportunity for entirely new trading institutions.

Acknowledgements We thank the NSF for funding the work under grants SBR-9617917 and SBR-9709874; Brian Eaton for programming assistance, and Alessandra Cassar, Svetlana Pevnitskya and Sujoy Chakravarty for help in running the experimental sessions. We benefited from the comments of three anonymous referees, Flavio Menezes, John Morgan, Shyam Sunder, and participants at the March 2000 Tokyo Experimental Economics Conference, the September 2000 Tucson ESA workshop,

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

247

the October 2000 Rio de Janeiro LACEA Conference and the January 2001 New Orleans ASSA meetings. We retain responsibility for any errors.

Appendix A. Abridged instructions for haggle market A.1. General Each time for buying and selling is called a TRADING PERIOD and will usually last 3 min or less. At the start of each period, sellers POST PRICES, i.e. each seller enters a list price for his or her units. Some of these prices then are shown to each buyer as explained below. Each buyer decides whether to search for other prices, to accept the price he or she is shown, to negotiate a better price with the seller, or simply not buy this period. Then the trading period is over. A.2. Sources of profit Each buyer can purchase a single unit of the good each period. All buyers have a VALUE for the unit shown at all times in the STATUS window on their screen (see Fig. 7a) . . . . Different buyers generally have different values.

Fig. 7.

248

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

A buyer with value v who purchases a unit at price p earns PROFIT5v 2 p that period . . . . Buyers who don’t buy a unit automatically earn a profit of zero that period. Each seller can sell up to 3 units of the good each period. All sellers have a COST for each unit of the good in this experiment that is displayed at all times in the STATUS on their screen (See Fig. 7b). Always check the cost at the beginning of each period because it can change from period to period as explained [two paragraphs] below. Different sellers may have different costs. A seller with cost c who sells a unit at price p earns PROFIT5p 2 c on that unit. Since sellers can sell to different buyers at different prices, total profit for a period is equal to the sum of the profits from each unit sold that period . . . . Sellers who don’t sell any units automatically earn a profit of zero that period. A seller does not pay the cost of a unit unless she sells that unit. Therefore, profit is simply zero for any unsold units. Each period, each seller’s cost is randomly changed from the previous period by an amount between 2 e and 1 e. In today’s experiment, part of this random change is the same for all sellers, and part of this random change is specific to each individual seller. The e for today’s experiment is $0.20, but $0.15 of the volatility is common to all sellers, and the remaining $0.05 is specific to each seller: Each period the computer randomly draws a number from the list 20.15, 20.14, . . . , 20.01, 0, 0.01, 0.02, . . . , 0.14, 0.15, and adds this same number to the previous period’s cost for all sellers. Then, the computer randomly draws a number separately for each seller from the list 20.05, 20.04, . . . , 20.01, 0, 0.01, 0.02, . . . , 0.04, 0.05, and also adds this to the previous period’s cost separately for each seller. Thus, for example, a seller with a cost of $1.51 in period 8 would have a cost of $1.69 in period 9 if the number drawn from the first list is 0.14 and the number drawn from the second list is 0.04. Each number on both lists are equally likely, random and INDEPENDENT. This means that the draw does not depend on the actions of any buyer or seller, nor on the draws for the other sellers (or buyers) that period or in any other period. A.3. How to buy or sell Each buyer is the CUSTOMER of a specific seller. In the first period, the computer randomly assigns all buyers to a seller. Buyers and sellers will be randomly rematched at several points throughout the experiment. If you are a Seller, you begin each period by setting a LIST PRICE that is then shown to your customers. You set the list price in the STATUS window (see Fig. 7b) by adjusting the scroll bar until the price you want is shown and then clicking on the ‘Change List’ button. This price becomes the maximum price in the negotiation window (see Fig. 8) shown on your screen and on that of your customers. After all sellers have set list prices and all buyers have forecast, a price negotiation window opens on the screen of each buyer showing his or her seller’s

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

249

Fig. 8.

list price and a price scroll bar. A similar window opens on the screen of each seller for each buyer that sees her list price. Note that sellers can have multiple price negotiation windows open at the same time if they are negotiating with more than one buyer. Buyers and Sellers each have three options when negotiating prices: – Accept the most recent offer of their trading partner by clicking on the ‘Accept $x.xx’ button. – Send a counter offer to their trading partner by adjusting the price scroll bar and clicking on the ‘Counter $x.xx’ button. – Drop out of the negotiation by clicking on the ‘Quit’ button. Note: buyers who quit must search for a new seller in order to purchase in that period Buyers have the additional option of paying a search cost to seek a new trading partner by clicking on the ‘Search’ button in the status window. Buyers who search are shown the list prices of all sellers in the Search window (see Fig. 9). They then may select a new trading partner by highlighting a list price and

Fig. 9.

250

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

clicking ‘OK’. Once the buyer selects, a new price negotiation window appears on the screen of both the buyer and the chosen seller. Price negotiation begins from that seller’s list price. Note: Buyers must pay the search cost for each search whether they end up purchasing that period or not. A clock in the STATUS window shows the remaining time in the trading period. If time expires on a negotiation then there is no transaction and no profit on it for either the buyer or the seller. Note that if a buyer switches she incurs the cost whether she completes a negotiation or not. The status window also keeps a running total of your profits for the session. Note that in today’s session there are more buyers than any single seller has units for. Therefore, some sellers may ‘sell out’ of their units before time expires. If the seller that you (as a buyer) are negotiating with sells her last unit before you are done negotiating, then your negotiation window will immediately switch to the message, ‘Transaction not in progress,’ and you can no longer negotiate with this seller for the current period. You may, of course, pay the search fee to search for another seller to negotiate with. If you ever try to switch to a seller that has no more units remaining to be sold, you will receive a message that this seller has sold all of her units and you will be prevented from initiating negotiations with her. A.4. At the end of the period 1. At the end of each trading period, a window will appear giving you information about your transactions that period as well as a history for the previous four periods . . . . The amount in the cost column includes the cost of any searches.

References Bertrand, J., 1883. Theorie mathematique de la richesse sociale. Journal des Savants, 499–508. Burdett, K., Judd, K., 1983. Equilibrium price dispersion. Econometrica 50, 955–970. Camera, G., Delacroix, A., 2001. Bargaining Or Price Posting. Purdue University Working Paper. Cason, T., Friedman, D., 1996. Price formation in double auction markets. Journal of Economic Dynamics and Control 20, 1307–1337. Cason, T., Friedman, D., 2001. Buyer Search and Price Dispersion: A Laboratory Study. Purdue University Working Paper. Cason, T., Friedman, D., 2002. A Laboratory Study of Customer Markets, Advances in Economic Analysis & Policy, Vol. 2, No. 1, article 1, http: / / www.bepress.com / bejeap / advances / vol2 / iss1 / art1. Purdue University Working Paper. Coase, R., 1960. The problem of social cost. Journal of Law and Economics 3, 1–44. Davis, D., Holt, C., 1994. The effects of discounting opportunities in laboratory posted offer markets. Economics Letters 44, 249–253. Davis, D., Holt, C., 1997. Price rigidities and institutional variations in markets with posted prices. Economic Theory 9, 63–80.

T.N. Cason et al. / Int. J. Ind. Organ. 21 (2003) 223–251

251

Davis, D., Holt, C., 1998. Conspiracies and secret discounts in laboratory markets. The Economic Journal 108, 736–756. De Meza, D., 1998. The Coase theorem. In: Newman, P. (Ed.), The New Palgrave Dictionary of Economics and the Law. Macmillan, New York, pp. 270–281. Diamond, P., 1971. A model of price adjustment. Journal of Economic Theory 3, 156–168. Eisenberg, A., 2000, WHAT’S NEXT; In online auctions of the future, it’ll be bot vs. bot vs. bot, New York Times, Circuits, August 17. Geertz, C., Geertz, H., Rosen, L., 1979. Meaning and Order in Moroccan Society. Cambridge University Press, Cambridge. Hoffman, E., Spitzer, M., 1985. Entitlements, rights, and fairness—an experimental examination of subjects concepts of distributive justice. Journal of Legal Studies 14, 259–297. Holt, C., 1995. Industrial organization: a survey of laboratory research. In: Kagel, J., Roth, A. (Eds.), The Handbook of Experimental Economics. Princeton University Press, Princeton, NJ, pp. 349–443. Hong, J., Plott, C., 1982. Rate filing policies for inland water transportation: an experimental approach. Bell Journal of Economics 13, 1–19. Julien, B., Kennes, J., King, I., 2000. Matching Foundations. University of Auckland Working Paper. Ketcham, J., Smith, V., Williams, A., 1984. A comparison of posted-offer and double-auction pricing institutions. Review of Economic Studies 51, 595–614. Kirchsteiger, G., Niederle, M., Potters, J., 2001. Markets With Endogenous Matching Structures. University of Vienna Working Paper. Klemperer, P., 1987. Markets with consumer switching costs. Quarterly Journal of Economics 102, 375–394. McKelvey, R., Page, T., 1997. An Experimental Study of Private Information in the Coase Theorem. Caltech Working Paper 1018. McKelvey, R., Page, T., 2000. Status Quo Bias in Bargaining: An Extension of the Myerson Satterthwaite Theorem With an Application To the Coase Theorem. Caltech Working Paper 1035 (Revised). Milgrom, P., Roberts, J., 1992. Economics, Organizations, And Management. Prentice-Hall, Englewood Cliffs, NJ. Myerson, R.B., Satterthwaite, M., 1983. Efficient mechanisms for bilateral trading. Journal of Economic Theory 29, 265–281. North, D., 1991. Institutions. Journal of Economic Perspectives 5, 97–112. Okun, A., 1981. Prices and Quantities: A Macroeconomic Analysis. The Brookings Institution, Washington, DC. Plott, C., Smith, V., 1978. An experimental examination of two exchange institutions. Review of Economic Studies 45, 133–153. Roth, A., 1995. Bargaining experiments. In: Kagel, J., Roth, A. (Eds.), The Handbook of Experimental Economics. Princeton University Press, Princeton, NJ, pp. 253–348. Roth, A., Murnighan, K., 1982. The role of information in bargaining: an experimental study. Econometrica 50, 1123–1142. Roth, A., Murnighan, K., Schoumaker, F., 1988. The deadline effect in bargaining: some experimental evidence. American Economic Review 78, 806–823. Scitovsky, T., 1952. Welfare and Competition: The Economics of A Fully Employed Economy. Unwin University Books, London. Taylor, C., 1999. Supplier Surfing: Competition and Consumer Behavior in Subscription Markets. Duke University Working Paper.