Can buyer “mobility” reduce aggregation failures in land-assembly?

Journal of Urban Economics 95 (2016) 16–30

Contents lists available at ScienceDirect

Journal of Urban Economics journal homepage: www.elsevier.com/locate/jue

Can buyer “mobility” reduce aggregation failures in land-assembly? R. Mark Isaac a, Carl Kitchens b,∗, Javier E. Portillo c a

Florida State University, 288 Bellamy Building, Tallahassee, FL 32306, USA Florida State University and NBER, 279 Bellamy Building, Tallahassee, FL 32306, USA c Florida State University, 039-A Bellamy Building, Tallahassee, FL 32306, USA b

a r t i c l e

i n f o

Article history: Received 19 January 2016 Revised 13 June 2016 Available online 23 June 2016 Classification codes: C9 D79 K11 R3 Keywords: Land assembly Urban sprawl Holdout problem

a b s t r a c t In this paper we examine whether site-development competition can be used to facilitate land assembly, in the absence of contingent contracts. In particular, we attempt to determine (1) whether competition can be induced among prospective sellers, (2) whether or not competition increases aggregation rates, and (3) what effects competition has on the distribution of surplus among the bargaining parties. We also study the incidence with which a buyer (endogenously) chooses to deal with a single “large parcel” owner vs. multiple “small parcel” owners. To do so, we make use of a laboratory experiment where all the relevant information about the project is common knowledge and landowner valuations are private information. Our results show that competition more than doubles aggregation rates, with aggregation rates of approximately 40% in the baseline, and at least 84% in the competitive treatments. We also find that developers have a strong preference to make transactions with landowners who have consolidated land holdings, doing so in 24/27 successful aggregations, providing empirical evidence that there is a link between the transactions cost associated with land-assembly and suburbanization, as suggested by Miceli and Sirmans (2007). © 2016 Elsevier Inc. All rights reserved.

1. Introduction Developers are often faced with consolidating disaggregated (and contiguous) parcels of land before real estate development begins. During the assembly process, landowners may behave strategically, such that Pareto improving developments are not completed (Munch, 1976; Posner, 1992; Miceli and Segerson, 2007; Miceli, 2011).1 While certain developers may be able to use ingenious methods to circumvent project failure, for instance, using a web of fronts, shell companies, and sub-holding companies, this is not always practical.2 In some cases, the development process occurs in the public domain, transmitting relevant details of the project to the targets of the development. Some projects must go through public hearings to acquire the necessary permits, or partially use

∗

Corresponding author. E-mail addresses: [email protected] (R.M. Isaac), [email protected] (C. Kitchens), [email protected] (J.E. Portillo). 1 This problem has also been referred to, in a broader sense, as an “anticommons” problem (Heller, 1998, Buchanan and Yoon, 20 0 0). It is meant to represent instances where resources are underutilized thanks to the existence of multiple rights to exclude. This situation is diametrically opposed to the more commonly known “tragedy of the commons” (Hardin, 1968). 2 A well-known example of such ingenuity is the assembly of land via shell corporations by Disney near Orlando, FL (Emerson, 2009). http://dx.doi.org/10.1016/j.jue.2016.06.001 0094-1190/© 2016 Elsevier Inc. All rights reserved.

public funds, making the details of the project public information (Kelly, 2006). Public information allows individuals to adjust their bargaining behavior in order to maximize their potential monetary gain, and likely exacerbates project failure. To mitigate the likelihood of project failure, scholars have suggested the use of contingent contracts or eminent domain.3 Both of these mechanisms have drawbacks. Experimental evidence has demonstrated that contingent contracts tend to shift surplus from the developer to property owners, such that, in expectation, the developer is just as well off using non-contingent contracts with high rates of project failure (Collins and Isaac, 2012). On the other hand, eminent domain may lead to excessive takings, as it reduces the price of public development (Epstein, 2001; Benson, 2005). In this paper, we explore another mechanism that may facilitate land assembly, site-development competition. Making use of laboratory experiments, we examine how competition (among sellers) affects the likelihood of successful aggregation for a Pareto improving project in the absence of contingent

3 Evidence for the benefits of contingent contracts has primarily come from the experimental literature. See for instance Swope et al. (2011), Collins and Isaac (2012), and Zillante et al. (2014). On the potentially correcting attributes of eminent domain use, see Seidenfeld (2008), Miceli (2011), Miceli and Segerson (2007), and Miceli and Sirmans (2007). On the potential inefficiencies of eminent domains, see Seidenfeld (2008), Lόpez and Clark (2013) and Miceli et al. (2008). For a discussion of the abuses associated with eminent domain, see Benson (2005, 2008).

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

contracts.4 Site-development competition is induced by varying the number of units needed by the buyer, relative to a given stock of available units.5 In particular, we attempt to determine (1) whether a competitive environment can be induced among prospective sellers, (2) how effective competition can be at increasing aggregation rates, and (3) what effects competition has on the distribution of surplus among the bargaining parties.6 We also study the incidence with which a buyer (endogenously) chooses to deal with a single “large parcel” owner vs. multiple “small parcel” owners. This investigation is done in an environment where all the relevant information about the project is common knowledge, while landowner values are private information. Our main findings suggest that site-development competition drastically increases the rate at which land is successfully aggregated. In our competitive treatments, the aggregate completion rate is always above 84%, relative to a baseline aggregation rate of approximately 40%. Reducing the number of units required by one is sufficient to observe the sharp increase in successful aggregations, as it eliminates the monopoly power held by the last seller and increases the number of possible successful combinations in a combinatorial fashion. This sharp increase in aggregation is not driven by timing effects. Additionally, competition increases the surplus retained by the buyer in expectation. In our competitive treatments, the buyer maintained at least 50% of the expected surplus, whereas in the baseline, the buyer retained less than 10%. This difference is driven by the differences in the aggregation rates between treatments. Conditional on aggregation, the retained surplus is similar between treatments. Given the high rate of aggregation and retained surplus in the competitive treatments, these results suggest that developers would likely prefer environments where they have multiple development options. Finally, when buyers are free to endogenously negotiate with preaggregated tracts or disaggregated parcels, they overwhelmingly make transactions (24/27 successful aggregations) to acquire the pre-aggregated tracts. This finding supports previous theoretical research by Miceli and Sirmans (2007), who combine a simple sequential aggregation game, highlighting the role of strategic delay, with results from the monocentric city model to predict that development will be biased toward the urban fringe. In our framework, there are no delay costs, however, we still demonstrate that there is a strong preference to transact with larger landowners. Thus, the transactions costs associated with multiple simultaneous negotiations alone may be sufficient to bias development toward the periphery. We believe that these results are highly policy relevant as cities seek to redevelop the urban core and seek to limit the externalities associated with urban sprawl. Our findings highlight several policy instruments that may be useful to mitigate the strategic incentives of landowners. First, our main result, that competition increases aggregation success, suggests that cities should seek to make more parcels available for development. In the short run, cities could re4 One concern with the use of a laboratory experiments is the external validity of the findings given that the design relies on undergraduate subjects making decisions over relatively small stakes. This criticism has been discussed at length in the experimental economics literature. Seminal in this discussion are Hong and Plott (1982) and Smith (1982). 5 Site-development competition should be distinguished from other “types” of competition; for example, competition among sellers to be the last seller, which would likely lead to failed aggregation. In this paper competition should be understood as the availability of multiple, or alternative, ways to achieve assembly (i.e. multiple sites for development). 6 Kominers and Weyl (2012) have theoretically discussed the role that competition may have in overcoming project failure. Miceli et al. (2008), Kominers and Weyl (2010) and Tanaka (2007) explore alternative mechanisms and schemes for improving the likelihood of successful aggregation. Seidenfeld (2008) has discussed limitations of commonly suggested techniques that tackle the problems faced in land assembly (e.g. secret purchases, options contracts).

17

duce the costs of re-zoning parcels, which will increase the number of potential properties suitable for commercial and industrial development.7 In the long run, cities could relax land use constraints. Or, even if they did not want to go this far, cities could attempt to create more contiguous zones, which may eliminate the need for developers to incur the transactions costs of re-zoning fragmented parcels with different zoning levels.8 To mitigate the development bias toward the outskirts of town, cities may consider, (i) the elimination or relaxation maximum lot size requirements, which may artificially increase the number of properties that must be aggregated for development (ii) consolidate adjacent foreclosed properties and make them available for future development. This policy measure has been followed in several metropolitan areas through the formation of nonprofit land banks. However, key to facilitating development, these land banks must be able to easily re-zone the assembled tracts, as their inability to do so may limit their long term success.9 2. Overview and relevant literature Problems associated with land aggregation have been studied both theoretically10 and experimentally.11 , 12 The experimental literature has provided important insights as to how the order of bargaining, type of contract, delay costs, number of sellers, and information affect project completion rates. Two of the most “robust” results found in this literature are that (i) project failure is consistently observed across experiments and (ii) the use of contingent contracts alleviates the problem. This paper investigates how competition among development sites may reduce project failure within different environments in the absence of contingent contracts. Previous studies have investigated how competition among property owners affects behavior. Our study is perhaps closest to Cadigan et al. (2011), Parente and Winn (2012), and Winn and McCarter (2014). Cadigan et al. (2011) study how increasing the number of required parcels affect proposals. They find that on average, increasing the number of required parcels tends to adversely affect the probability of successful aggregation, since transaction and strategic bargaining costs increase with the number of sellers. In one treatment they investigate the role that competition has on successful aggregation, whereby a buyer must acquire two of three available parcels. They found that competition led to increased aggregation rates and also increased the speed with which agreements occurred. Competition also tended to shift surplus from the sellers to the buyer. A second closely related paper by Parente and Winn (2012) studies the impact that the bargaining institution (i.e.

7 In our discussion, we limit the policy discussion to zoning, however, in general, it may include the cost and time delay associated with permitting, the cost of using local legal environment, or other land use restrictions discussed in Gyourko et al. (2008). Reducing these restrictions should facilitate development, for example, Suzuki (2013) demonstrates how more restrictive land use regulations increase the cost of commercial development in the hotel industry in Texas, leading to a reduction of entry in the market. 8 Other authors, such as McConnell and Walls (2009); McConnell et al. (2006) discuss the use of transferable development rights (TDRs) from the periphery to the urban core. 9 http://www.cuyahogalandbank.org/faq.php#zoningcodecompliance. 10 See Strange (1995), Menezes and Pitchford (20 04a, 20 04b), Miceli and Segerson (2007, 2012), Miceli et al. (2008), Miceli and Sirmans (2007), Shavell (2010), Miceli (2011), Sridhar and Mandyam (2013), and Lόpez & Clark (2013). 11 See Cadigan et al. (2009, 2011), Collins and Isaac (2012), Swope et al. (2011, 2014), Shupp et al. (2013), Parente and Winn (2012), and Kitchens and Roomets (2014). 12 Only but a handful of papers have studied strategic bargaining in land assembly using field data. See Brooks and Lutz (2013), Cunningham (2013), and Kitchens (2014).

18

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

simultaneous vs. sequential bargaining), willingness to pay signal (WTP), and input complementarity have on the negotiated prices and the likelihood of assembly failure when buyer values are not common knowledge. In a subset of their treatments, they reduce the number of inputs required for successful aggregation and find that assembly rates increase, regardless of the bargaining institution. Winn and McCarter (2014) study the roles that eminent domain and competition play in overcoming the assembly problem. In their “competition” treatment, a buyer needs to acquire two parcels out of two available to complete the project; they refer to these as the primary inputs of the project. Should the buyer fail to acquire either one, an outside option is available where the buyer can negotiate with a 3rd seller for his property; which if acquired, is equivalent to successfully aggregating the necessary land.13 Of relevance is their finding that introducing competition is equally effective at encouraging assembly as the threat of “eminent domain.” This will be similar in nature to one of our treatments, where we allow the buyer to endogenously negotiate with either the small, disaggregated, landowners or a single large property holder. While the previously described papers each explore competition to some extent, there are several key differences between our study and theirs. First, Cadigan et al. (2011), Parente and Winn’s (2012), and Winn and McCarter (2014) all use both contingent contracts and competition at the same time.14 In our design, contracts with sellers represent sunk costs, which allow us to isolate the treatment effect of increased competition without confounding it with another mechanism known to reduce assembly failure. This distinction may be critical, as Collins and Isaac (2012) demonstrated that contingent contracts effectively eliminate land assembly failures compared to a setting where investments are sunk. Second, our design allows for free-flow communication between the buyer and sellers, and among sellers whereas those aforementioned impose a rigid bargaining structure that use either alternate offer or ultimatum style bargaining. What effect communication is likely to have, if any, is ambiguous. Third, while Cadigan et al.’s (2011) experiment finds that negotiation with fewer sellers facilitates aggregation, they do not allow buyers to endogenously choose whether to bargain in places where land is fragmented or in places were land is consolidated, while our design does. Therefore, we believe that our design allows for a more analogous investigation of the hypothesis put forth by Miceli and Sirmans (2007), which suggests that the holdout problem biases development away from the city center, contributing to urban sprawl. Finally, our design makes the spatial component of land-assembly more salient than previous studies (see below). 3. Experimental design In what follows, we consider a scenario where a single buyer is charged with aggregating parcels of land owned by different individual sellers. Following the experimental design outlined in Collins and Isaac (2012), we allow communication between the buyer and sellers to facilitate informal negotiations, and among the sellers to facilitate information flows. However, a formal offer must be made by the buyer, and accepted by the seller, for it to 13 However, an interesting design difference is that Winn and McCarter (2014) incorporate heterogeneous parcel valuations for the buyer. In their competition treatment, as described above, if a buyer fails to acquire both primary parcels he can negotiate with a third seller. This third seller’s land may be used to complete the buyer’s project (as long an agreement is reached among buyer and seller), but it would be valued differently (lower) than the two primary parcels would have been, had they been acquired. Our design does not do this. 14 Parente and Winn (2012) implicitly impose contingent contracting by creating a decision rule whereby the automated buyer accepts proposals if the sum of seller demands is less than the value of the project, and rejects otherwise.

Fig. 1. Seller parcel arrangement in 2REQ, 3REQ, and 4REQ treatments.

be binding. We also follow their convention, whereby we allow sellers to sell below their stated value. Unlike Collins and Isaac (2012), who focus on the effect contingent and non-contingent contracts have on aggregation rates, our design manipulates the number of parcels that the buyer must aggregate. Therefore, we are effectively varying the number of possible parcel combinations that yield the aggregated property, which allows us to identify the effect that competition has on aggregation and the distribution of surplus. Furthermore, we only focus on non-contingent contracts where agreements represent sunk costs. We first describe the four treatments, and then present more details of the design below. In all treatments, parcels are placed in a 2 × 2 matrix as shown in Fig. 1. Each cell represents an individual seller’s parcel or unit, which is of value to the buyer. In the first treatment (4REQ), a buyer seeks to purchase all four units to successfully complete the project, which is a replication of Collins and Isaac’s capital constrained, public information treatment, and serves as the baseline for our experiment.15 In the second (2REQ) and third (3REQ) treatments, the buyer must purchase two and three out of four available units, where units must be adjacent to one other, yielding four possible aggregations in each treatment. For instance, in the 2REQ treatment, suppose the buyer initially purchases seller A’s parcel. He must then acquire either seller B’s or seller C’s parcel to successfully complete the project. He can no longer negotiate with seller D because acquisition of D would not yield a contiguously assembled property and the buyer would be in no better position than he was when he acquired A. In the 3REQ treatment the same adjacency restriction applies, yet it is not binding as in the 2REQ treatment. To see why this is the case, suppose a buyer purchases parcel A first. Unlike the 2REQ treatment, where parcel D would not yield a contiguous aggregation with parcel A, D could be combined with either parcel B or C to complete the project. Thus, no land becomes “unavailable” in this treatment, although the contiguous restriction is still implicitly present. Notice how in the 2REQ and 3REQ treatments, competition is always present. For expositional purposes, suppose that a buyer is in the 2REQ treatment, and has purchased parcel A. The buyer then makes an offer to seller C. C has an incentive to accept any offer above her value, as the buyer could also make an offer to seller B. If B were to accept, C would only receive her value. Thus, C is better off accepting any offer above her value. Assuming the buyer has not purchased any properties, all four sellers have an incentive to accept any price above their value. Similar analysis can be performed in the 3REQ treatment, where a buyer is not as restricted in his choices after the first purchase since all parcels remain available for negotiation. The fourth treatment (4v1) varies slightly from those previously described in that we introduce a 5th seller, who owns a plot of

15 We use the public information design because Collins and Isaac (2012) found that private information, with respect to the buyer’s valuation had little effect on aggregation rates empirically.

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

19

Fig. 2. Seller parcel arrangement in 4v1 treatment.

land whose area equals the aggregate area of the four disaggregated sellers (Fig. 2). In this treatment, the buyer is free to acquire the needed property either by aggregating the fragmented parcels from sellers A–D, or by purchasing the single parcel from seller E. In this setting, the competitive dynamic is now between multiple, small, parcel owners and the single, large, parcel owner. This treatment will allow us to study the hypothesis put forth by Miceli and Sirmans (2007) that failure to assemble is a contributing factor to urban sprawl. That is, assembling small properties has higher transactions costs relative to purchasing a single, large property, due to the number of contracts required, and the strategic incentives that disaggregated landowners have to be the last seller in order to capture the largest share of surplus. Negotiating with a single, large property owner, may reduce transactions costs, and potentially eliminates strategic behavior. We now describe the parameters of the experiment. First, seller values are randomly and independently drawn from a uniform distribution of integers between 100 and 300 cents. These serve as sellers’ initial valuations, which will also equal their payoff if they do not sell their property. Values are drawn for sellers A, B, C, and D for each of the four periods in the experiment and these are the same across all treatments. Seller E’s value for the 4v1 treatment is the summation of seller A–D’s values and this is common knowledge. This information is summarized in Panel A of Table 1. The buyer’s values are determined by adding the n-lowest seller values plus a buffer drawn independently from a uniform distribution of integers. The distribution for the 2REQ treatment ranges from 75 to 275 cents, in the 3REQ treatment, integers are drawn from 150 to 550 cents, and in the 4REQ (and 4v1) treatment they range from 300 to 1100 cents.16 We summarize the actual draws in Panels B–D of Table 1. As in Collins and Isaac, we impose a capital constraint on the buyer to create a natural environment whereby the distribution of seller surplus is unequal. The constraint is imposed to provide a strong incentive to sellers to be the last to sell, effectively encouraging strategic behavior during the assembly process. This constraint is operationalized by creating a starting capital fund that endows the buyer with enough money to buy the first n − 1 parcels in any of the four periods, where n is the number of required units. More specifically, we take the max of the sum of the lowest n – 1 seller values across treatments, and add 40 cents to that total. For example, in the 3REQ treatment, the sum of the lowest seller values in each period are {310, 329, 274, 434}. Thus the capital constraint is 434 + 40 = 474. In Panels B–D of Table 1, we report the sum of the lowest n-1 parcels in each treatment and each period, as well as the operationalized starting fund. After the necessary number of parcels are acquired, the constraint is relaxed,

16 The last distribution for 4REQ conforms to Collins and Isaac (2012). As experimenters we were faced with a choice between keeping this distribution fixed across treatments or not. If we did, the presence of a large surplus might lead to unsurprisingly higher aggregation rates in treatments where less than four properties were needed. Therefore, we halved the distribution from which the surplus was drawn in our baseline for 3REQ, and then halved it again for 2REQ.

and the buyer is able spend any amount up to his value, minus any outstanding and completed contracts. The capital constraint is akin to receiving a loan from a financial institution to complete the project once the probability of completion is high enough. Given the nature of the game, there are certain aggregations that generate more surplus for the buyer than others, should they occur. In Table 1, Panels B and C, we report the aggregations that would theoretically yield the maximum surplus for the buyer, and alternatively, the aggregation that yields the least surplus. These will serve as benchmarks in the empirical analysis. Throughout the experiment, the buyer’s project valuation, his capital constraint, his offers to other sellers, any accepted offers, as well the accepted amounts of completed contracts, are all common knowledge in all treatments. This environment resembles situations that are created by freedom-of-information and/or sunshine laws, for public agencies. By contrast, seller valuations are not common knowledge (i.e. they are private information and only known to each individual seller) but the distribution of their values is known. The payoffs subjects receive depend on their actions. A seller receives her offer (in cents) if accepted, or her initial valuation if no offer is accepted. This allows sellers to not sell, if they are not inclined to do so. A buyer’s payoff depends on whether the project is completed, and how much is spent in the process to acquire parcels. If a buyer completes the project, he earns his value minus all accepted offers. If a buyer fails to assemble the necessary parcels, his payoff is equal to his capital constraint minus any accepted offers. This should provide buyers with the proper incentives to (a) try to complete the project by assembling the required number of parcels, and (b) do so at the lowest cost. Each period lasts 12 min. Subjects are randomly assigned the role of buyer or sellers A–D (and E in 4v1), and remain in the same role throughout the experiment. Subjects are re-grouped every period according to a quasi-zipper rotation.17 Each session for the 2REQ, 3REQ, and 4REQ treatments consisted of 20 subjects, while the 4v1 sessions consisted of 24 subjects. In all periods, communication among subjects is continuous and free flowing. Sellers have the capacity to communicate with all other sellers (without the buyer knowing), and a buyer can communicate with all sellers, or a specific seller. The buyer must acquire the required number of

17 This algorithm ensures that no seller is ever grouped with a buyer with whom they have been previously paired and vice versa, and minimizes the extent to which sellers are grouped with other sellers with whom they have been previously paired. In theory, the observations may not be independent from one another using this rotation scheme. However, Collins and Isaac (2012), who previously used this rotation scheme, showed that the rotation scheme (be it random re-matching or the quasizipper rotation) had no effect on the aggregation results in this environment. Given that we replicate their results in our baseline treatment, we do not suspect that our rotation scheme will have an impact on our findings. As an additional check, we specify linear probability models where group aggregation is the outcome of interest. We then test for differences in means across treatments, while clustering the standard errors at the session level to allow for within session correlation. Due to the small number of clusters, we implement the wild bootstrap method proposed by Cameron et al. (2008). Results from these regressions show that the treatment effects for competitive treatments are qualitatively similar to those presented in the paper and are statistically significant (p < 0.001).

20

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30 Table 1 Summary of randomly drawn values for buyers and sellers. Variable

Period 1

Period 2

Period 3

Period 4

Panel A: Seller values (constant across all treatments) Seller A’s value Seller B’s value Seller C’s value Seller D’s value Seller E’s valuea

165 145 185 272 767

122 207 279 273 881

159 253 174 115 701

270 237 231 203 941

Panel B: 2REQ Lowest value Capital constraint Buyer’s value Sum of two lowest adjacent seller values Optimal unit combination Maximum possible surplus Buyer’s value Sum of two highest adjacent seller values Least optimal combination Left-over surplus

145 243 510 310 AB 200 510 457 CD 53

122 243 506 329 AB 177 506 480 (552) BD (CD) 26

115 243 526 289 CD 237 526 412 AB 114

203 243 694 434 CD 260 694 507 AB 187

Panel C: 3REQ Sum of two lowest values Capital constraint Buyer’s value Sum of three lowest adjacent seller values Optimal unit combination Maximum possible surplus Buyer’s value Sum of three highest adjacent seller values Least optimal combination Left-over surplus

310 474 655 495 ABC 160 655 622 ACD 33

329 474 762 302 ABD 460 762 759 BCD 3

274 474 930 448 ACD 482 930 586 ABC 344

434 474 924 671 DBC 253 924 738 ABC 186

Panel D: 4REQ/4v1b Sum of three lowest values Capital constraint Buyer’s value Sum of seller’s values Maximum possible surplus

495 711 1237 767 470

602 711 1929 881 1048

448 711 1436 701 735

671 711 1365 940 425

a

Only relevant for 4v1 treatment. The same values apply for 4v1 treatment. The only difference is that a buyer now has the option to deal with seller E, without being subject to a capital constraint. b

units before the 12 min elapse in each period, for successful aggregation to occur. All experiments were conducted at the Florida State University Experimental Social Sciences laboratory. Subjects were recruited through ORSEE, as described by Greiner (2004). All experimental sessions lasted between 90 and 120 min. Subjects were paid a $10 show-up fee in all treatments. In addition, subjects earned an average of $13.93 across all treatments. Appendix A contains an example of the instructions for the 2REQ treatment. The experiment was programmed and conducted using z-Tree (Fishbacher, 2007).18 A total of eight sessions were run, six consisting of 20 subjects each, and two sessions of the 4v1 treatment, consisting of 24 subjects each.19 This yields a dataset of observations from 168 subjects, with 128 group level observations. In this data, we have 32 buyer observations and 136 seller observations. In Appendix B, we discuss additional experimental procedures and data criteria, which lead us to drop 4 group observations from the 4v1 treatment. Ultimately, this restriction leaves us with 124 group level observations, 28 buyer observations, and 116 seller observations. We refer the reader to Appendix B to see how this restriction affects the results discussed below. 4. Results Result 1: We successfully replicate Collins and Isaac (2012). 18 In addition to the decision game, subjects were able to play tic-tac-toe against the computer to mitigate boredom as they waited for all groups to complete each round. Subjects were not paid for tic-tac-toe. 19 We also ran 1 pilot session, which we describe in Appendix B.

To ensure that our results are not driven by differences in the subject pool over time or slight deviations in experimental procedures (e.g. a different person reading the instructions), we test for differences between our 4REQ treatment and Collins and Isaac’s public information/capital constrained treatment. Collins and Isaac reported that buyers had an aggregation rate of 46%. In our replication, we find that buyers successfully aggregated parcels 40.63% of the time. The difference in aggregation rates of 5.37% is not statistically significant (p-value of 0.821 using Fisher’s exact test). We also find similar divisions of surplus among the buyers and sellers, as we discuss below. Result 2: Competition encourages aggregation. From Fig. 3, it is clear that treatments with competition (i.e. 2REQ, 3REQ, and 4v1) have much higher rates of successful aggregation than the baseline (4REQ). Aggregation rates were 100% (32/32), 84.38% (27/32), and 96.42% (27/28) in the 2REQ, 3REQ, and 4v1 treatments, respectively, whereas only 40.63% (13/32) of projects were aggregated in 4REQ. A Fisher’s exact test reveals that these aggregation rates are indeed statistically different (p = 0.001, or lower) from the baseline. The differences in aggregation rates are not only reflected in the aggregate figures, but can also be observed in every individual period following the first. Therefore, these results suggest, that not only is it possible to induce a competitive environment among prospective sellers, but that doing so is an effective avenue to overcome the strategic behavior of sellers. This finding closely mirrors that of Cadigan et al. (2011), Parente and Winn (2012), and Winn and McCarter (2014) who combine competition and contingent contracting. Given their use of

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

21

Fig. 3. Aggregation rates by treatment and period.

contingent contracts, baseline aggregation rates in their studies were relatively high (for instance 89–98, 81–98, and 71–96% respectively), limiting the potential role of competition. We find that competition has a larger impact on successful assembly (more than doubling aggregation rates), which makes it equally effective as contingent contracting based on the findings of previous work by Swope et al. (2011) and Collins and Isaac (2012), who find that contingent contracts increase aggregation rates from 46 to 94%. Result 3: The aggregation rates are not driven by the fixed negotiation time per period. A possible criticism of our design is that we impose a fixed time constraint (12 min) for each period, regardless of the treatment. Therefore, increased aggregation rates might not be driven by increased competition, but simply due to the fact that buyers have more time to negotiate for fewer parcels. While the statement regarding the time per parcel is true, we also note that in each treatment, the buyer has the option to negotiate with four or more sellers, so the time per potential partner is not affected. We also feel that this is a fair representation of the real world. When developers obtain permits from the municipal, county, or state governments, there are clauses that void the permits if certain deadlines or targets are not satisfied, that is, permits are not perpetual.20 Additionally, financial institutions set the term length of construction loans, which are often less than 5 years, creating a hard deadline, regardless of the number or parcels required for the final development.21 To examine the role of the fixed time constraint, we plot the distribution of the time required to aggregate for each group by treatment, conditional on successful aggregation (Fig. 4). Had we implemented a proportional time limit (based on the number of parcels required to aggregate) in the 2REQ and 3REQ treatments, 20 In Tallahassee, FL site-development permits expire automatically after 3 years if no additional permits have been filed to improve the site (see Municipal Codes Sec 9-158). In some instances, states extend these deadlines to promote economic development during economic downturns, see for example Massachusetts, The Permit Extension Act (Section 173 of Chapter 240 of the Acts of 2010). 21 http://www.hg.org/article.asp?id=18115.

the time constraint would have been 360 and 540 s respectively (denoted by solid vertical lines in the figure). From the figure, we see in the 2REQ treatment that 21 of 32 of the groups aggregated within 360 s, suggesting that even under the extreme assumption that none of the groups aggregated beyond that point, aggregation rates would still be higher than in the baseline.22 Similarly, in the 3REQ treatment, we see that 23 of 27 groups aggregated before the hypothetical 540 s time constraint. What is most striking, is that in the 4v1 treatment, the distribution shifts significantly to the left, even though the buyer faces a negotiation problem with five, rather than four, sellers and may still end up on a path to acquisition by four separate parcels.23 To further examine the role of the time constraint, we examine the time that it takes to negotiate for a parcel, net of beginning or end of period effects, in the 4REQ treatment. In the 4REQ treatment, the median (mean) time to acquire the first parcel was 162 (277) s, this however includes start of the period effects. To acquire two parcels, the median (mean) times was 249 (362) s.24 Using either the median or mean, the marginal time to acquire a parcel, net of beginning of the period effects is 85 or 87 s. That is, to communicate with a seller to finalize a contract takes about 87 s. Thus, the 720 s time constraint is never binding based on the time it takes to communicate. Accounting for beginning of the period effects, four parcels can be aggregated in 539 s (277 s for the first + 3 × 87 for parcels 2–4), 3 min less than the time constraint.

22 In another experiment, Portillo (2016) uses this paper’s 2REQ treatment as his baseline. He imposes a 7 min time constraint instead of a 12 min time constraint, and finds that aggregation occurred 87.5% of the time. The difference between aggregation rates is not statistically significant. While his baseline is a minute longer than actual proportional time limit, his results are at least suggestive that imposing a 6 min time constraint in this treatment would not drastically change our results. 23 A Man–Whitney test reveals that the time distributions in competitive treatments are significantly different and to the left of the baseline’s, conditional on aggregation: 2REQ (p = 0.0250), 3REQ (p = 0.0858), and 4v1 (p = 0.0056). The evidence is stronger when we do not condition on aggregation (p = 0.0 0 01, or smaller). 24 27/32 buyers purchased at least one parcel, while 22/32 groups buyers purchased at least two parcels.

22

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

Fig. 4. Bargaining time distribution by treatment, conditional on aggregation.

Fig. 5. Buyer’s surplus share by treatment and period.

Result 4: The expected buyer surplus is increased by competition. While the experimental literature has consistently found that contingent contracts help increase aggregation rates, instances where they are actually used in real estate development are rare. This may have to do with the fact that a high proportion of the available surplus is transferred to sellers (Collins and Isaac, 2012). As the expected buyer’s surplus is eroded, so too is the incentive to develop the project. Therefore, the division of surplus is an important consideration.

Fig. 5 reports the average share of the surplus captured by the buyer in each treatment (both by period and in aggregate). This fraction was computed by dividing the buyer’s payoff by the maximum possible surplus in a given period. Any expenditure that did not result in aggregation was considered negative surplus in the calculation. From Fig. 5, it is clear that the competitive treatments led to increases in the buyer’s expected surplus relative to the baseline, 4REQ. Given the increase in the aggregation rate, this result is perhaps not very surprising. We verify the differences in surplus using a Mann–Whitney test (p < 0.02 for all comparisons).

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

23

Fig. 6. Buyer’s surplus share by treatment and period, conditional on aggregation.

Looking at the aggregate statistics alone might not be the best comparison as failure is less common in the competitive treatments. In Fig. 6, we report the proportion of surplus captured by buyers conditional on successful aggregation. Once we condition on successful aggregation, there is little difference among the buyer’s surplus share across treatments as seen in the right panel of the figure. A Mann–Whitney test cannot reject equality. Combined, the results illustrated in Figs. 5 and 6 suggest that conditional on aggregation, there is little difference from the buyer’s perspective among the various treatments. However, because aggregation rates increase substantially when competition is introduced, ex-ante, the buyer would prefer to develop in a competitive environment. We show that even in the absence of contingent contracts that competition results in high rates of aggregation and surplus levels for the developer, which may ultimately incentivize development relative to a setting where landowners retain a large share of the surplus.25 While buyer surplus was generally high in the competitive treatments, it could have been higher in the 2REQ and 3REQ treatments had the buyer aggregated parcels from the lowest value sellers. In Table 2, we report the proportion of successful aggregations that were formed using the optimal combination. Even though seller values are private information, the buyer successfully completed the project using the lowest cost combination 71.88% of the time in the 2REQ treatment. In five instances, the lowest value seller was locked out of the negotiations because the buyer purchased the first unit from a higher value seller located diagonally from the lowest value seller. In the 3REQ treatment, the buyer aggregated using the lowest cost combination 85.19% of the time. These rates are substantially higher than the 25% chance that

25 This finding is consistent with Cadigan et al. (2011) who report statistically significant increases in buyer earnings when competition is introduced in their two seller treatments with contingent contracts. Winn and McCarter also find that competitive treatments improve efficiency (defined as realized surplus over maximum possible surplus). In their setting, the realized gains are driven by faster negotiations, which minimize delay costs, as well as reduced bargaining costs as sellers are more willing to accept any offer above their value.

Table 2 Frequency with which buyers acquire the optimal combination of unit in 2REQ and 3REQ.

N Completions Completions w/optimal combination Percent Random probability of optimal assembly

2REQ

3REQ

32 32 23 71.88% 25%

32 27 23 85.19% 25%

a buyer randomly assembles the optimal parcel combination. This suggests that buyers were actively seeking the least costly option. Result 5: Seller’s receive no more surplus in competitive treatments than in the baseline, conditional on aggregation. So far we have discussed the surplus of the buyer, we now turn to discuss the surplus captured by the sellers. Two natural questions arise: (i) does selling later increase the share of surplus for the seller and (ii) does the introduction of competition reduce the incentive to delay? To address these questions, we construct a measure of seller surplus as follows:

Sur plus =

Contract P r ice − Sel l er V al ue . Buyer ’s Maximum Sur plus

In Table 3, we report each seller’s surplus by selling position by treatment. We also compare selling in the N − 1 position with being the Nth seller. In the first column, we report the surplus from the baseline (4REQ) treatment. As in Collins and Isaac (2012), we find that the last seller gains a much larger portion of the surplus than those who sell earlier, which is consistent with the notion of holding out until the capital constraint is relaxed.26 Overall, the sellers obtained 28.13% of the surplus in the baseline, with the last seller 26 This result is intuitive, provided that the capital constraint is only relaxed once N − 1 parcels are purchased. Additionally, there is a large body of literature on

24

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30 Table 4 Number of additional units acquired in 4v1 treatment, conditional on aggregation.

Table 3 Seller’s surplus share, conditional on aggregation. Seller

Treatment 4REQ

3REQ

2REQ

4v1 (Seller A–D)

4v1 (Seller E)

1st 2nd 3rd 4th

0.34% 9.41% 3.81% 14.57%

3.12% 2.26% 2.23% –

2.86% 19.46% – –

0.42% 2.99% −5.08% 3.27%

– – – –

Total Early Last N

28.13% 4.52% 14.57 13

7.61% 2.69% 2.23% 27

22.32% 2.86% 19.46% 32

1.60% −1.67% 3.27% 3

22.73% – 22.73% 24

capturing 14.57%. Moving across the columns, we can see that sellers in the competitive treatments obtain a smaller share of the total surplus; 7.61, 22.32, 1.60, and 22.73% respectively. When we pool the competitive treatments and condition on successful aggregation, we find that the total surplus obtained by the sellers is not statistically different than the baseline (p-value 0.1302 using a Mann–Whitney test). This is consistent with the previous finding that the buyer’s surplus is not statistically different across the competitive and non-competitive treatments. Perhaps a more surprising results, is that there is marginal evidence that the expected seller surplus is higher in competitive treatments when we do not condition on aggregation (pvalue = 0.0958 using a Mann–Whitney test). While this may seem counterintuitive at first, this becomes less surprising when we recall that competition leads to improved aggregation rates. As more projects are completed, more surplus is available for both buyers and sellers. Thus, it appears that even sellers might benefit from the competitive dynamics. There are other interesting seller surplus patterns within the treatments. In the 3REQ treatment, we find that the surplus among the sellers was relatively flat by position, thus, there appears to be no benefit to holding out. In the 2REQ treatment, the second (and last) seller, obtained a premium for their property compared to the first seller. The value of this surplus is similar to the surplus captured by the last seller in the baseline. This result runs counter to our expectation, as we expected increased competition to reduce the seller’s surplus. However, the nature of the competition between the sellers is likely Bertrand. Therefore, the lower value seller of the two remaining should be willing to accept any offer below the other remaining seller’s value, and above their own value. This behavior would lead to a positive surplus for the second seller. Finally, we separately report the share of the surplus for the sellers in the 4v1 treatment for cases in which the buyer aggregated by negotiating with sellers A–D (column 4) or from seller E (column 5). It would appear from these results that seller E was able to obtain a premium, capturing an average surplus of 22.73%%, likely reflecting the reduced transactions costs associated with negotiating with a single party. Secondly, we note that the presence of an additional seller limits the surplus that can be extracted by the last seller when the buyer negotiates with sellers A–D, however, we do not stress this difference given the small sample size, as this only occurred in three instances. Result 6: Buyers prefer transacting with a single large landowner.

the land assembly problems that shows the last seller obtaining a large share of the surplus in the absence of contingent contracts due to the sunk cost feature of acquiring prior inputs (see for instance Miceli and Segerson, 2007, Micili and Sirmans, 2007).

Additional units acquired

Frequency

Percent

0 1 2 3 Total

16/24 7/24 0/24 1/24 24/24

66.67 29.17 0 4.17 100

Miceli and Sirmans (2007) conjectured that the land assembly problem and the transactions costs associated with aggregation may be causal factors of suburbanization and urban sprawl. We find support for this hypothesis, as buyers overwhelmingly transact with a single, consolidated landowner when given the option. In the 4v1 treatment, aggregation was successful in 27 out of 28 groups. Of these 27 successful aggregations, 24 projects were completed when the buyer purchased the parcel from seller E. Relative to the baseline 4REQ, the aggregation rates are 55.8 percentage points higher (a 130% relative increase) and the buyer keeps a much larger share of the expected surplus (60.04% in 4v1 vs. 4.71% in 4REQ). When dealing with seller E, buyers retained approximately 62% of the available surplus. In the three cases in which the buyer successfully negotiated with sellers A–D, the buyer obtained roughly 98% of the surplus (p = 0.0448, using to a Mann–Whitney test). The difference in the buyer’s surplus is driven by at least two factors. First, as we discussed in the previous subsection, the buyer paid seller E a premium for their property, likely associated with the reduced transactions cost. Secondly, there were several instances in which the buyer purchased excessive parcels, leading to “inefficient” aggregation. In 8 of the 24 cases where the buyer ultimately purchased seller E’s parcel, the buyer initially purchased parcels from sellers A–D. In Table 4 we report the frequency with which the buyer purchased extra parcels. Buyers who initially purchased units from sellers A–D, and then completed the project by purchasing the unit from seller E earned an average 42% of the surplus. Buyers who acquired the unit from seller E retained roughly 72% of the surplus (p = 0.0254), according to a Mann–Whitney test). 5. Discussion and conclusions The land assembly problem is difficult to overcome given the subjective nature of costs and benefits of the pertinent parties. Previous literature has shown that contingent contracts provide one mechanism to increase aggregation rates. However, contingent contracts have unattractive features, as they tend to shift surplus from the buyer to the sellers, reducing incentives to develop. Contingent contracts may also increase certain transactions costs, as more complex contracts require more administration and legal expertise. In this paper we propose that increasing the flexibility of siting, via induced site-development competition, may be an effective way to encourage aggregation. Our results show that competition among fragmented landholders drastically increases land assembly, even in the absence of contingent contracts. This finding is robust across our competitive treatments. Furthermore, not only do aggregation rates increase, competition also provides buyers with larger shares of (expected) surplus than settings lacking competition, which provides stronger incentives to develop. In addition to this finding, we also find evidence linking land assembly to suburbanization, as suggested by Miceli and Sirman’s (2007). When buyers were given the option to negotiate endogenously with a set of small, fragmented, landowners, or a landowner with consolidated holdings, they overwhelmingly

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

transacted with the large landholder. The speed of the negotiations also increased relative to the baseline, which suggests additional benefits if delay is costly. Finally, we find that developers are willing to pay a premium to consolidated landowners over fragmented ones within the treatment (a premium of approximately 20%). While our design is not a direct test of the suburbanization hypothesis per se, it is reflective, given that parcel sizes tend to be small and more disaggregated in the urban core. Our finding has direct implications for policy makers and city development councils. Overall, our results suggest that when developers have multiple development sites, or where there are existing, large parcels of land, development is more likely to occur. Therefore, cities interested in redeveloping blighted sections of the city or attracting new developments must carefully evaluate their permitting processes and existing land use restrictions. Relaxing land use restrictions would likely increase property values (Turner, Haughwout, and van der Klaauw, 2014), providing local governments new sources of property tax revenue, while promoting development at the same time. Eliminating land use restrictions, such as minimum/maximum lot sizes or other zoning regulations would likely increase the number of potential development sites, while a simpler permitting process would lead to reduced costs associated with cutting the red tape. Alternatively, cities may opt to form non-profit land banks, to buy and consolidate foreclosed and dilapidated properties that they then sell to developers.27 If such policies are successful, the need and prevalence of more controversial development tools, such as eminent domain, may be diminished. Acknowledgments The authors would like to thank The Quinn Eminent Scholar Chair, the Quinn Graduate Fellowship, and Florida State University for providing funding for the project. The authors also thank participants of the FSU Experimental Workshop for valuable comments. Appendix A. Sample of instructions (2REQ treatment) Instructions General information The purpose of this experiment is to study how people make market decisions. From now on and until the end of the experiment any verbal or written communication with other participants is not permitted, except through the software interface as explained below. If you have a question, please raise your hand and one of us will come over to answer it. You will receive $10.00 for showing up on time for the experiment. In addition, you will have an opportunity to earn more during the experiment. All currency in this experiment will be denominated in U.S. cents. At the end of the experiment you will be paid by check the sum of your show-up fee and other earnings. You will be paid privately and we will not disclose any identifiable information about your actions or your payment to the other participants in the experiment. Roles, groups, and periods During this experiment, you will be randomly assigned one of two roles, buyer or seller. Each group will consist of one buyer and

27 See for example the Cuyahoga Land Bank, serving the Cleveland, Ohio metropolitan area.

25

four sellers denoted as the Buyer and Seller A through Seller D respectively. You will remain in the same role throughout the experiment. This experiment will consist of four periods. In each Period, new groups will be formed from a buyer and four sellers using a computer algorithm. This algorithm will (1) ensure that no seller is ever grouped with a buyer with whom they have been previously paired and vice versa, and (2) minimize the extent to which sellers are grouped with other sellers with whom they have been previously paired.28 Description of the decision task(s) and payoffs The buyer will be paid for acquisition of units to build a hypothetical object. The units will be represented in a 2 × 2 matrix as shown below.

The object is comprised of two units (out of four available units). Each seller has one unit for which the buyer may make an offer. If a seller accepts the buyer’s offer, a contract has been made for the buyer to acquire the unit. To successfully build the object, a buyer must acquire two adjacent units. Properties adjacent to unit A, for instance, are units B and C. Properties adjacent to unit C, are units A and D. In particular, note that buying diagonal units A and D, or B and C will not suffice to build the object. Therefore, if a seller who is diagonally across from another (A and D for example) sells his or her unit first, then the seller who still owns his unit will no longer be part of the game and must wait for the time to elapse. For instance, if seller D sold his or her unit, then seller A would no longer be part of the game in the current period. If seller B sold his or her unit, then seller C no longer participates in the current period. Seller earnings. In each period, a seller’s value is an integer independently and randomly drawn from the interval 10 0–30 0 cents, with each number in between being equally likely. No seller’s value depends on the value of any other seller. If a seller does not sell his or her unit to the buyer, the seller’s earnings for that period equal that value. If a seller’s unit is sold to the buyer, the seller will receive the amount offered by the buyer (instead of the seller’s value). Each seller will receive a new, randomly drawn value every period. Buyer earnings. The buyer’s earnings are a little more complicated. At the beginning of each period, each buyer receives a starting fund for that period of 243 cents. In each period, the buyer has a value for acquiring two adjacent units of the object. The buyer’s value is an integer randomly drawn in the following fashion. Let “S” equal the sum of two lowest sellers values in a given period. Then, in that period the buyer’s value is drawn from the interval (S + 75 cents to S + 275 cents), with each value in between being equally likely. The buyer’s value

28 Sellers: Every period you will be grouped with a buyer with whom you have never been grouped before. At least half of the sellers in your group will not have been grouped with you in the previous period round, and at least one of those will be a seller with whom you have never been grouped before. Buyers: Every period you will be grouped with four sellers with whom you have never been grouped before.

26

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

is such that there is always, at least, one combination of two adjacent units for which the buyer will have a higher value than the sum of the individual seller values for the units, but that difference can be anywhere from 75 to 275 cents. Notice that there could be more than a single combination depending on the buyer’s random draw. If the buyer purchases two adjacent units, the buyer’s earnings will be calculated as follows: 243 cents starting fund PLUS the buyer’s value MINUS the amount paid to the two sellers for the units acquired. If the buyer does not purchase the two needed units to build the object, the buyer’s earnings will be calculated as follows: 243 cents starting fund MINUS the amount paid to another (one, if any) seller for the unit acquired. There is a limit upon the offers that a buyer can make. Because a buyer will receive his value only if he or she purchases two adjacent units, as long as the buyer has not yet purchased a unit, then all outstanding offers cannot exceed his or her starting fund, 243 cents. In a formula: The sum of all outstanding offers ≤ 243 cents If a buyer has purchased one unit, then the total of the payment he or she has already made plus any offers for a second unit cannot exceed the buyer’s value. The sum of the accepted offer + any additional outstanding offers ≤ the buyer’s value Each period will last 12 min (720 s) or until the buyer has acquired two adjacent units. Note that accepted offers cannot be rescinded or changed. This means that if the buyer gets close to exhausting his or her limit when acquiring units he or she may be unable to make a sufficient offer in order to acquire any additional units. If this occurs, the buyer will be unable to acquire the two needed units and must wait for the remaining time to elapse. Information that you and others will receive Prior to making your decisions, you will see your value. Over the course of the experiment, you will see all outstanding offers, whether or not a seller has sold their unit, and the number of units acquired and required. Everyone will be told the buyer’s value. Appendix B. Sample details In this appendix, we describe how we arrived at the main sample described in the paper. We include this description for two primary reasons, first during the initial 4v1 session, we had a software bug that affected chat communication for seller E. This mistake, led us to reboot the software following the scripted practice round, which led to a reassignment of roles for the subjects. The mistake appeared later during the game, as noted by subjects’ chat log, leading to us interrupting the session midway to inform all subjects of the error, although we did complete the session. Below, we describe how this data would have affected our results. Secondly, during the first batch of bug free treatments (specifically in the 4v1 treatment), we noticed a few sellers selling for significantly less than their values, which prompted us to do two things: first, we added an additional quiz question for all treatments and ran a second round of treatments, and secondly, in the analysis, we impose, what we believe are reasonable criteria to eliminate outliers in the data. We describe this procedure in more detail below. Given the addition of the quiz question, we have two separate batches of sessions, the first includes the standard set of comprehension

questions following the reading of the instructions, while the second batch adds one additional question to the quiz.29 Finally, we demonstrate the similarity of the samples, which ultimately allows us to pool across the two batches of treatments. Software bug and pilot session of 4v1 During the first session of the 4v1 treatment, as subjects finished a practice round, we (the experimenters) observed that the subject’s chat interface was not properly displaying Seller E’s name. We rebooted the z-Tree program, which (likely) changed the roles subjects had initially been assigned. That is, if a subject was Seller E or the buyer before, it was unlikely that they would be Seller E or the buyer again, after the reboot. After the reboot, subjects began the experiment. While we were monitoring the chat logs however, we observed that subjects were discussing the same issue again. At this point we interrupted the session to notify all subjects of the issue, but finished the session. Due to the interruption, we decided that the data should not be used because it would not be directly comparable to other sessions. We did however examine the data, and found high rates of aggregation with the buyer transacting with seller E in proportions similar to those reported in the text.30 In some regards this is surprising, because the bug made it difficult to identify seller E. Handling outliers Before we describe the sample restrictions imposed, we discuss how sellers selling for less than their value would affect the main results of the paper. If a seller agrees to a contract price that is less than their value, it will have several impacts (i) it will facilitate aggregation (ii) it will increase the surplus of the buyer (iii) decrease the surplus of the seller, and (iv) in the 4v1 treatment, it will bias the buyer toward a particular set of sellers. For instance, if seller E sells for less than their value, it will increase the likelihood that buyers negotiate with Seller E rather than sellers A–D. Each of these biases work in the direction of supporting our hypothesis, therefore, by removing the data associated with sellers selling below their value, it will make it more difficult for us to find significant differences between a given treatment and the baseline 4REQ. To identify systematic outliers, we followed the following procedure. First, we generate the mean normalized surplus by treatment for each seller, where the surplus is defined as

Sur plus =

Contract P r ice − Sel l er V al ue . Sel l er V al ue

We then calculate the standard deviation of the seller surplus within each treatment. Finally, to identify an outlier, we follow two rules: first, surplus must be negative and more than two standard deviations from the treatment mean, and secondly, the seller must have had negative surplus of this magnitude for more than a single period of the game. Following the first rule, we identify five observations had extremely negative surplus, satisfying the first criteria, they are as follows (Table B1). Of these five observations, only one subject repeatedly made a transaction for such a significant loss. Given this one subject who undersold, we drop all group level observations that include this seller. Therefore, in the 4v1 treatment, we have 28 group level observations. 29 During the quiz, subjects were able to repeatedly submit answers until they answered the question correctly. This format may have allowed subjects who did not fully comprehend the game to proceed by simply clicking repeatedly until the prompt advanced. 30 The authors are willing to make the data available upon request.

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

27

Table B1 Sellers with a negative surplus greater than 2 standard deviations from the mean. Batch

Treatment

Group

Period

Role

1 1 1 1 2

4v1 4v1 4v1 4v1 2REQ

1 3 4 2 1

1 2 3 4 1

Seller Seller Seller Seller Seller

E E E E D

Subject ID

Contract price

Value

Surplus

1_1_7 1_1_7 1_1_7 1_1_7 2_2_19

100 160 299 250 50

767 941 701 881 272

−0.869 −0.829 −0.573 −0.71 −0.816

Fig. B1. Aggregation rates by treatment and period, with outlier.

Ultimately, excluding this seller has little impact on our findings. Below we replicate our main figures from the paper so the reader can observe any differences in the aggregation rate attributable to the outlying data (Figs. B1 and B2). Visually, there is little difference in the aggregation rate within the 4v1 treatment across all periods of the experiment. This is confirmed by the aggregate statistics. Aggregation rates across all periods following this exclusion of the outlier were 96.4% (27/28 projects) in the 4v1 treatment, as reported in the text, while the aggregation rate was slightly higher, 96.8% (31/32 projects), when we include the outlier in the sample. The proportion of buyers buying from Seller E also increases slightly from 88.8% (24/27) to 90.3% (28/31). Similarly, excluding the outlier has little impact on our measure of buyer surplus. As we noted above, a seller who sells for significantly less than their value likely biases buyer surplus to be increasingly positive. Below we replicate Fig. 5 from the text to highlight any differences by including or excluding the outlier on buyer surplus. In Panel A, we present the surplus with the inclusion of the outlier, while in Panel B, we reproduce the figure from the text. What should be obvious from the aggregate data, is that the share of the surplus going to the buyer falls when we exclude the outlier, as predicted. With all observations, the buyer captures 74.37% of the surplus, unconditional on aggregation (80.28% conditional on aggregation). When we remove the outlier, the unconditional surplus falls to 60.05% (66.29 conditional on aggregation). Whether or not we exclude this data is ultimately of little consequence, as the buyer surplus in the restricted sample is still significantly different

from the baseline 4REQ treatment, where the buyer receives almost no share of the surplus (Figs. B3 and B4). Addition of a quiz question As we previously noted, to prevent significant underselling in subsequent sessions, we added the following quiz question to the comprehension quiz and re-ran all of the treatments. That is, in our sample, 64/124 group level observations subjects were asked the following additional question: Suppose a seller has a value of 236, receives an offer from the buyer for 200, and receives no other offer. If the seller does not accept the offer, the seller’s payment will be (i) (ii) (iii) (iv)

36 (that is, 236-200) 200 (the offer) 236 (the seller’s value) 436 (that is, 236 + 200).

To provide evidence of how the addition of this quiz question affects our sample, we present the p-values for a series of Fisher’s exact tests and Mann–Whitney tests that compare each treatment by batch (before and after the addition of the quiz question) in Table B2. The first row, compares the aggregation rate within treatment. It is easy to see that there are no significance differences between batches for any treatment. The remaining rows compare the earnings for specific subject roles (buyer, sellers A–D, and seller E). The only instance where there is a possible difference between the samples occurs for seller E in the 4v1 treatment when we include

28

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

Fig. B2. Aggregation rates by treatment and period, without outlier.

Fig. B3. Buyer’s surplus share by treatment and period, with outlier. Table B2 Test statistics between batches/pre and post additional quiz question. Variable

4v1 outlier included

4v1 outlier excluded

2REQ

3REQ

4REQ

Completion ratesa Buyer earnings Seller A earnings Seller B earnings Seller C earnings Seller D earnings Seller E earnings

1.0 0 0 0.534 0.5155 0.8175 0.7152 0.9083 0.1047

1.0 0 0 0.5617 0.4946 0.8499 0.6705 1 0.6089

–b 0.3363 0.8942 0.5682 686.58 0.3922 –

1.0 0 0 0.9549 0.7035 0.7756 0.3828 0.6918 –

1.0 0 0 0.7767 0.3735 0.7906 0.1913 0.2403 –

a This row reports p-values obtained after running Fisher’s exact test. All remaining rows report pvalues from Mann–Whitney tests. b All projects were completed in both 2REQ treatments and thus, yield the same aggregation rate.

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

29

Fig. B4. Buyer’s surplus share by treatment and period, without outlier.

the outlier as identified previously.31 When we drop the outlier from the data (p-values reported in column 2), there are no significant differences within the treatments between the batches. Therefore, in the body of the text, we pool the data between batches. References Benson, B.L., 2008. The evolution of eminent domain: a remedy for market failure or an effort to limit government power and government failure. The Independent Review XII (3), 423–432. Benson, B.L., 2005. The mythology of holdout as a justification for eminent domain and public provision of roads. The Independent Review X (2), 165–194. Brooks, L., Lutz, B., 2013. From Today’s City to Tomorrow’s City: An Empirical Investigation of Urban Land Assembly. George Washington University Mimeo. Buchanan, J.M., Yoon, Y.J., 20 0 0. Symmetric tragedies: commons and anticommons. Journal of Law and Economics 43 (1), 1–13. Cadigan, J., Schmitt, P., Shupp, R., Swope, K., 2009. An experimental study of the holdout problem in a multilateral bargaining game. Southern Economic Journal 76 (2), 444–457. Cadigan, J., Schmitt, P., Shupp, R., Swope, K., 2011. The holdout problem and urban sprawl: experimental evidence. Journal of Urban Economics 69, 72–81. Cameron, A.C., Gelbach, J.B., Miller, D.L., 2008. Bootstrap-based improvements for inference with clustered errors. The Review of Economics and Statistics 90 (3), 414–427. Collins, S.M., Isaac, R.M., 2012. Holdout: existence, information, and contingent contracting. Journal of Law and Economics 55 (4), 793–814. Cunningham, C. (2013). Estimating the Holdout Problem in Land Assembly. Tech. rep. Working Paper. Federal Reserve Bank of Atlanta. Emerson, C.D., 2009. Merging public and private governance: how Disney’s Reedy Creek improvement district re-imagined the traditional division of local regulatory powers. Florida State University Law Review 36, 177. Epstein, R.A., 2001. Intellectual property: old boundaries and new frontiers. Indiana Law Journal 76, 803. Fischbacher, U., 2007. z-Tree: Zurich toolbox for ready-made economic experiments. Experimental Economics 10 (2), 171–178. Greiner, B. (2004). An online recruitment system for economic experiments. Published in: Forschung und wissenschaftliches Rechnen. GWDG Bericht, Vol. 63, 79-93. Gyourko, J., Saiz, A., Summers, A., 2008. A new measure of the local regulatory environment for housing markets: the Wharton residential land use regulatory index. Urban Studies 45 (3), 693–729. Hardin, G., 1968. The tragedy of the commons. Science 162, 1243–1248. Heller, M., 1998. The tragedy of the anticommons: property in transition from Marx to markets. Harvard Law Review 111 (3), 621–688.

31

The p-value from a Mann–Whitney test is marginally significant, p = 0.1047.

Hong, J.T., Plott, C.R., 1982. Rate filing policies for inland water transportation: an experimental approach. The Bell Journal of Economics 13 (1), 1–19. Kelly, D., 2006. The ‘Public Use’ requirement in eminent domain law: a rationale based on secret purchases and private influence. Cornell Law Review 92, 1–66. Kitchens, C., Roomets, A., 2014. Dealing with eminent domain. Journal of Behavioral and Experimental Economics 54, 22–31. Kitchens, C., 2014. The use of eminent domain in land assembly: the case of the Tennessee valley authority. Public Choice 160, 455–466. Kominers, S.D., Weyl, E.G. (2010). Concordance Among Holdouts. Working Paper, April. Kominers, S.D., Weyl, E.G., 2012. Holdout in the assembly of complements: a problem for market design. American Economic Review Paper & Proceedings 102, 360–365. Lόpez, E.J., Clark, J.R., 2013. The problem with the holdout problem. Review of Law and Economics 9 (2), 151–167. McConnell, V., Walls, M., 2009. Policy monitor US experience with transferable development rights. Review of Environmental Economics and Policy 3 (2), 288–303. McConnell, V., Walls, M., Kopits, E., 2006. Zoning, TDRs and the density of development. Journal of Urban Economics 59 (3), 440–457. Menezes, F., Pitchford, R., 2004a. The land assembly problem revisited. Regional Science and Urban Economics 34 (2), 155–162. Menezes, F., Pitchford, R., 2004b. A model of seller holdout. Economic Theory 24 (2), 231–253. Miceli, T.J., 2011. Free riders, holdout, and public use: a tale of two externalities. Public Choice 148, 105–117. Miceli, T.J., Segerson, K., 2007. A bargaining model of holdouts and takings. American Law and Economics Review 9, 160–174. Miceli, T.J., Segerson, K., 2012. Land assembly and the holdout problem under sequential bargaining. American Law and Economics Review 14 (2), 372–390. Miceli, T.J., Sirmans, C., 2007. The holdout problem, urban sprawl, and eminent domain. Journal of Housing Economics 16, 301–319. Miceli, T.J., Segerson, K., Sirmans, C., 2008. Tax motivated takings. National Tax Journal LXI (4), 579–591. Munch, P., 1976. An economic analysis of eminent domain. Journal of Political Economy 84 (3), 473–498. Parente, M.D., Winn, A.M., 2012. Bargaining behavior and the tragedy of the anticommons. Journal of Economics Behavior & Organization 84 (2), 475–490. Portillo, J.E. (2016). Externalities, Free-Riding, and Land Assembly. Working Paper. Florida State University. Posner, R.A., 1992. Economic Analysis of Law, fourth ed. Little Brown and Company. Seidenfeld, M., 2008. In search of Robin Hood: suggested legislative responses to Kelo. Journal of Land Use & Environmental Law 23 (2), 305–332. Shavell, S., 2010. Eminent domain versus government purchase of land given imperfect information about owners’ valuations. Journal of Law and Economics 53, 1–27. Shupp, R., Cadigan, J., Schmitt, P.M., Swope, K.J., 2013. Institutions and information in multilateral bargaining experiments. B.E. Journal of Economic Analysis & Policy 13 (1), 485–524.

30

R.M. Isaac et al. / Journal of Urban Economics 95 (2016) 16–30

Smith, V.L., 1982. Microeconomic systems as an experimental science. The American Economic Review 72 (5), 923–955. Sridhar, U., Mandyam, S., 2013. A group utility maximizer mechanism for land assembly. The Journal of Real Estate Finance and Economics 47 (3), 466–488. Strange, W.C., 1995. Information, holdout, and land assembly. Journal of Urban Economics 38, 317–332. Suzuki, J., 2013. Land use regulations as a barrier to entry: evidence from the Texas lodging industry. International Economic Review 54 (2), 495–523. Swope, K.J., Cadigan, J., Schmitt, P., 2014. That’s my final offer! Bargaining behavior with costly delay and credible commitment. Journal of Behavioral and Experimental Economics 49, 44–53.

Swope, K., Wielgus, R., Schmitt, P., Cadigan, J., 2011. Contracts, behavior, and the land-assembly problem: an experimental study. In: Isaac, R.M., Norton, D. (Eds.), Research in Experimental Economics: Experiments on Energy, the Environment, and Sustainability, 14. Emerald Group Publishing Limited, pp. 151–180. Tanaka, T., 2007. Resource allocation with spatial externalities: Experiments on land consolidation. The BE Journal of Economic Analysis & Policy 7 (1), 1–31. Turner, M.A., Haughwouth, A., van der Klaauw, W., 2014. Land use regulation and welfare. Econometrica 82 (4), 1341–1403. Winn, A.M., McCarter, M.W. (2014). Who’s Holding Out? An Experimental Study of the Benefits and Burdens of Eminent Domain. SSRN Working Paper. Zillante, A., Schwarz, P.M., Read, D.C., 2014. Land aggregation using contingent and guaranteed payments. Southern Economic Journal 80 (3), 702–727.