The Effect of Relationships on Contract Choice: Evidence ...homes.chass.utoronto.ca/~horstman/corts.pdf · The Effect of Relationships on Contract Choice: Evidence from Offshore

The Effect of Relationships on Contract Choice:

Evidence from Offshore Drilling

Kenneth S. Corts and Jasjit Singh∗

Harvard University

March 18, 2002

Abstract

We argue that relationships and high-powered formal contracts canbe either substitutes or complements, depending on the relative impact ofrelationships on incentive problems and contracting costs. In the offshoredrilling industry, we find that oil and gas companies are less likely tochoose fixed-price contracts as the intensity of their relationship with adriller increases. This supports the conclusion that relationships and high-powered formal contracts are substitutes in this setting, indicating thatrelationships reduce incentive problems more than contracting costs. Wefurther test the effect of relationships on multiple types of wells that differin the severity of their contracting costs and find results consistent withour argument.

∗Harvard Business School, Boston, MA 02163; [email protected] and [email protected]. Wethank George Baker, Oliver Hart, Dale Jorgenson, Paul Oyer, Steve Tadelis, and seminarparticipants at Harvard, MIT, Carnegie Mellon, Michigan, USC, London Business School,and the 2002 AEA meetings for helpful comments.

1 Introduction

Oil and gas companies contract with independent drillers under two very dif-ferent contracts, known as dayrate and turnkey contracts. These correspond,respectively, to the cost-plus and fixed-price contracts used in construction, mil-itary procurement, and many kinds of professional services, including consultingand software development. In any setting, the choice between these two typesof contracts presents the buyer with a stark dilemma. On one hand, writinga fixed-price contract requires carefully enumerating many contingencies anddetailing the project specifications ex ante, making it very costly to change theproject specifications once the project is underway. On the other hand, a cost-plus contract is simpler to write and gives the buyer more flexibility in alteringthe specifications as the project proceeds; however, this flexibility comes at thecost of introducing a moral hazard problem, as the agent may bill the principalfor excessive materials and labor.The choice becomes even more complicated in a repeated relationship. For

example, having completed a bathroom renovation, does the homeowner nego-tiating with the same contractor for kitchen remodeling lean toward one type ofcontract or the other? Does the trust established in the repeated relationshipmore dramatically assuage fears of hold-up in renegotiation (making a fixedprice contract more attractive) or skepticism about the legitimacy of the cost-plus charges (making a cost-plus contract more attractive)? In the languageof contract theory, are relationships and high-powered (fixed-price) formal con-tracts substitutes or complements? We argue that the answer is in generalambiguous. Empirically, we examine the effect of relationships on the choice ofcontract type in the offshore drilling industry and conclude that, in this partic-ular context, relationships and high-powered formal contracts are substitutes.As a preview of our results, consider these striking figures: fixed-price contractsgovern 28% of projects between parties who have not worked together before,but only 15% of repeat contracts.Two papers have highlighted the importance of the trade-off between con-

tracting costs and moral hazard in procurement and construction contracts with-out considering the effect of relationships. Crocker and Reynolds (1993) empha-size that the optimal contract features a degree of completeness that strikes anappropriate balance between these costs and benefits. In a more formal model,Bajari and Tadelis (2001) show why contract completeness and strong incen-tives go together and suggest a number of reasons why optimal contracts forvarious projects may tend to fall into the dichotomous categories of fixed-priceand cost-plus as they do in the offshore drilling setting that we study.1

1Others have argued that other considerations beyond the realm of standard contracttheory play important roles in contract choice. For example, Oyer (2002) argues that com-pensation plans tied to stock performance serve to match compensation to outside offers overtime rather than to provide incentives, and Lafontaine and Masten (2002) argue that truckingcontracts are structured to economize on price-setting across heterogenous projects ratherthan to induce effort. While such alternative explanations may also apply to this industry,we focus here on the considerations suggested by the literature on moral hazard and transac-tion costs since these considerations seem the most likely to be mitigated through repeated

1

Papers that explicitly consider the impact of relationships on contract designoffer two different views on the role of relationships. One branch of the literature(which includes economists like Williamson (1985), Bull (1987) and Kreps (1990)as well as sociologists like Granovetter (1985), Sabel (1993) and Gulati (1995))argues that relational contracts represent an alternative governance mechanismto formal contracting. A second branch of the literature (e.g., Klein and Leffler(1981), Baker, Gibbons, and Murphy (1994) and Klein (1996)) argues thatthere are important interactions between the relational contract and the formalcontract, with the formal contract affecting the sustainability of the relationalcontract by defining the fallback positions of the parties.According to the first view, formal and relational contracts are substitutes:

if a relational contract is feasible then the parties may simply ignore formalcontracting altogether. In contrast, the second view implies that formal andrelational contracts may be either substitutes or complements. In particular,Baker, Gibbons, and Murphy (1994) show that the two may be substitutes whenformal contracts work so well that they provide the parties an attractive fallbackposition, thus undermining their incentives to adhere to the relational contract.In contrast, they may be complements when the formal contract sufficientlyincreases the future value of the sustained relationship, thus overcoming theincentive to violate the relational contract. Baker, Gibbons, and Murphy assumethat all performance measures are exogenously either perfectly contractible atzero cost or observable but not contractible (and hence usable only in a relationalcontract). In contrast, Klein (1996) argues that the degree of completeness ofthe formal contract is endogenous. By affecting contracting costs, relationshipscan therefore affect the design of the formal contract by altering the cost-benefittrade-off that defines the optimal degree of contractual completeness.In the spirit of Klein’s suggestion that the degree of completeness of the

formal contract should be affected by the strength of the relationship, we ex-plore the ramifications of relationships for the choice of contract form, focusingon the dichotomous choice between fixed-price and cost-plus contracts that isobserved in this and many other industries. We argue that whether relation-ships make high-powered (fixed-price or turnkey) formal contracts more or lessattractive relative to low-powered (cost-plus or dayrate) contracts depends onhow relationships affect both incentive problems and contracting costs.2 If rela-tionships sustain effort levels sufficiently close to those provided by formal high-powered contracts but do not lead to much savings in contracting costs, theytip the trade-off in favor of low-powered (cost-plus) contracts that have lowercontracting costs. In contrast, if relationships reduce contracting costs but donot provide significant improvements in incentives, they tip the trade-off to-ward stronger formal contracts. Empirically, we find that stronger relationshipsreduce the use of high-powered contracts in the offshore oil-drilling industry,suggesting that relationships and high-powered formal contracts are substitutes

contracting.2In the rest of the paper, for brevity, we will often refer to both ex ante contracting and ex

post renegotiation costs as simply “contracting costs”, though it should be clear from contextthat we are referring to both ex ante and ex post costs.

2

in this industry.Several empirical papers provide evidence that relationships and strong for-

mal contracts are substitutes. Gulati (1985) studies governance structures ininterfirm alliances and finds that repeated alliances between partners are lesslikely than other alliances to be organized using formal equity-based contracts.He argues that close interaction between firms over prolonged periods leadsto increased trust through mutual awareness and familiarity, making detailedequity-based contracts unnecessary. Banerjee and Duflo (2000) demonstratethat relationships have a significant effect on the choice between cost-plus andfixed-price contracts in the Indian software industry. Indian software firms thathave worked for the same US client before are more likely to work on cost-pluscontracts. Both of these results suggest that relationships and formal fixed-pricecontracts are substitutes.In contrast, other papers find evidence that relationships and formal fixed-

price contracts are complements. Based on an analysis of survey data from theUS information services outsourcing industry, Poppo and Zenger (2001) arguethat complex formal contracts and relationships are complements in the sensethat respondents perceive that the complexity of contracts increases with thelength of the relationship. Mayer and Kalnins (2002) study the contracts of aspecific US information technology services firm, and show that repeated con-tracts are more likely to be the more complete contract forms like fixed-pricecontracts. This again suggests that high-powered formal contracts and relation-ships are complements; Mayer and Kalnins suggest that this is precisely becauserepeated contracting reduces contracting costs for software development.Empirically, we contribute to the literature by using instrumental variables

estimation to account for potentially problematic simultaneity issues and en-dogenous matching of agents to projects. None of the aforementioned studiesaddress this problem, though it has been shown to be potentially serious inother contexts—specifically, agricultural contract choice—by Ackerberg and Bot-ticini (2002). We exploit the site-specific and asset-intensive nature of theoffshore drilling industry to construct instruments for agent characteristics thatare arguably exogenous to the choice of contract form. Comparing our IVresults with our preliminary models shows that endogenous matching, if unac-counted for, would have led us to underestimate the magnitude of the effect ofrelationships on contract choice.In addition, we extend the empirical literature on fixed-price and cost-plus

contracts to a new industry, using a dataset that offers several advantages overthose used in previous studies. First, we analyze a much larger set of projectsthan the aforementioned studies. Second, while our projects (offshore oil and gaswells) differ in some important observable ways, they are arguably considerablymore homogenous than the projects examined in previous studies (e.g., soft-ware development projects (Banerjee and Duflo, 2000), IT services outsourcingprojects (Poppo and Zenger, 2001, and Mayer and Kalnins, 2002), and alliances(Gulati, 1995)). Third, we have a substantial number of agent firms, whichprovides us more variation in the relationships of firms than was present in thetwo-contractor setting of Crocker and Reynolds (1993). Fourth, we have a panel

3

with multiple projects for almost all principals, which allows us to control forprincipal heterogeneity in a way that Banerjee and Duflo (2000) and Mayer andKalnins (2002) could not.Section 2 describes the offshore drilling industry and the two major kinds

of contracts employed there: dayrate (cost-plus) and turnkey (fixed-price) con-tracts. Section 3 describes the data and lays out our empirical hypotheses,which are motivated intuitively there and derived formally from a simple theorymodel in the appendix. Section 4 presents our empirical models and describesour results. Section 5 concludes by situating our results in the context of therelated empirical literature and by discussing directions for future research.

2 Offshore Drilling

Oil and gas exploration and production (E&P) companies lease offshore tractsfrom governments, typically through auctions. These E&P firms include theintegrated majors (Shell and BP Amoco, for example, with tens of billions ofdollars in assets), large independents (Anadarko and Vastar, for example, withassets in the $2-5 billion range), and many smaller firms (the smallest publicfirms in our data are Petroquest and Santa Fe Energy, each with assets lessthan $30 million). Having acquired the rights to a tract, these firms formulate aplan for its exploration. Typically, this plan involves drilling several exploratorywells to determine the extent (volume), composition (oil vs. gas), and economicviability (cost of extraction) of whatever fuels may be present within the tract.If the results are favorable, the exploratory wells are followed by developmentwells placed to efficiently extract these reserves.Only a very small number of state-owned E&P companies own and operate

their own offshore drilling rigs. All other firms, including all of the majors,contract for the services of drilling rigs with independent drilling contractors.3

These firms include industry giants like Transocean Sedco Forex, Noble Drilling,and Global Marine (each with 30-120 rigs and $2-6 billion in assets) as well asmuch smaller firms that own only a handful of rigs or even a single rig. Rigscan be classified into two types: those that rest on the ocean floor (shallow-water rigs), and those that float while drilling (deepwater rigs). By far the mostcommon type of shallow-water rig is the jackup rig, which accounts for abouttwo thirds of the global rig fleet. The replacement value of a jackup is $80-100million. A standard jackup houses a crew of 25-30 workers and can drill in

3The industry has been structured this way since its inception in the 1950s. These assetsare not specific to particular tracts, except inasmuch as they are difficult and costly to move.As a result, efficient use of these assets seems to be facilitated by the independent ownershipof rigs. This allows rigs to be used on nearby tracts of different E&P companies to minimizerelocation costs, without forcing the E&P firms to do business directly with each other, whichcould be problematic from antitrust and intellectual property standpoints. For highly special-ized rigs that are more tract-specific (ultra-deepwater rigs and harsh environment rigs), E&Pfirms often make long-term contracts for the financing, construction, and operation of newrigs. This is not an issue for the relatively homogenous shallow-water jackup rigs on whichwe focus.

4

150-300 feet of water, depending on the model. Deepwater rigs that drill in upto 10,000 feet of water may cost as much as four times that much and house aslightly larger crew.A drilling rig is used only for the drilling of the well, which requires 30 to

60 days in most cases. Once the well is drilled, the rig moves on to anotherjob. Lighter-duty equipment and specialized services companies (Schlumbergerand Halliburton, for example) then move in for the installation of productionequipment. Some of this production equipment then stays semi-permanentlyfixed at the well location. Such ‘production platforms’ are typically owned bythe E&P company, unlike the drilling rig.E&P companies contract with drilling contractors under two standard con-

tract forms: dayrate and turnkey. Under a dayrate contract, the drilling con-tractor agrees to provide a staffed and functional rig for the duration of theproject, in exchange for which it receives a daily payment called the dayrate.The contract typically specifies some minimal performance benchmarks thatmust be met to avoid penalties. Commonly, for example, the driller is penal-ized for downtime in excess of one day per month. Under such a contract, thedrilling process is managed by two workers on the rig, one representing theE&P company and the other representing the driller. The E&P’s “companyman” makes a number of decisions, in consultation with the E&P’s land-basedengineering staff, about the speed of drilling, the type of bit used, the weightand viscosity of the “drilling mud” pumped down the well, and a number ofother important technical dimensions of drilling. The drilling contractor’s “toolpusher” manages the rig’s crew and the maintenance of the rig. In the Gulfof Mexico, dayrate contracts govern the drilling of more than 80% of all wells.Most dayrate contracts involving jackup rigs are one-well contracts.The alternative contract form is a turnkey contract, under which an E&P

company pays a fixed price for a well drilled to its specifications. The drillingcontractor then manages the entire drilling process (there is no “company man”)and assumes all financial risk for cost overruns and delays in the completion ofthe well. Note that this contract does not shift the exploration risk: the E&Pcompany still bears the risk of a “dry hole” and still gains all the upside shouldmajor reserves be discovered. Most turnkey contracts are one-well contracts.The standard contract types (turnkey and dayrate) in offshore drilling closely

mirror classic fixed-price and cost-plus contracts. The fixed-price contract hasstrong incentive properties, but high contracting costs due to both (1) the needto specify the project relatively completely ex ante, and (2) the inefficienciesassociated with the recontracting process that will arise if the principal wishesto alter the specifications of the project after the initial contract is agreed to.The cost-plus contract has lower contracting costs on both counts, but providesvery weak incentives for the agent to provide effort to reduce costs.Typically, the E&P company’s staff formulates an exploration and devel-

opment plan and decides which of the planned wells should be drilled undereach type of contract. Next, they determine the drilling contractors likely tohave available rigs in the general area. They then solicit bids from a handful ofdrilling contractors and evaluate these bids based on rig capabilities, the rig’s

5

safety record, price, and other considerations.

3 Data and Hypotheses

Our data come from Offshore Data Services (ODS), a Houston-based firm thatgathers and disseminates data on offshore drilling rigs. E&P companies buy thisdata to track rig availability and aid in soliciting bids on projects; contractorsbuy it to track competitors’ activities, including fleet additions and movementsof rigs. The Offshore Rig Locator database contains monthly observations onevery offshore rig in the competitive world marketplace; it also covers a few thatare “non-competitive,” like those owned by the Indian state-owned explorationcompany, which are used solely for its own drilling activities. The data forthe present analysis include monthly observations from January 1998 throughOctober 2000.The Rig Locator database provides data on the technical specifications of the

rig, the rig’s ownership, and the rig’s contract status. It also gives characteristicsof the well the rig is working on, including the water depth, the well type(exploratory or development) and the identity of the E&P company that controlsthe lease where the rig is working. From this, one can construct variables thatcapture E&P company and driller characteristics (e.g., total number of projects).While the data are global in scope, turnkey drilling activities are not. In only

two geographic regions (as defined by ODS) do turnkey contracts account for anon-negligible fraction of observed rig-months. As a result, the present analysisfocuses on only these two regions—the US Gulf of Mexico, and Mexican offshorewaters (which, together, we refer to as simply the Gulf of Mexico). We alsorestrict our analysis to projects using jackup rigs, which account for over threefourths of projects in these regions over this time period. The homogeneity ofthe capabilities of jackup rigs ensures that the projects we study are relativelycomparable.The unit of analysis of this study is the project (or well), as the fundamental

question we ask is what determines whether a particular well is drilled under adayrate or turnkey contract. The data, in contrast, are organized by rig-monthand include both observations on idle rigs and multiple observations for a singlewell that takes more than a month to drill. Therefore, to create observationsat the project level, we examined changes in well characteristics, and also thelength of the project, to ascertain when a new project began. First, we droppedall rig-months in which the rig’s status was not “drilling.” Second, all rigs thatwere drilling in the first month of the data were marked as new projects. Wethen assumed work on a new well began if at least one of the following conditionswas satisfied: (1) the rig appeared in the data after not appearing in the previousmonth (i.e., its previous status was not “drilling”); (2) the E&P company onwhose lease the rig was drilling changed from the previous month; (3) the waterdepth of the well the rig was drilling changed from the previous month; or (4)the well type changed from the previous month.When a rig worked on an observationally identical well for more than two

6

months, every other month was marked as the beginning of a new project sincetwo months is the typical time required to drill a well. We then dropped allobservations not deemed to mark the beginning of a new project. This left 1874projects, 17% of which were drilled under turnkey contracts.Table 1 provides some descriptive statistics on contracting in this industry.

Specifically, it provides a matrix of the 25 largest E&P companies and the 10largest drillers, and all of the projects between each pair of them in our data (i.e.,all jackup wells in the Gulf). This table shows that virtually all E&Ps use awide range of drillers and drill at least some wells under turnkey contracts. Thenext subsection describes the variables used in the analysis, which are definedin Table 2. Summary statistics for these variables are given in Table 3.

3.1 Project Characteristics

The dummy variable exploratory is set equal to one if the well is exploratoryand zero if the well is a development well. This project characteristic is im-portant because exploratory and development wells differ dramatically in theopportunities to increase the value of the well by changing its specificationsduring drilling. Such opportunities arise most often as a consequence of re-cent advances in drilling technology that allow the underground structures andhydrocarbons to be monitored in real time (“measurement-while-drilling”) andallow this information to be used to guide the well along non-linear trajecto-ries to maximize the extraction of oil and gas (“directional drilling”). Thesetechniques are of much more importance on development wells, where efficientextraction of oil and gas is the primary objective. Thus, contracting costs arehigher for exploratory wells. As a result, we expect turnkey contracts to be rel-atively more attractive on exploratory wells. (This is derived as Observation 2in the theory model in the Appendix.) Consistent with this hypothesis, thesplit-sample means in Table 3 show that exploratory wells account for a largerproportion of turnkey wells than of dayrate wells.The other project characteristic is water depth, which measures the water

depth, in hundreds of feet, at the well site. Over time, the Gulf has been exploredfrom the shallower coastal waters out towards the deeper waters at the edge ofthe Outer Continental Shelf (OCS), and ultimately beyond. As a result, geolog-ical conditions are much more well-understood in shallower waters; moreover,much of this knowledge becomes public through the government-mandated pub-lication of “well logs” within a certain number of years after drilling. Because ofthe relatively poorer information that is available, there is greater uncertaintyon deeper water wells, leading both to more complex contingencies that mustbe specified ex and to a greater likelihood of an opportunity to profitably rene-gotiate the initial contract. Thus, contracting costs vary with water depth, withlarger values of water depth corresponding to higher contracting costs. As aresult, we expect larger values of water depth to lead to less turnkey contracting.(This is derived as Observation 3 in the theory model in the Appendix.) Con-sistent with this hypothesis, the split-sample means in Table 3 show that waterdepth is slightly higher for dayrate wells, though the difference is small.

7

3.2 Company & Market Characteristics

E&P company scale is the number of rig-months of drilling on all of a particu-lar E&P company’s leases around the world over the 34 months covered in thepresent data. It measures the overall scale of the E&P company’s drilling ac-tivities and is as a time-independent constant for each E&P company. SmallerE&P companies may prefer turnkey contracts for a number of reasons, includingshifting risk to more risk-tolerant drillers and avoiding the scale diseconomiesinherent in building up an engineering staff to manage a small number of drillingprojects.4 Based on analyses of E&P financial characteristics and of the in-teraction of E&P scale with project characteristics, Corts (2002) argues thatthe latter of these explanations is more plausible. In either case, we expectlarger values of E&P company scale to lead to less turnkey contracting. This isconsistent with the split sample means in Table 3.We define driller scale similarly. Larger drilling contractors may prefer

turnkey contracts for similar reasons to smaller E&P companies: larger drillersmay be more willing to accept the risk inherent in a turnkey contract and mayalso be better able to take advantage of scale economies in drilling management.5

It is also possible that larger drilling contractors have more well-established rep-utations; to the extent that reputations substitute for strong formal contracts,this would lead larger drillers to dayrate contracts. Since we believe the scaleeffects to be quite strong for both E&Ps and contractors, we expect larger valuesof driller scale to lead to more turnkey contracting, unless reputation effects ofthis sort are extremely strong.Dayrate is the average dayrate in the US Gulf of Mexico region for the cur-

rent month.6 Low dayrates in the market are likely to exacerbate incentiveproblems under dayrate contracts by lowering the value of the driller’s outsideoption (i.e., a driller who finishes a job quickly either gets very low rates ona new dayrate project or idles the rig while looking for a new job). It is alsopossible that contractors cut turnkey margins in times of low dayrates and uti-lization, making turnkey contracts more attractive (the total turnkey price willreflect lower market dayrates, in addition to the possibly lower turnkey margin).Either of these rationales implies that we expect higher values of dayrate to leadto less turnkey contracting. The split-sample means are consistent with thisclaim, though the difference of values for the two contract types is not too large.

4Including considerations like scale economies is consistent with recent papers—Lafontaineand Masten (2002) and Oyer (2002)—that emphasize that many considerations besides incen-tive and contracting costs may be important in determining optimal contractual form. Whilewe include these controls and acknowledge the importance of such considerations, we empha-size incentives and contracting costs because we think these considerations are the most likelyto be affected by relationships and repeated interaction, the effect of which is the centralconcern of our paper.

5According to industry sources, specialization of the geological and engineering expertiserequired to manage a drilling project implies a minimal staff of 3-6 professionals; keeping themfully employed requires 8-12 projects a year.

6We also used an alternate measure of market conditions—the monthly rig utilization rate inthe Gulf of Mexico. This measure is highly correlated with dayrate and gives almost identicalresults in the regressions. We therefore report only results using dayrate.

8

3.3 Relationships

We constructed a variable relationships to measure the intensity of the rela-tionship between a particular E&P company and a particular contractor in aparticular month. We define relationships as the number of projects any par-ticular E&P company-contractor pair have worked on together in the previoussix months.7 Naturally, this variable is undefined during the first six months ofdata, so all analysis involving relationships is based upon the data from July1998 (month 7) through October 2000 (month 34).Relationships is the main variable of interest, since we seek to explore the

impact of repeated contracting on contract choice. While many papers sug-gest that relationships can help to solve incomplete contracting problems, whatis not clear from this literature is whether this relationship should primarilymitigate incentive problems (making weak-incentive dayrate contracts more at-tractive) or contracting costs (making high-contracting cost turnkey contractsmore attractive). Where the primary effect of relationships resides will there-fore determine whether relationships and high-powered formal contracts aresubstitutes or complements, but in principle either case can prevail. (This isderived as Observation 1 in the theory model in the Appendix.) As a matterof definition, if larger values of relationships lead to less turnkey contracting,then relationships and high-powered formal contracts are substitutes; if largervalues of relationships lead to more turnkey contracting, then relationships andhigh-powered formal contracts are complements. The split-sample means in Ta-ble 3 show that the mean value of relationships for dayrate wells (5.7) is muchhigher than that for turnkey wells (3.0), which is suggestive of the finding thatrelationships and high-powered formal contracts are substitutes.In fact, this substitutes/complements framework suggests an additional hy-

pothesis having to do with the interaction of relationships with project charac-teristics. This is derived as Observation 4 in the theory model of the Appendix,but is easily grasped graphically. Figure 1 plots the net gain of adopting aturnkey contract over a dayrate contract (incentive gains less contracting costs)—which can be thought of as “probability of choosing a turnkey contract”—againstthe severity of contracting costs. As we argued above, this corresponds to projectheterogeneity in the exploratory and water depth dimensions, with exploratoryand shallower water wells lying further to the left in this graph. In the absenceof a relationship, the net gain to a turnkey contract is a downward sloping curve,since more severe contracting costs lower the gain to using a turnkey contract.The graph is drawn under the presumption that relationships mitigate contract-ing costs (in addition to mitigating incentive problems). Thus, in the presence

7As a robustness check, we also ran a subset of our regressions using a definition for rela-tionships based on shorter as well as longer time windows than 6 months. All our qualitativeresults remained unchanged. Also, a forward looking game theoretic model might suggestlooking at interactions in the near future rather than the recent past. However, we foundthat such a definition had much less explanatory power. One explanation could be that theactual future realizations were a poor proxy of the expected future interactions at the time ofcontract, and that past relatioships provide a better proxy of the expectations regarding thefuture.

9

of a strong relationship this line is flatter, as more severe contracting costs donot undermine the attractiveness of a turnkey contract as quickly. Note that tothe left of the intersection of these lines, relationships and high-powered formalcontracts are substitutes—a stronger relationship takes one to the lower line—andto the right they are complements.8 In addition, if relationships and high-powered formal contracts are substitutes, relationships have the largest effect onthe attractiveness of turnkey contracts precisely on the types of project (here,low contracting cost projects at the far left) where turnkey contracts are moreprevalent. The reverse relationship holds if they are complements.

3.4 Geographic Sub-Regions and Related Variables

The US Gulf of Mexico is divided into a number of sub-regions for purposesof tract leasing and management. The sub-regions of the Outer ContinentalShelf (OCS) are irregularly sized and shaped, following the natural contours ofthe shoreline and the OCS, while deepwater tracts are rectangular. The jackuprigs on which we focus work only on OCS tracts. While there are a few smallsubregions very near the coast, most of the 25 OCS regions are “slices” ofthe Gulf running from the coast to the edge of the OCS. Thus, most regionsencompass a wide range of water depths. Roughly, the sub-region of the Gulfthat a well is in indicates where it lies along the coast, while the water depthindicates its distance from shore.While we do not observe the exact location of the rig, we do know its sub-

region within the Gulf. We use this to construct measures of the characteristicsof drillers that have rigs in the region local to a particular project. Since mov-ing rigs long distances is expensive and time-consuming, this provides a way ofidentifying likely winners on particular projects and determining their charac-teristics. In fact, 56% of the rigs in our data stay in the same sub-region fromone project to the next.Specifically, we define two new variables as the expected value of relationships

and driller scale, respectively, that would obtain if the E&P company in questionwas randomly matched with one of the rigs that was already in the sub-regionin the previous month. We then use these as instruments for the characteristicsof the winning contractor in our IV specifications.

4 Empirical Analysis

This discussion suggests a natural way to proceed with the empirical analysis:apply a standard discrete choice model like logit or probit to the contract choiceproblem, controlling for the characteristics of the E&P company and the driller,controlling for observed project characteristics like water depth and well type,

8The intersection is interior to the graph if the vertical axis intercept of this curve shiftsdown with relationships. This occurs if relationships also mitigate incentive problems, thusundermining the advantage of turnkey contracts.

10

and including some measure of relationships. Subsection 4.1 presents resultsfrom this straightforward approach as a baseline.This simple approach assumes that the contractor is known with certainty

before the contract type is determined, as it treats driller-specific characteristicslike driller scale and relationships as exogenous explanatory variables. In fact,however, the timing in this industry seems to be inconsistent with this, as bidsfor a particular contract type are solicited from numerous contractors before adriller is chosen. Since (1) the type of contract affects which contractor wins thebidding, and (2) the E&P companies’ information regarding the identity of thelikely winning contractor (which is unobservable to us as econometricians) af-fects the choice of contract type, the driller and contract type should be treatedas simultaneously determined. It is also plausible that project characteristicsunobserved by the econometrician but observed by the E&P companies giverise to the “endogenous matching” problem documented by Ackerberg and Bot-ticini (2002) in similar models. Subsection 4.2 addresses these issues throughinstrumental variables estimation.

4.1 Contractor Determined before Contract Type

The most straightforward way to estimate the effect of relationships on contractchoice is to assume that the particular E&P company and driller who ultimatelysign a contract already know they will work together prior to the determina-tion of the type of contract that will govern their relationship. Though it seemsinconsistent with the actual timing in the industry (where E&P companies typ-ically ask for formal bids after having decided whether they want to execute theproject as dayrate or turnkey), this specification provides a useful baseline andis directly comparable to the approach employed in virtually all the existingliterature.

4.1.1 Empirical Model

Let t represent the contract type, where t = 0 represents a dayrate contractand t = 1 represents a turnkey contract; let X represent the vector of projectcharacteristics; let P be the vector of principal (E&P company) characteristics;and let D be the vector of driller characteristics other than the intensity ofrelationships r. (In fact, through most of the paper, D simply measures thedriller’s global scale.) Define the net gain to using a turnkey contract over adayrate contract as G(X,P,D, r) + , where is a symmetric mean zero error.The E&P company chooses the turnkey contract if G(X,P,D, r) + > 0 or,equivalently, < G(X,P,D, r). We impose a simple linear form for the net gainfunction G, i.e., G(X,P,D, r) = α0 + α1X + α2P + α3D + α4r, and assumethat has a cumulative logistic distribution F .9 This yields the standard logit

9In writing the econometric models, we omit interaction terms for simplicity of exposition,but we do include interaction terms in the actual empirical analysis. In addition to the linearform for G, we tried more complex functional forms. This did not add much to the predictivepower and did not change the qualitative results, so we have not reported those results in the

11

modelPr(t = 1) = F (α0 + α1X + α2P + α3D + α4r) .

Because of the panel nature of our data, the error may not be indepen-dently distributed or homoskedastic. In particular, choices made on projectsundertaken by the same E&P company are likely to be correlated. Therefore,in the simple pooled regression, we report Huber-White robust standard er-rors, allowing for clustering among the observations of each E&P firm to giveconservative standard errors in case the errors are not independent.We next exploit the panel structure of our data to estimate alternate mod-

els. The most conservative model would have been conditional fixed effects onE&P company-driller pairs. However, such a model is ruled out by the prac-tical consideration that we have a relatively large number of such pairs, witha small number of observations within most pairs. Therefore, we account forthe heterogeneity of the E&P companies and drillers in a number of alternativespecifications.First, we report results from random effects and conditional fixed effects logit

models for E&P company.10 These models follow from different assumptions onthe components of the error term for project j undertaken by E&P companyi. We assume an error structure of the form ij = µi + ηij , where µi is the“E&P company effect” for firm i and ηij is the error specific to project j. If weassume the µi’s are constants and the ηij ’s are independently and identicallydistributed with a logistic cumulative distribution, this yields the conditionalfixed effects logit model. If we assume the µi’s are independently and identicallydistributed draws from a common distribution and the ηij ’s are independentlyand identically distributed with a logistic cumulative distribution, this givesthe random effects logit model. We adopt the model with random effects forE&P companies as our preferred specification. It allows us to control for E&Pcompany heterogeneity and is a more efficient estimator than the fixed-effectsestimator when the random effects assumptions hold, which is supported by aHausman test. Specifically, the Hausman test fails to reject the equality of theresults from the conditional fixed effects and random effects models.Next, we turn to testing the robustness of our results to driller-specific effects.

We do this using the fixed effects model, which is more conservative. We do notbelieve random effects is an appropriate model to use for drillers since bifurcationof drillers into those who offer turnkey contracts and those who do not makesimplausible the random effects model’s assumption that the errors are drawnfrom a common distribution. As a final robustness check, we also report resultsfor a specification in which dummy variables for each driller are included in amodel with fixed effects for each E&P company.11

paper.10Because the discrete choice model is non-linear, we use conditional fixed effects (see Cham-

berlain (1984)). Whenever we use the phrase “fixed effects” in this paper, we refer to condi-tional fixed effects.11Since, in general, including dummies in a non-linear model leads to inconsistent estimates,

we do not include dummies for E&P companies in any specification. However, the inclusion ofdummies for drillers may be less problematic since the number of drillers is small and grows

12

4.1.2 Empirical Results: Regressions without Interactions

The results of empirical analysis based on the above specification are reported inTables 4 and 5. Table 4 reports the results from all the models mentioned abovewithout any interactions among variables, while Table 5 reports the resultswith interactions among various variables, using our models of choice for E&Pcompany (random effects) and driller (conditional fixed effects) respectively.In Table 4, column (i) is the simplest possible logit model, regressing turnkey

on relationships alone. Recall that, since we define the relationships variableas the number of times the two parties to a contract have interacted in theprevious six months, we must omit the first six months of data from this analysis.Therefore, only 1476 of the total of 1874 observations are used.The coefficient on relationships is negative and significant, indicating that

stronger relationships seem to reduce reliance upon turnkey contracts. Thestandard errors have been corrected for heteroskedasticity as well as possiblecorrelation within each E&P company. In order to get a sense of the magnitudeof this effect, we calculate the marginal effect at the mean value of relationships,and find that an increase of relationships by 1 leads to a decrease in probabilityof choosing a turnkey contract by about 1.8%. This is calculated at the meanvalues of all variables except driller scale, which is held at its median.12 Thisconvention is followed throughout the analysis for the computation of marginaleffects, which are presented near the bottom of the tables.Column (ii) repeats the regression from column (i), but now includes vari-

ables to control for the well type (exploratory vs. development), the water depthof the well, the average dayrate for the current month, and the scale of the E&Pcompany and the driller. Column (iii) includes the same set of independentvariables in a specification with random effects for each E&P firm.Column (iii) is our preferred model, so we discuss the results in some detail.

The coefficient on relationships is negative and significant, with a marginal ef-fect of around 1.4%. Additionally, we find that the coefficient on exploratory ispositive and significant, indicating that exploratory wells are more likely to bedrilled under turnkey contracts. The marginal effect of going from a develop-ment well to an exploratory well, other things being equal, is a 13.2% increase inthe probability of a turnkey contract. This positive effect reinforces the simpleobservation from the summary statistics that turnkey contracts are more preva-lent in exploratory wells than in development wells. It is also consistent withthe hypothesis in section 3.1 that suggests that the contracting costs inherentin turnkey contracting are especially severe for development wells.The effect of an increase in water depth is negative, though not significant.

slowly. (We effectively have dummies for the universe of drillers, whereas there are many moreE&Ps for which observations might be added as the sample size is hypothetically expanded.)Hence it can be argued that the estimator’s desirable asymptotic properties hold and thatdriller dummies do not cause inconsistency in the coefficients of interest.12Driller scale is held at its median because of its highly skewed distribution and its over-

whelmingly strong influence on contract choice when it is small in value (small drillers simplydo not do turnkey projects). Results remain statistically significant, though slightly smallerin magnitude, when driller scale is held at its mean too.

13

Recall that a significant negative effect would have been consistent with thehypothesis in section 3.1 that the increasing complexity of deeper water wellsexacerbates the contracting costs problem associated with turnkey contracts.The marginal effect on probability of a turnkey contract is a 1.5% decreasein probability of turnkey with a 100 feet increase in well depth, though thisis statistically insignificant. We will later see that water depth does becomesignificant when we consider its effect on development wells separately. Anincrease in the dayrate is found to have essentially no impact on the probabilityof using a turnkey contract.An increase in E&P company scale is found to decrease the use of turnkey

contracts, though the effect is statistically significant only at 10%. On the otherhand, driller scale is found to have a positive and significant effect. This sug-gests that large E&P companies may be less likely to employ turnkey contracts,while large drillers are significantly more likely to take on turnkey projects. Thisis consistent with the hypotheses in section 3.2 that emphasize that turnkey con-tracts shift both risk and certain technologically sophisticated decision-makingresponsibility to the driller. Both of these burdens are likely to be more cheaplyborne by larger firms, as larger firms are likely to be more risk-tolerant and arebetter able to achieve the scale required to keep the requisite staff of engineersfully employed in project management tasks.Column (iv) presents the results from a logit models with E&P company

fixed effects. While the first three columns use all 1476 observations from months7 through 34, the conditional fixed effects specification in column (iv) usesonly 1159 observations since it drops the E&P companies for which there is novariance in the dependent variable. The qualitative results from the previouscolumns remain essentially unchanged. Note that we do not report marginaleffects in column (iv) since the E&P company effects, which would be neededto compute the marginal effects because of the non-linearity of the logit model,are not consistently estimated in the fixed effects approach. A Hausman testfails to reject equivalence of coefficients from columns (iii) and (iv). Therefore,for reasons already discussed earlier, we choose the random effects specificationas our specification of choice in order to deal with E&P company effects. Wereturn to this specification later to examine interactions between variables inTable 5.Column (v) estimates a logit model with driller fixed effects. We do not run

a random effects specification for drillers because the large disparities in drillers’propensity to use turnkey contracts make it implausible that the driller-specificcomponents of the error are independent draws from a single distribution. Notethat column (v) uses only 791 observations as it drops the drillers for whomthere is no variance in the dependent variable. The qualitative results alreadydiscussed remain unchanged. We also use this driller fixed effects specificationlater in our discussion of interactions between variables in Table 5. Beforemoving to Table 5, however, we do a final robustness check in column (vi) bycombining conditional fixed effects on E&P companies with driller dummies.Because we now need to exclude both drillers and E&P companies that onlydo one of the two types of contracts in our sample, we are left with only 665

14

observations. However, our results that relationships decrease the use of turnkeyand that exploratory wells have more turnkey contracting both continue to hold.To summarize, all six specifications in Table 4 produce similar results. In

particular, relationships have a negative and significant impact on the fractionof projects that are executed as turnkey contracts. This indicates that, in thisindustry, relationships tend to encourage the adoption of low-powered dayratecontracts, which suggests that relationships and high-powered fixed-price con-tracts are substitutes. As we argued in section 3 and in the appendix, thisalso implies that relationships mitigate incentive costs more than they reducetransactions costs. We now use our specifications of choice, namely E&P com-pany random effects and driller fixed effects, to further explore the interactionsamong relationships, water depth, and well type.

4.1.3 Empirical Results: Regressions with Interactions

Table 5, column (i) reports the results from a logit model with random effects foreach E&P company, with an interaction term for exploratory and water depthnow included. The probability of employing a turnkey contract is still signif-icantly higher for exploratory wells, even after controlling for other exogenousvariables like well water depth (interacted with exploratory vs. developmentwells), the average dayrate, and the scale of the E&P company and driller.Note that the net effect of exploratory at the mean values of all variables (thatit is interacted with) can be read from the table as simply its direct coefficientdespite the presence of the interaction term, since means have been subtractedfrom the respective variables before interacting them. Similarly, the net effectof water depth at its mean value can be read from the table as simply its directcoefficient despite the presence of the interaction term.The new finding in this table is that interaction between water depth and

exploratory is significant. While water depth has a negative effect (significantat 10%) on the fraction of turnkey projects on an average, this effect is larger(more negative) for development wells and smaller for exploratory wells. Forease of interpretation, we have calculated the marginal effect of water depthseparately for exploratory and development wells, and reported them near thebottom of the table. We find that the effect of water depth on the use of turnkeyis negative and significant for development wells, consistent with the hypothesisin section 3.1 that the increasing complexity of deeper water wells exacerbatesthe contracting costs problems associated with turnkey contracts. Moreover,the effect turns out to be less negative (in fact, positive but insignificant) forexploratory wells. This is consistent with the hypothesis that such contractingcosts are more severe on development wells.Column (ii) is a variant of column (i) that now considers the differential effect

of relationships on turnkey usage on exploratory and development wells; thisspecification adds an interaction term for relationships and exploratory. Again,the respective means are subtracted before interacting the two so that the directcoefficients on the original variables can be interpreted as net coefficient valuesat the means. The effect of relationships on the use of turnkey contracts is

15

negative and significant for both exploratory and development wells. However,the effect is much stronger in the case of exploratory wells. Therefore, the impactof relationships on contract choice is largest exactly where turnkey contracts aremost attractive to begin with, consistent with the hypothesis of section 3.3.As the initial regressions show, exploratory wells have more turnkey projects

than development wells, presumably due to more severe contracting costs on de-velopment wells. If relationships then act as substitutes for high-powered formalcontracts and help to solve incentive problems, it will be most attractive to sub-stitute relationships for high-powered formal contracts on those projects wherehigh-contracting cost turnkey contracts were most in use, i.e., on exploratorywells.13 The calculated marginal effect of relationships, shown near the bottomof the table, is negative and significant for both exploratory and developmentwells, with the magnitude being higher for development wells.Figure 2 uses the estimated coefficients from column (ii) to plot the curve

between probability of choosing a turnkey contract and the degree of relation-ship separately for exploratory and development wells. The curve is downwardsloping for both kinds of wells. Additionally, its starting point is much higherfor exploratory wells and it also has a steeper downward slope. This illustratesthat relationships have the biggest impact on exploratory wells, which are muchmore likely to use turnkey contracts in the absence of relationships; however,both types of wells are drilled almost exclusively under dayrate contracts oncerelationships are sufficiently well-established.Column (iii) repeats the above analysis with the addition of an interaction

term between well depth and relationships. Similar logic to that just describedfor the relationships*exploratory interaction suggests that relationships shouldhave a larger effect on the choice of turnkey for shallower wells, where turnkeycontracts are more in use. However, the coefficient is insignificant.We also check the robustness of this subsection’s results to driller-specific

effects by repeating the analysis of columns (i) through (iii) using conditionalfixed effects for driller. The outcome is reported in columns (iv) through (vi).All the results discussed above are robust to controlling for driller effects.

4.2 Contract Type and Driller Determined Simultaneously

The model in section 4.1 is technically correct only under the assumption thatthe contract type and driller choice decisions are made sequentially by the E&P

13The fact that, even though relationships and formal contracts are substitutes, the highestmarginal effect of relationships occurs exactly where formal contracts are most used mightseem counter-intuitive. However, it is consistent with the model presented in the Appendix.Using the notation of the Appendix, assume α(r) = β(r) = e−r for ease of exposition. Thenthe gain to employing a turnkey contract is G = [s(e∗) − cee∗ − (1 + p(x, d))k]e−r, and themarginal effect of relationships on the attractiveness of turnkey contracts is ∂G

∂r= −[s(e∗)−

cee∗ − (1 + p(x, d))k]e−r . From this it follows that if high-powered formal contracts andrelationships are substitutes (∂G

∂r< 0), then relationships should have the largest impact on

contract choice for the types of projects that have the highest turnkey usage. To see this, notethat the well characteristics x and d that maximize the bracketed expression simultaneouslymaximize G and minimize ∂G

∂rsince ∂G

∂r= −G in our stylized model.

16

company. In particular, since driller characteristics (scale and relationships) aretreated as exogenous explanatory variables, that model assumes that the drilleris chosen first and then type of the contract. This section allows the possibilitythat the contract type and driller are determined simultaneously.Even if these decisions are not literally made at the same time, this is a

more appropriate model either if the decisions are made through an iterativeprocess involving informal discussions, renegotiation, and rebidding or if theE&P company has information (that we as econometricians do not observe)that influences its expectations of how the second decision will be affected byits choice in the first decision. In either case, the models of section 4.1 would beplagued by a correlation of the error with the variables of interest, inducing abias in our estimates. In this section we use instrumental variables to estimatea simultaneous choice model that addresses these concerns. In addition, theinstrumental variables approach also corrects for bias that may be induced by“endogenous matching” of agents to projects, which in general induces a prob-lematic correlation between the contract choice error and driller characteristicsif the matching between the E&P companies and the drillers is not random buta function of unobserved project characteristics.

4.2.1 Empirical Model

Consider a version of our empirical model in which the E&P company endoge-nously determines the contract type and the driller, and where the only rele-vant characteristic of the driller is the intensity of its relationship with the E&Pfirm.14 In reduced form, the choice of driller can therefore be equivalently seenas directly choosing the relationship level r. For ease of exposition, we simplifyby assuming that the variables are continuous rather than discrete and thattheir respective structural equations are linear. This simplified model yields atwo-equation simultaneous system.

t = β0 + β1X + β2P + β3r +

r = γ0 + γ1X + γ2Z + γ3t+ η.

Here, Z represents a vector of the characteristics of potential winning drillers;this affects the principal’s choice of driller on a particular project, but does notdirectly figure into the contract choice. This might include, for example, thenumber of rigs that particular drillers have available nearby with appropriatetechnical specifications. The errors and η are assumed to be uncorrelated withX, P , and Z. Note that if turnkey contracts and relationships are substitutes,then β3 and γ3 are both negative; if complements, both are positive.First consider the case in which and η are not correlated with each other;

for reasons that will become clear, we refer to this as the case of no endogenous

14In fact, in the model we estimate, driller scale is a second endogenous driller characteristic.We describe below how we handle this; for ease of exposition, we focus first on the model withone endogenous driller characteristic.

17

matching. The only problem with directly estimating the first equation is that,because of the endogeneity induced by the second equation, is not uncorrelatedwith r. In particular, r contains γ3t, which is negatively correlated with whenturnkey contracts and relationships are substitutes. This imparts a negativebias to β3, i.e., makes it more negative.

15

Now consider the further complication that arises if and η are correlatedas well. Ackerberg and Botticini (2000) term this “endogenous matching,” em-phasizing that unobserved project characteristics are likely to induce matchingof agents to projects in a way that induces bias in the coefficients. For example,suppose there is a well that the principal expects to have especially severe in-centive problems for reasons unobservable to the econometrician. In this case,the principal is likely both to choose a high-r driller and to do the job undera turnkey contract; that is, and η are positively correlated. Now direct es-timation of the contract choice equation suffers from an additional source ofcorrelation of r and . Specifically, r contains both γ3t, which is negativelycorrelated with when turnkey contracts and relationships are substitutes, andη, which is positively correlated with . Thus, it is impossible to unambigu-ously sign the bias when turnkey contracts and relationships are substitutes,since simultaneity as described above imparts a negative bias, but endogenousmatching imparts a positive bias.16

In order to address these problems of simultaneity and endogenous matching,we use instrumental variables techniques to eliminate the correlation betweenr and . Note that in fact the problem is more complex than the two-equationmodel above, since D—which we proxy by driller scale—is also endogenous. Wetherefore need instruments for both types of agent characteristics: variablesthat are correlated with the characteristics (the scale and the relationships) ofthe winning driller, but uncorrelated with contract choice (except through thiseffect on r and D).Specifically, we use two instruments: weighted averages of the two driller

characteristics (relationships and driller scale), where the weights are given bythe number of rigs present in the relevant sub-region of the Gulf in the previousperiod.17 The validity of these instruments relies on the fact that, becauserigs are costly to move, the choice of contractor depends on which contractorsalready have their rigs in the same local region as this well (the correlationof relationships with its instrument thus constructed is 0.65). Further, thepresence of these contractors should not affect the choice of optimal contractexcept to the extent that each of these is more likely to be chosen because oftheir proximity. That is, the characteristics of drillers with rigs in the region

15In contrast, if turnkey contracts and relationships are complements, r is positively corre-lated with , which imparts a positive bias to β3. We emphasize the result for the substitutescase, since this is the case indicated by our data.16In contrast, when turnkey contracts and relationships are complements, both sources of

correlation between r and induce positive bias. In either case, the effect of endogenousmatching itself is to create a positive bias in the measure of the effect of relationships onturnkey choice.17When no drillers had rigs in the relevant sub-region in the previous period, all drillers in

the data were assigned an equal weight instead.

18

should not affect the choice of contract type except through an effect on theexpected characteristics of the winning driller.While we would ideally estimate a discrete-choice model with instrumental

variables, estimation of our preferred random-effects specification (or even arobust standard errors specification) in an instrumental variables logit or probithas proved intractable. We therefore estimate an instrumental variables linearprobability model, which does allow us to incorporate random effects for E&Pcompanies, as in our earlier (non-IV) specifications.

4.2.2 Empirical Results

Table 6 reports the results from regressions using the instrumental variablesdescribed above. Columns (i) and (iii) report two basic linear probability modelswith random effects for E&P companies and without instrumental variables, firstwith simply relationships included and then with this measure also interactedwith exploratory and water depth. These regressions parallel columns (i) and (iii)respectively from Table 5 for the logit case. They establish the baseline resultsfor the linear model to provide a point of comparison for the IV results, since thecoefficients from the linear IV regressions cannot be directly compared to thediscrete choice models from previous tables. Note that all of the basic resultsfrom the earlier regressions are preserved: the probability of employing a turnkeycontract decreases with relationships, increases for exploratory wells, decreaseswith water depth for development wells, decreases in E&P company scale, andincreases in driller scale. In addition, the effect of relationships is significantlymore pronounced for exploratory wells, as in previous regressions. It should benoted that the estimated marginal effects from these linear regressions also turnout to be comparable to the effects computed in earlier tables.Columns (ii) and (iv) present the corresponding instrumental variables re-

gressions. The basic qualitative results described above persist in both cases.However, in both cases, instrumenting for relationships and driller scale makesthe coefficient on relationships more negative. This indicates that simultane-ity and endogenous matching had, on balance, imparted a positive bias to ourcoefficients. Following from the discussion above, this implies that endogenousmatching was the dominant econometric problem we were encountering, as thatis the only source of positive bias when relationships and high-powered formalcontracts are substitutes.

5 Conclusion

Fixed-price and cost-plus contracts embody a trade-off between contractingcosts and incentive problems. We argue that relationships may help to solveeither of these problems and may therefore serve as either a substitute or com-plement for strong formal (fixed-price) contracts, depending on which problemthey mitigate more effectively. Our empirical analysis shows that, in the offshoredrilling industry, stronger relationships lead to greater adoption of cost-plus

19

contracts (which have poorer incentives but lower contracting costs), suggestingthat relationships work primarily to mitigate incentive problems and thereforeact as substitutes for strong formal contracts. This finding becomes even morepronounced when instrumental variables are used to account for simultaneityand endogenous matching of drillers to projects.While we have focused on the determinants of contract choice and the ef-

fects of repeated contracting within this particular industry, it is interesting toconsider the evidence across industries provided by a review of the relevant em-pirical literature. Table 7 summarizes the results of four studies (including ours)that provide evidence on this question in different industries. The first columnof numbers shows significant variation in the use of fixed-price contracts acrossindustries. The last column of numbers in Table 7 demonstrates that the use offixed-price contracts goes hand in hand with the level of repeated interaction inthe various industries: when more than two thirds of contracts are repeat busi-ness (in offshore drilling and aircraft engines), fewer than half of contracts arefixed-price. Thus, the between-industry evidence, while crude, seems broadlyconsistent with the within-industry evidence provided in this paper: frequentinteractions go hand-in-hand with a reduced incidence of fixed-price contracts.Table 7 also shows that only one industry (IT services) exhibits within-industrycomplementarity between relationships and fixed-price contracts.These differences may be explained by differences in the technological condi-

tions in these industries. Project requirements for IT services and the relativelystandard software tasks outsourced to Indian contractors are likely to be simplerto specify ex ante than those for offshore drilling and aircraft engine projects.18

This could explain why fixed-price contracts are more prevalent in these indus-tries. These four industries also differ in the degree of project homogeneity.While IT services like network, storage, and mainframe maintenance are fairlyroutinized and do not change over time, each offshore well and each softwareproject is more idiosyncratic, less similar to the projects that came before, andmore in need of real-time adjustments to specifications in response to new in-formation. Thus, the argument that relationships may facilitate the writing ofcomplex fixed-price contracts may apply most clearly to IT services, where thestability of the technological specifications over time allows firms to apply learn-ing from the past to projects in the future. This could explain why only in thatindustry are relationships and fixed-price contracts found to be complements.There remain a number of questions that we have not addressed in this

paper, which should provide fruitful areas for future research. In this paper, wehave considered only the E&P company’s static problem of contract and drillerchoice, but not the dynamic effects these choices have on future contracts. Moregenerally, we should view the driller selection process as a combination of makingthe optimal choice for the current project and also a deliberate cultivation ofrelationships with drillers for long-run benefit.

18While, in general, many software projects might be difficult to specify ex ante, US firmstypically choose to contract with Indian software firms only for relatively simple and standard-ized tasks. Hence, the results from Banerjee and Duflo (2000) might not be representative ofthe software industry in the US.

20

While we have established that relationships do affect contract choice, de-termining the exact mechanism through which this happens remains an openproblem. In particular, at least two different types of economic models wouldbe consistent with our description. One set of models involves forward-lookinggame theoretic reasoning wherein a threat of future punishment is used to sus-tain a cooperative outcome under symmetric information. If we are willing to ac-cept past relationship frequency as a proxy for expected future relationships, thisclass of models is consistent with our empirical results. However, another classof economic models that can give similar predictions is the backward-lookingreputation-building models (e.g., Kreps and Wilson (1982) and Milgrom andRoberts (1982)) in which there is a small fraction of the players that are inher-ently “crazy” (i.e., not opportunistic) so that other players find it beneficial tobuild a reputation for being “crazy.” Beyond these two classes of economic mod-els, there are cognitive and sociological explanations of the meaning of “trust”in repeated relationships that also yield similar predictions regarding the effectsof relationships on contract choice. Distinguishing these empirically remains atask for future research.

6 References

Ackerberg, Daniel and Maristella Botticini. 2002. “Endogenous Matchingand the Empirical Determinants of Contract Form,” Journal of Political Econ-omy, forthcoming.Bajari, Patrick and Steven Tadelis. 2001. “Incentives versus Transaction

Costs: A Theory of Procurement Contracts,” Rand Journal of Economics, 32,387-407.Baker, George, Robert Gibbons, and Kevin Murphy. 1994. “Subjective

Performance Measures in Optimal Incentive Contracts,” Quarterly Journal ofEconomics 109: 1125-1156.Banerjee, Abhijit and Esther Duflo. 2000. “Reputation Effects and the

Limits of Contracting: A Study of the Indian Software Industry,” QuarterlyJournal of Economics 115: 989-1017.Bull, Clive. 1987. “The Existence of Self-Enforcing Implicit Contracts,”

Quarterly Journal of Economics 102:147-59.Chamberlain, G., 1984. “Panel Data,” in Handbook of Econometrics, vol.

II, edited by Z. Griliches and M. Intriligator, pp. 1247-318. Amsterdam: North-Holland.Corts, Kenneth. 2002. “Fixed-Price vs. Cost-Plus: The Determinants of

Contractual Form in Offshore Drilling,” working paper.Crocker, Keith and Kenneth Reynolds. 1993. “The Efficiency of Incomplete

Contracts: An Empirical Analysis of Air Force Engine Procurement,” RandJournal of Economics 24, 126-46.Granovetter, Mark. 1985. “Economic Action and Social Structure: The

Problem of Embeddedness.” American Journal of Sociology 91: 481-510.

21

Gulati, Ranjay. 1995. “Does Familiarity Breed Trust? The Implications ofRepeated Ties for Contractual Choice in Alliances,” Academy of ManagementJournal 38: 85-112.Klein, Benjamin. 1996. “Why Hold-ups Occur: The Self-Enforcing Range

of Contractual Relationships.” Economic Inquiry 34: 444-63.Klein, Benjamin and Keith B. Leffler. 1981. “The Role of Market Forces in

Assuring Contractual Performance,” Journal of Political Economy. 89: 615-641.Kreps, David. 1990. “Corporate Culture and Economic Theory.” In J. Alt

and K. Shepsle, eds. Perspectives on Positive Political Economy. CambridgeUniversity Press.Kreps, D. and R. Wilson. 1982. “Reputation and Incomplete Information,”

JET 27, 253-279.Lafontaine, F. and S. Masten. 2002. “Contracting in the Absence of Specific

Investments and Moral Hazard.” working paper.Mayer, K. and A. Kalnins. 2002. “Relationships, Incentives, and Measure-

ment: An Analysis of Contract Choice in Information Technology,” workingPaper.Milgrom, P. and J. Roberts. 1982. “Predation, Reputation, and Entry

Deterrence,” JET 27, 288-312.Oyer, Paul. 2002. “Why Do Firms Use Incentives That Have No Incentive

Effects?” working paper.Poppo, Laura and Todd Zenger. 2001. “Substitutes or Complements? Ex-

ploring the Relationship between Formal Contracts and Relational Governance,”working paper.Sabel, Charles. 1993. “Studied Trust: Building New Forms of Coopera-

tion in a Volatile Economy.” Chapter 4 in R. Swedberg (ed.), Explorations inEconomic Sociology. New York: Russell Sage Foundation.Williamson, Oliver. 1985. The Economic Institutions of Capitalism. The

Free Press.

22

7 Appendix: A simple model

In this section we develop a simple model that elucidates the role of incentiveproblems and contracting costs in determining the effect of relationships oncontract choice. We then derive our empirical hypotheses from this model.The E&P company contracts with a predetermined driller to drill a well at

water depth x.19 The completed well is worth v. The cost of drilling the well ist− s(e), which includes the opportunity cost of the rig’s time plus various otherinput costs (pipe, mud, fuel). Here, s(e) reflects the cost savings generatedby the driller’s choice of an effort e ≥ 0 at a cost ce.20 The driller exertsthis effort immediately after the contract is signed and before any subsequentrenegotiation of the contract.21 We assume s0(0) = ∞, s0(e) > 0, s00(e) < 0,and lime→∞ s0(e) = 0, which together ensure an interior effort level is jointlyefficient. We also normalize s(e) for simplicity by assuming that s(0) = 0.Under a turnkey contract, all input costs and rig opportunity costs are borne

by the driller, so these costs t−s(e) and also the cost of effort ce enter negativelyin the expression for the driller’s payoff. Under a dayrate contract, the drillerbears the opportunity cost of the rig as well as the cost of effort; the E&Pcompany bears the other input costs, and also pays a dayrate to the driller.We assume that the rig rental market is competitive in the sense that all E&Pcompanies pay the market dayrate, which is in turn exactly the opportunitycost of the rig (since another contract at the market dayrate is the foregoneopportunity).Subsequent to the signing of the contract and the determination of the effort

level, an opportunity may arise, with probability p(x, d), to increase the value ofthe well by an amount u by changing the project specifications from those of theoriginal contract. The argument d represents an indicator variable that is equalto one if the well is a development well and zero if it is exploratory. As describedin section 3, we expect development wells and wells in deep water to be moreprone to such a desire to recontract. Thus, with a slight abuse of notation,pd > 0, in the sense that p(x, 1) > p(x, 0), and px > 0. In addition, this effectis especially important for development wells, where directional drilling is morelikely to be employed; thus, pxd > 0.Dayrate and turnkey contracts differ in the severity of the contracting and

recontracting costs. Since the E&P company need not delineate project speci-

19For expositional simplicity, we model the E&P company’s choice between turnkey anddayrate contracts for a particular well with a particular driller. Our empirical models doaddress the issues that arise from the endogenous matching of drillers to projects.20For simplicity, the cost and benefit of the driller’s private effort are assumed to be inde-

pendent of well characteristics. Allowing s(e) and c to vary with the well characteristics (likewater depth and well type) would complicate our analysis; however, we do not believe thisdependence to be particularly important in practice. Allowing v and t to vary with thesecharacteristics is straightforward and does not change our results.21This could be generalized to allow the driller to adjust the effort level after renegotia-

tion, but that would leave the basic insights of the model unchanged, since two contractsare so starkly different. The driller would still choose the (new) efficient effort level afterrenegotiation under a turnkey contract and would continue to exert no effort under a dayratecontract.

23

fications ex ante, dayrate contracts are costless to write and never require re-contracting. Should opportunities arise (as they do with probability p(x, d)) forimprovements in the well specifications, the E&P company simply orders thosechanges made. In contrast, a turnkey contract costs k to write initially, as a fullset of engineering specifications must be prepared and the firms must contractover various contingencies having to do with geological formations encountered.In addition, changing the project specifications under a turnkey contract is alsoexpensive; for simplicity, we assume that renegotiation requires incurring thefull cost k again. We assume that k < u, so that there is always a net gain torenegotiation.Under a dayrate contract, the E&P company’s payoff is v + p(x, d)u −(t −

s(e)); the driller’s payoff is −ce. Note that, since we assume that the dayratepaid to the driller reflects exactly the market dayrate (and therefore the oppor-tunity cost of the rig), the driller’s revenue from rig rental and its opportunitycost exactly cancel each other out. The important consequence of this is thatthe driller does not benefit from reductions in the number days required, or fromany other sort of cost saving, and therefore exerts the minimal effort edr = 0.Thus, the joint payoff under a dayrate contract is πdr = v + p(x, d)u −t.Under a turnkey contract, in contrast, the E&P company’s payoff is v− z−

k + p(x, d)(u − k), where z is the fixed turnkey fee. The driller chooses e tomaximize its payoff z − (t − s(e)) − ce. As a result, the driller always choosesthe efficient effort level e∗ defined implicitly by s0(e∗)− c = 0. Finally, the jointpayoff under a turnkey contract is πtk = v−k−(t−s(e∗))−ce∗+p(x, d)(u−k).If there were, hypothetically, neither incentive problems nor contracting

costs, then the efficient effort level e∗ could be induced without incurring thecost k either up front or at the point of recontracting (which occurs with prob-ability p). This hypothetical first-best outcome would yield a joint payoff ofπ∗ = v−(t−s(e∗))−ce∗+p(x, d)u. By comparing the joint payoff under the twoalternative contracts with this first-best outcome, we can separate the inefficien-cies that arise due to incentive problems from those that arise due to contractingcosts. Specifically, turnkey contracts involve additional “contracting costs” de-noted T (x, d) = π∗ − πtk = (1 + p(x, d))k, while dayrate contracts involveinefficient effort levels leading to “incentive costs” I = π∗ − πdr = s(e∗)− ce∗.In the absence of any gains from relationships, this model implies that the

net gain from adopting a turnkey contract instead of a dayrate contract isπtk − πdr = I − T . The turnkey contract would be the preferred contractform if and only if this difference is positive, that is, when the incentive gains ofa fixed-price contract outweigh its contracting costs. We incorporate the effectof relationships by assuming that a stronger relationship, denoted by a highervalue of a variable r, mitigates both incentive costs I and contracting costs T .Since relationships need not have the same effect on both types of costs, wemodel the net gain to turnkey contracts, including the effect of relationships, asG = α(r)I − β(r)T (x). We assume that α(0) = β(0) = 1, that both α(r) andβ(r) are nonnegative for all values of r, and that both are decreasing in r.Since relationships may reduce the costs associated with both types of con-

tracts, both types of contracts approach the first-best outcome as r increases.

24

Thus, it is not clear whether increasing the strength of the relationship favorsthe choice of one specific contract type over the other. In the terminology of con-tract theory, it is not clear whether relationships and strong formal (fixed-price)contracts are complements or substitutes.Definition. Relationships and strong formal contracts are substitutes if

∂G∂r < 0. They are complements if ∂G

∂r > 0.

Note that ∂G∂r = α0(r)I − β0(r)T . Noting that turnkey contracts are chosen

if and only if G > 0 leads to our first observation. Recall that both α0 and β0

are negative.Observation 1. If incentive problems are severe enough relative to con-

tracting costs (I is sufficiently large relative to T ) or these incentive problemsare sufficiently more responsive to relationships than contracting costs (the mag-nitude of α0(r) is large enough relative to that of β0(r)), then relationships area substitute for strong formal (fixed-price) contracts. Similarly, if contractingcosts are severe enough relative to incentive problems or incentive problemsare sufficiently more responsive to relationships than contracting costs, thenrelationships are a complement to strong formal contracts.22

Expanding I and T yields G = α(r)[s(e∗)−ce∗]−β(r)[(1+p(x, d))k]. Againwith a slight abuse of notation, ∂G

∂d = −β(r)kpd (technically, this expression isthe discrete change as d goes from 0 to 1, not a derivative). Recall that pd > 0;this implies ∂G

∂d < 0, which yields the next observation.Observation 2. Turnkey contracts are less likely to be used on development

wells than on exploratory wells.Now note that ∂G

∂x = −β(r)kpx. Recall that px > 0, which implies ∂G∂x < 0.

To see how well water depth interacts with well type in determining contract

choice, note that ∂2G∂x∂d = −β(r)kpxd < 0.

Observation 3. Increases in well water depth reduce the use of turnkeycontracts, and this effect is stronger for development wells.Note that observations 2 and 3, which do not deal with the role of relation-

ships but address the direct trade-off between incentive costs and contractingcosts, are simplified versions of Proposition 1 in Bajari and Tadelis (2001). Wefix contractual completeness by assuming a particular function p(x, d), whilethey consider a more general model in which completeness in endogenously cho-sen along with contract type.To see the interaction of well type and relationships on contract choice, which

is not addressed by existing models, note that ∂2G∂r∂d = −β0(r)kpd > 0. Recall

that when relational contracts and strong formal contracts are substitutes, ∂G∂r <0, so that this positive second derivative implies that an increase in d reducesthe magnitude of ∂G

∂r .Observation 4. If relational contracts and strong formal contracts are

substitutes, then relationships reduce the use of turnkey contracts less dramati-cally on development wells than on exploratory wells. If relational contracts and

22Theoretically, it is therefore possible that relationships and turnkey contracts are substi-tutes for some values of x, p and r and complements for other values. But we do not find anyevidence of this in our data.

25

strong formal contracts are complements, then relationships increase the use ofturnkey contracts more dramatically on development wells than on exploratorywells.

26

Table 1Summary table of all E&P company-driller interactions

Each entry represents the number of wells in our data drilled by that column's driller for that row's E&P company.Number in parentheses is the number of these projects that were drilled under a turnkey contract.

Global Marine ENSCO

R&B Falcon

Noble Drilling Rowan

Diamond Offshore

Marine Drilling Pride Intl.

Parker Drilling

Cliffs Drilling Other

Total for E&P Company

Chevron 33(17) 17(0) 11(0) 28(3) 6(0) 28(0) 2(0) 8(0) 18(0) 151(20)

Vastar Resources 20(5) 8(0) 2(1) 2(2) 6(0) 36(0) 15(0) 1(1) 90(9)

Spirit Energy 76 31(2) 19(0) 15(0) 2(0) 1(0) 8(0) 8(0) 5(0) 89(2)

PEMEX 30(5) 52(6) 82(11)

Coastal O&G 22(2) 20(0) 24(0) 12(0) 78(2)

Ocean Energy 16(4) 4(0) 4(2) 6(0) 3(0) 9(0) 10(0) 52(6)

Newfield Exploration 18(17) 7(0) 7(7) 5(5) 3(0) 2(1) 9(0) 1(0) 52(30)

Samedan 4(4) 6(0) 5(0) 25(0) 1(0) 9(0) 1(0) 51(4)

Apache Corp 1(1) 7(0) 9(0) 22(0) 6(0) 4(0) 49(1)

Basin Exploration 15(15) 2(0) 8(0) 5(3) 7(0) 7(0) 1(0) 45(18)

Exxon 42(0) 42(0)

Exxon Mobil 5(0) 37(0) 42(0)

Burlington Resources 1(1) 3(0) 4(1) 11(7) 1(0) 7(0) 10(0) 4(0) 41(9)

Sonat Exploration 13(11) 5(0) 20(0) 1(1) 1(0) 40(12)

Houston Exploration 1(1) 3(0) 6(1) 10(0) 2(0) 13(0) 35(2)

BP Amoco 5(4) 12(0) 8(0) 9(0) 34(4)

Walter O&G 3(1) 5(0) 1(1) 5(0) 2(0) 15(0) 2(0) 33(2)

Stone Energy 6(1) 3(0) 4(0) 17(0) 2(0) 32(1)

Bois d'Arc 1(0) 5(3) 2(2) 4(0) 1(0) 1(0) 16(0) 30(5)

Union Pacific Res 2(1) 4(0) 6(0) 11(0) 5(0) 2(0) 30(1)

Equitable Resources 14(13) 4(0) 8(1) 2(0) 28(14)

Hall Houston 3(3) 1(0) 9(3) 6(0) 2(0) 5(0) 1(0) 27(6)

Spinnaker 5(5) 4(0) 7(0) 5(5) 4(0) 2(0) 27(10)

Anadarko 5(1) 21(0) 26(1)

Shell 1(0) 2(0) 1(0) 3(0) 1(0) 15(0) 2(1) 25(1)

Other 136(98) 132(0) 106(27) 58(22) 66(0) 21(1) 48(0) 20(0) 21(0) 25(5) 10(0) 643(153)

Total for Driller 360(207) 325(0) 242(45) 198(57) 192(0) 136(2) 135(0) 89(0) 79(0) 49(6) 69(7) 1874(324)

27

Table 2Description of variables

Variable Description

Turnkey Binary variable equal to one if well is drilled under a turnkey contract; zero if dayrate contract

Exploratory Binary variable equal to one if well the is exploratory; zero if development

Water depth Water depth at well site, measured to nearest foot, reported in hundreds of feet

Dayrate Average dayrate, in thousands of dollars, paid to drillers in the Gulf of Mexico in particular month

E&P company scale Total number of projects worldwide for particular E&P company from Jan 98 through Oct 00

Driller scale Total number of projects worldwide for particular driller from Jan 98 through Oct 00

Relationships Number of projects worldwide involving particular operator-driller pair in the preceding six months

28

Table 3Summary statistics for jackup wells in the US/Gulf of Mexico region

Number of months of data 34 (Jan 1998 to Oct 2000)

Total number of projects 1874

Number of E&P companies 127

Number of drillers 18

Project summary statistics:

Variable Obs Mean Std Dev

Turnkey 1874 0.17 0.38

Exploratory 1874 0.45 0.50

Water depth 1874 1.19 0.84

Dayrate 1874 24.53 9.81

E&P company scale 1874 112.90 113.84

Driller scale 1874 657.29 312.54

Relationships 1476 5.19 5.55

Project summary statistics by contract type:

Dayrate Turnkey

Variable Obs Mean Std Dev Obs Mean Std Dev

Exploratory 1550 0.40 0.49 324 0.66 0.48

Water depth 1550 1.20 0.84 324 1.15 0.87

Dayrate 1550 24.68 9.86 324 23.77 9.58

E&P company scale 1550 120.30 115.93 324 77.48 95.77

Driller scale 1550 605.93 311.26 324 903.01 169.17

Relationships 1210 5.67 5.75 266 3.00 3.80

29

Table 4Effect of relationships under alternate regression models

Dependent variable: turnkey i) Logit with robust

standard errors and

clustering on E&P

ii) Logit with robust

standard errors and

clustering on E&P

iii) Logit with random

effects on E&P

company

iv) Logit with conditional

fixed effects on E&P

company

v) Logit with conditional

fixed effects on driller

vi) Logit with conditional

fixed effects on E&P

company and dummies for

Relationships -0.1298 *** -0.1382 *** -0.1101 *** -0.0863 *** -0.2073 *** -0.1097 **(0.0439) (0.0498) (0.0286) (0.0285) (0.0326) (0.0429)

Exploratory 0.9895 *** 0.9809 *** 0.7604 *** 1.2520 *** 1.0625 ***(0.2021) (0.2109) (0.2198) (0.2012) (0.3003)

Water depth -0.2019 -0.1185 -0.0421 -0.3556 *** -0.0583 (0.1464) (0.1219) (0.1325) (0.1144) (0.1817)

Dayrate 0.0132 0.0000 -0.0143 0.0206 -0.0100 (0.0132) (0.0148) (0.0153) (0.0143) (0.0206)

E&P company scale -0.0011 -0.0022 -0.0003 (0.0012) (0.0013) (0.0012)

Driller scale 0.0097 ** 0.0100 *** 0.0086 *** (0.0045) (0.0011) (0.0011)

d(E[turnkey ])/d(relationships ) -0.0175 *** -0.0179 *** -0.0142 *** (0.0051) (0.0038) (0.0036)

d(E[turnkey ])/d(exploratory ) 0.1339 *** 0.1321 ***(0.0360) (0.0346)

d(E[turnkey ])/d(water depth ) -0.0261 -0.0152 (0.0171) (0.0159)

N 1476 1476 1476 1159 791 665

Notes: -Marginal effects calculated at means of all variables except Driller scale, which is held at its median.- *** = p-value<.01; ** = p-value< .05; * = p-value <.10

30

Table 5Interactions among explanatory variables

Dependent variable: turnkey Logit with random effects Logit with conditional fixedon E&P company effects on driller

(i) (ii) (iii) (iv) (v) (vi)Relationships -0.1131 *** -0.1231 *** -0.1231 *** -0.2088 *** -0.2069 *** -0.2094 ***

(0.0286) (0.0294) (0.0294) (0.0333) (0.0329) (0.0334)Exploratory 1.0146 *** 0.8391 *** 0.8384 *** 1.2827 *** 1.0949 *** 1.1145 ***

(0.2129) (0.2331) (0.2338) (0.2032) (0.2229) (0.2254)Water depth -0.2521 * -0.2419 * -0.2399 * -0.4839 *** -0.4840 *** -0.5765 ***

(0.1341) (0.1337) (0.1455) (0.1288) (0.1286) (0.1466)Exploratory * Water depth 0.8642 *** 0.8462 *** 0.8481 *** 0.9199 *** 0.9281 *** 0.8531 ***

(0.2518) (0.2523) (0.2587) (0.2408) (0.2414) (0.2466)Relationships * Exploratory -0.1052 ** -0.1055 ** -0.1204 ** -0.1098 **

(0.0509) (0.0515) (0.0547) (0.0556)Relationships * Water depth 0.0011 -0.0527

(0.0332) (0.0365)Dayrate 0.0011 0.0026 0.0026 0.0243 * 0.0227 0.0232

(0.0148) (0.0148) (0.0148) (0.0143) (0.0148) (0.0148)E&P company scale -0.0022 * -0.0023 * -0.0023 * -0.0004 -0.0008 -0.0009

(0.0013) (0.0013) (0.0013) (0.0012) (0.0012) (0.0012)Driller scale 0.0101 *** 0.0102 *** 0.0102 ***

(0.0011) (0.0011) (0.0011)

d(E[turnkey ])/d(exploratory ) 0.1323 *** 0.1011 *** 0.1010 ***(0.0337) (0.0336) (0.0337)

Exploratory wells: d(E[turnkey ])/d(relationships ) -0.0200 *** -0.0287 *** -0.0287 ***

(0.0049) (0.0063) (0.0063) d(E[turnkey ])/d(water depth ) 0.0405 0.0362 0.0366

(0.0281) (0.0255) (0.0291)

Development wells: d(E[turnkey ])/d(relationships ) -0.0100 *** -0.0066 ** -0.0066 **

(0.0027) (0.0029) (0.0029) d(E[turnkey ])/d(water depth ) -0.0560 *** -0.0535 *** -0.0534 ***

(0.0189) (0.0182) (0.0183)

N 1476 1476 1476 791 791 791

Notes:-Means have been subtracted before interacting, so uninteracted coefficients can be interpreted as overall coefficient at means.-Marginal effects calculated at means of all variables except driller scale , which is held at its median.- *** = p-value<.01; ** = p-value< .05; * = p-value <.10

31

Table 6Controlling for simultaneity and endogenous matching using

instrumental variables

Dependent variable: turnkey i) Linear regression

with random effects on

E&P company

ii) Linear regression

with random effects on

E&P company,

with IVs

iii) Linear regression

with random effects on

E&P company

iv) Linear regression

with random effects on

E&P company,

with IVsRelationships -0.0081 *** -0.0101 ** -0.0093 *** -0.0150 ***

(0.0020) (0.0047) (0.0020) (0.0051)

Exploratory 0.0988 *** 0.0909 *** 0.0943 *** 0.0797 ***

(0.0188) (0.0207) (0.0188) (0.0215)

Water depth -0.0149 -0.0080 -0.0146 -0.0069

(0.0110) (0.0130) (0.0110) (0.0134)

Exploratory * Water depth 0.0688 *** 0.0751 *** 0.0668 *** 0.0750 ***

(0.0204) (0.0213) (0.0207) (0.0226)

Relationships * Exploratory -0.0101 *** -0.0211 ***

(0.0034) (0.0070)

Relationships * Water depth 0.0000 0.0014

(0.0020) (0.0036)

Dayrate -0.0005 -0.0005 -0.0011 0.0000

(0.0011) (0.0016) (0.0011) (0.0016)

E&P company scale -0.0004 -0.0003 -0.0004 -0.0002

(0.0002) (0.0002) (0.0002) (0.0002)

Driller scale 0.0005 *** 0.0003 0.0005 *** 0.0003

(0.0000) (0.0002) (0.0000) (0.0002)

d(E[turnkey ])/d(exploratory ) 0.0988 *** 0.0909 *** 0.0943 *** 0.0797 ***

(0.0188) (0.0207) (0.0188) (0.0215)

Exploratory wells:

d(E[turnkey ])/d(relationships ) -0.0081 *** -0.0101 ** -0.0150 *** -0.0267 ***

(0.0020) (0.0047) (0.0030) (0.0076)

d(E[turnkey ])/d(water depth ) 0.0233 0.0339 * 0.0226 0.0348 *

(0.0153) (0.0181) (0.0156) (0.0194)

Development wells:

d(E[turnkey ])/d(relationships ) -0.0081 *** -0.0101 ** -0.0048 ** -0.0057

(0.0020) (0.0047) (0.0023) (0.0048)

d(E[turnkey ])/d(water depth ) -0.0455 *** -0.0414 *** -0.0442 *** -0.0402 ***

(0.0147) (0.0156) (0.0147) (0.0157)

N 1476 1476 1476 1476

Notes:-Means have been subtracted before interacting, so uninteracted coefficients can be interpreted as overall coefficient at means.-Marginal effects calculated at means of all variables except driller scale , which is held at its median.-Driller scale and relationships are treated as endogenous in IV specifications.- *** = p-value<.01; ** = p-value< .05; * = p-value <.10

32

Table 7Cross-industry comparison of contract incidence

% contracts intra-indIndustry fixed-price mixed cost-plus repeat bus comps/subs

Offshore drilling 17 0 83 86 substitutes (Corts and Singh)

Air Force engines 34 45 20 100 ---- (Crocker and Reynolds)

Indian software development 58 26 15 40 substitutes (Banerjee and Duflo)

IT services 57 11 32 64 complements (Mayer and Kalnins)

Notes:-Crocker and Reynolds have eight classes of contracts. We have included the two most complete contract types (those with ceilings on costs) in the fixed-price category and the two least complete types in the cost-plus category. Note also that Crocker and Reynolds have only two sellers, both of whom have sold to the US military under many contracts over a long period of time.-In our offshore drilling context, we determined the percent of repeat contracts by looking at how many new pairs of principals and agents formed in our data after the first six months. In this way, we attempt to account for the fact that we do not observe the full industry history (unlike Mayer and Kalnins). If we include the first six months, the proportion of new pairs rises, and the percent of repeat contracts falls, but only to 74%.

Percentage of contracts

33

net gain toturnkey contract

severity of contracting costs

no relationship

with relationship

substitutes complements

EXPL DEV

Figure 1: Effect of relationships

34

Figure 2: Contract choice and relationships

0.00

0.10

0.20

0.30

0.40

0 5 10 15

Relationships

E[Tu

rnke

y] Exploratory

Development

35