American Economic Association Contract Duration and Relationship-Specific Investments: Empirical Evidence from Coal Markets Author(s): Paul L. Joskow Reviewed work(s): Source: The American Economic Review, Vol. 77, No. 1 (Mar., 1987), pp. 168-185 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/1806736 . Accessed: 07/01/2013 16:52 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp . JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. . American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The American Economic Review. http://www.jstor.org This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions Contract Durationand Relationship-SpecificInvestments: EmpiricalEvidence fromCoal Markets By PAULL. JOSKOW* in This paper examines the importanceof specificrelationshipinvestments negotiatedbetweencoal suppliersand thedurationof coal contracts determining theanalysis.The are usedtoperform Data for277 coal contracts electricutilities. resultsprovidestrongsupportfor the viewthatbuyersand sellersmake longer to thetermsoffuturetradeat thecontractexecutionstage,and rely commitments are more investments less on repeatedbargaining,when relationship-specific important. This paper seeks to test empiricallythe investimportance of relationship-specific ments in determiningthe durationof coal contractsnegotiatedbetweencoal suppliers and electricutilities.'The analysismakesuse of informationfor a large sample of coal contractsthatwerein forcein 1979. It takes as a startingpoint OliverWilliamson's1983 definitionsand categorizationof relationship-specificinvestmentsand applies them to the characteristicsof coal markettransactions. It also follows Williamson and Benjamin Klein, Robert Crawford, and ArmenAlchian (1978) and assumesthatrisk factordeterminaversionis not an important being the structureof verticalrelationships tweencoal suppliersand electricutilities.2 *Departmentof Economics,MassachusettsInstitute of Technology,Cambridge,MA 02139. The preliminary whenI was a Fellow at versionof thispaper was written the CenterforAdvanced Studyin theBehavioralSciences. Leslie Sundt and Rafe Leeman provided valuable researchassistance.Keith Crocker,HenryFarber, Oliver Hart, Scott Masten,JeanTirole,and OliverWilliamson read preliminaryversions and made useful suggestions.Three anonymousrefereesprovidedhelpful suggestions.I benefitedfromseminarpresentationsat UCLA, Stanford,and MIT. SupportfromMIT and the Center forAdvanced Studyin the BehavioralSciences is gratefullyacknowledged.The views expressedhere and anv remainin errors are mv sole resnonsibilitv ' Electric utilities account for over 80 percent of domesticcoal consumption. 2Other empiricalworkin thistraditionincludesKirk Monteverdeand David Teece (1982) and Scott Masten (1984), which focus on vertical integration;Keith Crocker and Scott Masten (1986), J. Harold Mulhem to are interesting Coal markettransactions focus on because thereis considerablevariaof vertical tion in the durationand structure relationshipsbetweenbuyersand sellers.I observe spot market transactions,vertical integration,and a wide varietyof longerterm contractual relationshipswith durations rangingfromone year to fiftyyears.3 My relatedwork (1985) suggeststhatassetspecificityconsiderationsmay be an important factor affectingthe structureof vertical relationshipsin coal markets.The empirical results reported below provide strongsupportforthehypothesisthatbuyers and sellers make longer ex ante commitmentsto the termsof futuretrade,and rely overtime,when less on repeatednegotiations investmentsare more relationship-specific important. and Duration I. Contract Investments Transaction-Specific The reliance on relationship-specific exinvestmentsto supportcost-minimizing (1986), and Victor Goldbergand JohnErickson(1982) which focus on long-termcontracts;my paper (1985) and long-term whichexaminesboth verticalintegration contracts. 3About 15 percentof electricutilitycoal consumption is accounted for by transactionswith integrated suppliers,15 percentis accountedfor by spot market purchases, and about 70 percentis accounted for by contractswith durationsof one to fiftyyears. See my paper (1985, pp. 50-54). 168 This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. 1 JOSKOW: CONTRACTDURATION change is frequentlyadvanced as an important factor explainingwhy we observe the use of long-termcontractsthatestablish the termsand conditionsof repeatedtransactions between two parties ex ante.4According to transactionscost theory,5when in exchangeinvolvessignificant investments relationship-specific capital,an exchangerelationshipthatrelieson repeatedbargaining is unattractive.Once the investmentsare sunk in anticipationof performance, "holdup" or "opportunism"incentivesare created ex post which,if mechanismscannotbe designed to mitigatethe parties'abilityto act on these incentives,could make a socially cost-minimizingtransactionprivatelyunattractiveat the contractexecutionstage.6'7A long-termcontractthat specifiesthe terms and conditionsforsome set of futuretransactions ex ante,providesa vehicleforguarding againstex post performance problems.8 A coal contractgenerallyspecifiesin advance a method for determiningthe price that the buyeris obligatedto pay for each 4Williamson(1979, 1983), Klein et al., Oliver Hart and Bengt Holmstrom(1986, pp. 1-2; 86-101). Other reasons for the use of long-termcontractshave also been suggested.These includeinformation lags,income effectsand risk aversion,and improvedmonitoringof performance. 5I use the term"transactionscost theory"to refer generallyto the workof Williamson(1979, 1983, 1985) and Klein et al. 6As Klein et al. discuss,thesunkinvestments createa streamof quasi rentsthatgivesone partyor the other (or both) some ex post bargainingpower. 7The presenceof thesecontracting hazards and imperfectionsin the abilityof the transactingparties to protectagainstthemdoes not mean thata deal will not be made. It simplymeans thatthe costs of makingthe transactions-the cost of thecoal in thispaper-will be higherthan it would be if thesehazards could be fully mitigated.Both parties have an interestin tryingto structuretherelationshipso thata cost-minimizing deal can be struck. 8Asdiscussedin myearlierpaper(1985), reputational considerationsmay providea naturalmarketconstraint on "bad behavior" ex post. Reputationalconstraints reduce the need to writeinflexiblelong-termcontracts to supportcost-minimizing exchangein the presenceof asset specificity.Reputationalconstraintsare likelyto be imperfectin coal markets,however.At the other extreme,verticalintegration maybe chosento deal with ex post performanceproblemsif satisfactory contractual solutionscannot be found. 169 delivery(generallya formulafordetermining prices for deliveriesat each point in time),9 quantitiesthat the selleris obligatedto deliver and the buyeris obligatedto purchase at each point in time (usually monthly),10 the qualityof the coal (Btu, sulfur,ash, and chemicalcomposition),thesourceofthecoal, and the period of timeover whichthe contractualprovisionsare to governthe terms and conditionsof trade.11The primaryreadily quantifiablecharacteristics of coal contracts that appear to vary widely from contract to contract are the quantityand characteristics of thecoal contractedforand the lengthof time thatthe partiesagree ex ante to committhemselvesto the termsand conditionsspecifiedin thecontract.It is this length of time to which the parties agree ex ante to abide by the termsof a contract that I referto as the "duration"of the contract.12 My hypothesisis that the more importantare relationship-specific the investments, longerwill be theperiodof time(or number of discretetransactions)overwhichthe parties will establishthe termsof trade ex ante by contract. I thereforeexpect to observe that the variationin the agreed upon duration of contractualcommitments is directly related to variationsin the importanceof relationship-specific investments.13 9See myearlierpaper (1986). l?The typical contractspecifiesa monthlyand annual deliveryschedule subject to minimumand maximum production and take obligations.The allowed variationsfromthe contractedquantitiesin actual contractsthatI have reviewedis fairlysmall. " There are manyotherprovisionsas well,including arbitrationprovisions,force majeur provisions,resale provisions,etc. These provisionsare fairlystandardin long-termcoal contracts,however. 12The actual durationof a contractcould be longer or shorterthanthis.Buyersand sellersfrequently voluntarily negotiate an extensionof an existingcontract. Contractsmay also be brokenthroughbreachor mutual agreement.As faras I can tellfromthedata thatI have reviewed,however,coal contractsare rarelyterminated prematurelySee my paper (1986, p. 2). 13 To the extent that there are tradeoffsbetween contract duration and the incidenceand structureof other contractual"protective"provisions,such as the method for determiningprice adjustmentand quantities, these other provisionsshould be included in the analysis as well. As indicated above, however,there This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 170 THE AMERICAN ECONOMIC REVIEW Duration andtheContractual II. AssetSpecificity ofCoal SupplyRelationships Williamson (1983, p. 526) identifiesfour investdistincttypes of transaction-specific ments,threeof whichappear to be relevant to differenttypes of coal supply relationships. The threetypesof relevanceto coal markettransactionsare:"4 The buyerand seller (a) Site Specificity: are in a "cheek-by-jowl"relationshipwith ex ante decisionsto one another,reflecting exminimizeinventoryand transportation penses. Once sited theassetsin questionare highlyimmobile. Whenone (b) PhysicalAssetSpecificity: or both parties to the transactionmake investmentsin equipmentand machinerythat specificto the involvesdesigncharacteristics transactionand whichhave lower values in alternativeuses. (c) Dedicated Assets: General investmentby a supplierthatwould not otherwise be made but for the prospectof sellinga amountof productto a particular significant customer.If the contractis terminatedprematurely,it would leave the supplierwith significantexcess capacity. Although Williamsondoes not discussit,I thinkthatthere is probably a "buyer" side analogy to the dedicated asset storyas well. A buyerthat relieson a singlesupplierfora largevolume and costlyto of an input may findit difficult quickly replace these supplies if they are appears to be relativelylittlevariationin these provisions in my data base, especiallyrelativeto the very feelthat largevariationin contractduration.I therefore it is safe to assume forpurposesof thisanalysisthatwe have a sample of contractsthatessentiallyholds these otherprovisionsconstant.In any event,we can measure the utilizationof theseotherprovisionsonlyfora small fractionof the contractsin the data base and therefore cannot examine such tradeoffs directly.Note that the coal marketis not subject to the kinds of price regulationdiscussedin Masten-Crocker and Crocker-Masten regardingnaturalgas contracts. 14The fourthis whatWilliamsoncalls "human asset specificity"(1983, p. 526). JeanTirole has suggestedto me that Williamson's four types of relationship-(or inare simplydifferent transaction)specificinvestment stances of the same phenomenon.I believe thatthisis correct.However, I find the distinctionsto be quite usefulforempiricalapplications. MARCH 1987 withterminatedsuddenly and effectively drawn fromthe marketand, as a result,a large unanticipated demand is suddenly thrownon the market. As discussed in more detail in the Appendix,I have put togethera data base that includes informationfor nearly 300 contractsbetweenelectricutilitiesand coal suppliers that were in forcein 1979. The data base includes informationof various kinds of the individregardingthe characteristics ual coal contracts,the suppliers,the buyers, and the quality and quantityof the coal contractedfor. The strategywas to use the informationabout the individualcontracts in the data base and to attemptto measure, in the imporat least ordinally,differences investmentsof tance of transaction-specific one or more of the typesidentifiedby Williamson foreach contract. is Williamson'snotionof "site specificity" the easiest to capture explicitlyfor coal supplyrelationships.For mostelectricgenerating plants, coal is purchased in one of threemajor coal-producingregionsand then transportedby rail, barge, and/or truck (oftenat least two of thesetransportmodes are involved) to the powerplant whereit is burned.However,thereare a relativelysmall numberof plants thathave been sited next to specificminesin anticipationof takingall fromthatmine. or mostof theirrequirements These "mine-mouth"plants are generally developed simultaneouslywith the mines themselves.This appears to be a classic case of the cheek-by-jowlrelationshipthat Williamson has in mind when he discussessite specificity.15The potentialfor ex post opportunismproblems arising if the parties were to relyon repeatedbargainingappears to be especiallygreatin thiscase.16I there- 15This is discussed in muchmoredetail in my 1985 paper. 16 Williamson(1983) states that commonownership My work is the predominantresponseto sitespecificity. indicatesthat common with coal supply arrangements ownership(verticalintegration)is much morelikelyto plantsthanothertypesofplants, emergeformine-mouth but thatcontractsare also used to governexchangefor plantsconstructedsince about half of the mine-mouth 1960. We would probablysee moreverticalintegration This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. 1 JOSKOW: CONTRACTDURATION fore expect that contractsfor supplies for mine-mouthplantswillbe muchlongerthan the average contractinvolvingsupplies to othertypesof plants,otherthingsequal."7 Let us turnnextto physicalasset specificity.When coal-burningplantsare built,they are designed to burn a specifictypeof coal (see my 1985 paper; Richard Schmalensee and myself,1986; mypaper withSchmalensee, 1985). By "type" of coal, I mean coal with a specific Btu, sulfur,moisture,and chemical content.The type of coal that a generatingunitis designedto burnaffectsits constructioncost and its design thermal efficiency.Deviations from expected coal in perforqualitycan lead to a deterioration mance or requirecostlyretrofit investments. Thus when a plant is designed,the operator becomes "locked in" to a particulartypeof coal.18 The fact that a plant is locked in to a particulartype of coal does not necessarily implythatthebuyeris lockedin to a specific supplier,however.Whetheror not the plant lock in also leads design/coal characteristic to a lock in with the currentsupplierdeof thetransacpends on othercharacteristics tion. In particular,it is likelythat the relationshipbetweenthistypeof asset-specificity and ex post hold up or opportunismproblems is related to inter-and intraregional formine-mouthplantsif stateand federalregulationof electric utilitiesdid not discourageit. I have argued elsewhere(1985) that to the extentthat electricutility regulationbiases coal supplyarrangements at all, it is probably to make short-term purchasesmore desirable than theymightotherwisebe. Whileutilitiesmightalso like to integratebackwardsintocoal productionto shift profitsfroma regulatedto an unregulatedactivity,the regulatoryprocesshas discouragedthis. 17j have includedtwoplantsin thiscategory thatare not technicallymine-mouthplants but have economic characteristics thatare identicalto thoseof mine-mouth plants. For example, if a mine and a plant are connectedby a transportation facility(a slurrypipelineor a rail line) built and owned by the supplier or buyer specificallyto transportcoal froma specificmine to a specificplant, the associatedcoal contractwas grouped withthe mine-mouthplants. l8Exactly how locked in, is a variable of choice, however. 171 variationsin coal quality,least cost supply technology,and transportation alternatives.19 The characteristics of coal producedin the amongthe United States varysystematically three major coal-producingregions. The easterncoal-producingdistricts producehigh Btu coal of reasonablyuniformquality.The midwesterncoal-producingdistrictsproduce lower Btu coal thatgenerallyhas a veryhigh sulfur content. Coal quality is also more variable than that in the East. Finally,the western coal-producing districtsgenerally produce coal witha muchlowerBtu content and a verylow sulfurcontent.The qualityof the coal varies quite widelythroughoutthe westernregion.20 In addition to variationsin coal quality among the regions,thereare also systematic variations in the least-costtechnologyfor producing coal and in the transportation alternativeavailable. In the East, relatively small underground minesare economicaland the supply of easterncoal can be expanded fairlyquickly.Relativelyabundanttransportation alternatives,combinedwithrelatively shortaverage transportdistances,mean that transportation is not likelyto be a significant barrier to a buyer's obtaining alternative supplies. In the West, large surfacemines that can be most economicallyexpanded in large "lumps," are the least-costproduction technology.Transportationalternativesare poor, the average transportdistance quite long, large unit trainshipmentsare themost economical transportmethod,21and utilities often must rely on one or two railroadsto move the coal. The situationin the Midwest lies somewherebetweenthesetwoextremes.22 19Transportationcosts are on average a large fraction of deliveredcosts and liningup efficient transportation arrangementsforlargequantitiesof coal can be a time-consuming process. 20 The midwesternregion is sometimesbroken up into two subregions(eastern and westerninterior)in discussionsof coal supply.The westernregionis sometimes broken up into threeor more subregions.Texas, where lignite coal is produced,is often considereda separate producingregion.My data base has no contractsforTexas coal and I do not discussthatarea here. 21Unit train cars are oftenowned or leased by the utilityratherthan by the railroad. 22See MartinZimmerman(1981, pp. 17-36). This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 172 THE AMERICAN ECONOMIC RE VIEW in There are also systematicdifferences the relativeand absoluteimportanceof spot marketsin the three supply regions.23On average,from1974 through1982 spotmarket transactionsaccounted for roughly15 percent of total domestic coal purchases by electricutilities.In 1982, spot marketsales accounted forabout 10 percentof coal supplies or about 60 milliontons. However,in the westernregion,less than2 percentof the coal deliveredto electricutilitieswas purchased on the spot marketor less than 5 milliontons.24The spot marketis more active in the Midwest,accountingforabout 8 percent of deliveriesin 1982 or about 10 million tons per year. The spot marketis most active in the East whereabout 18 percent of deliveries went throughthe spot marketin 1982 or about 45 milliontons.25 These considerationsimplythe following: Coal suppliersare likelyto be less able to exploit the lock-in effectassociated with boilers designed to burn coal with specific in theEast thanin theWest.26 characteristics contract Thus the protectionof a long-term is likelyto be moredesirablefromthebuyers' 23Spot marketsales also varyfromyearto year.Spot markettransactionstend to be higherwhencoal miner strikes are anticipated as utilitiesseek to build up stockpiles or after coal miningstrikesare over and stockpilesare replenished.The volume of spot market transactionsalso varies in response to unanticipated changesin coal supplyand demand. 24The aggregatevolumeof spot markettransactions forwesterncoal is quite small comparedto the annual quantity of westerncoal contractedfor in a typical contract.The contractsforwesterncoal in mydata base have a mean quantityof about 1.8 milliontonsper and a maximumquantityof over 8 milliontonsper year. 251 believe that the wide variationin the importance regionsis largelyrelated of the spot marketin different to the same economic considerationsthat lead me to also conclude that the importanceof asset specificity varies fromregionto region. 26Generating plants located along the eastern seaboard that use coal essentiallyalwaysuse easterncoal. These plantsare also morelikelyto have unitsthathave multifuelcapabilitiesthanplantslocated elsewhereand can switchback and forthbetweencoal and oil or gas (often with some performancepenalty).See my paper with FrederickMishkin (1977). Purchasersof eastern coal with multifuelcapabilitieswill be less susceptible to opportunismproblemswhen coal and oil pricesare close together. MARCH 1987 for (and the sellers'-see below) perspective, transactionsinvolvingwesterncoal relative to transactionsinvolvingeasterncoal, with midwesterncoal fallingsomewherein between. Finally, let us turn to "dedicated asset" in considerations.The available information the data base does not make it possible for us to know specificallywhetherthe supplier made general investmentsthat would not have been made but for the prospect of amountof productto a selling a significant particular customerand if the contractis terminatedprematurelyit would leave the supplier with excess capacity(see Williamson, 1983, p. 526).27 Williamson'sconceptualization of dedicatedassetsimpliesthatthe coal importanceof thisfactorin structuring supply relationshipsshould vary with the quantityof coal that is initiallycontracted for,otherthingsequal. The largertheannual quantityof coal that is contractedfor,the it is likelyto be forthe seller more difficult to quicklydispose of unanticipatedsupplies (if the buyer breaches) at a compensatory will it be fora price, and the more difficult buyer to replace supplies at a comparable price if the sellerwithdrawsthemfromthe market. Thus, I expect that the greater the annual quantityof coal contractedfor the longerwill be the specifieddurationof thecontract. Because of systematicvariationsin the optimal scale and capital intensityof coal production across regions,dedicated asset considerationsare also likely to be more importantfor westerncoal than for eastern in coal supcoal. The greaterheterogeneity of obtainingsuitable plies and the difficulties transportationfor it in the West, suggest thatdedicatedassetproblemsare likelyto be more severein the westernthan the eastern region. The very thin spot marketin the 27Indeed, as a practicalmatter,it is unclear to me thisdirectly how one could ever go about determining given the data that are likelyto be available foranalysis. The coal contracts that I have reviewed do sometimeshave language thatappears to recognizethe dedicated asset notion directly,but the absence of an explicit statementcannot be assumed to imply that dedicated asset considerationsare not important. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. ] JOSKOW: CONTRA CTDURA TION western region should be especially problematical for both sellersand buyerswhen are breached large contractualcommitments of the coal, the because of the heterogeneity characteristicsof least-costproductionand alternatives.This the limitedtransportation all implies again that contractsforwestern coal should have longer contractualdurations thancontractsforeasterncoal. To summarize,if variationsin the impordo investments tance of relationship-specific in fact lead to variationsin the extentto whichthe partiesprecommitto the termsof futuretrade ex ante,I expectto findthatthe duration of contractualrelationshipsspecifiedat thecontractexecutionstagewillvary withthreeprimaryobservable systematically characteristicsof coal supply transactions. First,whetherthe plant takingthe coal is a plant or not. I expectto observe mine-mouth longer-termcontractsnegotiatedfor minemouthplants. Second,withtheregionof the countryin which the coal is produced. I expectthewesternregionto have thelongest contractsand the easternregiontheshortest withthe midwesternregionhavingcontracts with prespecifieddurationsthat lie in between. Third, with the annual quantityof coal contractedfor.I expectthatcoal supply arrangementsinvolvinglarge annual quanwillbe supportedby longer titycommitments involvcontractsthan supply arrangements ing smallerquantitiesof coal. III. ModelSpecification andEstimation I am primarilyinterestedin estimatinga betweentheduraset of simplerelationships tion of contractualcommitments(DURATION) specifiedby the partiesat the contract execution stage and, (a) the annual quantity(QUANTITY)28 of coal contracted 173 for,(b) a dummyvariable(MINE-MOUTH) that takes on a value of 1 fora mine-mouth plant and zero otherwise,and (c) dummy variablesthatindicatethecoal supplyregion in whichthe supplieris located(MIDWEST and WEST, so thatregionaleffectsare measured relativeto contractsforeasterncoal). Additional variables are consideredin the nextsection. The data base thatI use includesinformation forapproximately300 coal supplycontractsbetween domesticcoal suppliersand investor-ownedelectric utilities.The contractsincludedin thedata base werenegotiated in various years up through1979 and were in forceat least forpart of 1979. The data are discussed in more detail in the Appendix. Informationon all variables of primaryinterestfor this studyis available for 277 of the contractsin the data base. I present estimates using both the full 277 observationsample as well as a subsample consistingof 169 contractsthat involvedeliveriesdedicated to a singlepower plant.29 Table 1 provides the mean, standarddeviation, minimumand maximumvalues,and a briefdescriptionforall of thevariablesused in this section and subsequentsectionsfor both theprimarysampleand thesingle-plant subsample. Table 2 is a correlationmatrix for all of the variables in the two samples with the correlationsfor the 277 observation sample below thediagonaland thosefor the 169 observation subsample above the diagonal. I work firstwith three simple specifications of the contractdurationequation: (1) DURATIONJ= a0 + b1QUANTITYJ + b2QUANTITY7 + b3MINE-MOUTH1 + b4MIDWESTJ + b5WEST1 + Ui 28 Quantitiesare expressedin termsof the thermal (Btu) content of the coal. The basic resultsare not affectedwhen quantitiesare expressedin tons. For the extensions reported in the next section, normalizing quantities for Btu contentmakes it possible to more if any,of contractquanaccuratelyexamine theeffects, tities relative to total plant and utilityutilizationof coal. 29Severalpeople suggestedto me that relationshipspecificinvestmenteffectsare mostlikelyto be revealed forcontractsthatdedicateall suppliesto a singleplant. I also focus on this subsample to obtain the data necessaryto exploreissuesdiscussedin thenextsection. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 174 THE AMERICAN ECONOMIC REVIEW MARCH 1987 TABLE 1-SAMPLE STATISTICS Observations Variable 277 169 277 169 DURATION QUANTITY PLANT PROPORTION 169 UTILITY PROPORTION 169 PLANT QUANTITY 169 UTILITY QUANTITY 169 PLANT/UTILITY 169 MINE-MOUTH 277 169 277 169 277 169 277 169 277 169 277 169 277 169 WEST MIDWEST DA TE-71 DA TE-74 DA TE-78 YEAR Mean Description Standard Minimum Maximum Deviation 50 43 183.00 183.00 10.43 10.77 24.62 27.05 0.03 1.00 0.35 0.003 1.00 0.23 2.95 172.42 41.69 2.95 919.80 270.56 1.00 1.00 0.3696 0.3696 12.75 ContractDuration 14.18 (years) 20.45 AnnualContract 22.83 Quantity (trillionBtu's) 0.44 Fractionof Total PlantUse from Contract 0.19 Fractionof Total UtilityCoal Use fromContract 51.47 Plant Utilizationof Coal (trillionBtu's) 221.54 UtilityUtilization of Coal (trillionBtu's) 0.455 PlantUse as Fraction of Total UtilityUse Mine-MouthPlant DummyVariable WesternRegion SupplyDummy MidwesternRegion SupplyDummy ContractsSigned 1971-73: Dummy ContractsSigned 1974-77: Dummy ContractsSigned 1978-79: Dummy Year ContractExecuted 1974 1974 1.00 0.007 0.323 (D = 1; 14 Observations) (D = 1; 14 Observations) (D = 1; 54 Observations) (D = 1; 44 Observations) (D = 1; 68 Observations) (D = 1; 47 Observations) (D = 1; 43 Observations) (D =1; 29 Observations) (D = 1; 116 Observations) (D =1; 71 Observations (D = 1; 72 Observations) (D =1; 38 Observations) 1979 4.49 1955 1979 4.64 1955 can be foundin the Appendix. aData sourcesand variabledefinitions b The 169 observationsubsampleincludescontractsdedicatedto a singleplant. TABLE 2-CORRELATION MATRIX 277 Observation Sample DURA TION QUANTITY LOG-QUANTITY MINE-MOUTH MIDWEST WEST YEAR 169 ObservationSample DURATION - 0.60 0.64 0.54 0.14 0.41 -0.63 LOG- MINE- QUANTITY QUANTITY MOUTH 0.60 0.80 0.38 0.08 0.25 -0.39 0.68 0.78 0.27 0.11 0.26 -0.44 0.64 0.42 0.30 -0.06 0.39 -0.31 MIDWEST WEST YEAR 0.004 0.51 -0.57 -0.04 0.28 -0.35 0.02 0.32 -0.44 0.41 -0.37 -0.09 -0.37 -0.20 -0.28 -0.10 -0.27 -0.08 PLANT UTILITY PLANT/ PLANT UTILITY PROP. PROP. UTILITY QUANTITY QUANTITY 0.58 0.48 0.57 0.46 0.04 0.46 -0.28 0.43 0.43 0.42 0.41 -0.04 0.40 -0.11 0.61 0.29 0.41 0.45 0.10 -0.03 -0.06 -0.16 -0.24 -0.08 0.15 0.02 0.11 0.17 -0.05 -0.15 0.16 -0.11 -0.14 -0.36 -0.57 0.05 0.05 0.02 0.05 -0.06 0.16 0.13 -0.11 0.66 PLANT PROP. - - - - - - - - UTILITY PROP. - - - - - - - - - PLANT/UTILITY - - - - - - - - - - - - - - - - - - - - - - - - PLANT QUANTITY - 0.34 UTILITY QUANTITY - Note: Figuresbelow thediagonalare forthe277 observationsample,thoseabove thediagonal are forthe 169 observationsubsample. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. I (2) JOSKOW: CONTRACT DURA TION DURA TION1= a0 + b1LOG-QUANTITY1 + b3MINE-MOUTH, + b4MIDWEST1 + b5WEST1 + U1 (3) log(D URA TION1) = a 0 + b1LOG-QUANTITY1 + b3MINE-MOUTHi + b4MID WEST, + b5WEST1+ log(ui) wherei indexes contractsand ui is an error termwhose characteristics will be discussed further below. I have allowed (QUANTITY) to enter these relationshipsnonlinearlyby introducing a quadratic in quantity(QUANTITYSQUARED) in (1) and using the natural logarithmof quantity(LOG-QUANTITY) in equations (2) and (3). Since powerplants have usefullives of roughlyfortyyearsand the costs of breach are likelyto declineover time as plants and minesage, I expectthat the impact of quantityon contractualduration will diminishas quantityincreases.The following pattern of coefficientestimates for the three equations is implied by the hypothesized relationship between asset and contractduration: specificity (i) All of the b 's should be positive, exceptforb2, whichcould be positive,negative, or zero (no nonlinearity),althoughI expect thatit will be negative. (ii) b4 should be smallerthan b5. Equations (1), (2), and (3) are estimatedin threedifferent ways.First,I presentordinary least squares (OLS) estimatesof each equation for both samples.Next, I presentOLS estimates that introducedummy variables which indicate the date that the contracts were executed. Finally,I presentmaximum likelihood estimatesbased on the assumption thatwe have a truncatedsample drawn froma populationwitheithera normalor a log-normaldensityfunction.I discuss the rationaleand resultsforeach estimationapproach and also present OLS results for 175 contractswithsuppliersin each of the three regions. A. OLS Estimates If we assume thatthe u1in (1), (2), and (3) are independentlydistributedand drawn froma normal distribution withmean zero, thenOLS will yieldan unbiasedestimatorof the coefficientsof interest.I proceed first with this assumptionand provideestimates for alternativeassumptionsabout the error structurebelow. The OLS resultsare presented in Table 3. The firstthreecolumns are estimatesfor the threeequations using the 277 observationsample. Columns 4, 5, and 6 containestimatesforthe 169 observation subsample. The OLS estimatesare, in all cases, consistent with the hypothesizedrelationship betweenasset specificity and contractduration.The effectsof annual contractquantity, region,and mine-mouth plantshave thepredicted signs and are estimatedquite precisely. A mine-mouthplant is predictedto have a contractthatis about 16 yearslonger than those of other plants. Contractswith eastern producersare 3 to 5 years shorter than those forwesternand midwestern producers.Contractswithwesternproducersare 2 to 3 yearslongerthanthosewithmidwestern producers. The differencein duration between the WEST and the MIDWEST is generallynot significantat the 5 percent level for the 277 observationsample,but is significantfor the single-plantsubsample. An increase in annual contractquantityof 22 trillion Btu's (roughly1 million tons) yields about a 13-yearincreasein contract duration.As a generalmatter,the estimates are morepreciseforthesample of contracts dedicated to a singleplant. B. OLS Estimateswith ContractDate Dummies The desirable propertiesof the OLS estimates depend on the strong assumptions made about the errorstructure.Since the contractsin the data base were signed at manydifferent times,it is naturalto consider the possibility that contractingpractices This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 176 THE AMERICAN ECONOMIC REVIEW MARCH 1987 TABLE3-CONTRACTDURATIONa 277 Sample Independent Variables QUANTITY QUANTITYSQUARED LOG-QUANTITY MINE-MOUTH MIDWEST WEST 169 Sample LOG- DURATION LOG- DURATION DURATION 0.4289 (0.0373) - - -0.0024 (0.00030) - - 16.3300 (2.0496) 3.4267 (0.9682) 5.3550 (1.357) 4.4206 (0.3742) 16.4317 (2.0045) 3.8795 (0.9821) 5.2033 (1.1641) 0.5057 (0.0425) 0.5104 (0.2279) 0.5154 (0.1116) 0.6142 (0.1323) - - - - - - - - - - (1) - (2) (3) 2SLS DURATION DURATION DURATION DURATION 0.4091 (0.0040) - - - - - -0.0020 (0.00003) - - - - - - 4.2080 (0.4069) 16.2300 (1.8421) 2.7843 (1.1032) 5.6108 (1.2586) 0.4942 (0.0453) 0.4616 (0.2050) 0.5785 (0.1228) 0.6844 (0.1401) 4.2022 (0.4617) 16.3432 (1.9426) 2.7317 (1.1295) 5.6456 (1.3406) 4.2057 (0.4084) 16.2284 (1.8477) 2.7848 (1.1065) 5.6391 (1.2751) 5.1066 (0.812) 15.4391 (1.968) 2.4268 (1.153) 4.8916 (1.394) - - 0.9729 (1.9806) - - - - - -2.0570 (2.5832) - - - - (4) 15.9583 (1.9106) 2.7832 (1.0928) 5.9856 (1.2346) (5) (6) PLANT PROPORTION UTILITY PROPORTION PLANT/ UTILITY Constant Corrected R-squared Observations a 3.6770 -0.7902 (0.6586) (0.9579) 0.61 277 0.60 277 0.6014 (0.1089) 0.51 277 Estimate Duration (9) DURATION 3.9334 (0.8109) 0.71 169 0.0155 (1.0917) 0.70 169 (7) (8) -0.2246 (1.4265) 0.6242 - 0.0146 0.1157 - 1.8922 (0.1215) (1.0978) (1.2665) (1.852) - - 0.61 169 0.70 169 0.70 169 - 169 OLS estimates.Standarderrorsof coefficient estimatesare shownin parentheses. changedover time.Not onlymighttheduration of a typicalcontracthave changedover time,but such changesmayhave been correlated with changes in contractquantities, supply location, and the developmentof mine-mouthplants over time.Failing to include variablesindicatedcontractdatescould thenlead to a correlationbetweenthe independent variables and the errorterm.The OLS estimates would then be biased. To check to see if the estimatesare sensitiveto the presenceof a left-outvariablereflecting the contractingdate, in Table 4, I report estimatesof equations (1), (2), and (3) that have contractdate dummiesincluded.These contractdate dummyvariablesare DA TE-71, whichis equal to one forall contractssigned between1971 and 1973 inclusive,DA TE-74, which equals one for contractssigned between 1974 and 1977, and DATE-78 which equals one forcontractssignedin 1978 and 1979. Since a separatevariableforcontracts signed prior to 1971 is not included, the coefficientestimatesare all relativeto pre1971 contracts(i.e. the constantterm).This aggregationof signingdates was made to reflectmajor shocks to coal supply and/or demand.30 30The 1971-73 periodis just afterthe Clean Air Act Amendmentsof 1970 werepassed, the 1974-77 period is the period afterthe Arab oil embargoand includes the subsequent increases in fossil fuel prices. The 1978-79 period coincides with the beginningof a slowdown in utilitycapacity additions.These periods are discussed in more detail in my paper (1986). The aggregationchosen is the same used to analyzepricing behaviorin thatpaper. The resultsreportedhereare not sensitiveto thisaggregation, however.The same qualitative resultsare obtained if separate dummyvariables are used foreach yearduringthe 1970's plus a separate dummyvariable forpre-1965and 1966-70 contracts. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. I 177 JOSKOW: CONTRACT DURATION TABLE4-CONTRACTDURATIONa Independent Variables QUANTITY QUANTITY-SQUARED DURATION DURATION (1) (2) 0.3120 (0.03547) -0.0018 (0.00027) - LOG-QUANTITY 13.9437 (1.8482) 1.6814 (0.8785) 4.8429 (1.0301) MINE-MOUTH MIDWEST WEST LOGDURATION (3) - - - - 3.0482 (0.3655) 13.6701 (1.8260) 1.9761 (0.8952) 4.8662 (1.0549) 0.3245 (0.0380) 0.3140 (0.1899) 0.3029 (0.0931) 0.4831 (0.1097) DURATION (4) 0.3355 (0.0406) -0.0018 (0.00029) 14.6494 (1.8495) 1.4906 (1.0544) 5.1054 (1.2183) DURATION (5) LOGDURATION (6) - - - - - 3.3485 (0.4330) 14.6907 (1.8347) 1.6405 (1.0786) 5.1731 (1.2524) 0.3631 (0.0446) 0.3792 (0.1891) 0.4083 (0.1112) 0.5137 (0.1291) - - - - UTILITY PROPORTION - - - - - - PLANT/UTILITY - - - - - - 9.2341 6.1982 1.4220 (1.5185) (1.8612) (0.1918) -0.7282 -0.9103 -0.0311 (1.4890) (1.5057) (0.1552) -3.9714 -4.3786 -0.3149 (1.3483) (1.3685) (0.1410) -7.0789 -6.5193 -1.0317 (1.5540) (1.6324) (0.1682) 0.74 0.69 0.73 169 169 169 3.2847 3.3461 (0.4923) (0.4344) 14.5742 14.6800 (1.9423) (1.8403) 1.5543 1.6323 (1.1083) (1.0821) 5.0697 5.1258 (1.3338) (1.2676) 0.9930 (1.8907) -0.9138 (2.4975) 0.3785 (1.3660) 6.1583 6.0806 (1.8853) (1.9142) -0.9389 -0.9455 (1.5290) (1.5154) -4.3848 -4.4134 (1.3789) (1.3781) -6.4995 -6.5749 (1.6702) (1.6493) 0.73 0.73 169 169 coefficient estimatesare shownin parentheses. While the introductionof these contract date dummyvariablesmay help to control for contract date related correlationsbetween the error termand the independent variables,the estimatedcoefficients of these variables themselveshave no obvious economic meaning.This is because of the nature of the sample. Recall that I observe contractsinforcein 1979. If we thinkof the populationas consistingof contractswritten for particularplants (i) in a particularyear (t), we can observea contractonlyif (4) (8) - - a OLS/Contract date dummies.Standarderrorsof DURATION - - 12.0145 9.4184 1.7205 (1.2526) (1.5307) (0.1592) DA TE- 71 -2.3734 -2.4564 -0.0988 (1.2715) (1.2876) (0.1339) DA TE- 74 -6.6815 -7.3044 -0.5098 (1.2647) (1.1446) (0.1190) DATE-78 -10.5052 -10.6151 -1.3926 (1.2647) (1.2976) (0.1349) 0.70 0.69 CorrectedR-squared 0.67 277 Observations 277 277 (7) - PLANT PROPOR TION CONSTANT DURATION DURA TIONIt 2 (1979-Contract YEAR) This means thatof thosecontractssignedin 1970, I can observe,in 1979,only thosethat had durationsof at least 9 years,whileI will observeshortercontractsthatweresignedin lateryears.Even if therewereno changesin contractingbehavior over time, we would inevitablyfind that the coefficients of the contract date dummies indicate that the average lengthof a contractin the sampleis negativelycorrelatedwith the date of the contract.31 The coefficients of the contract date variables can therefore tell us nothing directlyabout the changes in contracting behaviorover time. With theseconsiderations in mind,we can turnto the resultsreportedin Table 4, columns 1 through6. The resultsobtainedare again consistentwith the hypothesizedrelationshipbetweenasset specificity and contractduration.The coefficients of the quantity, mine-mouth,and regional variables In the sample, the simplecorrelationbetweencontract date (YEAR) and DURATION is about -0.60. See Table 2. 31 This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 178 THE AMERICAN ECONOMIC REVIEW MARCH 1987 The resultsare again quite consistentwith the hypothesizedrelationshipbetween asset specificityand contractduration. The of signs and magnitudesof the coefficients QUANTITY, MINE-MOUTH, WEST, and MIDWEST are again as predicted.Withthe exception of the mine-mouthdummy in are estimated equation (3), the coefficients quite precisely.The magnitudesof the estiin thistableare difficult to matedcoefficients C. MaximumLikelihoodEstimates compare directlywiththosein Tables 3 and A third alternativefor estimatingthese 4 because of the need to incorporatethe truncationeffectsinto estimatesof contract equations is to follow Keith Crocker and duration.33Correctingfor the sample trunScott Masten (1986), who workwitha samcation, the estimatesreportedin Table 5, ple of naturalgas contractswithsimilarsampling properties,and assume that the sam- column 5, for example,yield the following pling procedurewhichchooses contractsin expected durations: The expected duration forcein a single year (1981 in theirpaper) evaluated at the means of the independent represents a classical sample truncation variablesis 10.5 years(comparedto a mean problem as discussed by G. S. Maddala of 14.2 years forthe truncated169 observa(1983, pp. 165-170). The populationof contion sample). Mine-mouthplants have contractsthenimplicitly consistsof all contracts tractswithan expecteddurationthatis about writtensince theearliestcontractin thedata 12 years longer than the averagenon-minebase. We obtain a truncatedsamplebecause mouthplant.Midwesterncontractsare about we observecontractsonlyif the durationof 3.5 years longer than eastern contracts. the contractis greaterthan or equal to L1, Westerncontractsare about 11 yearslonger whereLi is equal to 1979 minusthecontract than eastern contractsand 6 years longer than midwesterncontracts.The difference date. In thiscase, OLS estimatesof (1), (2), betweenwesternand midwestern contractsis and (3) would be biased, because the samat the5 percentlevel.34 plingprocessis likelyto inducea correlation generallysignificant between the independentvariablesand the errorterm. D. EstimatesforIndividualCoal We can obtain estimateswith desirable SupplyRegions thelikeasymptoticpropertiesby specifying lihood functionof the sample, given the Finally, in Table 6, I report estimates natureof the samplingtruncation, and then of the relationshipbetween DURA TION, solve for the maximumlikelihoodestimates LOG-QUANTITY, and MINE-MOUTH for of interest.Follow(MLE) of thecoefficients ing Maddala (p. 166), I assume that the population relationship between contract of Li = (1979- YEAR) thatrequiretakingthelogarithm durationand theindependentvariableshas a which is zero for contractswrittenin 1979. Since the normallydistributederrorand thateach obshortestcontractin thedata base has a durationof one servationis truncatedat Li. This leads to a year,we can impose a lowerbound betweenzero and 1 standardmaximumlikelihoodestimator. on Li to include all observations.Estimatesobtained The maximumlikelihoodestimatesare reusing different lower bounds does not change the results,so I simplyreportresultsforthesampleexcluding ported in Table 5, columns 1 through6.32 continue to be of the predictedsigns and difrelativemagnitudes.The onlyinteresting ferencebetween these resultsand those rebeported in Table 3 is that the difference tweenthe durationsof contractssignedwith westernand midwesternproducersis now at the5 pergenerallystatistically significant cent level forboth samples. 32 The estimateswereobtainedusingtheMLE routine in the StatisticalSoftwareTools package (Version1.0 as of October 1986) developed by Jeffrey Dubin and R. Douglas Rivers runningon an IBM XT. The estimates in cols. 3 and 6 assume thatthedensityfunction is log-normal.Contractsexecutedin 1979 have been dropped since the likelihood functionincludes terms the small numberof contractsin thedata base executed in 1979. 33See Maddala (p. 167). 34I have also produced maximumlikelihood estimates of equation (2) for subsamples consistingof pre-1974contractsand thosesignedbetween1974 and 1979. The hypothesizedrelationshipbetweencontract variationsin duration and the variables representing asset specificity persistsforboth subsamples. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. I JOSKOW: CONTRA CT DURA TION 179 DURATION a TABLE 5-CONTRACT 2SLS Independent Variables LOGLOGESTIMATE DURATION DURATION DURATION DURATION DURATION DURATION DURATION DURATION DURATION QUANTITY (1) (2) (3) 0.7948 - - - - (0.1046) QUANTITYSQUARED -0.0043 (0.00061) (4) (5) (6) (7) (8) (9) 0.5807 - - - - - -0.0030 - - - - - (0.0717) (0.0004) - 11.8527 0.6733 (1.5663) (0.0897) 16.5557 13.8763 0.2407 (4.2407) (3.8955) (0.5714) 8.3650 8.1377 0.4493 (2.8188) (2.5461) (0.2281) 15.4864 13.8045 1.0500 (4.1940) (3.4244) (0.0500) LOG-QUANTITY - MINE-MOUTH MIDWEST WEST PLANT PROPORTION 15.5484 (2.9435) 5.0050 (2.5018) 11.7771 (3.0124) 8.4367 (1.0816) 14.4004 (2.7764) 4.7042 (2.4511) 10.5506 (2.4220) 0.6340 (0.0816) 0.2736 (0.4614) 0.5376 (0.2763) 1.0522 (0.2416) - - - - - - - - - - - - - - - - - - 2.4401 (3.7927) UTILITY PROPORTION 8.2709 8.4505 (1.0854) (1.0815) 14.1237 14.3841 (3.0120) (2.7650) 4.4397 4.7468 (2.5221) (2.4820) 10.1238 10.4730 (2.6136) (2.4095) -2.0043 (3.1937) PLANT/ UTILITY Constant Log-Likelihood Observations - 8.5737 (3.5777) 15.4508 (2.1083) 2.4404 (2.1797) 6.9477 (1.8329) - - - - 1.4626 - (2.1945) -18.4532 - 35.6427 - 0.4699 - 6.9510 -19.0118 - 0.2789 -19.0377 - 19.7755 - 36.1521 (5.7203) (7.2690) (0.26709) (3.4094) (4.6758) (0.2448) (4.8545) (4.8645) (6.9717) - 781.08 - 769.29 -174.29 - 475.17 - 466.99 - 101.60 - 466.68 - 466.83 - 451.38 277 277 255 169 169 160 169 169 169 aMaximum likelihood estimates. Standard ofcoefficient errors estimates areshownin parentheses. TABLE 6-CONTRACT DURATION a DependentVariable: DURATION Independent Variables WEST (1) (2) LOG-QUANTITY 3.4825 (1.0714) 18.4179 (2.5064) 6.819 (3.1610) 0.68 54 5.1249 (0.7646) 11.8107 (5.1695) 1.365 (5.1695) 0.42 68 MINE-MOUTH Constant CorrectedR-squared Observations MIDWEST EAST EAST WEST (3) (4) (5) 4.2362 (0.4668) - 0.3975 (1.1159) 0.35 155 4.2968 3.5878 (0.4661) (1.0992) 17.6804 17.616 (3.7348) (2.3183) - 0.5265 7.2101 (1.1151) (3.3441) 0.43 0.73 44 158 MIDWEST EAST EAST (6) (7) (8) 4.0269 (0.6538) 12.8372 (3.6363) 3.4075 (1.8264) 0.52 47 4.408 (0.5798) 0.4106 (1.4256) 0.42 78 4.4992 (0.5786) 17.0887 (3.8111) - 0.6028 (1.4246) 0.54 81 a OLS by region.Standarderrorsof coefficient estimatesare shownin parentheses. each coal supply region.35The samples do not include any mine-mouthplants using eastern coal, so columns 3 and 7 simply report the estimated relationshipbetween 35I reportonly this variantof equation (2) to conservespace. It should be clear by now thatthe alternative specificationsof (1), (2), and (3) do not yield any importantdifferences in results. DURATION and LOG-QUANTITY. There are in fact a few mine-mouth plants in the East and I have some information forthree of them. These were not included in the sample because the data base did not have informationon annual contractquantities for them. However, I was able to obtain informationfordeliveredquantitiesforthree eastern mine-mouthplants and have aug- This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 180 THE AMERICAN ECONOMIC REVIEW MARCH 1987 dependenton suppliesprovidedpursuantto a specificcontract.These argumentsimply thatvariablesmeasuringplantand/or utility "dependence" on a specificcontractshould be introducedinto the contractdurationrelationship. To examine this possibility,I have included variables in (2)37 that measure the (PLANT fractionof a plant's requirements PROPORTION) and thefractionof thetotal coal requirementsof the utility(UTILITY that operates the plant PROPORTION) whichare accountedforby a particularcontract. As an alternative,I also estimate(2) introducinga variablethatis equal to total plant utilization of coal divided by total utilityutilization of coal (PLANT/UTILonly ITY). I can estimatetheserelationships for the contractsthat are for deliveryto a singleplant because it is onlyfortheseconmeatractsthatI can constructa meaningful sure of plant specificdependence(i.e., the 169 observationsubsamplemustbe used). IV. Altemative MeasureofAssetSpecificity The resultsare reportedin columns7 and 8 of Tables 3, 4, and 5. The coefficient Clearly,the hypothesizedrelationshipbeestimatesfor PLANT PROPORTION and tween contract duration and the variables UTILITY PROPORTION are veryimprethat I have chosen to capturevariationsin cise. They are of opposite signs and are asset specificity is quite robustto alternative neitherindividuallynorjointlysignificant at specifications, samples,and estimatingtech- conventionalsignificancelevels. The coeffinique. Nevertheless,it is natural to ask cient of PLANT/UTILITY is also veryimwhetherthere are alternativeor additional precise and varies in sign depending on factorsthatmightexplaintheobservedvari- estimatingtechnique.Introducingthesevariations in contractduration.One argument ables has no effecton the estimatesfor the thathas been suggestedto me is thatit is not primaryvariables of interest.These results only the size of the contractualcommitment implythat a plant or utilitythatrelieson a that is likelyto be important,but also the single supplier for a large fractionof its fractionof a plant's,and perhapstheutility's, requirements,or that depends heavilyon a requirementsobtained froma specificconspecificplant,does not encountersignificant tract.The argumentis thatas a largerfrac- hold-up problemsper se. The lock-ineffect tion of a plant's requirements is associated associated with designingplants to burn a with a specific supplier, " physical asset particulartype of coal becomes a potential specificity"attributesare likelyto be more contractualproblemonly to the extentthat importantand lead to longer contracts.It are characteristics the otherasset specificity has also been suggestedto me that opactive. portunismproblems are likely to be less severeif the utilityas a whole is not heavily mented the sample to include these plants, using deliveredquantitiesratherthan contractquantitiesas thevalues forthe QUANTITY variable.These resultsare reportedin columns4 and 8 of Table 6.36 The effectsof contractquantityand the mine-mouthdummy on contractduration are clearly not simplyassociated with the contractingbehaviorforcoal froma particular region.The expectedeffectsare foundin of each of the threeregions.The coefficients QUANTITY and MINE-MOUTH are of the expectedsignsand are estimatedpreciselyin in the all cases. While thereare differences of thesevarimagnitudesof the coefficients ables betweenthe threeregions,theyare not very large numericallyand equality of the coefficientsof QUANTITY and MINEMOUTH across regionscannot be rejected at standard significancelevels. Contracts basically simplyget longeras we move from East to West,otherthingsequal. The inclusion of these three additional observations does not change the aggregateresults. 36 37Includingthese variablesin (1) and (3) does not change the results and I reportthe resultsfor (2) in order to conservespace. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. ] JOSKOW: CONTRACT DURATION Quantity V. Contract 181 I estimatethe followingrelationshipusing the single-plant(169 observation)subsample to determineempiricallywhetherand how these considerationsaffectannual contract quantities. Before concluding,it is usefulto explore in determining the role of asset specificity annual contractquantitiessincethisvariable plays such an importantrole in the contract (5) QUANTITY = c0 duration equation. A relationshipbetween potencontractquantityand asset specificity + d1PLANT-QUANTITY tiallyemergesbecause of thepresenceof all First,other three types of asset specificity. + d2PLANT-QUANTITYI things equal, physical asset specificity considerationssuggestthata plant operator + d3UTILITY-QUANTITY, would like to rely on a specific supplier producing a particulartype of coal at a + d4UTILITY-QUANTITY7 particularlocationto thegreatestextentpossible.38This implies that contractquantity + d5MINE-MOUTHi should vary directlywith the coal requirements of the individual plant (PLANT + d6MIDWEST, + d7WEST, + ViQUANTITY).39 On theotherhand,themore I expect d1, d3, d5, d6, and d7 to be a utilitycomes to relyon a single supplier greaterthan zero, and d7 should be larger the morecostlya breachof contractmaybe. This suggeststhata utilitymaybe willingto than d6. Equation (5) is estimatedin two different relymoreon a singlesupplierforan individual plant the largeris totalutilityutilization ways. First, OLS estimatesare presented. of coal (UTILITY QUANTITY) given the Second, OLS estimateswitha correctionto reflectthe possibilitythatI have a censored utilizationof a specificplant. plants, Second, in the case of mine-mouth sample are also presented.The OLS estiof (5) maybe biased the nature of the ex ante location/invest- matesof the coefficients ment decision involvesthe mutualexpectaas a result of the samplingproceduredistion that all or most of a plant's require- cussed earlier.We observecontractquantity ments will be taken from the proximate only if contractdurationis greaterthan or equal to (1979-contract YEAR), so we have supplier.This impliesthat the quantityper plants, a censored sample. This implies that the contractwillbe largerformine-mouth random error (v) in the contractquantity otherthingsequal. Finally,as discussedabove, whenutilities equation (5) may be correlatedwiththerandom error(u) in thecontractdurationequadesign plants to closely matchspecificcoal quality attributestheywill have an interest tion. If thisis the case, therandomerror(v) in relyingmoreheavilyon a specificsupplier in the contractquantityequation will be a functionof the independentvariablesin the who contractsto supply coal froma seam This is likelyto be contract duration equation. The OLS estiwiththesecharacteristics. of the independent mates of the coefficients an especially importantconsiderationfor variablesin the quantityequation (5) would coal suppliesfromthewesternregion. thenbe biased if theyare correlatedwiththe independentvariablesin the durationequa38One mightalso argue that,other thingsequal, a tion.In particular,independent variablesthat buyerwould ratherrelyon a singlesupplierto conserve in the contract duration equation appear cost associated on moretraditionaltypesof transactions may appear to be significantwhen introwith negotiating,monitoring,and enforcingcontracts withmultiplesupplies. duced as independentvariablesin equation 39Ideally,we would like to look at specificgenerating (8) when in fact theyare not.40Since three plantswhereplants unitsratherthanspecificgenerating designcharacteristics. have multipleunitswithdifferent is not available coal supplyinformation Unfortunately, at the generatingunitlevel. 40See George Judgeet al. (1985, pp. 610-13) and JamesHeckman (1976, 1979). This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 182 THE AMERICAN ECONOMIC REVIEW MARCH 1987 TABLE 7-CONTRACT QUANTITYa DependentVariable: QUANTITY OLS IndependentVariables PLANT QUANTITY PLANT QUA NTITYSQUA RED UTILLITY QUANTITY UTILITY QUA NTITYSQUA RED MINE-MOUTH WEST MIDWEST H (2) 0.2501 (0.0441) 0.3133 (0.1573) - -0.00066 (0.0010) 0.0672 (0.0311) 0.00349 (0.00689) Corrected R-Squared Observations (3) 0.2128 0.0489 0.00393 (0.00685) - 29.7932 (6.8591) 13.5416 (4.6437) 5.2469 (4.1841) -0.00007 (0.00003) 27.1594 (6.8997) 13.5865 (4.6259) 4.856 (4.1708) 1.7329 (3.5879) - 4.2283 (5.0515) 30.3261 (6.8251) 8.4955 (5.4680) 2.5832 (4.4371) -66.7618 (38.7474) 10.6708 (6.2951) 0.33 169 0.34 169 0.34 169 - - Constant OLS/H (1) - (4) 0.2827 (0.1578) -0.00067 (0.00099) 0.06376 (0.03104) -0.000063 (0.000032) 27.775 (6.879) 8.9738 (5.4597) 2.4749 (4.4193) -60.6647 (38.5749) 4.0997 (7.3027) 0.35 169 aStandard errorsof coefficient estimatesare shownin parentheses. variables that appear in the quantityequation (5) also appear in thedurationequation, thisis a potentialproblemhere. We can obtain consistentestimatesof the coefficientsof (5) by obtainingmaximum conlikelihcodestimatesof a reduced-form tractdurationequation41to generatea sample selectioncorrectionH foreach observation,adding theestimatedvalues of H to (5) and thenestimatingtheaugmentedequation of H is thena (5) usingOLS. The coefficient consistentestimateof the covariance of u and v. The resultsare reportedin Table 7.42 The OLS resultsappear in columns1 and 2 and theOLS resultswithH introducedappear in columns 3 and 4. In both cases, estimates with and withoutquadratic termsin plant and utilityquantitiesare reported.The re- 41QUANTITY and LOG-QUANTITY are treatedas being endogenous. 42 The mean value forH is 0.076. sults are generallyconsistentwithmyexpectations. Larger plants tend to place larger ordersand utilitieswithlargeraggregaterequirementsdo so as well,althoughtheutility effectis generallysmall. The quadratic in different plant quantityis not significantly of themine-mouth fromzero.The coefficients dummyhas the predictedsign and is estimated fairlyprecisely.Mine-mouthplants have contractsthat are nearly 1.5 million tons larger (30 trillionBtu's) than other in plants ceterisparibus.Regionaldifferences supply characteristicslead to larger contractswithwesternsuppliersthan withsuppliers elsewhere. The coefficientof the correctionvariable H is negative,although not quite significantat the 5 percentlevel (two-tailedtest),implyingthatthe errorsin the durationand quantityequationare negatively correlated.Including this correction on theresults does not have dramaticeffects for the coefficients of interest, however.The primaryeffectis to reduce the magnitude of thecoefficient of WEST. and significance This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. I JOSKOW: CONTRACT DURA TION Finally,for the record,I reporttwo-stage least square (2SLS) estimatesof equation (2) in Table 3, column9, and the equivalent of two-stageleast squares forthe maximum likelihoodestimatesof equation(2) in Table 5, column 9.43 The estimatesdo not change in any importantway fromthose obtained using the other estimatingtechniques in eithercase. VI. Conclusions The purpose of this paper has been to examine empiricallyhypothesesabout the relationshipbetween the duration of coal contractsand thepresenceof thethreetypes of relationshipspecificinvestments discussed by Williamson (1983).4 I arguethatas rela- investmentsbecome more tionship-specific important,the partieswill findit advantageous to rely on longer-term contractsthat specifythe termsand conditionsof repeated 43I assumed that LOG-QUANTITY is endogenous, and use the right-handside variablesin (5) as instrumentsforthe 2SLS resultsreportedin Table 3, col. 9. Obtainingtheequivalentof 2SLS estimatesforthecase in whichI treatthe sample as being truncatedand use maximumlikelihood techniques,is more complicated. Following Maddala (pp. 234-40) and L. S. Lee et al. (1979), I proceeded in the followingway. First,I estimate a reduced-form durationequationusingmaximum likelihood techniques.This allows me to obtain consistentestimatesof Hi which can in turnbe used to obtain consistentestimatesof a contractquantityequation as discussed above. I use these estimatesof the quantityequation to obtainpredictedvalues of QUANTITY or LOG-QUANTITY which are then used in place of QUANTITY and LOG-QUANTITY to estimate equations (1) and (2) using maximumlikelihood techniques. The results reportedin Table 5, col. 9, assume that log(PLANT QUANTITY), log(UTILITY QUANTITY), MIDWEST, WEST, and MINEMOUTH are exogenousvariables.These variablesare used to estimatea reduced-form durationequationusing themaximumlikelihoodtechniquedescribedearlier.An equation forlog(QUANTITY) is thenestimatedusing log( PLA NT QUANTITY), log(UTILITY QUA NTITY), MIDWEST, WEST, MINE-MOUTH, and H (generatedfrom the reduced-form durationestimates) using OLS. The predictedvalues fromthisequationare thenused insteadof LOG-QUANTITY in (2) to obtain the maximumlikelihoodestimatesreported.Specification (1) has also been estimatedusing this approach. The resultsappear to be robust. "As well as similar considerationsof transactionspecificinvestments identified by Klein et al. 183 transactionsex ante, ratherthan relyingon repeated bargaining.I make use of a large sample of coal contractsto examine this hypothesis.The empiricalresultsobtained provide fairlystrongsupport for this hypothesis.They are quiterobustto alternative model specifications,samples,and estimating techniques.The resultstherefore provide additional empirical support for the view that the structureof verticalrelationships between buyers and sellers is stronglyaffected by variationsin the importanceof relationship-specific investments. APPENDIX Here I discuss the sources of the data and the constructionof the variables.The data base thatI rely on was constructedfroma varietyof sourcesforuse in a research project focusingon vertical relationships betweenelectricutilitiesand coal suppliers.This is one of threepapers thathas been producedso farfromthis project. The constructionof the data base began with the choice of contractsto use in the analysis. Contracts werechosenif theyappearedin boththe1981 and 1983 editionsof The Guideto Coal Contracts(Pasha Publications) and for which informationnecessary for the project was reported.Contractshad to appear in both thatwas desired publicationsbecause some information appeared in one or the otherpublication,but not both. This also made it possible to check for errorsand inconsistenciesin the contractcharacteristics reported. To appear in both publications,contractshad to be in forcein 1979. In fivecases, actual contractswere used to supplement the data available from the primary sources. This collectionprocedureresultedin a sampleof 296 contracts(of which 277 had enoughinformation to be used here) which generallyhad the followinginformation, some of whichis used in thispaper and some of whichI am usingin relatedwork: 1. Informationrequired to calculate the agreed upon durationof the contract(see discussionbelow). 2. Contractquantitiesfor1979,1980, and/or 1981 in tons. 3. The contractspecifications forthe Btu content and the sulfurcontentof thecoal. 4. The identityof thesellerand thelocationof the mine. 5. The identityof thebuyerand thedestinationof the coal. 6. The base price for the coal at the time the contractwas signed. 7. The actual price for the coal in 1979, 1980, and/or 1981. 8. Deliveredquantitiesin 1979,1980,and/or 1981. 9. Actual Btu and sulfurcontentof thecoal. Once the contractsample was selected,individual contractswere matchedwithspecificutilitiesand power This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions 184 THE AMERICAN ECONOMIC REVIEW plants (wherepossible). Two publicationswere utilized to obtain coal quantityand quality(Btu content)information by plant and utility:Cost and Qualityof Fuels for the Electric UtilityIndustry(U.S. Departmentof Energy,variousyears)and Steam ElectricPlant Factors (National Coal Association,variousyears). For public utility holding companies, coal utilization for subsidiaries was aggregated.Jointlyowned plants were assignedto the operatingcompany. The threecoal supplyregionsrepresentaggregations of smallerU.S. Bureau of Mines (BOM) districts.The West was definedas includingBOM districts16 through 23. The MidwestincludedBOM districts5, 9, 10, 11, 12, 14, and 15. BOM district15 includesTexas, but we have no contracts for coal produced in Texas. The East includes coal fromthe remainingBOM districts,pribetweenregions marilyin Appalachia. The differences is discussed in more detail in my 1985 paper. The KeystoneCoal IndustryManual (Mining Information Services)and an atlas wereused to help locate minesin specificBOM districts. and construction are as folThe variable definitions lows: DURATION: Contract Duration: The contract data base generallyprovidesinformationfor the date the contractwas executed,the terminationdate and (less frequently)the date of firstdeliveryof coal. A specificmonth and year is oftenprovided,but sometimes the source specifiesonly years.Because the contractexecutiondate was available moreoftenthan the date of firstdeliveryand because the two are generally quite close together,contractdurationwas measuredas contractterminationyear minusexecutionyear. Initial withdurationmeasuredusingthe date experimentation of firstdeliveryor usingmonthand yearindicatedthat the resultswere unaffected, so the definitionthatpreservedthe largestnumberof contractswas used. QUANTITY: Annual contractquantityin trillion Btu's. The contractedtonnagereportedfor1980 (if that was not available,or obviouslynot representative, 1979 1981 wereused instead)was multiplied or, alternatively, by the contractedBtu contentof the coal to arriveat the contractquantityvariable. MINE-MOUTH: Mine-mouthdummy variable thatis equal to one if the plant is a mine-mouth plant in my 1985 paper and zero otherwise.The information in the combined with the coal destinationinformation Guide To Coal Contractswas used to constructthis variable.The Navajo and Mohave plantswereincluded in thiscategoryas well since theyhave economiccharacteristicsverymuchlike a mine-mouth plant. MID WEST: A regional dummy variable that equals one if the coal is froma midwesternmine (as definedabove) and zero otherwise.The contractdata on minelocation. base providesinformation WEST: A regional dummyvariable that equals one if coal is froma westernmine (as definedabove) and zero otherwise. PLANT QUANTITY: Annual plant utilizationof coal. Coal utilizationby a plant to which a specific contractis dedicated (at least 90 percentof the coal deliveredto a single plant) for 1980 (1979 or 1981 if necessaryto match QUANTITY) in trillionBtu's. Ob- MARCH 1987 tained fromthe Departmentof Energyand National Coal Associationpublicationsidentifiedabove. UTILITY QUANTITY: Annual utilityutilization of coal. Coal utilizationin 1980 (or 1979 or 1981 if necessaryto matchotherdata) by theutilityoperatinga plant to which a contractis dedicated.Obtained from the Departmentof Energyand National Coal Association publicationsidentifiedabove. PLANT PROPORTION: Delivered contract quantityin Btu's divided by plant utilizationin Btu's. Delivered contractquantityin tonswas pulled fromthe contractinformationfor1980 (or 1979 if 1980 was not available), and multipliedby the deliveredBtu content of the coal to obtain quantitiesdeliveredto a specific plant under a contractdedicated to that plant. This figurewas thendividedby PLANT QUANTITY. UTILITY PROPORTION: Delivered contract quantity in Btu's (as definedabove in definitionof PLANT PROPORTION) dividedby utilityutilization of coal in Btu's for1980 (1979 or 1981 otherwise). PLANT/UTILITY: PLANT QUANTITY divided by UTILITY QUANTITY. YEAR: The yearspecifiedas theexecutiondate of the contract. DATE-71: A dummyvariablethatequals one for contractssignedin 1971, 1972, and 1973. DATE-74: A dummyvariablethatequals one for contractssignedin 1974, 1975, 1976, and 1977. DA TE-78: A dummyvariablethatequals one for contractssignedin 1978 and 1979. The mean, standarddeviation,minimum,and maximum values of these variablesforthe 277 observation sample and the 169 observation(singledeliverypoint) subsampleare containedin Tables 1 and 2. The data for the variables used in this paper are available upon request. REFERENCES Crocker,Keith J. and Masten, Scott E., "Miti- gating Contractual Hazards: Unilateral Options and ContractLength,"Working Paper No. 449, Graduate School of Business Administration,Universityof Michigan,March 1986. Dubin, Jeffrey and Rivers,R. Douglas, Statisti- cal SoftwareTools,Version1.0, Pasadena, March 1986. Goldberg, Victor and Erickson, John, " Long Term Contracts for Petroleum Coke," Working Paper No. 206, Departmentof Economics, University of CaliforniaDavis, 1982. Hart, Oliver and Holmstrom,Bengt,"The The- ory of Contracts," Departmentof Economics Working Paper No. 418, MIT, March 1986. Heckman,James,"The CommonStructure of This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions VOL. 77 NO. I JOSKOW: CONTRACT DURA TION StatisticalModels of Truncation,Sample Selection and Limited Dependent Variables and a Simple EstimatorFor Such Models," Annals of Economicand Social Measurement,1976, Vol. 5, 475-92. "Sample SelectionBias as a SpecificationError,"Econometrica, January1979, 47, 153-61. Joskow,Paul L., "Vertical Integrationand Long Term Contracts:The Case of CoalburningElectricGeneratingPlants,"Journal of Law, Economicsand Organization, Spring1985, 1, 33-80. "Price Adjustmentin Longer Term Contracts: The Case of Coal," mimeo., May 1986. and Mishkin, "Electric UtilFrederick, ity Fuel Choice Behaviorin the United States," InternationalEconomic Review, October 1977, 18, 719-36. andSchinalensee, "The PerforRichard, manceof SteamElectricGeneratingPlants in the United States: 1960-1989," Departmentof Economics Working Paper No. 379, MIT, July1985. Judge,GeorgeG. et al., The Theoryand Practice of Econometrics, New York: Wiley & Sons, 1985. Klein,Benjamin,Crawford, Robertand Alchian, Armen,"Vertical Integration,Appropriable Rents and the CompetitiveContracting Process,"JournalofLaw and Economics, October 1978, 21, 297-326. Lee, L. S., Maddala,G. S. and Trost,R. P., "Testing for StructuralChange By DMethodsin SwitchingSimultaneousEquationsModels," Proceedings of theAmerican StatisticalAssociation,Business and Economics Section,1979,461-66. Maddala,G. S., Limited-Dependent and Qualitative Variables in Econometrics,Cambridge:CambridgeUniversity Press,1983. Masten,ScottE., "The Organizationof Production: EvidencefromtheAerospaceIn- 185 dustry,"Journalof Law and Economics, October 1984, 27, 403-18. Adap_ andCrocker, Keith,J.,"Efficient tation in Long-TermContracts:Take-orPay ProvisionsforNatural Gas," American EconomicReview,December1985, 75, 1083-93. Kirkand Teece,David,"Supplier Monteverde, SwitchingCosts and VerticalIntegration in the AutomobileIndustry,"Bell Journal of Economics,Spring1982, 13, 206-13. J.Harold,"Complexityin Long-term Mulhern, Contracts: An Analysis of Natural Gas Contract Provisions," Journal of Law, Economicsand Organization,Spring1986, 2, 105-117. Schmalensee,Richardand Joskow,Paul L., "Estimated Parameters as Independent Variables: An Applicationto the Costs of Steam ElectricGeneratingUnits,"Journal April 1986, 31, 275-305. of Econometrics, OliverE., "Transaction-CostEcoWilliamson, nomics: The Governanceof Contractual Relations,"JournalofLaw and Economics, October 1979, 22, 233-61. _ " Credible Commitments: Using Hostages to SupportExchange,"American Economic Review, September 1983, 73, 519-40. of Capital, The EconomicInstitutions ism,New York: Free Press,1985. Zimmerman, Martin,The U.S. Coal Industry: Economics of Policy Choice, Cambridge: MIT Press,1981. Services, KeystoneCoal InMiningInformation dustryManual, New York, 1981. SteamElectricPlant NationalCoal Association, Factors,Washington,variousyears. Pasha Publications, Guide To Coal Contracts, Washington,1981; 1983. of Energy, U.S. Department EnergyInformation Costand QualityofFuelsfor Administration, Electric UtilityPlants, DOE/IEA-0191, Washington,variousyears. This content downloaded on Mon, 7 Jan 2013 16:52:23 PM All use subject to JSTOR Terms and Conditions