Measuring Market Inef®ciencies in California’s

Measuring Market Inef®ciencies in California’s Restructured

Wholesale Electricity Market

By SEVERIN BORENSTEIN, JAMES B. BUSHNELL, AND FRANK A. WOLAK*

We present a method for decomposing wholesale electricity payments into produc-tion costs, inframarginal competitive rents, and payments resulting from the exer-cise of market power. Using data from June 1998 to October 2000 in California, we&nd signi&cant departures from competitive pricing during the high-demand sum-mer months and near-competitive pricing during the lower-demand months of the&rst two years. In summer 2000, wholesale electricity expenditures were $8.98billion up from $2.04 billion in summer 1999. We &nd that 21 percent of thisincrease was due to production costs, 20 percent to competitive rents, and 59percent to market power. (JEL L1, L9)

In the spring of 2000, the momentum behinda dramatic restructuring of the electricity indus-try appeared to be irresistible. There were fourregions of the United States with independentsystem operators running spot markets forwholesale electricity—California, PJM (majorparts of Pennsylvania, New Jersey, Maryland,Delaware, Virginia, and the District of Colum-bia), New England, and New York. Severalother states were undertaking initiatives to re-structure their electricity sector along similar

lines. Beginning in summer 2000, however,soaring wholesale electricity prices in Califor-nia made international news and threatened the9nancial stability of the state. The disruptions inCalifornia slowed, and threatened to reverse, themovement toward restructured electricity mar-kets in the United States and elsewhere.

In the aftermath of California’s electricitycrisis, policy makers debated the correct lessonsto take from the state’s restructuring as well asthe proper regulatory response to the crisis.Many of the answers to the questions beingdebated depend upon a proper diagnosis of theproblems that disrupted California’s power sec-tor during this period. Were soaring power coststhe result of market “fundamentals” such asrising fuel prices, environmental cost, and ascarcity of generating capacity? Or were powersuppliers able to exercise signi9cant marketpower? In this paper we estimate the extent towhich each of these factors—input costs, scar-city, and market power—in@uenced market out-comes in the California power market from1998 through 2000. We analyze input and out-put prices, generator variable costs, and actualproduction quantities to measure the degree towhich California wholesale electricity pricesexceeded competitive levels. We also addressthe question of the ef9ciency impacts of marketpower in this market.

While market power has been studied andestimated in many industries, there has been

* Borenstein: Haas School of Business, University ofCalifornia, Berkeley, CA 94720, University of CaliforniaEnergy Institute, and NBER (e-mail: [email protected]); Bushnell: University of California EnergyInstitute, 2539 Channing Way, Berkeley, CA 94720 (e-mail: [email protected]); Wolak: Department ofEconomics, Stanford University, Stanford, CA 94305, andNBER (e-mail: [email protected]). Borenstein is amember of the Governing Board of the California PowerExchange. Bushnell was a member of the Power Ex-change’s Market Monitoring Committee during 1999–2000. He is currently a member of the California Indepen-dent System Operator’s Market Surveillance Committee,of which Wolak is the Chair. The views expressed in thispaper do not necessarily re@ect those of any organizationor committee. For helpful discussions and comments, wethank Carl Blumstein, Roger Bohn, Peter Cramton, DavidGenesove, Rob Gramlich, Paul Joskow, Ed Kahn, AlvinKlevorick, Christopher Knittel, Patrick McGuire, CatherineWolfram, an anonymous referee, and participants in numer-ous seminars. Jun Ishii, Matt Lewis, Erin Mansur, StevePuller, Celeste Saravia, and Marshall Yan provided excel-lent research assistance.

1376

little attention paid to intertemporal variation inthe ability to exercise market power. For indus-tries in which the good is storable, such inter-temporal variation is necessarily small, becauseinventories greatly reduce intertemporal supplyvariation, and possibly, demand variation. Inmarkets for nonstorable goods, including elec-tricity and most services, this is not possible.The problem is exacerbated in electricity be-cause demand is very inelastic in the short run,and supply becomes very inelastic as productionapproaches the system-generation capacity. Rec-ognizing the dynamics of market power is likelyto be important in both determining its causes andcrafting remedies as part of the evolving publicpolicy toward electricity restructuring.

Luckily, due to the history of regulation inelectricity markets, data exist on the hourly out-put of all generating units and transmissionpower @ows. In addition, information collectedon the technical characteristics of each Califor-nia generating unit during the regulated monop-oly regime allows very accurate estimation ofgenerating unit-level variable costs.

We 9nd that, due to rising input costs, even aperfectly competitive California electricity mar-ket would have seen wholesale electricity ex-penditures triple between the summers of 1998and 2000.1 Market power, however, also playeda very signi9cant role. In summer 1998, 25percent of total electricity expenditures could beattributed to market power, a 9gure that in-creased to 50 percent in summer 2000. Theincreased percentage margins due to marketpower combined with substantial productioncost increases for marginal producers to create adrastic rise in absolute margins and, thus,pushed the market into a crisis later in thesummer of 2000.

In Section I, we discuss the issues raised inestimating market power in electricity marketsand the consequences of market power. Wepresent an overview of California’s electricitymarket in Section II. In Section III, we describethe estimation technique in detail in the contextof the California market, addressing each com-ponent of the market and outlining the assump-

tions made in implementing the analysis. We tryto take a conservative approach, interpreting thedata in a way that would be likely to understatethe degree of market power exercised. In Sec-tion IV, we present estimates of premia of ac-tual prices over the competitive levels. InSection V, we attempt to parse changes in com-petitive revenues between changes in actualcosts and changes that re@ect rents to inframar-ginal competitive sellers. We conclude in Sec-tion VI.

I. Market Power Analysis in theElectricity Industry

During most of the 1990’s, regulatory evalu-ation of short-run horizontal market power inelectricity focused on concentration measures,such as the Her9ndahl-Hirschman index. Unfor-tunately, such measures are a poor indicator ofthe potential for, or existence of, market powerin the electricity industry, because the industryis characterized by highly variable price-inelasticdemand, signi9cant short-run capacity constraints,and extremely costly storage.2 It is easy to showthat in such circumstances, 9rms with verysmall market shares could still exercise signi9-cant market power.

We use data collected on the technologicalcharacteristics of generating units located inCalifornia to construct a competitive marketcounterfactual that we compare to actual mar-ket outcomes. This competitive counterfactualmodels each 9rm as a price-taker that would sellpower from a given plant so long as the price itreceived was greater than its incremental cost ofproduction. Of course, the cost of selling a unitof electricity can be greater than the simpleproduction costs if the 9rm has an opportunitycost that is greater than its production cost, suchas the revenue the 9rm would get from sellingpower or reserve capacity in a different locationor market. On the other hand, a high price in analternative market can re@ect market power inthat market, resulting in the transmittal of high

1 For the purpose of this analysis, we de9ne the summerto be June through October of each year.

2 See Borenstein et al. (1999) for a more detailed dis-cussion of the applicability of concentration measures tomarket power analysis in electricity markets and citationsto regulatory decisions that have relied on concentrationindices.

1377VOL. 92 NO. 5 BORENSTEIN ET AL.: ELECTRICITY MARKET POWER

prices across markets by the response of com-petitive suppliers. We discuss these alternativeopportunities below and how we account forthem in our analysis.

Thus far, we have discussed only situations inwhich a 9rm unilaterally exercises marketpower. Antitrust law is most often concernedwith collusive attempts to exercise marketpower. Unfortunately, many of the attributesthat facilitate collusion are present in electricitymarkets: In most electricity markets, 9rms playrepeatedly, interacting on a daily basis, so thereis opportunity to develop subtle communicationand collusive strategies. The payoff from cheat-ing on a collusive agreement may be limited dueto capacity constraints on production, thoughfor the same reason, the ability to punish defec-tors may be limited. Finally, the industry hasfairly standardized production facilities, so ho-mogeneous costs may make it easier for 9rms toattain tacit or explicit collusive outcomes. Allthat said, we have not explored the question oftacit or explicit collusion among 9rms in theCalifornia market as a potential cause of pricesin excess of competitive levels.3 Rather, in thispaper we focus on the competitiveness of mar-ket outcomes.

In focusing on market outcomes, there aretwo indicators that clearly distinguish marketpower, and each leads to a distinct estimationtechnique. First, in a competitive market, a 9rmis unable to take any action, including outputdecisions or offer prices, that signi9cantly af-fects the price in a market. This suggests amethod of estimation that involves studying thebidding and output supply decisions of each9rm in the market to detect successful attemptsto affect prices. This is the general approachused by Wolak and Robert H. Patrick (1997),Catherine D. Wolfram (1998), Roger Bohn etal. (1999), Bushnell and Wolak (1999), Wolak(2000), and Puller (2001).

The second empirical approach is at the mar-ket level, and this is the one that we adopt here.We examine whether the market as a whole issetting competitive prices given the productioncapabilities of all players in the market. As

such, this approach is less vulnerable to thearguments of coincidence, bad luck, or igno-rance that may be directed at analysis of theactions of a speci9c generator. It is less infor-mative about the speci9c manifestations ofmarket power, but it is effective for estimatingits scope and severity, as well as identifyinghow departures from competitive outcomesvary over time. This is the general approachused in Wolfram (1999). At least two papers,Erin T. Mansur (2001) and Paul L. Joskow andEdward Kahn (2002), utilize both of theseapproaches.

A potential drawback of the market-level ap-proach is that it captures all inef9ciencies in themarket, some of which may not be due to mar-ket power. If, for instance, low-cost generatorswere systematically held out of production sim-ply due to a faulty dispatch algorithm, thatwould impact the estimate of market power.During the period we study, the California mar-ket clearly still had a number of design @awsthat may have contributed to inef9cient dispatchand market pricing. For the great majority ofthese, however, the @aw would be benign if9rms acted as pure price-takers, rather than ex-ploiting these design @aws to affect the marketprice. Furthermore, we 9nd that, over substan-tial periods of time, prices did not signi9cantlydiffer from our estimates of marginal cost, in-dicating that there were no systemic inef9cien-cies raising prices in all periods. Still, ourestimates must be taken with the caveat thatthey include failures to achieve competitivemarket prices for reasons other than marketpower, including bad judgment and confusionon the part of some generators or market-making institutions.

The Consequences of Market Power

In analyzing the ef9ciency consequences ofmarket power in electricity, one must beginfrom the recognition that short-run electricitydemand currently exhibits virtually zero priceelasticity. Almost none of the customers in Cal-ifornia, or anywhere else in the United States,are charged real-time retail electricity pricesthat vary hour-to-hour as wholesale prices do.Because the extent of market power varies tre-mendously on an hourly basis, the absence of

3 See Steven L. Puller (2001) for an analysis of thisissue.

1378 THE AMERICAN ECONOMIC REVIEW DECEMBER 2002

very-short-run elasticity is critical to under-standing its consequences.4 In studying Califor-nia speci9cally, one must also consider that,during the 1998–2001 transition period, end-use consumers were insulated from energy price@uctuations by the Competition TransitionCharge (CTC). The CTC was implementedalong with the restructuring of the industry inorder to allow the incumbent utilities to recovertheir stranded generation costs. Due to the CTC,the vast majority of end-use consumers faced9xed retail rate schedules during the transitionperiod.5 Thus, the CTC greatly lessened eventhe monthly elasticity of 9nal consumer elec-tricity demand.

Due to the extreme short-run inelasticity ofdemand, market power in electricity marketshas little effect on consumption quantity orshort-run allocative ef9ciency. As described be-low, however, generating companies in Califor-nia vary markedly in their costs and generationcapacity, so the exercise of market power byone 9rm can lead to an inef9cient reallocationof production among generating 9rms: a 9rmexercising market power will restrict its outputso that its marginal cost is below price (andequal to its marginal revenue), while other 9rmsthat are price-takingwill produce units of outputfor which their marginal cost is virtually equalto price. Thus, there will be inef9cient produc-

tion on a marketwide basis as more expensivecompetitive production is substituted for lessexpensive production owned by 9rms with mar-ket power. This is the outcome Wolak andPatrick (1997) described in the U.K. market,where higher-cost combined-cycle gas turbinegenerators owned by new entrants provide base-load power that could be supplied more cheaplyby coal-9red plants that were being withheld bythe two largest 9rms. Joskow and Kahn (2002)9nd evidence of withholding by large 9rms inthe California market.

In addition, several recent analyses havedemonstrated that the exercise of market powerin an electricity network can greatly increase thelevel of congestion on the network.6 This in-creased congestion impacts negatively both theef9ciency and the reliability of the system. Mar-ket power can also lead 9rms to utilize theirhydroelectric resources in ways that decreaseoverall economic ef9ciency.7

Lastly, electricity prices in@uence long-termdecision-making in a way that can seriouslyimpact the economy and generation investment.While it has been pointed out that high pricesshould spur new investment and entry in elec-tricity production, these investments may not beef9cient if motivated by high prices that arecaused by market power, which may indicate aneed not for new capacity, but for the ef9cientuse of existing capacity. Arti9cially high pricesalso can lead some 9rms not to invest in pro-ductive enterprises that require signi9cant useof electricity, or to inef9ciently substitute to lesselectric-intensive production technologies.

Beyond the ef9ciency considerations, marketpower has potentially large and important redis-tributional effects. The California electricity cri-sis of 2000–2001 illustrates both the immensepotential size of these effects and the dif9cultyof analyzing them. The transitional retail ratefreeze associated with the CTC meant that theutilities bore the brunt of the wholesale priceincreases. The utilities’ eventual response wasto declare bankruptcy in one case, and threatento in another, so the ratepayers or taxpayers

4 In California and elsewhere, time-of-use rates are com-mon for large users. These price schedules generally havepreset peak, shoulder, and off-peak rates, which are changedonly twice per year. They do not distinguish, for instance, aweekday afternoon with extremely high wholesale pricesfrom a more moderate weekday afternoon. Borenstein(2001, 2002) argues that time-of-use rates are an extremelypoor substitute for real-time electricity pricing and thatreal-time pricing would greatly mitigate wholesale pricevolatility. Patrick and Wolak (1997) estimate the within-dayprice responsiveness of industrial and commercial custom-ers facing real-time half-hourly energy prices in the Englandand Wales electricity market. They argue that an electricitymarket would be much less susceptible to the exercise ofmarket power if even one-quarter of peak demand had theaverage level of price responsiveness that they estimate.

5 Even “direct access” consumers, who bought energyfrom some source other than their incumbent utility, wereinsulated from wholesale energy price @uctuations in theshort run by the CTC. This is because the stranded costcomponent paid by all consumers was calculated in a waythat moved inversely to the energy price: the higher theenergy price, the lower the CTC payment for that hour.

6 See Judith B. Cardell et al. (1997), Bushnell (1999),Borenstein et al. (2000), and Joskow and Jean Tirole (2000).

7 See Bushnell (2003).


ultimately became the bearers of much of thesecosts. At this writing, it is still unclear who willbear what share of the expense, and how muchof the revenues paid to generators will be re-funded to buyers under orders from federalregulators.

II. The California Electricity Market

Through December 2000, the two primarymarket institutions in California were the PowerExchange (PX) and the Independent SystemOperator (ISO). The PX ran a day-ahead andday-of market for electrical energy utilizing adouble-auction format. Firms submitted bothdemand and supply bids, then the PX set themarket-clearing price and quantity at the inter-section of the resulting aggregate supply anddemand curves. In the PX day-ahead market,which was by far the largest market in Califor-nia, 9rms bid into the PX offers to supply orconsume power the following day for any or allof the 24 hourly markets. The PX markets wereeffectively 9nancial, rather than physical; 9rmscould change their day-ahead PX positions bypurchasing or selling electricity in the ISO’sreal-time electricity spot market.8

The PX was not the only means for buyersand sellers to transact electricity in advance ofthe actual hour of supply. A buyer and sellercould make a deal bilaterally. All institutionsthat scheduled transactions in advance, includ-ing the PX, were known as “scheduling coordi-nators” (SCs).9 Because SCs use the transmissiongrid to complete some transactions, they arerequired to submit the generation and loadschedules associated with these transactions tothe ISO.

The ISO is responsible for coordinating theusage of the transmission grid and ensuring thatthe cumulative transactions, or schedules, donot constitute a reliability risk, i.e., are not

likely to overload the transmission system.10 Asthe institution responsible for the real-time op-eration of the electric system, the ISO must alsoensure that aggregate supply is continuouslymatched with aggregate demand. In doing so,the ISO operates an “imbalance energy” market,which is also commonly called the real-time, orspot, energy market. In this market, additionalgeneration is procured in the event of a supplyshortfall, and generators are relieved of theirobligation to provide power in the event thatthere is excess generation being supplied to thegrid. Like the PX, this market is run through adouble-auctionprocess, although of slightly dif-ferent format. Firms that deviate from their for-mal schedules are required to purchase (or sell)the amount of their shortfall (or surplus) on theimbalance energy market. During our sampleperiod, no further penalties were assessed fordeviating from an advance schedule. The imbal-ance energy market therefore serves as the defacto spot market for energy in California. Dur-ing our sample period, the ISO imbalance en-ergy market constituted less than 5 percent oftotal energy sales with the PX accounting forabout 85 percent and the remainder taking placethrough bilateral trades.

The ISO also operates markets for the acqui-sition of reserve, or stand-by, capacity. Reservecapacity is used to meet unexpected demandpeaks and to adjust production at differentpoints on the grid in order to relieve congestionon the transmission grid while still meeting alldemand. These reserves, known as “ancillaryservices,” are purchased through a series ofauctions that determine a uniform price for thecapacity of each reserve purchased. Most of thereserve capacity is still available to provideimbalance energy in real time, and thereforewill impact the spot price. A production unitcommitted to provide reserve capacity duringan hour would therefore earn a capacity pay-ment for being available and, if called upon inreal time, would earn the imbalance energyprice for actually providing energy.

“Regulation reserve,” the most short-term re-8 Though the transaction costs of trading in the PX and

ISO differed, these differences were negligible relative tothe costs of the underlying commodity, electrical energy.

9 In January of 2001, the PX ceased operation and theCalifornia Department of Water Resources assumed respon-sibility for the bulk of wholesale purchases on behalf of allinvestor-owned utilities in California, negotiating bilateralpurchases and operating as its own SC.

10 Unlike the PX, the ISO continued to function in ap-proximately its original role through the 2000–2001 elec-tricity crisis.


serve, is treated differently. Regulation reserveunits are directly controlled by the ISO andadjusted second-by-second in order to allow theISO to continuously balance supply and de-mand, and to avoid overloading of transmissionwires. For this reason, we treat it differently inour analysis as described later.

A. Market Structure of California Generation

The California electricity generation marketat 9rst glance appears relatively unconcentrated.The former dominant 9rms, Paci9c Gas & Elec-tric (PG&E) and Southern California Edison(SCE), divested the bulk of their fossil-fuel gen-eration capacity in the 9rst half of 1998 andmost of the remainder in early 1999. Most of thecapacity still owned by these utilities after thedivestitures was covered by regulatory sideagreements, which prescribed the price theseller was credited for production from theseplants independent of the PX or ISO marketprices. These divestitures left the generationassets in California more or less evenly distrib-uted between seven 9rms. The generation ca-pacity of these 9rms that was located within theISO system during July 1998 and July 1999 islisted in Table 1. “Fossil” includes all plantsthat burn natural gas, oil, or coal to power theplant, but over 99 percent of the output fromthese plants is fueled by natural gas. The vastmajority of capacity listed as owned by “other”9rms was composed of small independentpower projects. The market structure during2000 was largely unchanged from that of 1999.

As can be seen from Table 1, PG&E was thelargest generation company during the summerof 1998. The seemingly dominant position ofPG&E is offset to a large extent by its otherregulatory agreements. All of its nuclear gener-ation in California, for instance, is treated underrate agreements that do not depend on marketprices. More importantly, the incumbent utili-ties in California were the largest buyers ofelectricity during this time period.11

B. Analyzing Market Power in California’sElectricity Market

Critical to studying market power in Califor-nia is an understanding of the economic inter-actions between the multiple electricity marketsin the state. Participants moved between mar-kets in order to take advantage of higher (forsellers) or lower (for buyers) prices. For in-stance, if the ISO’s real-time imbalance energyprice was consistently higher than the PX day-ahead price, then sellers would reduce theamount of power they sell in the PX and sellmore in the ISO imbalance energy market.These attempts to arbitrage the PX/ISO pricedifference would cause the PX price and ISOimbalance energy price to converge. For thisreason, it is not useful to study the PX market,or any other of the California markets, in isola-tion. The strong forces of 9nancial arbitragemean that any change in one market that affects

11 The utilities had no incentive to raise market pricesbecause they were net buyers of electricity and the revenuefrom power that they sold into the PX was just netted outfrom their power purchase costs. In fact, the CTC mecha-nism paid the three investor-owned utilities the difference

between 9xed wholesale price (implicit in their frozen retailrate) and the hourly wholesale price per unit of energyconsumed in their distribution service territory.

TABLE 1—CALIFORNIA ISO GENERATION

COMPANIES (MW)

July 1998—online capacity

Firm Fossil Hydro Nuclear Renewable

AES 4,071 0 0 0Duke 2,257 0 0 0Dynegy 1,999 0 0 0PG&E 4,004 3,878 2,160 793Reliant 3,531 0 0 0SCE 0 1,164 1,720 0SDG&E 1,550 0 430 0Other 6,617 5,620 0 4,267

July 1999—online capacity

Firm Fossil Hydro Nuclear Renewable

AES 4,071 0 0 0Duke 2,950 0 0 0Dynegy 2,856 0 0 0PG&E 580 3,878 2,160 793Reliant 3,531 0 0 0SCE 0 1,164 1,720 0Mirant 3,424 0 0 0Other 6,617 5,620 430 4,888

Source: California Energy Commission (www.energy.ca.gov).


that market price will spill over into the othermarkets.12

This interaction of the different Californiaelectricity markets means that we must studythe entire California energy market in order toanalyze market power in the state. For thisreason, in the analysis below we look at allgeneration in the ISO control area regardless ofwhether the power from a generating plant isbeing sold through the ISO, the PX, or someother scheduling coordinator.

Recognition that the California power marketis effectively an integrated market due to arbi-trage forces yields two other important insights.First, although the California market has somelarge buyers of electricity directly purchasingfrom the transmission network who may re-spond to hourly wholesale prices, the large util-ity distribution companies (UDCs) cannotcontrol the level of end-use demand of theircustomers because these customer face priceschedules that do not vary with the hourlywholesale price. The UDCs cannot thereforereduce end-use consumption in a given hour inorder to lower overall power purchase costs.They did have some limited freedom as towhich market they used for purchase of theirrequired power, choosing between buying day-ahead in the PX and spot purchases from theISO imbalance energy market. Nonetheless, be-cause sellers could move between markets aswell, ultimately the buyers had no ability toexercise monopsony power, because they couldnot reduce their hourly demand for energy.

The second insight from a recognition ofmarket integration involves the impact of pricecaps in the various markets. Because the ISOimbalance energy market was the last in a se-quence of markets, the level of the price cap inthe imbalance energy market fed back to forman implicit cap on prices in the other advancemarkets. That is, knowing that the maximumone might have to pay for power in real timewas capped at $250 per megawatt-hour (MWh),for example, no buyer would be willing to paymore than $250 for purchases in advance. Thus,

the aggregate demand curve in the day-aheadPX market became near horizontal at pricesapproaching the level of the price cap in the ISOimbalance energy market.13

Many of the suppliers that compete in theISO or PX also are eligible to earn capacitypayments for providing ancillary services, aswell as energy payments for generating in realtime, if they bid successfully into one of theancillary services markets. Ancillary servicestherefore represent an alternative use of much ofthe generation capacity in California. It is there-fore necessary to consider the interaction be-tween the energy and ancillary servicesmarkets. In the case of the California market,the relevant consideration is that the provisionof ancillary services in most cases does notpreclude the provision of energy in the real-timemarket. Thus, for the bulk of generation, thedecisions to sell into ancillary service capacitymarkets and real-time energy markets are notmutually exclusive.

It is important to recognize that the pool ofsuppliers available to ancillary services marketsis very similar to that available to the energymarkets. The main difference is that some gen-erators are physically unable to provide certainancillary services. Thus, there are fewer poten-tial suppliers for some ancillary services thanthere are for energy. We therefore would expectthat the energy market would be at least ascompetitive as the ancillary services markets,and probably more so. In fact the ancillaryservices markets, for a variety of reasons, ap-pear to have been signi9cantly less competitivethan the energy market during the time period ofour study.14

The other prominent opportunity for the us-age of California generation is the supply ofpower to neighboring regions. Higher prices forelectricity outside of California could produce aresult in which generators within Californiawere able to earn prices above their marginalcost, even if all generators behaved as price-takers. For this to be the case, however, theCalifornia ISO region would have to be a netexporter of power. During our sample period,

12 Borenstein et al. (2001) 9nds that although signi9cantprice differences between the PX and ISO did occur duringindividual months, overall, there was no consistent patternof the PX price being higher or lower than the ISO price.

13 See Bohn et al. (1999).14 See Wolak et al. (1998).


such conditions arose in only 17 hours out of the22,681 hours in the sample. Even in these hours,the maximum net quantity of energy exportedout of the ISO control area in any hour was amodest 608 MWh. Therefore, if we assume thattrading in these markets was ef9cient, the netexport opportunities for producers within Cali-fornia were very limited relative to the Califor-nia market.

Even if this were not the case, however, ouranalysis would fully account for the opportunitycost of exports, because under the Californiamarket structure 9rms from other states had theoption to purchase power through markets runby the PX and ISO. Thus, power exported toArizona, for example, would raise the quantitydemanded in the California market, and there-fore would increase the competitive market-clearing price within the California PX and ISOmarkets. If transmission became congested,then further exports would be infeasible and thequantity demanded in the California marketwould include only the exports up to the trans-mission constraint. Thus, the competitive mar-ket prices we estimate incorporate opportunitiesfor export from California.

III. Measuring Market Power in California’sElectricity Market

The fundamental measure of market power isthe margin between price and the marginal costof the highest cost unit necessary to meet de-mand. As discussed above, if no 9rm wereexercising market power, then all units withmarginal cost below the market price would beoperating. Even in a market in which some9rms exercise considerable market power, themarginal unit that is operating could have amarginal cost that is equal to the price. When a9rm with market power reduces output from itsplants or, equivalently, raises its offer price forits output, its production is usually replaced byother, more expensive generation that may beowned by nonstrategic 9rms.

In estimating a price-cost margin in this pa-per, we therefore must estimate what the systemmarginal cost of serving a given level of de-mand would be if all 9rms were behaving asprice-takers. In the following subsections wedescribe the assumptions and data used for gen-

erating estimates of the system marginal cost ofsupplying electrical energy in California.

A. Market-Clearing Prices and Quantities

As described above, the California electricitymarket in fact consists of several parallel andoverlapping markets. Given that generation anddistribution 9rms, as well as other power trad-ers, can arbitrage the expected price of energyacross these commodity markets, we rely uponthe unconstrained PX day-ahead energy price asour estimate of energy prices in any givenhour.15 We chose to rely upon the PX uncon-strained price because the PX handled over 85percent of the electricity transactions during oursample period and the unconstrained PX pricerepresents the market conditions most closelyreplicated in our estimates of marginal costs. Inparticular, we do not consider the costs of trans-mission congestion or local reliability con-straints in our estimates of the marginal cost ofserving a given demand. The PX unconstrainedprice is also derived by matching aggregatesupply with aggregate demand without consid-ering these constraints. The resulting market-clearing price therefore re@ects an outcome thatwould occur in the absence of transmission con-straints, just as our cost calculations re@ect theoutcome in a market in which all producers areprice-takers and there are no transmissionconstraints.16

It has been argued that the day-ahead PXprice should be expected to systematically over-state the marginal cost of energy supply becausesellers in the day-ahead market would include a

15 One might be concerned that this arbitrage would nothold in light of the requirement during our sample periodthat the three investor-owned utilities buy all of their energyfrom the PX. Given the 9nancial nature of the PX marketand availability of a number of forward market products tohedge the day-ahead PX price risk, the full meaning of thisrequirement was ambiguous. More importantly, Borensteinet al. (2001) 9nd that the PX and ISO prices track quiteclosely throughout most of our sample period.

16 We would like to emphasize again that we use the PXprice as representative of the prices in all California elec-tricity markets. This is not a study of the PX market and themarket power we 9nd is not limited to the PX market. It isthe amount we estimate to be present in all Californiaelectricity markets. Quantitatively similar results obtain us-ing day-ahead or real-time zonal prices.


premium in their offer prices to account for theopportunity of earning ancillary services reve-nues, which require that the units not be com-mitted to sell power in a forward market.However, if this were true, then the PX pricewould also be systematically higher than theISO real-time price, which would not includesuch a premium because suppliers of ancillaryservices are also eligible to sell energy in thereal-time market. Empirically this is not thecase. Over our sample period, the PX averageprice was not signi9cantly greater than the ISOaverage price.17

The interaction of these energy markets alsorequires us to use the systemwide aggregatedemand as the market-clearing quantity uponwhich we base our marginal cost estimates. Thislevel therefore includes consumption throughthe PX, other SCs, and any “imbalance energy”demand that is provided through the ISO imbal-ance energy market. Consumption from all ofthese markets is in fact metered by the ISO,which in turn allocates imbalance energycharges among SCs during an ex post settlementprocess. We are therefore able to obtain theaggregate quantity of energy supplied each hourfrom the ISO settlement data.

The acquisition of reserves by the ISO alsorequires discussion here. Since the ISO is effec-tively purchasing considerable extra capacityfor the provision of reserves, it might seemappropriate to consider these reserve quantitiesas part of the market-clearing demand level.However, with the exception of regulation re-serve, as described below, all other reserves arenormally available to meet real-time energyneeds if scheduled generation is not suf9cient tosupply market demand.18 Thus, the real-time

energy price is still set by the interaction ofreal-time energy demand—including quantitiessupplied by reserve capacity—and all of thegenerators that can provide real-time supply.Therefore, we consider the real-time energy de-mand in each hour to be the quantity that mustbe supplied, and capacity selected for reserveservices to be part of the capacity that can meetthat demand and, as such, to be part of ouraggregate marginal cost curve.

Unlike the other forms of reserve, regulationcapacity is, in a way, held out of the imbalanceenergy market and its capacity could thereforebe considered to be unavailable for additionalsupply. For this reason we add the upwardregulation reserve requirement to the market-clearing quantity for the purposes of 9nding theoverall marginal cost of supply.19

B. Marginal Cost of Fossil-FuelGenerating Units

To estimate the marginal cost of productionfor an ef9cient market, we divide productioninto three economic categories: reservoir, must-take, and fossil-fuel generation. Reservoir gen-eration includes hydroelectric and geothermalproduction. These facilities differ from all oth-ers in that they face a binding intertemporalconstraint on total production, which implies anopportunity cost of production that generally

17 See Borenstein et al. (2001). There is also a funda-mental theoretical @aw in this argument. Though optionvalue would cause a 9rm to offer power in the day-aheadmarket at a price above its marginal cost, arbitrage on thedemand side (and by sellers that do not qualify to provideancillary services) would still equalize the market prices.The equilibrium outcome would just have a reduced shareof power sold through the day-ahead market due to theforgone option value.

18 In other words, all reserve capacity that is economic atthe market price is assumed to be used to meet energydemands in real time. Due to reliability concerns, the ISOoccasionally has not utilized some types of reserve (“spin-

ning” and “nonspinning”) for the provision of imbalanceenergy even when the units are economic (see Wolak et al.,1998). The conditions under which this occurs are some-what irregular and dif9cult to predict. For the purposes ofthis analysis we have assumed that these forms of reserveare utilized for the provision of imbalance energy.

19 Regulation reserve is procured for both an upward(increasing) and downward (decreasing) range of capacity.The amount of upward regulation reserve at times reachedas high as 10.8 percent of total ISO demand, although themean percentage of upward regulation was 2.2 percent overour sample. Because the generation units that are providingdownward regulation are, by de9nition, producing energy,the capacity providing downward regulation should not beconsidered to be held out of the energy market. Note alsothat by adding regulation needs to the market demand, weare implicitly assuming that all regulation requirements aremet by generation units with costs below the market-clearing price. To the extent that some units providingregulation would not be economic at the market price, thisassumption will tend to bias downward our estimate of theamount of market power exercised.


exceeds the direct production cost. Must-takegeneration operates under a regulatory side agree-ment and is always inframarginal to the market.Because of the incentives in the regulatoryagreements, these units will always operatewhen they physically can. All nuclear facilitiesare must-take, as well as all wind and solarelectricity production. We discuss below ourtreatment of reservoir and must-take generation.

For fossil-fuel generation, we estimate mar-ginal cost using the fuel costs and generatoref9ciency (“heat rate”) of each generating unit,as well as the variable operating and mainte-nance (O&M) cost of each unit. For units underthe jurisdiction of the South Coast Air QualityManagement District (SCAQMD) in southernCalifornia, we also include the cost of NOxemissions, which are regulated under a trade-able emissions permit system within SCAQMD.The cost of NOx permits was not signi9cant in1998 and 1999, but rose sharply during thesummer of 2000. The generator cost estimatesare detailed in Appendix A. Figure 1 illustratesthe aggregate marginal cost curve for fossil-fuelgeneration plants located in the ISO control areathat are not considered to be must-take genera-tion and shows how it increased between 1998and 2000 due to higher fuel and environmentalcosts.20 Note that because the higher-cost plants

tend to be the least fuel ef9cient and the heavi-est polluters, increases in fuel costs and pollu-tion permits not only shift the supply curve, butalso increase its slope since the costs of high-cost plants increase by more than the costs oflow-cost plants.21

The supply curves illustrated in Figure 1 do notinclude any adjustments for “forced outages.”Generation unit forced (as opposed to sched-uled) outages have traditionally been treated asrandom, independent events that, at any givenmoment, may occur according to a probabilityspeci9ed by that unit’s forced outage factor. Inour analysis, each generation unit, i, is assigneda constant marginal cost mci—re@ecting thatunit’s average heat rate, fuel price, and its vari-able O&M cost—as well as a maximum outputcapacity, capi. Each unit also has a forced out-age factor, fofi , which represents the probabil-ity of an unplanned outage in any given hour.

Because the aggregate marginal cost curve isconvex, estimating aggregate marginal cost usingthe expected capacity of each unit, capi z (1 2fofi), would understate the actual expected costat any given output level.22 We therefore sim-ulate the marginal cost curve that accounts forforced outages using Monte Carlo simulationmethods. If the generation units i 5 1, ... , Nare ordered according to increasing marginalcost, the aggregate marginal cost curve pro-duced by the jth draw of this simulation, Cj(q),is the marginal cost of the kth cheapest gener-ating unit, where k is determined by

(1) k 5 arg min5x Oi 5 1

x

I~i! z capi $ q6.

20 Costs of generation shown in Figure 1 are based onmonthly average natural gas and emissions permit prices.Here and throughoutour analysis, we assume that gas pricesare competitively determined and accurately reported. If gasmarkets were not competitive or reported prices exceeded

the actual prices paid by electricity generators, as recentFederal Energy Regulatory Commission 9ndings have sug-gested, then we have underestimated the full impact ofmarket power on the wholesale price of electricity.

21 Our estimates assume that the rated capacities of theplants, capi, are strictly binding constraints. It has been pointedout to us that the plants can be run above rated capacity, butat the cost of increased wear and more frequent mainte-nance. If we incorporated this factor—about which thereseems to be very little detailed information—it would shiftrightward the industry supply curve and would increase ourestimates of the extent to which market power was exercised.

22 For any convex function C(q), of a random variableq , we have, by Jensen’s inequality, E(C(q)) $ C(E(q)).

FIGURE 1. CALIFORNIA FOSSIL-FUEL PLANTS MARGINAL

COST CURVES, SEPTEMBER


where I(i) is an indicator variable that takes thevalue of 1 with probability of 1 2 fofi , and 0otherwise. For each hour, the Monte Carlo sim-ulation of each unit’s outage probability is re-peated 100 times. In other words, for eachiteration, the availability of each unit is basedupon a random draw that is performed indepen-dently for each unit according to that unit’sforced outage factor. The marginal cost at agiven quantity for that iteration is then the mar-ginal cost of the last available unit necessary tomeet that quantity given the unavailability ofthose units that have randomly suffered forcedoutages in that iteration of the simulation. If,during a given iteration, the fossil-fuel demand(total demand minus hydro, must-take, and thesupply of imports at the price cap) exceededavailable capacity, the price was set to the max-imum allowed under the ISO imbalance energyprice cap during that period. Note that in nearlyall cases, the PX price in that hour did not hitthe price cap, so such outcomes were counted as“negative market power” outcomes in the anal-ysis. Thus, these outcomes are not driving, andif anything are reducing, our 9nding of marketpower.

We did not adjust the output of generationunits for actual outages, because the schedulingand duration of planned outages for mainte-nance and other activities is itself a strategicdecision. Wolak and Patrick (1997) present ev-idence that the timing of such outages was ex-tremely pro9table for certain 9rms in the U.K.electricity market. It would therefore be inap-propriate to treat such decisions as randomevents. Because we 9nd market power in thesummer months—high-demand periods in Cal-ifornia in which the utilities have historicallyavoided scheduled maintenance on most gener-ation—it is unlikely that scheduled maintenancecould explain these results in any case.23 Wewould expect scheduled maintenance to takeplace in the autumn, winter, and spring months,which is the time period over which we 9ndlittle, if any, market power.

The operation of generation units entails

other costs in addition to the fuel and short-runoperating expenses. It is clear that sunk costs,such as capital costs, and periodic 9xed capitaland maintenance expenses should not be in-cluded in any estimate of short-run marginalcost. More dif9cult are the impacts of variousunit-commitment costs and constraints, such asthe cost of starting up a plant, the maximumrates at which a plant’s output can be ramped upand down, and the minimum time periods forwhich a plant can be on or off. These constraintscreate nonconvexities in the production costfunctions of 9rms. For a generating unit that isnot operating, these costs are clearly not sunk.On the other hand, it is not at all clear how, orwhether, a price-taking, pro9t-maximizing 9rmwould incorporate such costs into its supply bidfor a given hour. In fact, it is relatively easy toconstruct examples where it would clearly notbe optimal to incorporate start-up costs in asupply bid.24 We do not attempt to capturedirectly the impacts of these constraints on ourcost estimates. Below, we discuss how noncon-vexities could affect the interpretation of ourresults.

C. Imports and Exports

One of the most challenging aspects of esti-mating the marginal cost of meeting total de-mand in the ISO system is accounting forimports and exports between the ISO and othercontrol areas. We can, however, observe the netquantity of power entering or leaving the ISOsystem at each intertie point, as well as thewillingness of 9rms to import and export to andfrom California.

If the power market outside of California

23 Scheduled maintenance on must-take resources andreservoir energy sources was accounted for under the pro-cedures outlined in the following subsections.

24 Consider a generator that estimates it will be “in” themarket for six hours on a given day and bids into the marketin each hour at a level equal to its fuel costs plus one-sixthof its start-up cost. Consider the results if the market pricein one hour rises to a level suf9cient to recover all start-upcosts, but in all subsequent hours remains at a level abovethe unit’s fuel costs, but below the sum of its fuel cost plusthe prorated start-up costs. If this unit committed to operatein the one hour that it covers its fuel costs plus one-sixth ofits start-up costs, but stayed “out” of the market in subse-quent hours, it is not maximizing pro9ts, because it couldhave earned an operating pro9t at market-clearing prices inthe 9ve remaining hours.


were perfectly competitive, then the marginalgenerator that is importing into Californiawould, absent transmission constraints, have amarginal cost about equal to the market price inCalifornia. When market power is exercisedwithin California, this would mean that, in aneffort to drive up price, some in-state generatorsare withdrawing (or raising the offer price on)their marginal generation and allowing moreexpensive imported power to be substituted forit. Thus, in the absence of market power, wewould see lower imports. This means that thecost of serving the demand that remains afterthe competitive level of imports is netted outwould be higher than the cost of serving thedemand that remains after the true level of im-ports is adjusted for.25

Figure 2 illustrates a hypothetical marginalcost curve of the in-state generation, excludingmust-take (qmt) and reservoir energy resources(qrsv). The market demand is qtot, and the ob-served price is Ppx. At a price of Ppx, we seeimports of qimp (5qtot 2 qr) that shift theremaining demand to the left to a quantity qr. If

the price were instead set at the competitiveprice of Pcomp, we would see imports at somelevel less than or equal to those seen at Ppx.This would shift the residual in-state demand toa quantity q*r. Thus, in order to estimate theprice-taking outcome in the market, we need toestimate the net import or net export supplyfunction.

Estimating the Net Import/Export SupplyFunctions.—One of the primary responsibilitiesof the California ISO is to ensure the reliableusage of the system’s transmission network.This requires that the ISO operate a market forrationing transmission capacity when its use isoversubscribed. This market is implementedthrough the use of schedule “adjustment” bids,which are submitted by scheduling coordinatorsto the ISO along with their preferred day-aheadschedules.

Scheduling coordinators submit their pre-ferred import or export quantities and the ISOchecks to see whether these @ows exceed trans-mission capacity limits. If these proposed power@ows are feasible, no further adjustments arerequired. In the event that the net of proposedimport and export schedules does exceed trans-mission capacity on some intertie, the ISO ini-tiates a process of congestion relief by adjustingschedules according to their adjustment bids.Adjustment bids establish, for each schedulingcoordinator, a willingness-to-pay for transmis-sion usage. Schedules are adjusted according tothese values of transmission usage, starting atthe lowest value, until the congestion along theintertie is relieved. A uniform price for trans-mission usage, paid by all SCs using the intertie,is set at the last, or highest, value of transmis-sion usage bid by an SC whose usage wascurtailed.

Adjustment bids reveal the willingness-to-supply imported energy of out-of-state suppliers(and exported energy of in-state suppliers) ateach intertie over a wide range of quantities, notjust at the observed net import/export quantity.For the vast majority of hours the aggregate net@ow is into California for the relevant pricerange, so we refer to this as the import supplycurve, but negative import supply (net export) ispossible. Let the import supply curve of sched-uling coordinator sc at import zone z be the net

25 Capacity constraints on both the transmission intertiesinto California and the production capacity of non-Californianproducers complicate this intuition somewhat. If such acapacity constraint were binding at the observed Californiamarket-clearing price, then the marginal production cost ofimports would most likely be below this market-clearingprice and, thus, a perfectly competitive price within Cali-fornia would yield only weakly lower imports.

FIGURE 2. IMPORT ADJUSTMENTS AND EFFICIENCY LOSSES


of its preferred import quantity and all of itsincremental and decremental adjustment bidsinto California from z.

(2) qzsc~ p! 5 qz

sc ,init 1 Op̂ , p

qzsc ,inc~p̂!

2 Op̂ . p

qzsc ,dec~ p̂! .

In other words, the preferred level of importsfrom sc at z at a price of p, would be itsscheduled imports, which are independent ofprice, plus the amount of additional supply it iswilling to provide in exchange for receiving apayment less than or equal to p, minus theamount of reduction in supply that it wouldagree to in exchange for making a payment thatis greater than or equal to p. The aggregate netimport curve into the California ISO system forany hour can be calculated by summing thevalue of qz

sc( p) over all interties and SCs:

(3) qimp ~p! 5 Osc

Oz

qzsc~ p!.

This aggregation constitutes an upper boundon the responsiveness of net imports to changesin the California price. The ISO is in practiceprevented from substituting import adjustmentbids across individual scheduling coordinatorsor across transmission interties, so that the ac-tual import supply curve will be a signi9cantlysteeper function of price than the curve con-structed as described. The ISO will only act inthe event that the initial schedules indicate thatcongestion will arise, even though the adjust-ment bids may indicate a potential Pareto-improving import adjustment. Thus, while ouraggregate import supply curve assumes that allimports from all locations are perfect substi-tutes, and that these imports are priced at mar-ginal cost, reality falls short of this level ofimport ef9ciency.26

Our analysis assumes that wholesale electric-ity suppliers outside of California are price-takers,so that the import supply curve represents theaggregate marginal cost curve of suppliers out-side of California net of their native load obli-gations. Some observers have argued that thesuppliers of electricity outside of Californiamay exercise market power (i.e., offer power atabove their marginal cost) when selling into theCalifornia market. If this is the case, then animport supply curve that re@ected no marketpower from out-of-state suppliers would indi-cate a greater supply of imports at every priceand therefore a leftward shift in the residualdemand curve in Figure 6, which would lowerthe competitive benchmark price. Thus, ourtreatment of imports will tend to bias downwardour estimates of the extent of market power.

D. Hydroelectric and Geothermal Generation

Reservoir generation units (i.e., hydro andgeothermal units) present a different challengebecause the concern is not over a change inaggregate output relative to observed levels butrather a reallocation over time of the limitedenergy that is available to them. Thus, the bidsof hydro units do not re@ect a production costbut rather the opportunity cost of using thehydro energy at some later time.

In the case of a hydro 9rm that is exercisingmarket power, this opportunity cost would alsoinclude a component re@ecting that 9rm’s abil-ity to impact prices in different hours.27 It isimportant to note that even the actual observedbid prices of a small, price-taking hydroelectric9rm operating in an oligopoly market wouldprovide little information about its opportunitycost of the energy if the entire market wereperfectly competitive, because the actual oppor-tunity cost of water for these units will be in-@uenced by the expectation of future prices,

26 One consequence of this is that the import quantitiesimplied by the aggregate of the adjustment bids do notexactly equal the imports that are actually observed. Torealign the import supply curve implied by the adjustmentbids with the observed import-price pair for each hour, wecalculate the change in imports in each hour as Dqimp( p) 5

qimp( p) 2 qimp( pactual), where pactual is market priceduring the hour under consideration. This adjustment en-sures that at prices equal to the actual observed price for thathour, there would be no change from the observed level inimports when performing our counterfactual price calcula-tion. A positiveDqimp( p) implies an increase in net importsrelative to the observed price.

27 See Bushnell (2003).


which in turn will be impacted by the ability ofother 9rms to raise those prices.

For these reasons, we make the assumptionthat the actual, observed output of these re-sources is the output that would be produced bya price-taking 9rm participating in a perfectlycompetitivemarket. In practice, this assumptionmeans that in constructing our estimate of themarginal cost of meeting load in any given hourwe apply the observed production of hydro andgeothermal resources for each hour and thencalculate the marginal cost of satisfying theremaining demand with the state’s fossil-fuelresources. We give the intuition here for whythis assumption biases downward our estimatesof productive inef9ciency and market power. Amore detailed explanation is in Appendix C.

For the purpose of calculating the impact ofmarket power on total production cost, it is easyto see that this is a conservative assumption, onethat will produce downward-biased estimateson the ef9ciency effects of market power. Theoptimal hydro schedule will, by de9nition, leadto weakly lower production cost than any otherhydro schedule. To the extent that actual pro-duction differed from the optimal schedule, itcould only raise total production cost. Thus ourassumption will bias upward our estimate ofperfectly competitive production cost.

For the purpose of measuring market power,we need to consider the impact of our assump-tion on our estimates of marginal productioncost. Of concern is the possibility that the ob-served hydro schedule (which may include aresponse by hydro 9rms to the exercise of mar-ket power by others)—when combined with acounterfactual perfectly competitive productionof fossil-fuel resources—could produce a lowermarginal cost estimate on average than the op-timal hydro schedule. However, it is straightfor-ward to show that when system marginalproduction costs from nonhydro sources areconvex in quantity, any reallocation of hydroenergy away from the least-cost allocation willraise marginal costs more in the hours from whichenergy is removed than it will reduce marginalcost in the hours to which energy is added.28 Thus

our assumption of optimal hydro production canonly bias our time-weighted estimates of mar-ginal cost upwards, and therefore our estimatesof price-cost margins downward.

We present results in which price-cost mar-gins are weighted by the market volumes ineach hour. To consider the effect of our hydroassumptions on these results, we need to ad-dress the possibility of a reallocation of hydroenergy between off-peak and peak hours rela-tive to the optimal schedule.A hydro 9rm that isattempting to exercise market power wouldlikely allocate less hydro energy during peakhours than would be the case for a price-taking9rm (see Bushnell, 2003). This strategic hydroallocation, when combined with competitivefossil-fuel production, would produce a higherweighted average of marginal cost than wouldthe optimal schedule. To the extent the 9rmscontrolling hydro resources attempted to exer-cise market power with those resources, ourresults will therefore understate the overall levelof market power.

However, the vast majority of reservoir re-sources were controlled by the PG&E and SCE,each of which had a fairly strong incentive tolower wholesale power costs. Therefore it ispossible that these 9rms responded to an in-crease in market power with an overconcentra-tion, relative to perfect competition, of energyduring high-demand periods. As argued above,this reallocation (if allowed by the @ow con-straints) would raise off-peak marginal costsmore than it would lower on-peak marginalcosts. However, since (non-must-take) marketvolumes are likely to be higher on-peak, theimpact on the quantity-weighted average ofmarginal cost is uncertain.

We examined this issue empirically by askingwhether our estimates of marginal costs produceopportunities for a reallocation of hydro energythat would result in a lower weighted-averagemarginal cost. Such an opportunity would existif fossil-fuel marginal costs during some high-demand period were lower (due to “overproduc-tion” from hydro sources) than the marginal costin some lower-demand period. If, by contrast,

28 This is because at the least-cost allocation of hydroenergy, marginal fossil-fuel costs will be equalized over all

hours for which hydro @ow constraints allow a discretionaryuse of hydro energy.


our marginal cost estimates (over a period withstable input prices) were monotonic in marketdemand, then no systematic opportunity for low-ering the weighted average of marginal cost existsand any bias from our hydro assumption is in thedirection of raising costs and lowering marketpower.

To examine this possibility, we estimated akernel regression of our estimated marginal cost(i.e., competitive price) on system demand (pre-sented in Appendix C) in order to detectwhether in aggregate there are systematic devi-ations from a monotonically increasing relation-ship between demand and our estimate ofsystem marginal cost. Such regressions for eachof the three summers in our sample period showthat our system marginal cost estimates weremonotonically increasing in demand for each ofthese time periods. This leads us to concludethat it is highly unlikely that our assumption thatthe actual schedule of production reservoir re-sources was the cost-minimizing schedule cre-ates a signi9cant negative bias on the weighted-average estimates of system marginal costs.

E. Calculating Cost Increase Relative toCompetitive Outcome

Utilizing the assumptions outlined in the pre-vious sections, we estimated the perfectly com-petitive market price in the California energymarkets for each hour of market operation fromJune 1998 through October 2000. The residualmarket demand to be met by in-state fossil-fuelunits within the ISO system in hour t, qff

t , isestimated to be

(4) qfft ~ p! 5 qtot

t 1 qregt 2 qmt

t 2 qrsvt

2 qimpactt 2 Dqimp

t ~ p!.

Here qtott is the actual ISO metered generation

(including net imports), including generationscheduled through all energy markets associ-ated with the ISO control area, including thePX, ISO imbalance energy market, and otherSCs. qreg

t represents the addition to demand dueto the need for capacity dedicated to regulationreserve. The quantities qmt

t and qrsvt represent

the amount of energy produced by must-take

generation and by reservoir generation, respec-tively. These quantities are all treated as priceinelastic. The term qimpact

t 2 Dqimpt (p) is the

level of imported energy adjusted by the re-sponse to changes in the market-clearing price,as described above.

For each hour, we make 100 fossil-fuel gen-eration marginal cost curve estimates, each re-@ecting a combination of independent MonteCarlo draws for the outage of each generationunit. For each of these draws from the system-wide fossil-fuel marginal cost curves we com-pute the intersection of this marginal cost curvewith the residual market demand curve qff

t ( p).This yields an estimated marginal cost and anin-state market-clearing quantity qrj

t for MonteCarlo draw j. We denote the marginal costassociated with this quantity as Cj

t. We can thencompute an estimate of the expected value ofthe marginal cost of meeting the in-state de-mand that results from price-taking behavior byin-state generators as:

(5) P# compt 5

¥j 5 1

100

~Cjt!

100.

Note that there are cases in which Ppxt 2 P# comp

t

is negative in our simulations. Absent an oper-ational error or an attempt at predatory pricing,9rms will not actually be willing to sell power atprices below their true economic short-run mar-ginal costs. In other words, prices will not bebelow the perfectly competitive price. Nonethe-less, during some hours, particularly June 1998and during the winter and spring of 1999, PXprices were below our estimates of the perfectlycompetitive market price. At least three factorscontribute to these outcomes.

First, our cost estimates can exceed the actualmarginal cost because we do not consider thedynamic effects of unit commitment con-straints, such as start-up costs, ramping rates,and minimum down times. These constraintscan create opportunity costs of shutting downunits that, in essence, lower the true marginalcost of operating that plant. Of course thesesame constraints also can create opportunitycosts that, at other times, raise the true marginal


cost. This is one reason why we include thenegative markups in our results; we did not wantto exclude the off-peak impact of these constraintson our cost estimates, since there is an oppositeeffect on our estimates during peak hours.

Second, cost information for generating unitsare not exact data on which all parties agree. Forthe most part, we used values submitted to stateand federal regulatory agencies under theformer regulated regime. For this reason, ourestimate of a unit’s marginal cost may beslightly higher than the cost level at which it iscapable of operating in a market environment.Therefore we include negative price-cost differ-ences in order to prevent truncating the effect ofdata uncertainty on our cost estimates.

Third, and probably most important, our cal-culations do not control separately for theoutput levels of reliability must-run (RMR)generation. Some fossil-fuel generation unitshave been declared must-run for local grid re-liability under certain system conditions. Thesegenerators get separate nonmarket paymentswhen they are called under the RMR contractsthey have signed with the ISO. RMR units arenot dispatched through the price-setting pro-cess. Because they are held out and paid adifferent price, the resulting price in the PX canbe below the marginal cost to the system if thepower provided by RMR units were insteadprovided as part of the full dispatch of thesystem. In fact, due to the level of RMR calls bythe ISO during some periods during our sample,particularly the spring of each year, it is possi-ble that no other fossil-fuel generation was eco-nomic during these time periods. Under thesecircumstances, the highest (opportunity) costunits selling in the PX could be hydro or out-of-state coal plants, either of which have lowermarginal cost than any of the fossil-fuel plantswe examine. However, these periods are likelyto occur when the PX price is extremely low,not extremely high. In such cases, import en-ergy with costs below those of in-state fossil-fuel generation could be the marginalgeneration, and the actual PX price could belower than the marginal costs of any of thefossil-fuel units we have examined. Because wedon’t account for the RMR units, our estimatescould still indicate that a fossil-fuel unit is mar-ginal and its cost is the system marginal cost, so

our estimated system marginal cost would beabove the actual PX price due to unaccountedfor RMR calls.29

If the estimated MC is above the PX price foreither the 9rst or second reason, then it seemsthat the most accurate estimate of market powerwould come from including the “negative mar-ket power” outcomes in our calculations. How-ever, total start-up costs for the fossil-fuel unitsin California are about $39 million during oursample period, less than 1 percent of total fossil-fuel generation production costs during the pe-riod and less than 1 percent of the market powerrents we 9nd.30 In addition, there are otherreasons to think that start-up costs explain onlya minor part of the deviations from marginalcost pricing. First, the units that turn on to meetpeak demand during the summer have littlestart-up costs (fuel oil or jet fuel units) or noneat all (hydroelectric units), so the impact at thetimes we 9nd the greatest market power is likelyto be low.31 Second, our estimates of marketpower are substantially greater in summer 2000than in summer 1999, but the amount of elec-tricity produced per start-up is 5.7 percent lowerin summer 1999, implying that start-up costswould likely be a greater factor in 1999 than in2000. Similarly, the ratio of start-up costs to ourestimated fossil-fuel production costs was

29 This implies that neglecting RMR calls could under-estimate market power. In addition, it appears that the initialRMR agreements exacerbated some of the local marketpower problems that they were designed to mitigate. SeeBushnell and Wolak (1999).

30 For 65 of the 92 units in our fossil-fuel cost curve, theunit-speci9c formulae for determining the cost per start-up(as a function of both input fuel costs and the price ofelectricity) were submitted to the ISO as part of the RMRcontract renegotiation process. For the remaining 27 units,we estimated the unit-speci9c start-up formula by usingparameters from a similar unit that did have an RMRcontract. We used the daily price for the input fuel used bythat unit and daily average annual retail price to industrialcustomers for the electricity price in the start-up costformula.

31 Additionally, if marginal cost functions turn upwardsmoothly around the rated capacity, rather than having astrict L-shape, the typical argument that a competitive plantwould bid its start-up costs for the “single hour” it wouldrun are incorrect. In that case, even the last plant turned onwould run for many hours because it would be replacinghigher-cost output from other plants that would otherwisebe producing along the steepest parts of their MC curves.


higher in summer 1999 than in summer 2000,0.91 percent versus 0.37 percent.

Likewise, it is unlikely that much of the neg-ative market power outcomes could be the resultof cost-data errors. Many PX prices in June1998, for instance, were well below the coststhat anyone has claimed for operation of fossil-fuel generating units.32 Thus, it is most likelythat the cost estimates that exceed the PX priceoccur because there were no fossil-fuel gener-ating units that were economic to run at thetime. Only fossil-fuel units running under RMRcontracts were active. In that case, the marginalcost of the system, and thus the market price, isbeing set by much cheaper out-of-state coalplants, by nuclear plants, or by hydro or geo-thermal plants. If this is the case, then the propertreatment would be to truncate the results, re-setting any 9nding of “negative market power”to set marginal cost equal to price. Still, in orderto avoid biasing the results in favor of 9ndingmarket power, we do not truncate the negativeoutcomes in the primary results we report.

IV. Results

We computed the expected perfectly compet-itive price each hour for the months of June1998 through October 2000 using the algorithmdescribed above. From the import adjustmentbids, the median hourly reduction in importsfrom the observed level at the PX price versusthe level at our estimated competitive price was2.4 percent. For each hour, we can calculate anarc elasticity implied by the adjustment bids forthe import response from the change betweenthe competitive and actual price and the result-ing change in imports. The median arc elasticityof import supply for these hours is 0.63.33

The added wholesale cost of energy due todepartures from a competitive market, DTC, iscalculated by taking the difference between thePX price and our estimate of competitive bench-mark price and multiplying it by the total ISO

metered generation less the must-take energyfor that hour.34 That is, for hour t,

(6) DTC t 5 @Ppxt 2 P# comp

t # z @qtott 2 qmt

t #,

where P# compt is the expected competitive price

in period t. This expectation is taken with re-spect to the distribution of generating unit out-ages, as shown in (5).

For any set of hours , our measure of mar-ket performance is

(7) MP~ ! 5

¥t [

DTCt

¥t [

TCt .

MP( ) is the proportional increased wholesalecost of electricity during all hours in . De9n-ingMP( ) in this manner is consistent with theview, re@ected in our competitive benchmarkMonte Carlo simulation, that the observed mar-ket price is conditional on a realization fromthe joint distribution of generating unit outages.To re@ect this fact, let P̂px

t denote the ob-served PX price for hour t and E(P̂px

t ) the ex-pectation of this magnitude with respect to thejoint distribution of generating unit outages.Unlike the counterfactual case of price-takingbehavior, we cannot draw from the distributionof generating unit outages and compute a dis-tribution of market prices that re@ect the currentlevel of market power. This would require amodel for the strategic interaction among play-ers in the California market. However, by de-9ningMP( ) as shown in equation (7), we cantake advantage of the law of large numbers toprove that our measure is a consistent estimateof the proportional cost increase. To show this,rewrite the index as:

(79) MP~ !

5

1/Card~ ! ¥t [

@P̂pxt 2 P# comp

t # z @qtott 2 qmt

t #

1/Card~ ! ¥t [

P̂pxt z @qtot

t 2 qmtt #

,32 If we were to ignore any “negative market power”

outcomes for prices below, say, $18/MWh, virtually all ofthe “negative market power” effects would be eliminated.

33 We calculate the arc elasticity as~P1 1 P2 !/ 2

~Q1 1 Q2 !/ 2

~Q2 2 Q1 !

~P2 2 P1 !.

34 By taking the observed quantity as the market de-mand, we are, for the reasons discussed earlier, implicitlyassuming that demand is price inelastic.


where Card( ) is the cardinality or number ofelements (hours) in the set . For sets with alarge number of elements, the index is approx-imately equal to

(70) MP~ !

5

¥t [

@E~P̂pxt ! 2 P# comp

t # z @qtott 2 qmt

t #

¥t [

E~P̂pxt ! z @qtot

t 2 qmtt #

which is equal to the ratio of the expected costincrease relative to the perfectly competitivebenchmark, due to the current level of marketpower and market imperfections, divided by theexpected cost of purchasing electricity undercurrent market conditions.

Table 2 reports the PX price, estimated mar-ginal cost, and the added cost of power due toprices that exceeded marginal cost for eachmonth in the sample period. As is evident fromTable 2, June 1998 produced very idiosyncraticresults, with an average PX price considerablybelow our estimate of marginal cost. The mar-ket was only in its third full month of operationat this time and a number of fossil-fuel gener-ation units were going through ownership trans-fer and regulatory approval of these transfers.The CTC mechanism provided the threeinvestor-owned utilities with an incentive toinduce low energy prices and the utilities werestill operating and bidding many of these unitsthrough June 1998. As described above, the useof reliability must-run contracts, which paidsome fossil-fuel units to run in exchange for

TABLE 2—ACTUAL PRICE AND ESTIMATED MARGINAL COST

Month Year

Mean of ActualProductionPer Hour(MWh)

Mean ofPX Price($/MWh)

Mean ofMarginalCost

($/MWh)

Sum ofDTC

($ million)

AggregateDTC/TC(percent)

June 1998 24,134 12.09 22.55 244 251July 1998 28,503 32.41 27.33 103 28August 1998 31,256 39.53 27.71 220 39September 1998 28,209 34.01 26.28 134 33October 1998 25,043 26.65 26.21 13 5November 1998 24,107 25.74 27.53 24 22December 1998 24,953 29.13 25.40 45 17

January 1999 24,480 20.96 22.41 25 22February 1999 24,079 19.03 21.20 212 27March 1999 24,734 18.83 20.80 212 27April 1999 24,763 24.05 24.50 4 2May 1999 24,625 23.61 25.34 0 0June 1999 27,081 23.52 25.89 13 5July 1999 29,524 28.92 27.12 63 17August 1999 29,813 32.31 30.64 56 14September 1999 28,573 33.91 30.25 63 16October 1999 27,558 47.63 34.38 186 31November 1999 26,046 36.91 28.87 105 26December 1999 26,647 29.66 27.73 30 9

January 2000 26,377 31.18 27.66 48 13February 2000 25,961 30.04 29.52 10 3March 2000 25,618 28.80 31.38 217 26April 2000 25,728 26.60 32.43 243 216May 2000 27,038 47.22 40.43 150 25June 2000 30,644 120.20 53.59 1,152 63July 2000 30,343 105.72 59.37 801 50August 2000 32,310 166.24 76.19 1,475 56September 2000 29,981 114.87 76.86 577 36October 2000 27,422 101.51 68.06 443 34


payments that were above the market price, wasalso widespread during June. For these reasons,we believe the market results from June 1998 donot provide much meaningful information onthe state of competition in the California mar-ket. Nevertheless, for completeness, we haveincluded these results. For the set of all hoursover the entire 29-month period from June 1998to October 2000, the MP( ) is equal to 33percent, amounting to total payments in excessof competitive levels equal to $5.55 billion witha standard error of $1.19 billion.35

Having generated estimates of price-costmargins for each hour of the 29-month sampleperiod, we can examine subsets of the data togain insight into the underlying dynamics of themarket. One test of the credibility of our resultsis whether our estimates of market power varyin the way that economics would predict. Wewould expect market power to be quite lowduring the off-peak months, December throughApril. Electricity demand is low in these monthsand supply is relatively large due to the resur-gence in hydro production from winter rains. InDecember 1998–April 1999, we 9nd an averageMP( ) of 1.9 percent, and in December 1999–April 2000, we 9nd an average MP( ) of 1.8percent, neither of which is signi9cantly differ-ent from zero. Thus, we 9nd that there wasessentially no margin between prices and mar-ginal cost during the period in which supply wasmost abundant compared to demand and sellershad the least ability to exercise market power.In addition, these results provide evidence thatsigni9cant short-run operating costs are notmissing from our cost estimates, because nega-tive or zero margins would not be observed oversuch an extended period of time.

The series of events that led to the Californiaelectricity crisis in 2000–2001 began with dra-matic price increases during the summer of 2000.Many policy makers and regulators have arguedthat the competitive performance of the marketfundamentally changed during summer 2000,thereby initiating the crises. In order to make suchcomparisons, however, one must account for dif-

ferences in the relative levels of demand duringthese periods. As described above, we would ex-pect the estimated market power to increase as thedemand faced by in-state nonutility sellers risesrelative to the capacity of these players. Figure3 shows a kernel regression of this relationship forlate summer of 1998, 1999, and 2000.36 The hor-izontal axis of Figure 3 is the demand faced bythese 9rms after accounting for actual imports,must-take, and reservoir production. The verticalaxis is the ratio DTCt/TCt, which is equal to theLerner index for that hour.37

The results summarized by this 9gure showthat market power steadily increased with thedemand faced by the nonutility in-state suppli-ers consistent with the earlier discussion of thenature of competition in the electricity industry.During lower demand hours and months, aswell as springtime months when signi9cant hy-dro energy is available, no single 9rm can affectprices signi9cantly. During higher demandhours, however, competitive sources of energybegin to reach their capacity limits and the pool

35 Appendix B outlines our procedure for computing thisstandard error, which accounts for the error associated withthe randomness of forced plant outages.

36 To be precise, the data are for August 7 through Sep-tember 30 of each year. We focus on this period becausethe ISO energy price cap varied during our sample period, butit was set at the same level, $250/MWh, for August 7 throughSeptember 30 of all three years. August and September arehistorically two of the highest demand months of the year inCalifornia, and they can exhibit the lowest supply availabilitydue to declining hydro resources late in the summer.

37 Because the Lerner index is not symmetric aroundzero, negative values of the ratio are set to zero in estimatingthe kernel regressions.

FIGURE 3. KERNEL REGRESSIONS OF LERNER INDEX,AUGUST 7–SEPTEMBER 30


of potential competitors for additional supplydwindles. Because of the lack of signi9cantstorage capacity and the inelasticity of demand,9rms can take advantage of the capacity limitsof their competitors during these high-demandhours. This is consistent with the effects de-tected from the oligopoly equilibrium simula-tions in Borenstein and Bushnell (1999). Thissequence of events does not imply a shortage ofgenerating capacity to serve the energy or an-cillary services needs of the California ISO con-trol area. However, the combination of theconcentration of ownership of generating assetsand the level of demand did combine to createcircumstances where one or more market par-ticipants recognized that their capacity wasneeded to meet the ISO’s energy and ancillaryservices needs regardless of the actions of othermarket participants. Under these circumstances,9rms 9nd it in their unilateral interest to bid toraise prices even though there is suf9cient ca-pacity available to meet the California ISO’stotal energy and ancillary services requirements.

Our results also indicate that, given the sup-ply and demand conditions during that period,the performance of the market was not dramat-ically different in 2000 from that in 1998 and1999. Figure 4 shows the cumulative distribu-tion functions for the demand met by in-statefossil-fuel generation for the late-summer pe-riod during 1998–2000. Although total marketdemand was only 6 percent higher during latesummer 2000 than in late summer of 1999 and

5 percent higher during that period in 1998, thedemand met by in-state fossil-fuel plants, nearlyall of which were unregulated by 1999, in-creased from an average of 6,639 MWh during1998 and 5,690 MWh during 1999 to 10,007MWh during 2000. This is largely due to asubstantial decline in imports from an averageof 5,069 MWh in 1998 and 6,764 MWh in 1999to 3,627 MWh in 2000. Thus, although theperformance of the market controlling for thedemand faced by in-state fossil-fuel generationdid not change signi9cantly during 2000, thedistribution of this demand did change. Farmore hours spent at higher residual demandlevels created larger average margins during2000. This combined with the fact that margi-nal costs also nearly tripled between 1999and 2000, which meant that similar Lerner in-dices re@ected much larger absolute dollarmargins, producing extremely large wealthtransfers.

V. Deadweight Loss and Rent Division

Even without a market power analysis, it isclear that the extraordinary prices that began inthe summer of 2000 created large transfers ofwealth. The analysis we have carried out, how-ever, allows us to parse the changes in whole-sale payments for electricity between threemutually exclusive and exhaustive categories:changes in the competitive cost of generatingelectricity, changes in the level of competitiveinframarginal rents (which would have occurredwithout any market power), and changes inseller rents due to the exercise of market power.Some of the rents due to market power becamepro9ts of electricity producers or marketers, butsome were dissipated in production ef9ciencylosses: ef9ciency losses resulting from the op-eration of higher-cost production units when a9rm with lower-cost production exercises mar-ket power and restricts output.

A. Deadweight Loss

We begin by estimating the loss in economicef9ciency due to the imperfections in the mar-ket. Because the demand for electricity in theCalifornia market was effectively perfectly in-elastic with respect to the wholesale market

FIGURE 4. CUMULATIVE DISTRIBUTION FUNCTIONS (CDFS)OF DEMAND MET BY IN-STATE FOSSIL-FUEL PLANTS,

AUGUST 7–SEPTEMBER 30


price, ef9ciency losses would stem primarilyfrom the inef9cient allocation of production.38

With asymmetric producers, there will be ef9-ciency losses from the substitution of higher-cost production from price-taking 9rms, or evensmaller strategic 9rms, for the lower-cost pro-duction of the larger 9rms that are exercisingmarket power.

This would be a fairly straightforward calcu-lation if there were no imports into the state.With no imports and perfectly inelastic demand,we could simply compare the ef9cient produc-tion costs of a given quantity of power, usingthe approach described in the previous section,with the actual production cost of that quantityof power. Due to imports that vary in quantitywith the exercise of market power in state,however, we also need to account for the sub-stitution of higher-cost imports for lower-costin-state generation.

Thus, we divide the ef9ciency loss into thesetwo components: the loss due to misallocationof a given production quantity of output amongthe fossil-fuel plants inside the ISO system, andthe loss due to misallocation of production be-tween fossil-fuel plants within the ISO andplants outside the ISO (imports). The vast ma-jority of power imported into California origi-nates from regulated or publicly owned 9rms,and most of these 9rms have substantial nativedemand obligations. We have therefore as-sumed that the adjustment bids from these 9rmsre@ect the actual opportunity cost of their pro-duction (i.e., that they are price-takers). Whencalculating wealth transfers, this is a conserva-tive assumption. However, when calculating theimpact of market power on ef9ciency losses, itis not. By assuming that import bids re@ect themarginal cost of the supplier, we assume thatincreased production from these imports due tomarket power exercised by 9rms within Cali-fornia creates an increase in total production

cost. If the adjustment bids from 9rms outsideof California contain margins over their ownmarginal cost, this margin will be counted as anef9ciency loss, when it is in fact a transfer fromconsumers to those producers.

The two components of deadweight loss areillustrated in Figure 2. The inef9ciency from thereallocation of the actual production quantityamong fossil-fuel resources inside the ISO isillustrated by the solid gray area between thecompetitive marginal cost curve and the “ac-tual” marginal cost curve just above it.39 Ourestimates of the total in-state fossil-fuel produc-tion inef9ciency for June through October of1998, 1999, and 2000 are shown in Table 3. Theexpected cost from additional imports is illus-trated in the striped area of Figure 2 and re@ectsthe difference between producing the quantityDqimp

t (pcomp) from imported production andproducing that same quantity from in-state pro-duction along the marginal cost curve MCcomp.Again, we have assumed that the adjustmentbids of importing 9rms re@ect their actual mar-ginal production costs. Our estimates of thetotal production inef9ciency due to higher-than-optimal imports for the summer months of oursample are shown in Table 3.

In Figure 5, we illustrate the relationship be-tween our estimated in-state productive inef9-ciency and aggregate demand faced by Californiafossil-fuel plants using a kernel density regres-sion. Given our 9ndings in the previous section,it is not surprising that we observe low levels of

38 This is not true if the exercise of market power causedsome “interruptible customers”—customers that haveagreed to curtail consumption upon request from the utilityin return for lower electricity rates overall—to signi9cantlyreduce their demand. This happened on 26 occasions duringour sample (5 in 1998, 1 in 1999 and 20 in January–October2000, but we have no way of estimating the deadweight lossfrom these demand reductions.

39 This is a rough representation since the cost differenceneed not rise with the quantity of in-state fossil-fuel pro-duction, as we discuss in what follows.

TABLE 3—PRODUCTION COSTS AND RENT DISTRIBUTION

($ MILLION) JUNE–OCTOBER

1998 1999 2000

Total actual payments 1,672 2,041 8,977Total competitive payments 1,247 1,659 4,529Production costs—actual 759 1,006 2,774Production costs—competitive 715 950 2,428Competitive rents 532 708 2,101Oligopoly rents 425 382 4,448Oligopoly inef9ciency—in state 31 31 126Oligopoly inef9ciency—imports 13 24 221


production inef9ciency at low levels of systemdemand, when there are low levels of marketpower. What may be surprising at 9rst is thatproductive inef9ciency declines as demandnears system capacity even though our esti-mates of market power continue to increase.When the system is near capacity, however,very small output reductions can yield enor-mous price increases. Thus, while exercise ofmarket power at those times may cause largewealth transfers, the resulting productive inef-9ciency is small because nearly all resources arerunning in any case.

B. Rent Division

With the calculation of deadweight loss dueto productive inef9ciency, we are now in aposition to parse the total wholesale marketpayments into costs, competitive rents, andrents due to the exercise of market power. InFigure 6, the quantity qtot 2 qmt represents theamount of power traded in the wholesale marketin a given hour, qmt being must-take power thatis not compensated at the market price.

Total wholesale market payments are the sumof all the shaded areas. When the areas labeledMkt. Power Rents and Import Loss (together thearea above Pcomp) are removed from the total,the result is the total wholesale payments thatwould have resulted if the market were perfectlycompetitive. The quantity qrsv represents hydroand geothermal production during the hour. Forpurpose of calculating the change in rents dur-ing summer 2000 and how those rents were

divided, we assume that hydro and geothermalpower have zero marginal cost, though this as-sumption has no effect on the calculation of thechange in rents.

Under competition, the quantity q*r is pro-duced by in-state generation units and the quan-tity qtot 2 q*r is imported. The area labeledComp. Total Cost is the variable productioncosts of in-state units (other than must-take pro-duction) and of imported power for their respec-tive shares of production. Competition generatesinframarginal rents equal to the sum of the areaslabeled Comp. Rents 1 and Comp. Rents 2 forin-state fossil-fuel and reservoir generators, andthe area Comp. Rents 3 for imports.40 Together,these areas—Comp. Rents and Comp. TotalCost—account for all wholesale market pay-ments under perfect competition.

With market power, the quantity qr is pro-duced by in-state generation units and the quan-tity qtot 2 qr is imported. The areas labeledComp. Rents 2 and Import Loss are the addi-tional variable production costs of the importedpower under the assumption that imports are bidcompetitively. In addition to the inframarginal

40 We have not calculated the import elasticity for pricesbelow Pcomp so we assume that marginal costs decrease ina linear fashion and that the marginal costs of imports arezero when the import quantity is zero. The qualitative re-sults do not change if we use the actual import bids down tozero and assume that all imports that made no (or negativeprice) adjustment bids had marginal costs equal to zero.

FIGURE 5. EFFICIENCY LOSSES

FIGURE 6. CALCULATION OF DIVISION OF RENTS


rents represented by the area Comp. Rents 1,in-state producers receive the area Mkt. PowerRents 1, but some of these rents are dissipatedthrough inef9cient production as described pre-viously. Imports receive inframarginal rentsComp. Rents 3 as well as the area labeled Mkt.Power Rents 2. Together, these areas accountfor all wholesale market payments with marketpower.

Between the summers of 1998 and 2000, thewholesale market cost of power rose from $1.67billion to $8.98 billion. Ef9cient productioncosts more than tripled between these periodsand with the marginal unit having higher costs,competitive rents for lower cost units quadru-pled. Oligopoly rents, however, increased by anorder of magnitude, from about $425 million to$4.44 billion between these summers. Thus,while a substantial portion of the increased mar-ket cost of power was due to rising input costsand reduced imports, these factors also in-creased the dollar magnitude of the marketpower that was exercised by suppliers. As theresults in the previous section indicate, the un-derlying competitive structure of the marketdoes not appear to have changed substantiallybetween 1998 and 2000. Rather the higher de-mand and lower import levels in 2000 createdmore frequent opportunities for in-state fossil-fuel producers to collect large margins on in-creased costs, leading to the tenfold increase inoligopoly rents to suppliers.

The inef9ciencies that resulted from the real-location of production within California weremuch more modest, remaining at about 3–5percent of total production costs through allthree summers. The inef9ciencies due to in-creased imports in power did grow substantiallyduring our study, rising from 2 percent to 8percent of total production costs by the summerof 2000. To the extent that prices from import-ing 9rms did not re@ect their actual productioncosts, but their own market power, this 9gurewill include oligopoly rents earned by produc-ers outside of California as well as actual pro-ductive inef9ciencies.

VI. Conclusions

Restructuring of electricity industries hasbeen predicated on the belief that workably

competitive wholesale electricity markets canbe attained. The debate over whether that as-sumption is correct and what must be done toensure competition in electricity generation isongoing. We have attempted here to reliablyestimate the degree to which California’swholesale electricity market has deviated fromthe competitive ideal.

Though a great deal of cost data are availablefor electricity generation units, we still had tomake a number of assumptions in order to reachan estimate of the extent of market power inCalifornia. In most, though not all, cases, wehave made assumptions that, if anything, arelikely to produce results indicating less marketpower than actually exists.

The results indicate that market power inCalifornia’s wholesale market was a signi9cantfactor during the summers of 1998, 1999, and2000, though somewhat less so in 1999. Theseestimates should serve as a reminder that theproblem of producer market power that wasaddressed in a purely regulatory framework formost of the twentieth century has not com-pletely disappeared with the recent restructur-ing. Our results demonstrate that market poweris most commonly exercised during peak de-mand periods, which is not at all surprisinggiven the current inability of wholesale demandto respond to high hourly spot prices. This un-derscores the importance of designing whole-sale electricity markets that maximize thelikelihood that wholesale price signals will bere@ected in retail electricity rates.

These estimates demonstrate the degree towhich prices exceed system marginal costs, theprice level that would occur if all 9rms behavedas competitive price-takers. We have not at-tempted to assess the pro9tability of any gener-ation 9rms selling in California, because suchpro9ts are not necessarily an indication of mar-ket power, just as the absence of pro9ts is not anindicator of competitive behavior. In all marketswith durable assets, such as is the case in thisindustry, there are likely to be periods of highand low (or negative) pro9ts regardless of thecompetitiveness of the market.41

41 It is also worth noting that we have analyzed only theenergy markets in California. Most generation units were


Finally, we want to emphasize again that thisstudy is intended to develop an index of theextent and severity of market power. This isseparable from the important debate over whatindex levels indicate a need for some form ofmarket intervention. Years of electricity regula-tion con9rmed the belief that government inter-vention can be costly and can result in veryinef9cient production and pricing. The balanc-ing of the costs and bene9ts of such interventionwill require a great deal more study in thisindustry as restructuring proceeds.

APPENDIX A: DATA SOURCES

Fossil-Fuel Generation Data

Heat rates for fossil-fuel generation units thatare not must-take and are located within the ISOcontrol area are primarily taken from the Cali-fornia Energy Commission’s data set on gener-ation within the Western States CoordinatingCouncil for use with General Electric’s multi-area production cost model. This is the data setused in Borenstein and Bushnell (1999). Someunit heat rates were taken from the data set usedby Southern California Gas Company in its1995 performance-based rate-making simula-tion studies (Luis Pando, 1995). This data setwas also used by Kahn et al. (1997) in theirsimulation analysis of the WSCC. Capacitiesof all units are the California ISO’s “avail-able capacity” 9gures. For NOx emission creditcosts in the SCAQMD air basin, we used thequantity-weighted monthly average price paidfor emissions credit trades registered withSCAQMD. Emissions rates of generation unitsare taken from the Environmental ProtectionAgency’s ContinuousEmissionsMonitoring data.

An overwhelming share of California fossil-fuel generation uses natural gas as the energysource. For the time period studied, we useddaily average natural gas spot prices reported by

Natural Gas Intelligence at PG&E citygate andthe California–Arizona border. The former wereused for generation units north of path 15 whilethe latter were used for generation units in thesouth. Both sets of prices were adjusted by thedistribution rates of the gas utility serving eachgenerator.

A small number of California generators useeither fuel oil or jet fuel as their primary fuel.We use the Energy Information Administra-tion’s reported monthly average Los Angelesspot price for jet fuel and no. 2 fuel oil.

Unit forced outage factors are taken fromthe National Electricity Reliability Council’s(NERC) 1993–1997 Generating Unit StatisticalBrochure, which reports aggregate generationunit performance data by fuel type and name-plate capacity. The forced outage factor that weused for our Monte Carlo simulations werederived from the NERC-reported unit Equiva-lent Availability Factors (EAF) and unit Sched-uled Outage Factors (SOF). The former givesthe fraction of total hours in which a generationunit was available, including an adjustmentfor partial outages, while the latter gives thefraction of hours in which each unit was un-available due to scheduled maintenance proce-dures. Our derived forced outage factor (FOF),which re@ects the fraction of time a unit was notavailable for production for unplanned reasons,was

(A1) FOF 5 1 2EAF

1 2 SOF.

Demand and Generation Output Data

Total ISO quantity for every hour is basedupon the ISO’s real-time metered generationand is taken from ISO settlement data. Theoutput of must-take, hydro, and geothermalgeneration for each hour is also taken fromthese data. Imports are calculated from the set-tlement data as metered imports minus exportsaggregated over all transmission interties con-necting the ISO’s control area with neighboringcontrol areas. The Mohave generation plant,although located outside of California, appearsin metered data as a must-take generating facil-ity and not as an import. Production from all

eligible to earn additional revenues under reliability must-run contracts and from the sale of ancillary services. Totalancillary services plus RMR revenues were approximately$977 million in 1998, $879 million in 1999, and $1.859billion in 2000.


other generation units owned by SCE, but lo-cated outside of California, appear as imports inthe settlement data.

APPENDIX B: CALCULATION OF STANDARD ERROR

FOR DTC( )

The observed PX price in hour i and day d,P̂pxid , depends on a single realization of the joint

distribution of generation unit-level outagesduring that hour. A different realization of unit-level outages for that hour could result in a verydifferent observed PX price for that hour. Be-cause we do not know the precise nature of thecompetitive interaction among market partici-pants, speci9cally the bidding strategies of bothgenerators and loads that gave rise to the ob-served market-clearing PX price, we are unableto replicate the actual price-setting process for alarge number of draws from the joint distribu-tion of unit-level outages in order to computethe expected value of the PX price, E(P̂px

t ). Incontrast, we can compute the perfectly compet-itive market counterfactual equilibrium becauseit entails the assumption of marginal cost bid-ding by all in-state fossil generation unit own-ers. This allows us to compute the realizedvalue of the marginal cost of the highest costunit operating in that hour for a large number ofdraws from the joint distribution of unit-leveloutages which we can use to compute an esti-mate of the expected value of this marginalcost to an arbitrary degree of precision for eachhour. We found that the 100 realizations fromthe joint distribution of unit-level outages led toa very precise estimates of this expected mar-ginal cost for each hour. Consequently, the re-maining source of uncertainty in DTC( ), thetotal cost difference due to deviations fromcompetitive prices over time period , that ourstandard error estimate accounts for is the un-certainty in PX prices caused by actual forcedoutages.

In constructing the standard error estimate forDTC, we account for arbitrary correlation in[P̂px

id 2 P# compid ] across the 24 hours of the day

and general forms of autocorrelation in these 24prices across days. We can write the total costdifference due to deviations from competitiveprices over time period as

(B1)

DTC~ ! 5 Od [

Oi 5 1

24

@P̂pxid 2 P# comp

id #@qtotid 2 qmt

id #

5 Od [

Oi 5 1

24

mkupidqnid,

where qnid 5 [qtotid 2 qmt

id] and mkupid 5 [P̂pxid 2

P# compid ]. This expression can be rewritten as:

(B2) DTC~ ! 5 Od [

Oi 5 1

24

E~mkupid!qnid

1 Od [

Oi 5 1

24

« idqnid,

where «id 5 mkupid 2 E(mkupid) and E(z)denotes the expectation taken with respect to thejoint distribution of unit-level forced outages.Therefore, the variance of DTC( ) can be writ-ten as:

(B3) Var~DTC~ !! 5 VarX Od [

Oi 5 1

24

« idqnidD5 VarX O

d [

ZdD ,where Zd 5 ¥i51

24 «idqnid. Under suitable reg-ularity conditionson the sequenceZd, for examplethose assumed in Whitney K. Newey and Ken-neth D. West (1987) or Donald W. K. Andrews(1991),we can show that (DAY( ))21/2(¥d[ Zd)converges in distribution to a (0, V) randomvariable, as DAY( ) tends to in9nity, whereDAY( ) equals the number of days in timeperiod . A consistent estimate of V can beconstructed as follows:

(B4) V̂ 5 gZ ~0! 1 2 Ot 5 1

q

k~t/~q 1 1!!gZ ~t! ,


where gZ(0) 5 1/DAY( ) ¥d51DAY( ) (Zd)

2, gZ(t) 51/DAY( ) ¥d5t11

DAY( ) (ZdZd2t), and k(t) is aweight function satisfying restrictions given inAndrews (1991). Using this asymptotic distri-bution result, an estimate of the variance ofDTC( ) is (DAY( )V̂), which implies a stan-dard error of (DAY( )V̂)1/2.

To operationalize this procedure we need toconstruct an estimate of the unobservable, «id,the difference between mkupid, and its expec-tation. As discussed above, in order to computethe exact expectation of mkupid we would needto know how all market participants bid into thePX (and other markets) as a function of currentconditions in the market (system load and localreliability energy levels) and the realization ofunit-level outages. Because we do not know thebidding strategies of even a single market par-ticipant and we do not observe actual generationunit-level outages during our sample period, areduced-form approach to construct the estimateof E(mkupid) is necessary to compute an esti-mate of «id. We use a linear predictor ofmkupid

constructed using hour-of-day, day-of-the-week,and month-of-sample period dummies, alongwith the level and square of both the forecastISO load and day-ahead total RMR require-ments for that hour of the day. Our estimate of«id is the residual from the regression ofmkupid

on these variables for all hours in our sampleperiod. For the same reason that the squareddifference between a random variable and itsconditional expectation is always less than thesquared difference between that random vari-able and its best linear predictor using thosesame conditioningvariables, the variance of ourestimate of «id should be larger than the truevariance of «id. Consequently, we view ourstandard error as a very conservative estimate ofthe uncertainty in DTC( ) due to unobservableforced outages and their impact on realizationsof the PX price.

Applying this procedure with the Barlett ker-nel, k(s/(q 1 1)) 5 s/(q 1 1), for a value ofq equal to 10 yields a standard error for DTC( )for our entire sample period of $1.19 billion onthe estimated DTC( ) of $5.55 billion. For theJune–October periods of 1998, 1999, and 2000,the estimates of DTC( ) and their standarderrors (all in $ million) are $425 ($46), $382($41), and $4,448 ($430), respectively.

APPENDIX C: COST IMPLICATIONS OF RESERVOIR

ENERGY ASSUMPTION

In this Appendix we discuss the implicationsof our assumption that the observed productionfrom reservoir resources (hydroelectric and geo-thermal) is equal to the optimal schedule thatwould be produced by rational price-taking9rms in a perfectly competitive market. Weargue that, although this assumption is surelynot true, any bias it creates in our estimates ofthe competitive benchmark price will be in anupward direction, leading us to understate theamount of market power. We do this by 9rstcharacterizing an optimal hydro schedule, andthen discussing the cost impacts of deviationsfrom that optimal schedule.

Characterizing the Optimal Hydro Sched-ule.—Assume that there are n producers thatcontrol both hydro and thermal generation re-sources, where thermal resources include nuclearand fossil-fuel generation. Let qit 5 qit

Th 1 qith

represent the total output of 9rm i in time t,where qit

Th is the thermal output and qith the

hydro output of 9rm i. The thermal output of a9rm is required to be nonnegative and also nomore than its total thermal capacity, qi,max

Th .Each producer i 5 1, ... , n has a portfolio of

thermal generation technologieswith an associ-ated aggregate production cost of Ci(qi

Th)and marginal cost of ci(qi

Th). We assume thatc(z) is a strictly monotone increasing functionof qTh.

We can characterize the hydro systems of thesuppliers as having a reservoir of q# i

h units ofavailable water (measured in units of energy),qi,minh units of required minimum @ow in each

period, and a maximum @ow of qi,maxh per pe-

riod. We assume that any in@ows that occurduring the time periods modeled (say a week ora month) do not disrupt the aggregate hydrooutput decisions of each 9rm. In other words,the limits on the total reservoir capacity are notbinding during this relatively short-term plan-ning horizon, so that any unexpected in@ows areadded to storage. We also assume that demand,although responsive to price and varying withtime, is deterministic.

Let pt(Qt) represent the inverse demandfunction for the market at time t. Given the


output of the other 9rms, a price-taking 9rm ihas an optimal production problem de9ned as

(C1) Maxqith ,qitTh Ot

pt qit 2 Ci ~qitTh!

subject to the constraints

qi ,minh # qit

h # qi,maxh @t

qitTh # qi,max

Th @t

qith , qit

Th $ 0 @t

Ot

qith 5 q# i

h

where qit 5 qitTh 1 qit

h, and Qt 5 ¥i qit, thetotal market output in that period. Note that thisproblem would be separable in t, except for thelast constraint, which limits the total hydro pro-duction over the T periods. We assume that each9rm’s single-period pro9t is concave in its ownoutput. For decreasing price functions,pt, it can beshown that this problem has a concave objectivefunction so that the problem as a whole is convex.

In the above formulation, we do not allow forthe “spilling,” or free disposal, of hydro energy.Under extreme circumstances it may be pro9t-able for a 9rm to withhold energy by spillingwater, but this would only occur if the 9rm hadenough reservoir quantity that it had drivenmarginal revenue to zero for all periods inwhich it was not at a maximum @ow constraint.

To characterize the optimal solutions,we assignLagrange multipliers to each of these constraints.The multipliers of interest are cit for the thermaloutput limits, git and dit for the hydro productionlimits, and si on the total available water to thestrategic hydro producer. The term si is thereforethis 9rm’s marginal value of water in this model.This value represents the additional pro9t to the9rm that would arise if an additional unit of watercould be used for generation during the time frameof the optimization. The optimal solution is char-acterized by the following conditions:

(C2)L

qftTh 5 pt 2 ci ~qft

Th! 2 cft

# 0 qftTh $ 0 @t;

(C3) pt 5 2gft 1 dft 1 s f ;

(C4) c it ~qitTh 2 qi,max

Th ! 5 0 @i,t;

(C5) g it ~qi ,minh 2 qit

h ! 5 0 @i,t;

(C6) d it ~qith 2 qi ,max

h ! 5 0 @i,t;

(C7) s iX Ot

q ith 2 q# i

hD 5 0 @i;

(C8) qitTh # qi,max

Th , qi,minh # qit

h ,

qith # qi,max

h , Ot

qith # q# i

h @t;

(C9) qitTh, c it , git , dit , s i $ 0 @i, t

where the symbol indicates complementarity.Combining (C2) and (C3) shows that ci(qit

Th) 1cit $ 2git 1 dit 1 si for all t. When met withequality, conditions (C2) and (C3) represent thecondition that price, pt, equals marginal cost.

Each price-taking 9rm will schedule its hydroreleases so as to equate its marginal costs acrossall periods in the time horizon, where marginalcosts include a component for the shadow priceof generation capacity when the capacity con-straint is binding. In the hours in which there areno binding @ow constraints on hydro produc-tion, prices will be set equal to the marginalvalue of water, si , which is constant across thetime periods of the planning horizon. Let Y [{0, ... , T} denote the subset of hours whenneither @ow constraint binds. Then,

(C10) pt 5 s 5 ci ~qtTh! @t [ Y .

Deviations from the Optimal Hydro Sched-ule.—In order to evaluate the potential impactof a suboptimal hydro allocation on the equilib-rium conditions, we need to consider possibledeviations from the optimal hydro allocationdescribed above. These fall into four potentialcategories:

1. A reallocation from a period when the max-imum @ow constraint was binding to a pe-riod when the minimum @ow was binding;


2. A reallocation from a period when the max-imum @ow constraint was binding to a pe-riod in which no @ow constraints werebinding;

3. A reallocation from a period when no @owconstraints were binding to a period in whichthe minimum @ow constraint was binding;

4. A reallocation between periods in which no@ow constraints were binding.

A reallocation of energy away from hourswhen the maximum @ow constraint was bindingunder the optimal allocation can only raise mar-ginal cost since by condition (C3), hours inwhich this constraint was not binding wouldhave lower prices (marginal cost) than hours inwhich it was binding. The same is true for areallocation from an hour in which no @owconstraints bind to an hour in which the mini-mum constraint was binding. Last, by condition(C3), marginal costs in hours in which no @owconstraints bind must be equal. A reallocationfrom one such hour to another would necessar-ily raise the average marginal cost if the costcurve were convex. Therefore any hydro sched-ule that deviates from the optimum schedule canonly raise the unweighted average marginal costof production.

To address how a suboptimal hydro schedulemight impact the volume-weighted average ofmarginal cost over a given time period, weutilize the following result.

RESULT: Any feasible allocation of energythat is not equal to the optimal allocation andthat produces a monotonic relationship betweenmarginal cost, c(qt

Th), and total demand, qt,will produce a higher demand-weighted aver-age marginal cost than the optimal schedule.

To prove this result, consider again the fourpossible types of reallocations of energy awayfrom the optimal schedule that are listed above.Any reallocation of energy away from a higher-demand period to a lower-demand period, cases1, 2, and 3, will clearly raise the demand-weighted average marginal cost. Similarly a re-allocation away from an unconstrained hourwith higher demand to an unconstrained hourwith lower demand would also raise the demand-weighted average of marginal costs. However, a

reallocation from an unconstrained hour withlower demand to one with higher demand wouldnecessarily raise marginal cost in the lower de-mand hour to a level that is higher than themarginal cost in the next highest demandlevel since the marginal costs in the two hourswere previously equal to each other. Such areallocation would therefore create a nonmono-tonic relationship between marginal cost anddemand.

In light of the above result, we feel con9dentthat the observed hydro schedule that we utilizein our calculationswill produce a higher demand-weighted marginal cost than the unobservedoptimal hydro schedule as long as we observethat our calculations of marginal cost are mono-tonically increasing in aggregate demand. Fig-ure C1 illustrates this with graphs that are theresult of kernel density regressions of this rela-tionship for August–September of 1998, 1999,and 2000, respectively. In each case we doin fact observe a monotonic relationship be-tween our estimates of marginal cost and de-mand, leading us to conclude that ourassumption about the allocation of energy fromreservoir resources is a conservative one thatwill understate the degree of market power thatwe 9nd.

REFERENCES

Andrews, Donald W. K. “Heteroskedasticity andAutocorrelation Consistent Covariance Matrix

FIGURE C1. KERNEL REGRESSIONS OF COMPETITIVE PRICEVS. ISO DEMAND (JUNE–SEPTEMBER)


Estimation.” Econometrica, May 1991, 59(3),pp. 817–58.

Bohn, Roger; Klevorick, Alvin and Stalon, Charles.“Second Report on Market Issues in the Cali-fornia Power Exchange Energy Markets.” Pre-pared for the Federal Energy RegulatoryCommission by the Market Monitoring Com-mittee of the California Power Exchange,March 1999. Available at http://www.ucei.org/Restructuring_Archive/pxMMC_report99.pdf .

Borenstein, Severin. “Frequently Asked Ques-tions about Implementing Real-Time Elec-tricity Pricing in California for Summer2001.” Mimeo, University of California En-ergy Institute, March 2001. Available athttp://www.ucei.org/PDF/faq.pdf .

. “The Trouble with ElectricityMarkets:Understanding California’s Restructuring Di-saster.” Journal of Economic Perspectives,Winter 2002, 16(1), pp. 191–212.

Borenstein, Severin and Bushnell, James B. “AnEmpirical Analysis of the Potential for Mar-ket Power in California’s ElectricityMarket.”Journal of Industrial Economics, September1999, 47(3), pp. 285–323.

Borenstein, Severin; Bushnell, James B. and Knit-tel, Christopher R. “Market Power in Electric-ity Markets: Beyond ConcentrationMeasures.”Energy Journal, October 1999, 20(4), pp.65–88.

Borenstein, Severin; Bushnell, James B.; Knittel,Christopher R. and Wolfram, Catherine D.“Trading Inef9ciencies in California’s Elec-tricity Markets.” POWER Working PaperPWP-086, University of California EnergyInstitute, October 2001. Available at http://www.ucei.org/PDF/pwp086.pdf .

Borenstein, Severin; Bushnell, James B. and Stoft,Steven. “The Competitive Effects of Trans-mission Capacity in a Deregulated ElectricityMarket.” RAND Journal of Economics, Sum-mer 2000, 31(2), pp. 294–325.

Bushnell, James B. “Transmission Rights andMarket Power.” Electricity Journal, October1999, 12(8), pp. 77–85.

. “A Mixed Complementarity Model ofHydro-Thermal Electricity Competition inthe Western U.S.” Operations Research, Jan-uary/February 2003, 51(1).

Bushnell, James B. and Wolak, Frank A. “Regu-lation and the Leverage of LocalMarket Power:

Reliability Must-Run Contracts in the Cali-fornia Electricity Market.” POWER WorkingPaper PWP-070, University of California En-ergy Institute, September 1999. Available athttp://www.ucei.org/PDF/pwp070.pdf .

Cardell, Judith B.; Hitt, Carrie C. and Hogan,William W. “Market Power and Strategic In-teraction in Electricity Networks.” Resourcesand Energy Economics, March 1997, 19(1–2), pp. 109–38.

Joskow, Paul L. and Kahn, Edward. “A QuantitativeAnalysis of Pricing Behavior In California’sWholesale Electricity Market During Summer2000.” Energy Journal, 2002, 23(4), pp. 1–35.

Joskow, Paul L. and Tirole, Jean. “TransmissionRights and Market Power on Electric PowerNetworks.” RAND Journal of Economics,Autumn 2000, 31(3), pp. 450–87.

Kahn, Edward; Bailey, Shawn and Pando, Luis.“Simulating Electricity Restructuring in Cal-ifornia: Interactions with the Regional Mar-ket.” Resources and Energy Economics,March 1997, 19(1–2), pp. 3–28.

Mansur, Erin T. “Environmental Regulation inOligopoly Markets: A Study of ElectricityRestructuring.” POWER Working PaperPWP-088, University of California EnergyInstitute, November 2001. Available at http://www.ucei.org/PDF/pwp088.pdf .

Newey, Whitney K. and West, Kenneth D. “ASimple, Positive Semi-de9nite, Heteroske-dasticity and Autocorrelation Consistent Co-variance Matrix.” Econometrica, May 1987,55(3), pp. 703–08.

Pando, Luis. Testimony and Workpapers inSouthern California Gas Company, Califor-nia Public Utilities Commission ApplicationNo. 95-06-002, 1995.

Patrick, Robert H. and Wolak, Frank A. “Esti-mating the Customer-Level Demand forElectricity Under Real-Time Market Prices.”Mimeo, Stanford University, 1997. Availableat http://www.stanford.edu/˜wolak .

Puller, Steven L. “Pricing and Firm Conduct inCalifornia’s Deregulated Electricity Market.”POWER Working Paper PWP-080, Univer-sity of California Energy Institute, July 2001.Available at http://www.ucei.org/PDF/pwp080.pdf .

Wolak, Frank A. “An Empirical Analysis of theImpact of Hedge Contracts on Bidding Be-


havior in a Competitive Electricity Market.”International Economic Journal, Summer2000, 14(2), pp. 1–40.

Wolak, Frank A.; Nordhaus, Robert and Shapiro,Carl. “Preliminary Report on the Operation oftheAncillary Services Markets of the CaliforniaIndependent System Operator (ISO).” Preparedfor the Federal Energy RegulatoryCommissionby the Market Surveillance Committee ofthe California ISO, August 1998. Available athttp://www.ucei.org/PDF/isomsc.pdf .

Wolak, Frank A. and Patrick, Robert H. “TheImpact of Market Rules and Market Structureon the Price Determination Process in the

England and Wales Electricity Market.”POWER Working Paper PWP-047, Univer-sity of California Energy Institute, April1997. Available at http://www.ucei.org/PDF/pwp047.pdf .

Wolfram, Catherine D. “Strategic Bidding in aMultiunit Auction: An Empirical Analysis ofBids to Supply Electricity in England andWales.” RAND Journal of Economics, Win-ter 1998, 29(4), pp. 703–25.

. “Measuring Duopoly Power in theBritish Electricity Spot Market.” AmericanEconomic Review, September 1999, 89(4),pp. 805–26.


Documents

Measuring Market Inef®ciencies in California’s