power marketer pricing behavior in the california power exchange

8
Power marketer pricing behavior in the California Power Exchange Tyler Hodge a , Carol A. Dahl b, c, , 1 a U.S. Energy Information Administration, United States b Colorado School of Mines, United States c King Saud University, Saudi Arabia abstract article info Article history: Received 1 February 2009 Received in revised form 16 March 2011 Accepted 8 May 2011 Available online 13 May 2011 JEL classication: L19 L94 Q49 Keywords: Power Marketer Pricing Electricity Wholesale Various studies have examined whether market power abuses by independent electricity generators contributed to the demise of the California Power Exchange (PX). However, the behavior of wholesale power marketers has generally been overlooked. To ll this gap, our paper focuses on the pricing behavior of ve major power marketers in the California PX during 2000: Duke Energy Trading & Marketing, Reliant Energy Services, Dynegy Power Marketing, Enron Power Marketing, and Williams Energy Marketing & Trading. Our unique data set, collected by the Federal Energy Regulatory Commission during an investigation of energy market pricing manipulation, allows us to assess the level of market power using the conduct parameter pricing model. The estimated conduct parameter allows us to determine power marketer pricing behavior is competitive, Cournot, or collusive. Our results indicate that Duke Energy and Reliant were exercising market power when pricing the wholesale electricity they sold in the California PX during 2000. No statistical evidence was uncovered to show that the smaller marketers Dynegy, Williams and, Enron were setting prices at a level higher than those consistent with a competitive market. © 2011 Elsevier B.V. All rights reserved. 1. Introduction The investigation of price manipulation in emerging electricity markets has been a popular avenue of research in recent years. In particular, the California Power Exchange (PX) market has attracted most of the U.S. research attention because of its spectacular demise in 2001 and the wide availability of trading data. Authors such as Joskow and Kahn (2002), Wolak (2003), Kim and Knittel (2004) and Puller (2007) have implicated price manipulation by merchant generating rms as an important cause of the price spikes leading to the collapse of the California power market. However, none of these studies looked at the pricing behavior of power marketers, which are vital for a well- functioning electricity market, assuming they act competitively. In this paper, we use an untapped dataset for power marketer transactions in the California Power Exchange. We focus on the pricing behavior of ve large wholesale power marketers (Duke Energy Trading & Marketing, Reliant Energy Services, Dynegy Power Marketing, Enron Power Marketing, and Williams Energy Marketing & Trading) to analyze whether they exerted signicant market power in the California Power Exchange during the year 2000. In order to investigate this issue, we apply a conduct parameter model to analyze the power marketers' strategic pricing behavior. California Assembly Bill (AB) 1890, the Electric Utility Industry Restructuring Act, was enacted in 1996 with the goal of breaking up the vertical structure of the industry and creating a competitive electricity market, which many believed would lower the cost of electricity for retail consumers. One of the most signicant provisions of AB 1890 was the establishment of a wholesale electricity market trading mechanism. This mechanism consisted of two related markets: the California Power Exchange (PX) and the CAISO balancing market. The California PX was intended to act as the primary market for wholesale electricity. It operated as a day-ahead market in which hourly demand and supply bids were submitted for the next day's trades, and an equilibrium price was set by the interaction between the supply and demand schedules. The CAISO received information about the planned supply schedule and the expected load and checked for any strains to the transmission system. In addition, CAISO ran a real-time balancing market to match actual realized load with available power supply. The California wholesale market system worked smoothly for its rst two years of operation with prices in the California PX averaging about $33 per MWh compared to retail rates of $65 per MWh. Energy Economics 34 (2012) 568575 The views in this paper are those of the authors and do not necessarily reect the views of the U.S. Energy Information Administration, the Colorado School of Mines, or King Saud University. Corresponding author at: Colorado School of Mines, United States. E-mail address: [email protected] (T. Hodge). 1 The authors' views and opinions expressed herein do not necessarily state or reect those of the United States government or any agency thereof. All errors and omissions are the sole responsibility of the authors. 0140-9883/$ see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.eneco.2011.05.003 Contents lists available at ScienceDirect Energy Economics journal homepage: www.elsevier.com/locate/eneco

Upload: tyler-hodge

Post on 05-Sep-2016

213 views

Category:

Documents


0 download

TRANSCRIPT

Energy Economics 34 (2012) 568–575

Contents lists available at ScienceDirect

Energy Economics

j ourna l homepage: www.e lsev ie r.com/ locate /eneco

Power marketer pricing behavior in the California Power Exchange☆

Tyler Hodge a, Carol A. Dahl b,c,⁎,1

a U.S. Energy Information Administration, United Statesb Colorado School of Mines, United Statesc King Saud University, Saudi Arabia

☆ The views in this paper are those of the authors anviews of the U.S. Energy Information Administration, thKing Saud University.⁎ Corresponding author at: Colorado School of Mines

E-mail address: [email protected] (T. Hodge).1 The authors' views and opinions expressed herein do

those of the United States government or any agency thare the sole responsibility of the authors.

0140-9883/$ – see front matter © 2011 Elsevier B.V. Adoi:10.1016/j.eneco.2011.05.003

a b s t r a c t

a r t i c l e i n f o

Article history:Received 1 February 2009Received in revised form 16 March 2011Accepted 8 May 2011Available online 13 May 2011

JEL classification:L19L94Q49

Keywords:PowerMarketerPricingElectricityWholesale

Various studies have examined whether market power abuses by independent electricity generatorscontributed to the demise of the California Power Exchange (PX). However, the behavior of wholesale powermarketers has generally been overlooked. To fill this gap, our paper focuses on the pricing behavior of fivemajor power marketers in the California PX during 2000: Duke Energy Trading & Marketing, Reliant EnergyServices, Dynegy Power Marketing, Enron Power Marketing, and Williams Energy Marketing & Trading. Ourunique data set, collected by the Federal Energy Regulatory Commission during an investigation of energymarket pricing manipulation, allows us to assess the level of market power using the conduct parameterpricing model. The estimated conduct parameter allows us to determine power marketer pricing behavior iscompetitive, Cournot, or collusive. Our results indicate that Duke Energy and Reliant were exercising marketpower when pricing the wholesale electricity they sold in the California PX during 2000. No statisticalevidence was uncovered to show that the smaller marketers – Dynegy, Williams and, Enron – were settingprices at a level higher than those consistent with a competitive market.

d do not necessarily reflect thee Colorado School of Mines, or

, United States.

not necessarily state or reflectereof. All errors and omissions

ll rights reserved.

© 2011 Elsevier B.V. All rights reserved.

1. Introduction

The investigation of price manipulation in emerging electricitymarkets has been a popular avenue of research in recent years. Inparticular, the California Power Exchange (PX) market has attractedmost of the U.S. research attention because of its spectacular demise in2001 and the wide availability of trading data. Authors such as Joskowand Kahn (2002), Wolak (2003), Kim and Knittel (2004) and Puller(2007) have implicated price manipulation by merchant generatingfirms as an important cause of the price spikes leading to the collapseof the California powermarket. However, none of these studies lookedat the pricing behavior of power marketers, which are vital for a well-functioning electricity market, assuming they act competitively. Inthis paper, we use an untapped dataset for power marketertransactions in the California Power Exchange.

We focus on the pricing behavior of five large wholesale powermarketers (Duke Energy Trading & Marketing, Reliant Energy

Services, Dynegy Power Marketing, Enron Power Marketing, andWilliams Energy Marketing & Trading) to analyze whether theyexerted significant market power in the California Power Exchangeduring the year 2000. In order to investigate this issue, we apply aconduct parameter model to analyze the power marketers' strategicpricing behavior.

California Assembly Bill (AB) 1890, the Electric Utility IndustryRestructuring Act, was enacted in 1996 with the goal of breaking upthe vertical structure of the industry and creating a competitiveelectricity market, which many believed would lower the cost ofelectricity for retail consumers. One of the most significant provisionsof AB 1890 was the establishment of a wholesale electricity markettrading mechanism. This mechanism consisted of two relatedmarkets: the California Power Exchange (PX) and the CAISO balancingmarket. The California PX was intended to act as the primary marketfor wholesale electricity. It operated as a day-ahead market in whichhourly demand and supply bids were submitted for the next day'strades, and an equilibrium price was set by the interaction betweenthe supply and demand schedules. The CAISO received informationabout the planned supply schedule and the expected load andchecked for any strains to the transmission system. In addition, CAISOran a real-time balancing market to match actual realized load withavailable power supply. The California wholesale market systemworked smoothly for its first two years of operation with prices in theCalifornia PX averaging about $33 per MWh compared to retail ratesof $65 per MWh.

$0

$100

$200

$300

$400

$500

$600

$700

$800

$900

$1,000

Jan2000

Feb2000

Mar2000

Apr2000

May2000

Jun2000

Jul2000

Aug2000

Sep2000

Oct2000

Nov2000

Dec2000

Jan2001

Pri

ce (

$/M

Wh)

Fig. 1. Daily California Power Exchange prices.Source: University of California Energy Institute(2006).

569T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

During the summer of 2000, many problems converged simulta-neously causing wholesale prices in the California PX to reachstaggering levels. As Fig. 1 shows, in the last week of May peak pricesbegan to reach levels 2–3 times higher than during the previousmonths. Prices settled at slightly higher prices for about a month.Then July saw a number of price spikes. From August throughSeptember, prices again became extremely volatile at high levels. Thecrisis in the California electricity market reached its pinnacle inDecember of 2000 and January of 2001 when peak prices reachedaverage levels in excess of $500 per MWh on many days. UltimatelyCalifornia's largest utility, PG&E, declared bankruptcy due to the hugelosses it incurred from purchasing wholesale electricity at the high PXrates and reselling at the frozen retail rates. In mid-January, theCalifornia Department of Water Resources began contracting private-ly with generating firms to purchase electricity on behalf of the threelarge California utilities. Finally, on January 31, 2001, the CaliforniaPower Exchange disbanded and ceased operations.

In response to the price spikes in late 2000–2001, the FederalEnergy Regulatory Commission began a major fact-finding investiga-tion into potentialmanipulation of energy prices in thewesternUnitedStates for 2000 and 2001 (Federal Energy Regulatory Commission[FERC], 2003b). This investigation (Docket No. PA02-2) examined thepricing behavior of participants in the electricity and natural gasmarkets by using data on all electricity transactions during the periodunder investigation, including transactions in the California PX. FERCuncovered a large amount of evidence documenting questionable, andin some cases criminal, pricing practices by a number of firmsincluding Enron, AES/Williams, Dynegy/NRG, Mirant, and Reliant. Inthe next section of this paper, we develop a model to examine thepricing behavior, which we then apply to the California PX data forthese four power marketers and another major player, Duke Energy.

2. Conduct parameter model

We analyze pricing conduct in the California Power Exchangeduring 2001 using a conduct parameter model of an industry with noentry and producing a homogenous good, which are two character-istics of the California wholesale power market. Bresnahan (1989)“conduct parameter”model measures of the level of competitiveness,

frequently denoted by the Greek letter theta (θ). Our analysis ofpricing behavior in this paper relies heavily on the conduct parametermeasure. The concept behind θ can be illustrated with the generalizedprofit-maximizing condition for firm i at time t:

Pt = CQ Qit ;Wi;Zi;γi; εcitÞ−DQ Qt;Y;δ; εdtÞQitθitð�

ð2:1Þ

where Pt ismarket price,CQ is themarginal cost function,DQ is thepartialderivative of the inverse demand functionwith respect to quantityQ,Wis a vector of input prices, Z includes other variables that shift themarginal cost curve, Y includes variables that shift the inverse demandcurve,γ and δ are theunknownparameters of the two functions, and theε terms represent the errors between the observed data and themodeled economic relationship. The parameter θ represents the firm'sconduct parameter in terms of pricing behavior, and it generalizes theprofit-maximizing condition for all types of market structures.

Eq. (2.1) is a supply relation since each firm's optimal supplyquantity occurs at the point where marginal cost CQ is equal to“perceived” marginal revenue (P+DQQθ)—i.e., the firm's addedrevenue assuming it understands that competitors may react to itschoice of quantity. When analyzed from an industry perspective, asopposed to a firm-level perspective, the conduct parameter takes on avalue of zero if the industry is perfectly competitive (so that P=CQ),and it takes on a value of one if the industry is monopolistic or if firmsperfectly collude (so that P+DQQ=CQ). Values between zero and oneindicate a level of competitiveness between perfect competition andmonopoly—if all n firms are identical and have Cournot pricingbehavior then θ=1/n. The interpretation of θ is related to theconjectural variations model explored by Iwata (1974).

Three general market structures can be summarized into thefollowing generalized first-order condition:

P +dPdQ

θqi = MCi: ð2:2Þ

Each firm chooses the quantity in which its “perceived” marginalrevenue equals its marginal cost. The industrial organization literatureoften refers to the value of θ as the firm's conduct parameter. Thethree possible structures can thus be represented by different values

570 T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

for the conduct parameter: θ=0 for perfect competition, θ=1 for aCournot industry, or θ=n and qi=Q for a cartel of n firms. Oneinterpretation of the firm's “conduct” comes from the theory ofconjectural variations. This theory postulates that, when choosing theprofit-maximizing quantity, a firm accounts for the (conjectured)response of other firms to its supply decisions—that is, the value forθ=∑∂qj/∂qi may or may not equal zero. Although the estimatedconduct parameter can theoretically be a continuous value, for ourstatistical work we nested hypotheses that the estimated value willtake on one of three values: 0, 1, or n corresponding to perfectcompetition, Cournot behavior, or collusive behavior under a cartel.

The first of the nested hypotheses will be an initial null hypothesisthat the electricity sellers price their product at perfectly competitivelevels (i.e., at marginal cost). If the initial null hypothesis of perfectcompetition is rejected, the nested hypotheses will move onto higherlevels of market power: first testing Cournot behavior, and thentesting the possibility of collusive behavior with other wholesalers.

Since marginal costs are not directly observable, economists havedeveloped various models which utilize available data in order torecover the cost function and analyze firm-level pricing behavior. Themodel we use in this paper is similar to the Bresnahan-style conductparameter model used by other researchers who analyze pricingbehavior in electricity markets. For example, Wolfram (1999) uses theconduct parameter methodology to analyze the pricing behavior ofgenerators in the England and Wales spot electricity market. Herstudy interprets the conduct parameter as an elasticity-weightedLerner index. The resulting values are very small, indicatinggenerators may be behaving competitively. She argues that thefirms may be pricing near marginal cost in order to deter entry and toavoid regulatory investigation into pricing practices. A similar studywas conducted by Kim and Knittel (2004) for the California electricitymarket. The authors developed detailed measures for marginal cost,calculated Lerner indices and then estimated conduct parameters.Although their conduct parameter estimates are significantly differentfrom zero, they reject the methodology since the estimated valuesdiffer from the direct Lerner index calculations.

Wolak (2003) examines the ability of the five largest suppliers inthe CAISO balancing market to exercise market power. He calculatesthe residual demand elasticity for each firm based on actual bid data,and uses the inverse elasticity to calculate a Lerner index. The studyresults indicate that market power was exhibited by all five firmsduring 2000, when compared against pricing behavior during 1998.Puller (2007) adopts the conduct parameter methodology in hisexamination of the pricing behavior of independent power producersbefore and during the California crisis of 2000. He uses a conjecturalvariations approach, similar to Iwata (1974) and derives the estimateof conjectural variations using parameters from the demand equation.The results of his static analysis indicate that the spot market inCalifornia was not perfectly competitive, but more in line with thenon-cooperative behavior of a Cournot oligopoly consisting of 4–5firms. Of the numerous studies examining market power inCalifornia's electricity market, only one directly addresses the pricingof power marketing firms. Wolak (2003) identifies the five-largestCalifornia “suppliers” in his analysis as AES/Williams, Duke, Dynegy,Mirant, and Reliant. Although these five firms owned generating unitsin California, the CAISO bid schedules used in Wolak's study werelikely submitted by the firms' power marketing affiliates.

Market demand in the California PX was ultimately derived fromend-use electricity consumption, through retail sales to the residen-tial, commercial, and industrial sectors. Load-serving entities such asthe state utilities PG&E and SDG&E were the primary purchasers onthe PX. Power marketers might also have purchased electricity fromthe PX, possibly to arbitrage against other markets such as the CAISObalancing market.

We start with a model of wholesale electricity demand in theCalifornia PXwith quantity demanded as a function of wholesale price

and other relevant variables. We model inverse market demand usingthe following linear representation:

Pt = α0 + α1DHt + α2D

Ct + α3Wt + α4Yt + α5Rt + α6Ht + γQD

t + εDt :

ð2:3Þ

The explanatory variables in this simple demand model capturemost of the demand determinants. The quantity that wholesalebuyers wish to purchase is dependent primarily on the wholesalemarket price Pt. If power marketers purchased from the PX, then theprice of other regional markets, especially the CAISOmarket, might bea relevant explanatory variable. However, for our model we assumethat the affect of power marketer arbitrage between the California PXand CAISO is relatively small since about 90% of the volume in theCalifornia markets traded on the day-ahead PX market during 2000(Borenstein et al., 2001).

Retail electricity load fluctuates widely depending on the regionalweather patterns. A large proportion of retail load met by Californiautilities consists of climate control in residential households,specifically for cooling requirements during hot weather. For ourmodel we use a measure of relative temperature, cooling degrees (Dt

C)and heating degrees (Dt

H), as a proxy for weather. If the meantemperature on a given day exceeds the baseline “comfortable”temperature of 65° then cooling degrees is calculated as the meantemperature minus 65°; otherwise Dt

C=0. Similarly, heating degreesare defined as 65° minus the observed mean temperature during theday; Dt

H=0 if this difference is negative.Businesses also use electricity for climate control, as do residential

consumers, but a primary determinant of commercial and industrialelectricity consumption is the level of economic activity. Daily data formacroeconomic production in California is unavailable. As a proxy forcommercial economic activity, we use monthly California employ-ment, Yt (California Employment Development Department, 2006).Industrial economic activity is modeled using weekly Californiarefinery gasoline production, Rt (California Energy Commission,2001). Electricity use for the commercial and industrial sectorsdrops on weekends. Therefore, the model includes a dummy variable,Wt, equal to one if the specific observation occurs on a Saturday orSunday, to account for this daily variation. Finally, a large proportionof the electricity delivered by California utilities is generated byhydroelectric projects, the largest of which is Shasta Dam on theSacramento River. Higher levels of hydropower releases reduce theneed for utilities to purchase electricity from the California PX; thuswe include releases from the Shasta Dam, Ht, as an explanatoryvariable in the demand equation.

The slope of the inverse demand curve,γ, can be used to analyzethe price elasticity of electricity demand. This electricity demand canfluctuate widely in the summer months depending on regionaltemperatures. For this paper, we hypothesize that hotter tempera-tures relative to the baseline 65° temperature will impact demandresponsiveness: γ=γ0+γ1Dt

C. Temperature effects during wintermonths have a smaller impact on day-to-day changes in electricitydemand since consumption of electricity for space heating inCalifornia is less than a quarter of that used for summer airconditioning (Energy Information Administration, 2008). This mod-ification of elasticity based on temperatures is an important facet ofthe model; it will allow us to identify the level of market power.Substituting the elasticity relationship, we can rewrite the inversemarket demand Eq. (2.3) with an interaction term between quantitydemanded and cooling degrees:

Pt = α0 + α1DHt + α2D

Ct + α3Wt + α4Yt + α5Rt + α6Ht + γ0Q

Dt

+ γ1QDt D

Ct + εDt :

ð2:4Þ

571T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

A firm's supply behavior is determined by the marginal cost ofproducing a given quantity of product. The largest component of totalcost for a power marketer, and most likely the strongest determinantof marginal cost, is the cost of purchasing the electricity that will beresold on the wholesale market. Electricity is purchased through bothlong-term contracts and short-term purchases. However, a marketer'ssupply decision is influenced primarily by the variable cost ofmarginal short-term purchases. These wholesale purchases are oftentransacted through bilateral negotiation with wholesale generatorslocated outside of California, primarily on the borders with Oregonand Arizona.

Although measures of marginal cost are unobservable, a repre-sentation of the marginal cost function in statistical form allows us toset up the econometric supply relation representing pricing behavior:

MCit = β0 + β1qSit + β2Cit + β3Tt + β4Gt + εSit ð2:5Þ

where C represents the short-run input price for purchased electricity.Although this electricity was often purchased at control areas outsidethe jurisdiction of the California market, the PX price, P, is likely to bestrongly correlated with the input cost, C, on any given day. Theindividual firm's quantity supplied and the market input price arestandard economic determinants of marginal cost. Another importantvariable that impacts a power marketer's marginal cost of obtainingelectricity is the level of congestion on the transmission grid. Pricesamong control zones within a specific market are normally equalexcept during times of transmission congestion. Transmission lineconstraints result in wholesale price premiums or direct transmissionfees in regions affected by the constraint. Herewemeasure the level oftransmission congestion encountered by all firms in the California PXusing the number of hours during the day, Tt, that price in thesouthern SP-15 transmission control area differed from price in thenorthern NP-15 control area. Many power marketers are affiliatedwith independent power producers who own generation resources,and some of the electricity sold to the California PX may ultimatelyhave been produced by these resources. Therefore, we include theaffiliated generator's heat rate, Gt, as a proxy for the cost of self-generated power.2

Direct observations of a firm's marginal cost, MCit, are notavailable, but this functional representation of marginal cost (2.5)can be paired with the firm's profit maximization condition to derivethe power marketer's supply relation. Assuming the market is inequilibrium, i.e.:

QD�t ≡ QS�

t

where QtS is the sum of all sellers submitting bids into the market

(QtS=∑qit), the general first-order condition for profit maximization

that perceived MR=MC allows us to create a statistical supplyrelation by substituting the partial derivative of the demand Eq. (2.4)with respect to total demand, Qt

D, and the marginal cost Eq. (2.5):

Pt + θ γ0 + γ1DCt

� �qit = β0 + β1qit + β2Cit + β3Tt + β4Gt + εit

which we can simplify to:

Pt = β0 + β1−θγ0ð Þqit−θγ1DCt qit + β2Cit + β3Tt + β4Gt + εit : ð2:6Þ

2 The source of the electricity sold by power marketers is often difficult todetermine. Presumably, some of the power was generated by affiliated generators, inwhich case a measure of generation cost is an important component of the marketmodel. We appreciate the reviewer's suggestion of heat rate as a standard proxy forgeneration cost.

This supply relation contains two exogenous right-hand-sidevariables, Dt

C and Tt. Transmission congestion is exogenous becauseit is determined by actual load conditions after the (day-ahead) PXmarket has settled. The firm's choice of supply quantity, qit, and thepurchase cost, Cit, are assumed to be right-hand-side endogenousvariables. Supply quantity is determined simultaneously to pricingdecisions, and shocks to California PX prices are likely to haverelatively strong effects on wholesale electricity prices (costs) inneighboring regions. However, these neighboring region prices arealso determined by generator fuel costs in these regions. Somegenerators in Arizona and the Pacific Northwest use natural gas forelectricity generation just as California generators do, but coal andhydropower are also common fuels. Ourmodel uses the dailymarginalfuel cost from Arizona and hydropower releases from Bonneville Damin Oregon as instruments for the wholesale purchase cost variable.

The full model is a system of equations comprising the supplyrelation (2.6) and the demand Eq. (2.4). Note that price in the demandequation is a function of the total market quantity, Qt

D, and in thesupply relations price is a function of firm-level quantity, qit. In orderto accurately estimate the parameters of both equations, we need toconnect these two variables using the following identity:

QDt = qit + ∑

j≠iqjt

where the summation term on the right-hand side is the quantitysupplied into the market by all firms other than the one(s) beingmodeled. We assume that this “other” supply quantity is exogenous(i.e., all other suppliers behave competitively). Note that subtractingthis quantity from both sides illustrates the firm's equilibriumcondition that residual demand equals the firm's quantity supplied.The full system for the pricing behavior model can be summarized as:

ðDemandÞPt = α0 + α1DHt + α2D

Ct + α3Wt + α4Yt

+ α5Rt + α6Ht + γ0QDt + γ1Q

Dt D

Ct + εDt

ðSupplyÞPt = β0 + β1−θγ0ð Þqit−θγ1DCt qit + β2Cit + β3Tt + β4Gt + εSit

ðEquilibriumÞQDt ≡ qit + ∑

j≠iqjt :

ð2:7Þ

Market demand (QtD), residual demand (qit), PX selling price (Pt),

and the input cost of purchased electricity (Cit) are the endogenousvariables in the system of equations. All other variables act asexogenous variables. The parameter of most interest for this paper isthe conduct parameter, θ, in the supply relation, which provides anumerical indicator of market power for the firm or group of firms.

One of our goals is to identify pricing power abuses among leadingpower marketers in the California PX market during 2000. There are avariety of ways to define a “leading firm.” Presumably, firms withlarge market shares are able to manipulate prices without the worrythat consumers will retaliate by patronizing competing firms. TheFederal Energy Regulatory Commission [FERC] (2003b) conducted aninvestigation of the pricing strategies employed by thirty-five firmswho reported electricity sales to the California PX during 2000–2001.Besides the three primary investor-owned utilities (PG&E, San DiegoGas & Electric and Southern California Edison), which were requiredto supply all of their generated electricity to thewholesalemarket, thefive firms with the highest sales volume to the California PX werepower marketers (see Table 1).

Duke Energy Trading & Marketing alone controlled more than 35%of the non-utility sales in the California PX, and the five powermarketers together account for 75% of the market. If all five werecolluding, theywould potentially have had enough power to influenceprices substantially. Even two or three of the firms working togetherwould control a large part of the residual market. On the other hand,

Table 1California Power Exchange sales, 2000.Source: Data from FERC Docket No. PA-02-2-000.

Power marketer Total sales Share ofnon-utility sales(MWh)

Duke Energy Trading & Marketing, LLC 13,926,827 38.25%Reliant Energy Services Inc. 5,836,292 15.49%Enron Power Marketing Inc. 3,041,558 8.55%Dynegy Power Marketing Inc. 2,976,598 8.08%Williams Energy Marketing & Trading 1,802,666 4.75%

572 T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

these five firms may have charged prices close to the market price,indicating a competitive or Cournot-type market. Thus, consideringonly individual market shares may fail to accurately describe firms'strategic behavior.

We perform nested hypothesis tests on the conduct parameter inorder to more accurately describe the market structure within whichthe five firms are operating and to test for any pricing abuses. If thefirms are acting competitively, then the conduct parameters shouldnot be significantly different from zero. Any level of market powershould cause us to reject the null hypothesis that the firms actcompetitively (i.e., calculated θ=0). If so, we next test for thepossibility that each firm is behaving as a Cournot firm (θ=1).

Obtaining accurate measures of the conduct parameter coeffi-cients (θ) requires an appropriate estimation technique. Theendogeneity of price and quantity in both the demand equationand the supply relation require a systems estimation technique sincethe error term for each individual equation is correlated with aregressor variable. Furthermore, the supply relation coefficients θ andβ1 require estimates of the elasticity parameters, γ0 and γ1, from thedemand equation in order to be identified. Coefficients that appear inboth equations must be restricted to take on identical valuesduring estimation so that the conduct parameter can be separatelyestimated.

The endogenous variables in each system include the California PXprice (Pt), purchase cost (Cit), the PX market demand (Qt

D), quantitysupplied (qit), and the interaction terms between the quantityvariables and cooling-degrees. Each equation in Eq. (2.7) excludes asufficient number of non-endogenous variables for the rank and ordercondition of identification to hold. Instrument variables used in theestimation procedure include: employment (Yt), refinery production(Rt), Shasta Dam hydropower releases (Ht), weekend dummy variable(Wt), level of cooling degrees (Dt

C) and heating degrees (DtH), hours of

transmission constraint (Tt), and the quantity sold by all other firms.In addition, Bonneville dam releases and the marginal fuel cost ofArizona power plants are used as instruments for the cost ofpurchasing electricity.

3. Empirical results

The wide availability of data has been one of the primary reasonsthat firm pricing behavior in the electricity industry has beenextensively analyzed. Both price and cost data for the regulatedinvestor-owned utilities are easily accessible, and those markets withcentralized power pool trading provide transparent data on equilib-rium market prices. For this paper, we use a relatively untappedsource of California PX transaction data. OnMarch 5, 2002, the FederalEnergy Regulatory Commission (FERC) directed all wholesale sellers(whether under the jurisdiction of FERC or not) to provide data abouthistorical electricity sales during 2000–2001 in the Western SystemsCoordinating Council (WSCC) region. These standardized reportsincluded detailed information about short-term wholesale electricitysales, including the price, quantity, buyer, seller, and location of eachtransaction during the specified period.

We use data from the short-term files submitted by Duke, Dynegy,Enron, Reliant and Williams to investigate those firms' pricingbehavior. Only sales transactions showing the purchaser as CaliforniaPower Exchange are considered. The two primary variables in theanalysis are the quantity (qit) and price (Pt) of the electricity sold tothe California Power Exchange by each of the five power marketers,measured in megawatt-hours (MWh) and dollars per megawatt-hour($/MWh). Our analysis workedwith daily observations for the sampleperiod January 1, 2000–December 31, 2000. We calculated a meanprice for each day for each firm using the transaction quantity as theweight. Observations for quantity were obtained for each firm bysumming all transaction quantities for each day.

A dummy variable was created to indicate whether the day was ona weekend (Wt=1) or not (Wt=0). A variable for the daily cost ofpurchasing electricity, Ct, was also obtained from the short-term FERCdata files. Calculating observations for daily cost consisted of twosteps. First, all transactions listing each of the five power marketers aspurchasers for the sample period were extracted from the files of allfirms that submitted data to FERC. Some of these electricity trans-actionswere delivered to locations bordering California or just outsideof the state, such as the Palo Verde hub in Arizona. However, since alltransactions occurred in the WSCC, it is reasonable to assume theelectricity was available to the purchaser for resale in the California PXsince transmission lines in the WSCC are all interconnected. We onlyused data from the short-term files because transactions for electricitydelivered for longer periods are unlikely to affect marginal supplydecisions. The next step in the calculation of each firm's cost variableconsisted of determining the weighted mean of all purchase prices fora given day. A very small number of transactions with pricesexceeding $10,000 per MWh were excluded from the data set. Thesetransactions had quantity values equal to zero MWh, so it isreasonable to assume that these transactions may have representedflat-fee charges instead of prices for delivered electricity. Compilingthe cost data for the Duke, Dynegy, Enron, Reliant, and Williamsrequired downloading the electronic files of nearly 200 companies(few of which followed the standardized FERC template) andsearching each file for transactions in which one of the five powermarketers was listed as a purchaser.

The University of California Energy Institute (UCEI) internetwebsite (2006) posts aggregate data for the California PowerExchange. We used this data to calculate daily values for totalquantity of electricity demanded in the PX market. Each observationfor the daily equilibrium quantity demanded, Qt

D, is calculated as thesum of the hourly quantities traded in the PX during the day.

The UCEI data set provided information for calculating the numberof hours of transmission constraint. Constrained transmission causesprices in neighboring control areas to diverge. To calculate variablevalues, we compared the equilibrium prices between the SP-15 andNP-15 control areas in California and assigned a value of one for eachhour in which the prices were not exactly equal. The transmissionconstraint variable, Tt, was figured by summing all hours each day inwhich the prices were unequal. Hydroelectric release data for ShastaDam was obtained from the California Department of Water Re-sources (2006), and release data for Bonneville Dam was obtainedfrom the U.S. Army Corps of Engineers (2003). Both sets ofhydropower data are measured in average cubic feet per second foreach day. The variable for daily marginal fuel cost of Arizonagenerators is represented by the maximum hourly system lambdain the Arizona Public Service Control Area (Federal Energy RegulatoryCommission [FERC], 2003a). The term “lambda” refers to thegeneration cost of the most expensive unit dispatched in atransmission control area during any given hour. Monthly data forthe heat rate variable, Gt, was compiled for all firms except Enronusing the EIA-906 files from the Energy Information Administration.

Finally, the National Climatic Data Center provided source data forcalculation of the cooling degree (Dt

C) and heating degree (DtH)

Table 2Model estimation results.

Residual demand Duke Dynegy Enron Reliant Williams

Independent variable=price

α0 Constant −5583 −3771 −4943 −4806 −3905(−14.71) (−9.68) (−15.00) (−11.30) (−8.93)

α1 Heating degrees 5.13 4.01 3.54 2.53 3.66(5.12) (3.94) (4.10) (2.23) (3.35)

α2 Cooling degrees −130.41 −36.34 −151.48 −164.55 −153.93(−1.92) (−0.51) (−3.02) (−2.23) (−1.84)

α3 Weekend transaction 6.87 −33.76 −0.72 −7.55 −6.50(0.46) (−2.50) (−0.11) (−0.50) (−0.43)

α4 Employment 0.00035 0.00025 0.00032 0.00031 0.00026(15.66) (12.64) (16.09) (12.32) (10.44)

α5 Refinery production 0.01154 0.00818 0.01043 0.01560 0.01457(2.56) (1.72) (2.08) (3.45) (3.20)

α6 Hydropower releases −0.00080 −0.00044 −0.00063 −0.00039 −0.00085(−3.41) (−1.81) (−3.16) (−1.29) (−2.90)

γ0 Quantity demanded −0.00017 −0.00041 −0.00036 −0.00050 −0.00061(−0.86) (−1.46) (−1.93) (−1.75) (−2.84)

γ1 Quantity demanded×cooling degrees 0.000227 0.000097 0.000282 0.000306 0.000285(1.94) (0.78) (3.16) (2.34) (1.95)

β0 Constant 96.66 −33.27 −8.85 −15.44 62.45(1.75) (−2.17) (−0.85) (−1.49) (1.56)

β1 Quantity supplied 1.4E−4 0.00516 8.2E−4 0.00116 4.8E−4(0.47) (1.38) (0.94) (3.35) (0.32)

β2 Cost of electricity 0.7887 0.9117 0.8644 0.9212 0.6352(8.53) (15.62) (14.21) (18.73) (9.42)

β3 Hours constrained 0.3642 −0.9116 0.3799 −0.5623 −0.1019(1.03) (−2.20) (1.15) (−1.98) (−0.25)

β4 Generation heat rate −10.70 2.76 N/A 0.90 −4.49(−2.13) (2.16) – (0.82) (−1.55)

θ Conduct parameter 0.5446 −7.5932 0.0716 0.4780 −0.2865Standard error 0.3014 9.9259 0.3832 0.2656 0.4257Minimum objective function 23.1242 50.0871 16.7403 8.3872 17.3223Number of observations 308 333 365 328 323

Note: t-statistics shown in parentheses except for the standard error listed on the conduct parameter coefficient.

Table 3Residual demand elasticities.

Power marketer Elasticity

Duke Energy Trading & Marketing, LLC −10.40Dynegy Power Marketing Inc. −31.12Enron Power Marketing Inc. −29.39Reliant Energy Services Inc. −11.50Williams Energy Marketing & Trading −29.11

573T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

variables. As the previous section indicated, these variables aremeasured as deviations between each day's average temperature anda baseline of 65°. To compute statewide temperatures for each day wecomputed a population-weighted average of the minimum andmaximum daily temperatures for representative weather stationsthroughout California. These stations correspond to the 16 climatezones used by the California Energy Commission (2006) for analyzingenergy use: China Lake/Barstow, El Centro, Eureka, Fresno, LosAngeles, Mount Shasta, Oakland, Pasadena, Red Bluff, Riverside,Sacramento, San Diego, San Jose, Santa Ana, Santa Maria, and SantaRosa. These sixteen stations capture the variety of climates inCalifornia and include most of the major population centers.

Our analysis of the estimation results begins by applying thetraditional conduct parameter market power model to the data foreach of the five power marketers. We estimate the demand equationand supply relation shown in themodel (2.7) with a conduct parameterthat is unaffected by future economic expectations. Table 2 shows eachfirm's parameter estimates for the system of supply and demandequations estimated using the GMM methodology. Price is thedependent variable for each equation; the independent variables andtheir associated parameters are listed in the table rows.

Themajority of the estimated coefficients in the demand equationsare relatively significant with t-statistics greater than 1.96 in absolutevalue, the value associated with a 5% level of significance in a two-tailed hypothesis test (H0: α=0). One exception is the insignificanceof the weekend dummy variable, although the values for all of the fivefirms are negative as we would expect. The number of weekendobservations in each sample was relatively low, as most powermarketers focus on selling power during the more profitable weekdays. Employment, acting as a proxy for commercial electricitydemand, is positive and highly significant as we would expect.

Similarly, the effect of refinery production, the proxy for industrialelectricity demand, on California PX prices is also positive, however,the coefficient significance for some of the firms is questionable. Theestimated coefficient for hydropower releases from Shasta Dam isnegative as wewould expect. This result reflects the lower demand forwholesale electricity of municipal utilities on those days whenhydroelectric generation is abundant.

Temperature deviations from a “comfortable” average of 65°, asmeasured by heating and cooling degrees, have the expected positiveimpact on wholesale electricity demand. This effect is directly evidentin the signs of the coefficients for heating degrees (α1). The negativecoefficient values for the level of cooling degrees (α2) appearcounterintuitive at first. However, the full marginal effect of anincrease in cooling degrees is represented by both the level of coolingdegrees and its interaction with quantity demanded (∂Pt/∂Dt

C=α2+γ1Qt

D). In Duke's case, this marginal effect is positive wheneverQtDN557, 589, slightly more than the average level of demand in the

California PX. Days with higher temperatures (and above-averagedemand) require more air conditioning and thus higher electricityconsumption and higher prices.

574 T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

The slope coefficients for quantity demanded (γ0) are negative foreach of the powermarketers as wewould expect, according to the lawof demand. Each of the coefficients is quite close to zero, implying thatthe demand is very inelastic. The linear functional form that we haveassumed for the demand equation allows elasticity to vary dependingon quantity, price, and cooling degrees. After solving the inversedemand Eq. (2.4) for Qt

D, we can calculate an average price elasticity ofdemand at comfortable temperatures (i.e., Dt

C=0) using the meanselling price and quantity for each firm's sample: ε = 1 = γ0ð ÞP =Q

Dt

� �. Table 3 shows the calculated average residual demand

elasticities for each of the five firms given the estimated parameters inthe model. All of the values are elastic (|ε|N1) as we would expect forfirm-level residual demand. Dynegy, Enron and Williams have themost elastic demand, while Duke and Reliant have significant lowerelasticity values. Firms with less elastic demand have the opportunityto raise price high above marginal costs and reap excessive profits.The estimated values for the coefficient γ1, the effect of coolingdegrees on demand elasticity, has the expected sign and is significantin most cases. During hot days, as households utilize more airconditioning, the (negative) elasticity of market demand becomessmaller (i.e., less elastic), thus creating a formula for possible pricemanipulation during the peak summer months. This effect is mostpronounced for Dynegy. The γ1 coefficient also allows us to identifythe level of market power as represented by the conduct parameter θin the supply relation.

The supply relation coefficients are estimated with less precisionthan the coefficients in the demand equations. The most significantcoefficient is associated with the cost of purchased electricity, with t-statistics generally exceeding 10, as we would expect since this is theprimary component of total cost for a power marketer. For everydollar increase in the short-run cost of electricity, the selling pricerises between 0.71 $/MWh and 0.96 $/MWh, depending on the firm.The coefficient on sales is generally insignificant, indicating that thequantity supplied has no impact on marginal cost. This result isconsistent with assumptions made in other electricity market powerstudies. The variable representing transmission constraints positivelyimpacts the selling price for only two of the five firms. Our measure ofthe number of hours each day in which the SP-15 prices differ fromNP-15 prices appears to be a poor proxy for transmission constraintssince the coefficients often assume the incorrect sign and theestimates are statistically insignificant.

The differences in elasticity and temperature impacts imply thatdifferences in pricing behavior and observed market power, asmeasured by the conduct parameter, are likely to exist between thefive power marketers. The estimated values for θ in the supplyrelations are all relatively low. A test for H0: θ=0 indicates theparameters are statistically significant only for Duke and Reliant at the10% level of significance. The application of similar hypothesis tests forDynegy, Enron, and Williams result in parameters that are notsignificantly different from zero, which is more consistent withcompetitive pricing behavior.3 Likewise, we are unable to concludethat they were exhibiting Cournot or cartel behavior. These surprisingresults, which conflict with FERC's investigative findings, may reflectthat these firms were using more sophisticated dynamic pricingstrategies that this model was unable to detect.

For Duke and Reliant, we can test the hypothesis that they areacting as Cournot firms by using a two-tailed test for H0: θ=1. Theassociated t-statistics for this null hypothesis are −1.6 for Duke and−2.0 for Reliant. We fail to reject the null hypothesis of Cournotbehavior for Duke at the 10% level of significance, but we can reject

3 Although not significant, the negative estimated conduct parameters for Dynegyand Williams seem puzzling. As the reviewer pointed out, a true negative coefficientwould indicate a firm was setting its price less than marginal cost.

the null hypothesis for Reliant at the 5% level of significance. Themodel results indicate that both of these firms were exercisingmeasurable market power, with Duke showing stronger evidence ofCournot pricing behavior.

4. Conclusion

When we applied our econometric model to the pricing data foreach of the five firms, we determined that Duke Energy, and possiblyReliant, appeared to have behaved as Cournot firms in pricing theelectricity they sold to the California PX during 2000. This statisticalevidence of market power for Reliant coincides with the FERC findingsduring their investigation of potential manipulation of energy prices.The results for Dynegy, Enron, and Williams showed no statisticalevidence of pricing above competitive levels.

The latter results showing competitive pricing behavior amongthree of thefirms are inconsistentwith the results uncoveredbyWolak(2003), who found the five largest California suppliers were exhibitingsubstantial market power during 2000, and Puller (2007), who foundthe five firms acting as a Cournot oligopoly. One possible cause for thisinconsistency may be the static model employed in this paper. TheBresnahan-style model used here portrays pricing behavior by meansof an estimated conduct parameter, which is based on the theory ofconjectural variations. However, this theory is a dynamic concept thatimplicitly assumes that firms form pricing strategies based on thebehavior of other firms over time. Some authors such as Corts (1999)have pointed out that the traditional static conduct parameter modelmay underestimate market power. Addressing this issue would be aproductive avenue for future research.

The unique dataset compiled during FERC's investigation of pricemanipulation during 2000–01 has the potential to provide someinteresting insights into the pricing behavior of power marketers.However, researchers who use this dataset should exercise caution.Our result that Enron was pricing competitively in the California PXmarket conflicts with the substantial legal evidence that the firm wasclearly utilizing anti-competitive (and illegal) pricing strategies. Partof the cause for this inconsistent result may be the limitations of thestatic modeling approach discussed above. However, there also maybe issues with the reliability of Enron's data. While most other firmsquickly provided the standardized data that FERC requested in itsMarch 2002 letter, Enron's initial response was to provide thousandsof hardcopy pages detailing its numerous transactions. This responsewas deemed insufficient. The company claimed that its database wasnot capable of creating files in the standard Excel format. EventuallyEnron did submit a comma-delimited data file, yet the company'sresistance raises the question of whether the submitted dataset trulyreflects all of its transactions.

Lastly the data do not distinguish the times at which transactionstake place. Presumably the market power exists during peak periodsof the day. Lumping all transactions together maywell be masking themonopoly behavior suggested by other studies and suggests theimportance of data that is granular enough to reveal monopoly pricingthat is intermittent.

References

Borenstein, Severin, James Bushnell, Christopher R. Knittel, and Catherine Wolfram.2001. Trading inefficiencies in California's electricity market. POWER WorkingPaper PWP-086.

Bresnahan, Timothy F., 1989. Empirical studies of industries with market power. In:Schmalensee, Richard, Willig, Robert (Eds.), Handbook of Industrial Organization.North-Holland, New York, pp. 1010–1057.

California Department of Water Resources, 2006. Data Query Tools. http://cdec.water.ca.gov/queryTools.html.

California Employment Development Department, 2006. LaborMarketInfo Data Library.http://www.labormarketinfo.edd.ca.gov.

California Energy Commission, 2001. 2001 Weekly Fuels Watch Report. http://www.energy.ca.gov/database/fore/ Last updated July 9, 2003.

575T. Hodge, C.A. Dahl / Energy Economics 34 (2012) 568–575

California Energy Commission, 2006. California Building Climate ZoneMapLast updatedMay 16, 2006 http://www.energy.ca.gov/maps/climate_zone_map.html.

Corts, Kenneth S., 1999. Conduct parameters and the measurement of market power.Journal of Econometrics 88 (2), 227–250.

Energy Information Administration, 2008. Residential Energy Consumption Survey2005. http://www.eia.doe.gov/emeu/recs/contents.html. Accessed January 30,2009.

Federal Energy Regulatory Commission [FERC], 2003a. Annual Electric Control andPlanning Area Report 2001. http://www.ferc.gov/docs-filing/eforms/form-714/data.asp. Last updated January 23, 2003.

Federal Energy Regulatory Commission [FERC], 2003b. Final Report on PriceManipulation in Western Markets. Docket No. PA02-2-000. Federal EnergyRegulatory Commission, Washington, DC.

Iwata, Gyoichi, 1974. Measurement of conjectural variations in oligopoly. Econometrica42 (5), 947–966.

Joskow, Paul L., Kahn, Edward, 2002.Aquantitative analysis ofpricingbehavior inCalifornia'swholesale electricity market during summer 2002. Energy Journal 23 (4), 1–35.

Kim, Dae-Wook, Knittel, Christopher R., 2004. Biases in Static Oligopoly Models?Evidence from the California Electricity Market, CSEM Working Paper. 131.

Puller, Steven L., 2007. Pricing and firm conduct in California's deregulated electricitymarket. Review of Economics and Statistics 89 (1), 75–87.

U.S. ArmyCorps of Engineers, 2003. DataQuery: Bonneville Flows. http://www.nwd-wc.usace.army.mil/nwp/index.html. Last updated December 3, 2003.

University of California Energy Institute, 2006. California Electricity Market Data. http://www.ucei.berkeley.edu/ Last updated March 14, 2006.

Wolak, Frank A., 2003. Measuring unilateral market power in wholesale electricitymarkets: the California Market, 1998–2000. American Economic Review 93 (2),425–430.

Wolfram, Catherine D., 1999. Measuring duopoly power in the British electricity spotmarket. American Economic Review 89 (4), 805–826.