PhD Thesis Presentation

23
Doctoral Examination f h i h i Doctoral Examination Assessment of Short-term Strategic Behavior in Electricity Markets Introduction Introduction Analysis Modeling Results Conclusions Ing. Pablo Frezzi San Juan, 25/04/2008

description

PhD Thesis Presentation hold on 4.25.2008 in Argentina

Transcript of PhD Thesis Presentation

Page 1: PhD Thesis Presentation

Doctoral Examination

f h i h i

Doctoral Examination

Assessment of Short-term Strategic Behavior in Electricity Markets

IntroductionIntroductionAnalysisModelingResultsConclusions

Ing. Pablo Frezzi

San Juan, 25/04/2008

Page 2: PhD Thesis Presentation

1

Motivation (1)

Introduction

Motivation (1)

Characteristics of the present electricity marketsImpossibility to store economically large amounts of electricityLow price elasticity of demandRepeated interactionRepeated interaction Significant economies of scaleTransmission constraintsM k t t ti f i ffi i t di tit dMarket concentration as a consequence of inefficient divestiture and consolidations

Electricity markets are not perfectly competitive

In contrast to perfectly competitive markets, market participants do not play a passive role as „price takers“

Price and market dynamic depend on market participants‘ strategies to maximize profits

Strategic behavior: individual or group action to increase profits by means of overt or tacit agreements influencing the market variables

Market power: individual profit-maximizing actiona ke po e d dua p o ax g ac oCollusion: cooperative profit-maximizing action

Page 3: PhD Thesis Presentation

2

Motivation (2)

Introduction

Motivation (2)Consequences of strategic behavior:

Wealth transfer from customers to producerspDeadweigh loss and and reduction of social welfareSupply shortages pursue

l lPrice volatilityDistortion of price signals which may lead to inefficient investments

Ph i l ithh ldi E i ithh ldi

Price

Physical withholding Economic withholdingDemand OfferWealth transfer

Price

Demand OfferWealth transfer

Costs

PVM

PSM

Costs

PVM

PSM

Deadweigh lossWithheld capacity

PVMDeadweigh loss

PVM

Withheld capacity

Quantity0 QVMQSM Quantity0 QVMQSM

Strategic behavior may affect the benefits pursued by liberalizing processes

Need of models to reproduce actual strategic behavior in electricity markets

Page 4: PhD Thesis Presentation

3

Research Aim

Introduction

Research Aim

Development of a simulation model of electricity markets to reproduce and

f

p y passess the strategic behavior of market participants

Specific aims:

Indentification and proof of exercise of strategic behavior in electricity marketsQuantification of the influence of strategic behavior on the electricity priceAnalysis of the influence of individual behavior on the short-term dynamic of electricity markets, specially with signs of concentrationy , p y gIdentification of the most relevant causes of strategic behaviorAnalysis of the influence of transmission constraints on the individual behavior of the market participants and on the exercise of strategic behaviorthe market participants and on the exercise of strategic behavior

Application fieldppCompetition authorities Regulators

Page 5: PhD Thesis Presentation

4

Strategic Behavior

Analysis

Strategic Behavior

Market powerpMaximization of benefits by means of exploitation of market dominanceStatic contextU il t l d i d d t b h iUnilateral and independent behaviorOwn theory and well understoodComprehensively researchedo p e e e y e e eWell defined indices to quantify market power potential

Tacit collusionTacit collusionMaximization of benefits by means of tacit coordination of strategiesDynamic contextMultilateral and interdependent behaviorNo own theoryNot enough researchedNot enough researchedHardly any successful prosecution of tacit collusion due to lack of analysis models

Tacit collusion has not been comprehensively researched in electricity markets p y yyetNeed of models to detect and assess tacit collusion in electricity markets

Page 6: PhD Thesis Presentation

5

Tacit Collusion (1)

Analysis

Tacit Collusion (1)

Necessary conditionsyMarket concentration

• Easy to coordinate and reach a tacit agreement• T an mi ion con t aint inc ea e ma ket concent ation• Transmission constraints increase market concentration

Repeated interaction• Coordination of strategies by means of learning processes• Daily repetition intensify the learning process

Barriers to entry and exit• sunk costs“ nonreversible investments• „sunk costs nonreversible investments• No contestable market

Coordination capacity• Coordination on specific collusive equilibria

Punishment of deviation from collusive agreements• Discouragement of deviations from collusive agreements• Discouragement of deviations from collusive agreements

Electricity markets fulfill the necessary conditions for a tacitly collusive agreement to emerge and remain stable over time

Page 7: PhD Thesis Presentation

6

Tacit Collusion (2)

Analysis

Tacit Collusion (2)

Facilitating factorsSymmetrical firms

• Easy to achieve collusive agreement among firms with similar production costsHomogeneous productHomogeneous product

• Product variety reduces competition and thus increases concentration and coordination

TransparencyTransparency• Increases coordination and detection of deviations from collusive equilibria

Stable and predictable demand• Revisionary processes with decreasing prices• Revisionary processes with decreasing prices• Low price elasticity of demand

Fragmented demand-sidell d f d• Small and frequent orders

• Less incentives to defect• Short time-lags encourage coordination among market participants

Uniform-price auction• Difficult detection of collusion

Present electricity markets fulfill necessary conditions and facilitating factors and are thus prone to tacit collusion

Page 8: PhD Thesis Presentation

7

Tacit Collusion (3)

Analysis

Tacit Collusion (3)

Learning abilities of agents Learningprocess Collusion

Dynamic of electricity markets• Repeated interaction• Short-time lags

p

• Short-time lags• Adaptable behavior DeviationPunishment

Bids ResultsAgents

MarketMarket

Reward

Market participants learn the market dynamic and adapt their behavior

Page 9: PhD Thesis Presentation

8

Tacit Collusion in Liberalized Electricity Markets

Analysis

Tacit Collusion in Liberalized Electricity Markets

England & WalesgTacit collusion between the two biggest generation companies in the 1990‘s90% of the time, the price was set by the two biggest generation companies

CaliforniaCalifornian energy crisis between 1998 and 2001

ld d fEconomic withholding exercised 60% of the time

GermanyHigh level of market concentrationSome research reports prices much higher than cost estimators as a consequence of tacit collusion

European Transmission System Operators (ETSO)Advice about the importance of market monitoring in Europe in order to ensureAdvice about the importance of market monitoring in Europe in order to ensure adequate market conditions

Tacit collusion has become a worldwide problem

Page 10: PhD Thesis Presentation

9

Description of the Model

Modeling

Description of the Model

Classic oligopolistic modelsg pIdentification of equilibria, i.e. Nash equilibriaQuantity and price competitionSt ti d i l i d d lStatic and single-period modelsFor market power assessment suitable

d i h i f bli i f iRepeated games with imperfect public informationDynamic coordination among market participantsImperfect public information:Imperfect public information:

• Price and quantityNon-public information:

• Cost structure and past actionsPresent actions depend on public and non-public informationStrategy function: dynamic behavior of market participantsStrategy function: dynamic behavior of market participants

Repeated games with imperfect public information are adequate toRepeated games with imperfect public information are adequate to reproduce tacit collusion

Page 11: PhD Thesis Presentation

10

Simulation Model

Modeling

Simulation Model

Hourly assessment of tacit collusion on the generarion-side

Transmission constraintsRegulatory frameworkMean nodal demand

Availability of generation unitsFuel pricesThermal efficienciesG i f li

Generation Agent Market Agent

Generation portfolios

Off

Decision-making:

M i i ti f b fit

Generation scenarios Demand scenarios

OffersMaximization of benefitspolicy function

Assessment of rewards:

Minimization of generation costs

Market settlementResults

iterativerepetition

Assessment of rewards:reward function Market settlement

Updating of informationaction-value function Database

Time limitsSimulation horizon: 1 month – 1 yearPeriodicity: 1 hour

Page 12: PhD Thesis Presentation

11

Decision making of Generation Agents

Modeling

Decision-making of Generation Agents

Portfolios with thermal plantsDifferent thermal generation technologiesFuel prices exogenous variablesStartup costsStartup costs

Objective Function max [ Earnings from energy sales – Variable costs ]Assessment of rewards

Short-time uncertaintiesA il bilit f ti itAvailability of generating unitsStochastic fluctuations of demandDecision of other generation agents [€/MWh]

120Supply functionMarginal cost curveg g

StrategyPrice competition (Bertrand competition)

f l f

60

80Marginal cost curve

Percent increase of the supply functionPrice increase = 0 „price taker“

0

20

40

0 100 200 4000

Generation capacity[MW]

Page 13: PhD Thesis Presentation

12

Strategy Actualization

Modeling

Strategy Actualization

Game theory with artificial intelligence (Reinforcement-Learning)Effi i t i l f ti l t t i t i i fitEfficient appraisal of optimal strategies to maximize profitsConsideration of the characteristics of social behavior:

• Exploitation of past actionsp p• Exploration of new actions• Recency

Strategy act ali ation Softmax algorithmStrategy actualization Softmax algorithm

Agent

π(o)=σ

Strategy

f(S)

Probabilityfunction

I f ti Policy function

Soptimal

Strategy

A tiStrategies

P li f ti

Reward

Information Policy function Action π: Policy functiono: Vector of informationσ: Strategy mix

Environment

The policy function and strategy actualization allow to reproduce the actualThe policy function and strategy actualization allow to reproduce the actual behavior of generation agents

Page 14: PhD Thesis Presentation

13

Short term Uncertainties

Modeling

Short-term Uncertainties

Availability of generating unitsy g gTwo-state Markov model

Unit Unitλ Failure

Stochastic determination of generation scenarios

Unitoperable

Unitfailedμ Reparation

S o e e o o ge e o e o

Generation

45

43[GW]

Generationcapacity

404142

Stochastic demand scenarios

1st w 2nd w 3rd w 4th w0

40

[GW]

Stochastic Gauss-Markov modelStatistical information from system

10

20

Saturday SundayWorking d

JanuaryJuly

Demand

system0

1 7 13 19 1 7 13 19 1 7 [h] 19hour

Saturday SundaydayJuly

Page 15: PhD Thesis Presentation

14

Market Agent

Modeling

Market Agent

Spot marketp

Opening of market and reception of energy bids from generation agentsHourly bidsy

Demand scenariosstochastic Gauss-Markov modelstochastic Gauss Markov model

Clearance of the market and calculation of the hourly price through minimization of generation costs considering generation and transmission constraintsof generation costs considering generation and transmission constraints

Lagrange Relaxation:

i i [ ∑(G i )min L = min [ ∑(Generation costs) + β [ ∑(Demand) + ∑(Losses) - ∑(Generation) ] +∑ ŋ (Transmission constraints) + ∑ ε (Generation constraints) ]

Price calculationPrice = β [ node factor ] - ∑ ŋ [ PTDF ]

Losses Transmission constraints

Page 16: PhD Thesis Presentation

15

Model System

Results

Model System

6 thermal generation technologies100 generation plants with usual capacities in actual systemsTotal installed capacity 44,4 GWEmissions certificate 12 €/EUAEmissions certificate 12 €/EUA3 market concentration levels

100 GA: unconcentrated10 GA: moderately concentrated5 GA: highly concentrated Generation

costs

Generation marginal cost curve120

[€/MWh]

80Zusammensetzung des KraftwerksparksGeneration Technology Mix

40

60Hard coal

CCGT (gas/oil)

Lignite

0

20Steam turbine (gas/oil)

CCGT (gas/oil)

NuclearGas turbine (Gas/oil)

0 10 20 [GW] 500

Aggregate generation capacity30

( / )

Page 17: PhD Thesis Presentation

16

Simulated Hourly Prices (1)

Results

Simulated Hourly Prices (1)

a) Constant available generation capacity and deterministic demandg p y

120 120January July

80

[€/MWh]

80

[€/MWh]

40

60

Pric

e

40

60

Pric

e

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

PCM 100 GA 10 GA 5 GA

Simulated prices considering coordination abilities are higher than generation marginal costs

100 GA 10 GA 5 GA

The higher the market concentration is, the higher prices are

Page 18: PhD Thesis Presentation

17

Simulated Hourly Prices (2)

Results

Simulated Hourly Prices (2)

b) Stochastic availability of the generating units and deterministic demandy g g

120 120January July

80

[€/MWh]

80

[€/MWh]

40

60

Pric

e

40

60

Pric

e

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

PCM 100 GA 10 GA 5 GA

Simulated prices considering coordination abilities are higher than generation marginal costsg

The higher the market concentration is, the higher prices are

Page 19: PhD Thesis Presentation

18

Simulated Hourly Prices (3)

Results

Simulated Hourly Prices (3)

c) Stochastic availability of the generating units and demand fluctuationsy g g

120 120January July

80

[€/MWh]

80

[€/MWh]

40

60

Pric

e

40

60

Pric

e

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

0

20

1 7 13 19 1 7 13 19 1 7 [h] 19

Working day Saturday Sunday

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

1 7 13 19 1 7 13 19 1 7 [h] 19

hour

PCM 100 GA 10 GA 5 GA

Price differences are reduced due to information uncertainties

PCM 100 GA 10 GA 5 GA

Information uncertainties restrain influence of market concentration

Page 20: PhD Thesis Presentation

19

Comparative Analysis of Results (1)

Results

Comparative Analysis of Results (1)

Monthly revenues and producer surpluses

[Mio. €]

1400

1800

[Mio. €]

1400

1800January July

1000

1200

1400

1000

1200

1400

400

600

800

400

600

800

0

200

a b c a b c a b c a b c0

200

a b c a b c a b c a b c

PCM

Producer surpluses

100 GA 10 GA 5 GA PCM 100 GA 10 GA 5 GA

a) Constant available generation capacity and deterministic demand

h l b l f h d d d d Generation costsb) Stochastic availability of the generating units and deterministic demand

c) Stochastic availability of the generating units and demand fluctuations

Market concentration and information uncertainties play a key role when tacit collusion occurs

Page 21: PhD Thesis Presentation

20

Comparative Analysis of Results (2)

Results

Comparative Analysis of Results (2)

Assessment of collusion by means of the Lerner IndexL I d (P i M i l ti t)/P iLerner Index=(Price-Marginal generation cost)/Price

0,5January July

0,5Lerner- Lerner-

0,4 0,4

LernerIndex

LernerIndex

0,2

0,3

0,2

0,3

0,1

0,2

0,1

0,2

0

S i

100 GA 10 GA 5 GA

S i b S i

PCM0

100 GA 10 GA 5 GAPCM

Tacit collusion even with low levels of concentration

Scenario a Scenario b Scenario c

Information uncertainties reduce extraordinary surpluses

Page 22: PhD Thesis Presentation

21

Conclusions

Conclusions

ConclusionsResearch aim:Development of a simulation model of electricity markets to reproduce and assess theDevelopment of a simulation model of electricity markets to reproduce and assess the strategic behavior of market participants

AnalysisCharacteristics and consequences of strategic behavior in electricity markets Necessary conditions and facilitating factors of tacit collusionElectricity markets are prone to suffer tacit collusionElectricity markets are prone to suffer tacit collusion

ModelingMixed Model:

• Game theory: repetitive game with imperfect public information• Artificial intelligence: Reinforcement Learning

Results:Results:Market concentration and information uncertainties play a key role in cases of tacit collusionTacit collusion even with low concetration levelsTacit collusion even with low concetration levels

Main contributionsComprenhensive analysis of tacit collusion in electricity markets and its dynamicIdentification of main influencing factors and their assessment on the marketThe simulation model is suitable to reproduce short-term strategic behavior

Page 23: PhD Thesis Presentation

Doctoral ExaminationDoctoral Examination

Assessment of Short-term Strategic BehaviorAssessment of Short-term Strategic Behavior in Electricity Markets

Ing. Pablo Frezzi

San Juan, 25/04/2008