The Efficiency of Uniform- Price Electricity Auctions: Evidence from Bidding Behavior in ERCOT

Transcription

1 The Effcency of Unform- Prce Electrcty Auctons: Evdence from Bddng Behavor n ERCOT Steve Puller, Texas A&M (research jont wth Al Hortacsu, Unversty of Chcago) Tele-Semnar, March 4,

2 Outlne of Presentaton What makes an electrcty market effcent? Do balancng auctons provde ncentves for generators to bd n a manner that promotes effcency? Evdence from ERCOT: How much neffcency? What s the cause of neffcent bddng behavor? Implcatons for two types of stakeholders: Market montors Generators 2

3 What Is an Effcent Wholesale Market? Prce s the margnal cost (MC), or the cost of last MWh requred to serve load Short run: If most effcent unts not scheduled day-ahead, sgnals optmal reshufflng of generaton to get least cost dspatch. Long run: Sgnals value/locaton of new generaton 3

4 Problems wth Generators Submttng Bds Margnal Cost Typcal concern: large generator bddng (INC) substantally above MC Prces hgh May wthhold low cost unts to rase the prce Small generator bddng (INC) substantally above MC Prces hgh Ineffcent dspatch Any generator bddng (DEC) substantally below MC Prces too low! Ineffcent dspatch 4

5 Smple Example Unts have MC varyng from $30 to $120 Gven load, least cost dspatch Prce=$80 Generator A: Owns $100 unt Sgned blateral deal so schedules unt day-ahead Generator B: Owns $60 unt Only scheduled half of capacty day-ahead Effcency: A wll DEC, B wll INC Bad outcome: A submts a (low) $40 DEC bd B submts a (hgh) $120 INC bd 5

6 Market Power n a Balancng Market Suppose no further contract oblgatons upon enterng balancng market INCremental demand perods Bd above MC to rase revenue on nframargnal sales Just monopolst on resdual demand DECremental demand perods Bd below MC to reduce output Make yourself short but drve down the prce of buyng your short poston (monopsony) 6

7 Texas Electrcty Market ERCOT balancng market opened August 2001 Incumbents Implct contracts to serve non-swtchng customers at regulated prce Varous merchant generators 7

8 Electrcty Market Mechancs Forward contractng Generators contract w/ buyers beforehand for a delvery quantty and prce Day before producton: fxed quanttes of supply and demand are scheduled w/ grd operator (Generators may be net short or long on ther contract quantty) Spot (balancng) market Centralzed market to balance realzed demand wth scheduled supply Generators submt supply functons to ncrease or decrease producton from day-ahead schedule 8

9 Balancng Energy Market Approx 2-5% of energy traded ( up and down ) up bddng prce to receve to produce more down bddng prce to pay to produce less Unform-prce aucton usng hourly portfolo bds that clear every 15-mnute nterval Bds: monotonc step functons wth up to 40 elbow ponts (20 up and 20 down) Market separated nto zones f transmsson lnes congested we focus on uncongested hours 9

10 Who are the Players? 10

11 Prce Emprcal Strategy MR 1 MR = Revenue from sellng one more MWh MC = Cost of producng one more MWh RD = Max market wllng to pay for one more MWh purchased B MC (q) A RD 1 QC Quantty 11

12 Prce Emprcal Strategy MR = Revenue from sellng one more MWh MC = Cost of producng one more MWh RD = Max market wllng to pay for one more MWh purchased MR 2 MR 1 B C MC (q) A RD 1 QC RD 2 Quantty 12

13 Prce MR 2 Emprcal Strategy MR 1 S * (p) MR = Revenue from sellng one more MWh MC = Cost of producng one more MWh RD = Max market wllng to pay for one more MWh purchased S = Proft-maxmzng bd curve B C MC (q) A RD 1 RD 2 Quantty QC 13

14 Prce MR 2 Emprcal Strategy MR 1 B S * (p) S actual (p) MR = Revenue from sellng one more MWh MC = Cost of producng one more MWh RD = Amount market wllng to pay for one more MWh purchased C MC (q) A RD 1 RD 2 Quantty QC 14

15 50 Relant on June 4, :00-6:15pm Relant s MC Relant s Resdual Demand Prce ($/MWh) Ex Post Optmal Bd Schedule Relant s Bd Schedule Balancng Market Quantty (MW) 15

16 Prevew of Results Largest frm bds close to benchmarks for optmal bddng Small frms sgnfcantly devate, but there s some evdence of mprovement over tme Effcency losses from unsophstcated bddng at least as large as losses from market power 16

17 Unform-Prce Aucton Model of ERCOT Setup Statc game, N frms, costs of generaton C t (q) Contract quantty (QC t ) and prce (PC t ) ~ Total demand Dt = D + εt Generators bd supply functons S t (p) Market-clearng prce (p c ) gven by (removng t subscrpt from now on): N = 1 S ( p c ) = ~ D 17

18 Ex-post proft: Model (cont d) c c c c π = S ( p ) p C( S( p )) ( p PC) QC Informaton Structure C (q) common knowledge Prvate nformaton: QC PC but does not affect maxmzaton problem D ~ s unknown mportant sources of uncertanty from perspectve of bdder Rval contract postons (QC - ) and total demand (ε) 18

19 Characterzaton of Bayesan Nash Equlbrum* Strateges : S( p, QC ) ~ QC, D have jont dstrbuton F( QC ~, D QC ) (possbly correlated) Followng Wlson's (1979) share aucton model, defne the probablty dstrbuton of market - clearng prce, condtonal on supply functon Sˆ ( p) and QC, gven that other frms follow strategy profle S ( p, QC ) : H ( p, Sˆ ( p)) = = Pr{ QC j ε S 1{ j Pr{ p ( p, QC j S j j c ) + ( p, QC p ~ Sˆ ( p) D j ) + QC, Sˆ ( p)} Sˆ ( p) D QC, Sˆ ( p)} + ε} df( QC, ε QC ) * NE = A bddng outcome where bdders can t mprove proftablty by changng ther bd. 19

20 Equlbrum (cont d) Bdders choose supply functons to maxmze expected profts max S$ ( p ) p p ps$ ( p ) C ( S$ ( p )) ( p PC ) QC dh ( p, S$ ( p ); QC ) If H(.) s dfferentable, necessary condton for pontwse * optmalty of S ( p) : * S(, ( ); ) * p(, ( ); ) * * p C ( S( p)) = ( S( p) QC) H p S p QC H p S p QC 20

21 Equlbrum (cont d) CLAIM: If we restrct the class of supply functons: S ( p) = α ( p) + β( QC ) then (ex ante) equlbrum bds are ex post best responses: where * p C ( S ( p)) = RD( p) QC RD ( p) RD ( p) = D( p) S ( p) j j 21

22 Computng Ex Post Optmal Bds Ex post best response s Bayesan Nash Eqbm Uncertanty shfts resdual demand parallel n & out (observed realzaton of uncertanty provdes data on RD '(p) for all other possble realzatons) Can trace out ex post optmal/equlbrum bd pont for every realzaton of uncertanty (dstrbuton of uncertanty doesn t matter) * p MC( S ( p)) = 123 Unknown 678 Unknown * S ( p) QC RD ( p) ("nverse elastcty rule") 22

23 Sample Bddng Interface 23

24 Data (Sept 2001 thru Jan 2003) 6:00-6:15pm each day Bds Hourly frm-level bds Demand n balancng market assumed perfectly nelastc Margnal Costs for each operatng fossl fuel unt Fuel effcency average heat rates Fuel costs daly natural gas spot prces & monthly average coal spot prces Varable O&M SO2 permt costs Each unt s daly capacty & day-ahead schedule 24

25 Measurng Margnal Cost n Balancng Market Use coal and gas-fred generatng unts that are on and the daly capacty declaraton Calculate how much generaton from those unts s already scheduled == Day-Ahead Schedule Prce Balancng MC Total MC Day-Ahead Schedule 0 MW 25

26 Relant (bggest seller) Example Relant on February 26, :00-6:15pm Resdual Demand Ex-post optmal bd MC curve Actual Bd curve Prce ($/MWh) Balancng Market Quantty (MW) 26

27 TXU (2 nd bggest seller) Example Resdual Demand Ex-post optmal bd MC curve Actual Bd curve TXU on March 6, :00-6:15pm Prce ($/MWh) Balancng Market Quantty (MW) 27

28 Guadalupe (small seller) Example 50 Guadalupe on May 3, :00-6:15pm Prce ($/MWh) Resdual Demand Ex-post optmal bd 5 MC curve Actual Bd curve Balancng Market Quantty (MW) 28

29 29

30 What the Traders Say about Suboptmal Bddng 1. Lack of sophstcaton at begnnng of market Some frms bdders have no tradng experence; are employees brought over from generaton & dstrbuton 2. Heurstcs Most don t thnk n terms of resdual demand 3. Newer generators If a unt has debt to pay off, bdders follow a formula of % markup to add 4. TXU old school would prefer to serve t s customers wth own expensve generaton rather than buy cheaper power from market 5. Small players (e.g. muns) scared of market afrad of beng short w/ hgh prces 30

31 Possble Explanatons for Devatons from Benchmarks 1. Unmeasured adjustment costs 2. Transmsson constrants 3. Colluson / dynamc prcng 4. Type of frm 5. Stakes matter 31

32 Adjustment Costs? 1. Flexble gas-fred unts often are margnal 70-90% of tme for frms servng as own bdders 2. Bd-ask spread smaller for frms closer to benchmark Decreases over tme for hgher-performng frms 32

33 Transmsson Constrants? Does bddng strategy from congested hours spllover nto uncongested hours? 1 std dev ncrease n percent congeston only 3% Pct Acheved 33

34 Colluson? Colluson not consstent wth large bd-ask spreads Colluson smaller sales than ex-post optmal Bd-ask spread no sales Would be small(!) players - unlkely 34

35 Do Stakes Matter? Fracton from No Bddng to Optmal Bryan** South TX Elec Coop** Cty of Garland** Cty of Austn** LCRA** TXU* Calpne Relant * Volume of Optmal Output * = Investor Owned Utlty ** = Muncpal Utlty/Cooperatve 35

36 Effcency Losses from Observed Bddng Behavor Whch source of neffcency s larger? Exercse of market power by large frms? Bddng to avod the market by unsophstcated frms? Total effcency loss = 27% Fracton strategc = 19% Fracton unsophstcated =81%!! 36

37 Implcatons for Market Desgn Producton neffcences arse from both: Sophstcaton ( market power ) and lack of sophstcaton ( avod the market ) More partcpaton by small players less market power for large players Market montor should concern tself wth: Bddng by smaller players Market power on DEC sde Market desgn If strategc complexty mposes large partcpaton costs, may wsh to choose mechansms wth domnant strategy equlbrum (e.g. Vckrey aucton) 37

38 For more detals, see: Thank You! 38