A Structural Model to Evaluate the

Transition from Self Unit Commitment to

Centralized Unit Commitment
Alvaro Riascos Sergio Camelo
University of los Andes, Stanford University, USA
Colombia, and
Quantil, Colombia Quantil, Colombia

Luciano de Castro Shmuel Oren

University of Iowa, USA University of California
Berkeley, USA
Anthony Papavasiliou
Catholic University of Louvain,
Belgium

October 27 2016
Plan of presentation
 Introduction

 Model and methodology

 Marginal and opportunity costs
 Counterfactuals

 Results: Competitive benchmark vs Bid-based

simulation
 Hydro and thermal generation
 Economic efficiency

 Conclusions and extensions

Introduction

 Different dispatch models: Self Unit commitment vs.. Centralize

Unit Commitment.
Introduction: Self Unit Commitment

 Before 2009 generators submitted simple energy bids (one

price for 24 hours and declared hourly generating capacity).

 System operator (XM) solved optimization problem: Reduce

hourly total generating variable costs (similar to uniform auction).

Plants self commit: Risk of not covering startup costs (self unit
commitment).
Introduction: Centralized Unit Commitment

 After 2009 generators submitted “complex bids”: “simple bids” +

startup and shut down costs.

 System operator (XM) solved optimization problem: Reduce day

total costs.

 System operator commits plants: No risk of not recovering startup

costs (centralized unit commitment).
Introduction: Inneficiency Problem

 Self Unit Commitment:

 In the presence of non-convexities, self-committed uniform

price auctions with energy only offer prices can lead to productive
inefficiencies:

 Thermal units face an unnecessary risk when restricted to

submit energy only offer prices.

 Turning off thermal plants that are already running and

turning on a lower marginal cost unit could result in inefficient
production due to ignoring startup costs.
Introduction: Incentives Problem

 Centralized Unit Commitment:

The mechanism is meant to improve productive efficiency.

 But the mechanism used to solicit generator data, upon which

the market clearing prices and settlements are based, may compel
generators to overstate costs.

 This incentive to overstate costs is also true of self-commitment in

an energy exchange, but complex bids allow for further strategic
behavior.
Introduction: Empirical Problem

There are no theoretical studies with clear-cut results that rank

the performance of one design relative the other, so the question
remains an empirical one.

 It is an empirical problem: This study proposes a structural

model of the dispatch to evaluate empirically the ultimate benefits (if
any) of the 2009 regulatory intervention in Colombia.
Introduction: Related Literature

 Centralized unit commitment: Sioshansi, O’Neill and Oren (2008),

(2008b), (2010), O’Neill, Sotkiewicz, Hobbs, B.F., Rothkopf, and
Stewart, ( 2005).

 Incentives: Sioshani, Oren and O’Neill (2010) provide a stylized

example which shows that self-commitment in an energy exchange
can result in inefficient production of energy even if generators are
price takers. This is a phenomenon due only to non-convexities in
the cost structure of some generating units.

 Theory: Sioshansi and Nicholson (2011)

 Econometric: Riascos, Bernal, de Castro and Oren (2015). The

Energy Journal. (2015).
Model and Methodology: General View
Model and Methodology
Model and Methodology
Model and Methodology: Efficient Dispatch

 Optimization problem:

min ෍ ෍ 𝑃𝑜𝑓𝑖 × 𝑝𝑖,𝑡 + 𝑃𝑎𝑟𝑖 𝑠𝑖,𝑡

𝑠𝑜𝑎𝑘 𝑑𝑖𝑠𝑝 𝑑𝑒𝑠 𝑠𝑜𝑎𝑘 𝑑𝑖𝑠𝑝 𝑑𝑒𝑠
𝑝𝑖,𝑡, 𝑝𝑖,𝑡 ,𝑝𝑖,𝑡 ,𝑝𝑖,𝑡 ,𝑠𝑖,𝑡 ,ℎ𝑖,𝑡 ,𝑢𝑖,𝑡, 𝑢𝑖,𝑡 ,𝑢𝑖,𝑡 ,𝑢𝑖,𝑡
𝑡=0,…,23 𝑖

 Subject to many restrictions (seventeen thousand equations):

demand, thermal plants technical restrictions, hydro plants
environmental restrictions.
Model and Methodology: Efficient Dispatch

 The marginal price 𝑀𝑃𝑂𝑡 is calculated as the price bid of the

marginal plant that is flexible.

 The hourly spot price 𝑃𝑡 is defined as:

𝑃𝑖 = 𝑀𝑃𝑂𝑡 + ∆𝐼

 ∆𝐼 is the uplift (transfer payments):

σ𝑖 max{0, 𝐶𝑖 − 𝑅𝑖 }
∆𝐼 =
σ24
𝑡=1 𝐷𝑡

 𝑅𝑖 revenue, 𝐶𝑖 cost.
Model and Methodology: Data

 National demand

 50 plants (30 thermal plants, 20 hydro).

 Marginal costs

 Opportunity costs,

 Start up and shut down costs.

 Technical parameters (thermal plants)

 Enviormental restrictions.
 0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9

0
1
2000/01
2000/04
2000/07
2000/10
2001/01
2001/04
2001/07
2001/10
2002/01
2002/04
2002/07
2002/10
2003/01
2003/04
2003/07
2003/10
2004/01
2004/04
2004/07
2004/10
2005/01

Hydraulic generation (% of total generation)

2005/04
2005/07
2005/10
2006/01
2006/04
2006/07
2006/10
2007/01
2007/04
2007/07
2007/10
2008/01
2008/04
2008/07
2008/10
2009/01
2009/04
2009/07
Model and Methodology: Data

Thermic generation (% of total generation)

2009/10
2010/01
2010/04
2010/07
2010/10
2011/01
2011/04
2011/07
2011/10
2012/01
2012/04
2012/07
Spot price (right axis)

2012/10
2013/01
2013/04
2013/07
0
50
100
150
200
250
Model and Methodology: Costs

 Marginal costs

Heat Rate
Marginal Cost = ∗ P + VOM + TAXES
Calorific Value

 Opportunity costs

෣
𝑀𝐶𝑔𝑡 = min 𝑇ℎ𝑒𝑟𝑚𝑜𝑀𝑃𝑂𝑔𝑡 , 𝐵𝑖𝑑𝑔𝑡 ,
Model and Methodology: Costs
Model and Methodology: Startup Costs
Model and Methodology: Restrictions
Model and Methodology: Model’s fit

Week by Week Fit of the Model's Market Price

200

150
MP
Price

Simulated
Actual
100

50

2010−07 2011−01 2011−07 2012−01 2012−07

Date
Model and Methodology: : Model’s fit

Weekly Total Costs Fit

100

Scenario
Cost

75
Simulated
Actual

50

25
2010−07 2011−01 2011−07 2012−01 2012−07
Date
Counterfactuals: Competitive Benchmark vs
Bid-based simulation

 We perform three simulations:

(1) The competitive benchmark for the whole period of study.

(2) The simulated real scenario before 2009, result of using our
structural model of the dispatch under self-unit commitment.

(3). The simulated real scenario after 2009, result of using our
structural model of the dispatch under centralized-unit commitment.

Main hypothesis: Agents bid the same when faced with our
dispatch model (before or after 2009) rather than SO dispatch
model.
Model and Methodology
Results: Economic Efficiency

24 24
𝑅 𝑅
𝑅𝑒𝑎𝑙 𝐶𝑜𝑠𝑡𝑠 = ෍ ෍ 𝑞𝑔𝑡 𝑝𝑔𝑡 + ෍ ෍ 𝑢𝑔𝑡 𝑠𝑔𝑡
𝑡=1 𝑔∈𝐺 𝑡=1 𝑔∈𝑇

24 24
𝐶 𝐶
𝐶𝑜𝑚𝑝𝑒𝑡𝑖𝑡𝑖𝑣𝑒 𝐶𝑜𝑠𝑡𝑠 = ෍ ෍ 𝑞𝑔𝑡 𝑝𝑔𝑡 + ෍ ෍ 𝑢𝑔𝑡 𝑠𝑔𝑡
𝑡=1 𝑔∈𝐺 𝑡=1 𝑔∈𝑇
 Results: Economic Efficiency

Weekly Total Costs

100

80

Simulation
Cost

Real
60
Competitive

40

2006 2008 2010 2012

Date
 Results: Economic Efficiency

Table 6: Average weekly deadweight loss ratios

Year 2006 2007 2008 2009BR 2009AR 2010 2011 2012

Deadweight 3.87% 10.90% 17.95% 18.70% 19.04% 14.69% 4.23% 10.26%

Table 7: Average weekly deadweight loss ratios with exclusion of a period of very high
fuel prices

Reform Before After

Deadweight 12.12% 8.80%

Conclusions

In the presence of non-convexities, self-committed uniform price

auctions with energy only offer prices can lead to productive
inefficiencies.

This paper capitalizes on the recent transition in Colombia from

self-commitment to centralized unit-commitment to empirically
evaluate the relative economic efficiency under the two regimes.

 We estimate the observed relative deadweight loss reduction

of at least 3.32% after reform.

 This can be explained in part be by the fact that, before 2009,

there was an underproduction of thermal energy.