Simulation of Alberta Pool Prices

1
Simulation of Alberta Pool Prices

• Scott MacDonald

2
Outline

• High level observations of Alberta power market
  since deregulation
• Load growth, Generation, Historical pool prices
• Challenges of simulating prices
• What drives price spikes?
• Albertas merit order
• Description of model
• Results

3
(No Transcript)
4
(No Transcript)
5
(No Transcript)
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
Summary of Off Peak Prices
10
Summary of On Peak Prices
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
System Marginal Price

• Pool Price is the time weighted average of the
  SMP
• Intersection of Alberta Internal Load (AIL) and
  merit curve determines the price
• SMP remains constant as long as AIL is on the
  price setting block
• Many factors effect the merit order (Wind,
  Imports, DCR)

16
Sample SMP Calculation
Adjusted Load AIL SK WIND 0.44DCR
7496 MW 7496 MW (intersect
with merit curve) implies SMP 52.80 /MWh
17
Description of Price Simulation Model

• Very simple (perhaps too simple?)
• Good
• Reproduces historically reasonable CDFs
• Stats are inline with history
• Spikes happen in the right hour endings (8-23)
• Back tests calculate margins with 1 to 1.5 std of
  mean
• Bad
• Fails to reproduce the large number of 10/MWh
  hours. (Coal units marginal price)
• Over parameterized?
• Assumes that history is likely to repeat itself

18
Model Process

• Query historical data from SQL database
• Create shaping matrix
• Create noise vector
• Create price path

19
Shaping Matrix(for one month)
Notation P_i,124
Day Occurrence
Hours 1 through 24
20
Shaping Matrix(aggregate)
Notation, M mean() P_14,124,i
Year Label
Day Occurrence
Hours 1 through 24
21
(No Transcript)
22
Noise Vector

• Skip the details
• Broken out into peak (HE 8-23) and off peak
  (everything else)
• Remove seasonality by doing
• log(P_i) mean(log(P_month))

23
Create Price Path (pseudocode)

• Tmp_(i,j) Smat(i,j) XSmat_std(i,j)
• where X a bU(0,1) random variable
• 2. Tmp flatten(Tmp)
• 3. S Sc
• where c is scaling constant
• 4. On_peak_chain emprand(Nvec_on)
• Off_peak_chain emprand(Nvec_off)
• 6. IF (HE is peak)
• P log(S) On_peak_chain
• Else
• P log(S) Off_peak_chain
• P exp(P) X
• where X a bU(0,1) random variable

24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
Conclusion

• Thin supply stack causes wild price variation
• Generators dispatching intelligently
• Scarcity pricing, shadow bidding
• Naïve approach seems to yield reasonable results
  (back tested over 1 year of data)
• Flexible tool to allow users to assess the risk
  of various hedge scenarios (7x24, 6x16, etc)