Transcript PSERC SEMINAR - University of Wisconsin–Madison

10/03/00
PSERC Seminar
Are Price Spikes Predictable,
Reproducible and Avoidable?
Tim Mount
Cornell University
© Cornell University. All rights reserved
.
1
PSERC Researchers at Cornell
Faculty & Research Staff
Bob Thomas
Hsiao-Dong Chiang
Tim Mount
Carlos Murillo-Sanchez
Dick Schuler
Bill Schulze
Jim Thorp
Ray Zimmerman
Current Students
Jie Chen
Yumei Ning
Hyungua Oh
Manu Parashar
Trent Preszler
Hongye Wang
Zhifang Wang
Former Students
John Bernard
Simon Ede
Bob Ethier
2
Objectives
1) To understand the causes of price
volatility in spot markets
2) To determine how to make spot markets
less vulnerable to strategic behavior by
traders
3
Outline
1) A statistical model of price behavior
2) Analyzing actual offer curves for power
3) Properties of optimum offer curves
4) Testing market structures with PowerWeb
5) Replicating price spikes in experiments
6) Conclusions
4
The Data are from the PJM (Pennsylvania,
New Jersey, Maryland) Market East
(http://www.pjm.com/market_system_data/market_data_index.html)

Hourly and on-peak daily prices from 4/1/97
to 8/31/00

Hourly offer curves by company from
4/1/99 to 8/31/99
5
Two Important Disclaimers
1) The PJM market was chosen because
offer data are available to the public.
2) The PJM market is working relatively
well compared to other markets.

Take one step at a time

Respect commercial obligations
6
PJM daily average on-peak spot price and max load
$/MWh
MW
800.00
78000
Price
Maxload
600.00
70000
400.00
62000
200.00
54000
0.00
46000
-200.00
38000
-400.00
-600.00
30000
-800.00
4/97
22000
6/97
8/97 10/97 12/97 2/98
4/98
6/98
8/98 10/98 12/98 2/99 4/99
6/99
8/99 10/99 12/99 2/00
4/00
date
7
PJM Market Rules
1. Day-ahead offers with hourly
settlements
 Cost-based offers 4/97 to 3/99
 Market-based offers 4/99 to 5/00
2. Day-ahead market plus an hourly
balancing market 5/00 to current

Market-based offers 6/00 to current
8
Statistical Model
Markov Regime Switching
Conditional Distributions
2
yt is N (1t , 1 ) if St = 1 (high price regime)
2
yt is N ( 2 t ,  2 ) if St = 2 (low price regime)
Transition Probabilities
P r[S t  1 | S t 1  1]  P1t
P r[S t  2 | S t 1  1]  1  P1t
P r[S t  2 | S t 1  2]  P2 t
P r[S t  1 | S t 1  2]  1  P2 t
exp( c i  d i z t )
Pit 
for i = 1,2
1  exp( c i  d i z t )
where yt is the logarithm of price
 it   i   i y t 1   i x t is the conditional mean
xt

2
i
is the logarithm of forecasted load
is the variance
i, i and i are parameters
where
zt
is the logarithm of actual load
c i and d i are parameters
9
Probability (price switches to high regime)
1.00
Estimated with
Pre - 4/99 Data
0.80
0.60
0.40
0.20
0.00
4/97
7/97
10/97
1/98
4/98
7/98
10/98
1/99
4/99
7/99
Long run conditional Mean Prices in PJM $/MWh
300
High
Low
250
200
150
100
50
0
4/97 6/97 8/97 10/97 12/97 2/98 4/98 6/98 8/98 10/98 12/98 2/99 4/99 6/99 8/99
Long Run Weighted Average Price ($/MWh)
160
140
120
100
80
60
40
20
0
4/97 6/97 8/97 10/97 12/97 2/98 4/98 6/98 8/98 10/98 12/98 2/99 4/99 6/99 8/99
10
Probability (price switches to high regime)
1.00
Estimated with
Post - 4/99 Data
0.80
0.60
0.40
0.20
0.00
4/99 5/99
6/99 7/99
8/99
9/99 10/99 11/99 12/99 1/00
2/00 3/00
4/00
Long run conditional Mean Prices in PJM East $/MWh
250
High
Low
200
150
100
50
0
4/99 5/99 6/99 7/99 8/99 9/99 10/99 11/99 12/99 1/00 2/00 3/00 4/00
Long Run Weighted Average Price ($/MWh)
140
120
100
80
60
40
20
0
4/99
5/99
6/99
7/99
8/99
9/99 10/99 11/99 12/99 1/00
2/00
3/00
4/00
11
Probability(price switches to high regime)
versus daily maximum load (MW)
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
Min (observed)
0.00
25000
30000
Max (observed)
Pre
Post
35000
40000
45000
50000
55000
60000
Long Run Weighted Average Price
$/MWh
150
average-post
125
average-pre
100
75
50
25
0
6/01
6/11
6/21
7/01
7/11
7/21
7/31
8/10
8/20
8/30
9/09
9/19
9/29
Daily on Peak Average Spot Prices in PJM summer 1999
$/MWh
600
500
400
300
200
100
12
0
6/01
6/15
6/29
7/13
7/27
8/10
8/24
9/07
9/21
Conclusions about switching from cost-based
to market-based offers in 1999
1) Probability of switching to the highprice regime is lower for a given load
with market-based offers.
2) Average price in the high-price regime is
much higher and more sensitive to load
with market based offers.
13
$/MWh
PJM daily average on-peak spot price and max(load)
MW
800.00
Price
78000
Maxload
600.00
70000
400.00
62000
200.00
54000
0.00
46000
-200.00
38000
-400.00
-600.00
30000
-800.00
4/97 6/97 8/97 10/97 12/97 2/98 4/98 6/98 8/98 10/98 12/98 2/99 4/99 6/99 8/99 10/99 12/99 2/00 4/00 6/00 8/00
22000
date
14
$/MWh
Estimated with
data from 4/99 to
4/00
Predicted average
price $66/MWh
Long Run Weighted Average Price Projected from 1999 data
150
125
100
75
50
25
0
6/01
6/11
6/21
7/01
7/11
7/21
7/31
8/10
8/20
8/30
Daily on Peak Average Spot Prices in PJM Summer 2000
$/MWh
150
Observed average
price $34/MWh
125
100
75
50
25
0
6/00
$/MWh
6/00
6/00
7/00
7/00
7/00
7/00
8/00
8/00
8/00
PJM Daily on Peak Average Day Ahead and Spot Prices
Summer 2000
120
r_price
100
d_price
80
60
40
20
0
6/1
6/11
6/21
7/1
7/11
7/21
7/31
8/10
8/20
8/30
15
Conclusions about switching from hourly
settlements to a two stage market
1) Prices predicted by the 1999 model for Summer
2000 are higher than the actual prices
2) Two-Stage markets may be effective in reducing
susceptibility to price spikes.
3) Not enough data to estimate model for Summer
2000.
16
17
PJM Offer Curves at 5pm from 7/25/99(Sun) to 8/01/99 (Sun)
Offer Price
7/25 (Sun) : $59.8/MWh 44.2GW/h
7/29 (Thr) : $999.0/MWh 47.3GW/h
Offer Price
1200
1200
1000
800
600
400
200
0
1000
800
600
400
200
0
0
5
10
15
20
25
30
35
40
45
50
55
60
7/26 (Mon) : $220.8/MWh 47.2GW/h
Offer Price
0
65
5
10
Offer Price
15
20
25
30
35
40
45
50
55
60
65
50
55
60
65
50
55
60
65
50
55
60
65
7/30 (Fri) : $999.0/MWh 47.7GW/h
1200
1200
1000
800
600
400
200
0
1000
800
600
400
200
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
7/27 (Tue) : $935.0/MWh 49.2GW/h
Offer Price
0
5
10
Offer Price
1200
1000
800
600
400
200
0
15
20
25
30
35
40
45
7/31 (Sat) : $430.4/MWh 46.3GW/h
1200
1000
800
600
400
200
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
7/28 (Wed) : $935.0/MWh 48.8GW/h
Offer Price
0
5
10
Offer Price
1200
1000
800
600
400
200
0
15
20
25
30
35
40
45
8/01 (Sun) : $71.8/MWh 44.8GW/h
1200
1000
800
600
400
200
0
0
5
10
15
20
25
30
35
40
45
50
55
60
65
0
5
10
15
20
25
30
35
40
45
18
PJM Offer Curves at 5pm from 7/25/99(Sun) to 8/01/99(Sun)
Offer Price
7/25 (Sun) : $59.8/MWh 44.2GW/h
7/29 (Thr) : $999.0/MWh 47.3GW/h
Offer Price
1200
1000
800
600
400
200
0
1200
1000
800
600
400
200
0
0
2000
4000
6000
8000
10000
7/26 (Mon) : $220.8/MWh 47.2GW/h
Offer Price
0
12000
2000
4000
6000
8000
10000
12000
10000
12000
10000
12000
10000
12000
7/30 (Fri) : $999.0/MWh 47.7GW/h
Offer Price
1200
1000
800
1200
1000
800
600
400
200
0
600
400
200
0
0
2000
4000
6000
8000
10000
12000
7/27 (Tue) : $935.0/MWh 49.2GW/h
Offer Price
0
2000
6000
8000
7/31 (Sat) : $430.4/MWh 46.3GW/h
Offer Price
1200
1000
800
4000
1200
1000
800
600
400
600
400
200
0
200
0
0
2000
4000
6000
8000
10000
12000
7/28 (Wed) : $935.0/MWh 48.8GW/h
Offer Price
0
2000
6000
8000
8/01 (Sun) : $71.8/MWh 44.8GW/h
Offer Price
1200
1000
4000
1200
1000
800
800
600
400
600
400
200
0
200
0
0
2000
4000
6000
8000
10000
12000
0
2000
4000
6000
8000
19
PJM Offer Curves at 5pm from April to August (last Tuesday)
Offer Price
1200
1000
800
600
400
200
0
0
April (4/27/99) : $29.4/MWh 28.2GW/h
10
20
0
10
20
0
10
20
Offer Price
1200
1000
800
600
400
200
0
0
0
60
70
30
40
50
60
70
30
40
50
60
70
50
60
70
50
60
70
July (7/27/99) : $935.0/MWh 49.2GW
10
20
30
40
August (8/24/99) : $33.7/MWh 38.5GW
Offer Price
1200
1000
800
600
400
200
0
50
June (6/29/99) : $59.5/MWh 48.1GW
Offer Price
1200
1000
800
600
400
200
0
40
May (5/25/99) : $25.9/MWh 30.3GW/h
Offer Price
1200
1000
800
600
400
200
0
30
10
20
30
40
20
<-- 5 biggests
Offer Price
1200
1000
800
600
400
200
0
0
2000
April (4/27/99) : $29.4/MWh 28.2GW/h
4000
Offer Price
1200
1000
800
600
400
200
0
0
2000
0
2000
4000
0
2000
4000
0
2000
10000
12000
6000
8000
10000
6000
8000
10000
4000
6000
8000
10000
12000
6000
8000
10000
0
12000
2000
2000
0
2000
4000
0
2000
8000
10000
12000
6000
8000
10000
12000
June (6/29/99) : $59.5/MWh 48.1GW
4000
6000
8000
10000
12000
July (7/27/99) : $935.0/MWh 49.2GW
4000
Offer Price
1200
1000
800
600
400
200
0
6000
May (5/25/99) : $25.9/MWh 30.3GW/h
Offer Price
1200
1000
800
600
400
200
0
12000
4000
Offer Price
0
12000
2000
smallers & import -->
April (4/27/99) : $29.4/MWh 28.2GW/h
Offer Price
1200
1000
800
600
400
200
0
August (8/24/99) : $33.7/MWh 38.5GW
4000
0
1200
1000
800
600
400
200
0
July (7/27/99) : $935.0/MWh 49.2GW
Offer Price
1200
1000
800
600
400
200
0
8000
June (6/29/99) : $59.5/MWh 48.1GW
Offer Price
1200
1000
800
600
400
200
0
6000
Offer Price
1200
1000
800
600
400
200
0
May (5/25/99) : $25.9/MWh 30.3GW/h
Offer Price
1200
1000
800
600
400
200
0
PJM Offer Curves at 5pm from April to August (last Tuesday)
6000
8000
10000
12000
August (8/24/99) : $33.7/MWh 38.5GW
4000
6000
8000
10000
12000
21
PJM Offer Curves on July 27, 1999 (Tue)
2AM: $19.8/MWh 30.5GW/h
2PM: $83.3/MWh 46.3GW/h
8AM: $32.2/MWh 33.8GW/h
8PM: $619.7/MWh 47.0GW/h
1200
1200
1200
1200
1000
1000
1000
1000
800
800
800
800
600
600
600
600
400
400
400
400
200
200
200
200
0
0
0
0
10
20
30
40
50
60
70
0
10
20
30
40
50
60
70
0
0
20
30
40
50
60
70
0
4PM: $935.0/MWh 48.7GW/h
10AM: $41.2/MWh 39.4GW/h
4AM: $15.9/MWh 28.2GW/h
10
1200
1200
1200
1000
1000
1000
1000
800
800
800
800
600
600
600
600
400
400
400
400
200
200
200
200
0
0
10
20
30
40
50
60
70
10
20
30
40
50
60
70
0
12PM: $55.6/MWh 43.5GW/h
6AM: $15.4/MWh 29.1GW/h
10
20
30
40
50
60
0
70
1200
1200
1200
1000
1000
1000
1000
800
800
800
800
600
600
600
600
400
400
400
400
200
200
200
200
0
0
10
20
30
40
50
60
70
10
20
30
40
50
60
70
50
60
70
10
20
30
40
50
60
70
0
0
0
40
0AM: $30.8/MWh 38.0GW/h
6PM: $935.0/MWh 49.2GW/h
1200
0
30
0
0
0
20
10PM: $82.6/MWh 45.2GW/h
1200
0
10
0
10
20
30
40
50
60
70
0
10
20
30
40
50
60
70
22
23
Conclusions about Offers
1) The kinked slope of the offer curves is consistent
with the stochastic regime-switching model of price
behavior.
2) Two or three big companies set prices in the highprice regime, and small companies do not speculate.
3) Withholding capacity is also an important issue.
4) The total offer curve is fairly stable from hour to
hour and from day to day.
5) The total offer curve shifts from month to month, but
the kinked shape does not change.
6) Price responsive load would be an effective way to
limit price spikes.
7) (Total capacity offered/load) is potentially a better
variable than load for explaining price behavior.
24
Optimum Offers for a Single Supplier When
Other Suppliers are Competitive
(submit cost-based offers).

Load is stochastic

Capacity from other suppliers is always
sufficient to meet load

Small supplier with 4% of capacity

Large supplier with 20% capacity
25
Optimum Offer Curves
for Two Individual Suppliers
60
50
Price ($/MWh)
40
30
20
10
0
0
10
20
30
40
50
60
Percentage of Capacity
70
80
90
100
Large
Small
Marginal Cost
26
33 for max.
profit
54 for max.
profit
27
Expected Excess Profit for a Large Supplier
54 for max.
profit
28
Conclusions about Optimum Offers
1) Not very sensitive to the expected load
2) Small suppliers are punished for being
greedy
3) Highest optimum offer is relatively low
compared to actual offers
4) Large suppliers are indifferent about
having marginal units dispatched
5) Offers submitted by large supplier are
sensitive to market rules
29
Why Use Experiments to Test Markets?
1. Market structures for electricity auctions are too
complicated to derive analytical results.
2. Experiments are inexpensive compared to
experimenting directly on the public.
3. The effects of specific market characteristics can be
tested.
4. PowerWeb supports a full AC network so that the
engineering complications of congestion and ancillary
services as well as real power can be studied.
5. Paying players in experiments on the basis of
performance duplicates market behavior effectively.
30
PowerWeb
31
Completed Series of Experiments
1. Does the choice of auction rule affect economic
efficiency? Not as much as the number of
competitors.
2. Can players exploit market power in load
pockets? Yes.
3. Is self-commitment as economically efficient as
optimal unit commitment? Yes.
4. Is it easy to generate price spikes in an auction?
No, unless: a) Load is stochastic
b) Standby costs are charged
32
Withholding Capacity
33
Price Spikes
34
Current Experiments on Price Spikes
1. Sensitivity to market rules about capacity
shortfalls
a) Price set to maximum allowed price
b) Price set to highest offer
c) Price set to highest offer and idle capacity is
recalled with a cost penalty
2. Effectiveness of alternative market structures
a) Price responsive load
b) Forecasting price before the final settlement
c) Day-ahead market plus a balancing market
d) Use a discriminatory auction (pay actual
offer) instead of a uniform price auction.
35
Summary
Predicting Price Spikes

Stochastic regime switching models describe
price behavior well for financial analyses

Predicting the probability of switching to a
high-price regime as a function of load (or a
similar variable) gives an effective
quantitative measure for anticipating price
spikes.
36
Summary (Continued)
Replicating Price Spikes

Stochastic load and standby costs are
necessary for getting players to produce
price spikes in auction experiments.

Current experiments are focusing on:
a) market rules for capacity shortfalls
b) market characteristics
 price-responsive load
 forecasting price
 two-stage market
 discriminatory auction
37
Summary (Continued)
Avoiding Price Spikes

There is no silver bullet

Price spikes are not always bad

Need research on Temporally Integrated
Markets (TIM) for energy and reserves.
Reserves should be used for both economic
and engineering contingencies.

Need a new type of participant in a balanced
market. A DERALCo (Distributed Energy
Resources and Active Load Company)
should determine the net demand/supply of
energy and ancillary services for a load
center.
38