ERCOT Credit Loss Model
Mark Ruane
Credit Work Group / Market Credit Work Group
ERCOT Public
April 22, 2015
Former ERCOT Credit Loss Model
ERCOT received Potential Future Exposure (PFE) estimates from
Oliver Wyman in February 2008. In 2009, subsequent to the adoption
of the Market Credit Risk Standard, PFEs were re-estimated quarterly
using the Wyman model.
Estimates were derived from two models:
– Credit scoring model
•
Uses Counter-Party financial data to generate synthetic Counter-Party credit
ratings and probabilities of default, which was an input to:
– Credit loss model
•
•
•
Generates probability distribution of credit losses
Excel front end with VB code
Versions for zonal and nodal markets
ERCOT Public
Credit Scoring Model
The credit scoring model estimated proxy ratings based on CounterParty financial ratios and qualitative assessment.
Quantitative Inputs
• Working Capital / Sales
• Cash / Assets
• Current Ratio
• EBITDA / Interest Expense
• FCF / Debt
• Total Debt / Total Capital
• Equity / Assets
• EBITDA / Sales
• Net Income / Assets
• Total Assets
• Sales / Assets
ERCOT Public
Qualitative Inputs
• Policies
• Management Quality
• ERCOT Relationship
• Performance / Strategy
• Industry Characteristics
Credit Scoring Model
– Allowance for warning signal adjustments
– Ratings of subsidiaries tied to parents with “group logic” scoring
Group Logic Determinants
• QSE and parent have same primary line of business
• QSE covers a key geographical area for the parent
• QSE covers a key customer segment for the parent
• QSE fulfills an essential activity for the parent within a service
line
• QSE is clearly controlled by the parent through influence on
management
• QSE and parent share the same name
• The parent has a minimum 45% ownership in the subsidiary
– Default probabilities for ratings calibrated based on ERCOT
historical average loss rate and assumed worst case default
probability.
ERCOT Public
Credit Loss Model
High-level model schematic
Price
module
Simulated
prices per day
per hub1
Simulates daily prices
per hub over the
specified
time horizon
Default
module
List of
defaulted
QSEs by
scenario
Generates correlated default
scenarios over the specified
time horizon
1. Hub refers to a zone, settlement point, location or market
NYC-ECO00111-007
ERCOT Public
Volumetric
exposure
module
Exposure
by
QSE
Calculates exposure for
defaulted QSEs using
simulated prices
and volumes
Collateral
module
Collateral
by
QSE
Calculates collateral
for each of the
defaulting QSEs
Loss
calculation
Aggregate
losses
across all
QSEs
Based on exposure and
collateral of defaulting
QSEs, calculates loss
(if any) for each
simulation and
summarizes results
across all simulations
Credit Loss Model
Price module schematic
Forward
ERCOT
price inputs
Monthly
natural gas
futures
price
Convert natural
gas monthly time
series to daily
time series
 Dependent on
time horizon
Store daily heat
rate by hub
Historical
ERCOT
price
parameters
Calculate
implied
ERCOT
forward prices
by hub
 Alternatively, these
can be used directly
once available
ERCOT
log forward
prices by hub
 Nat gas price *
heat rate
 Convert to log price
Generate daily
log price series
and convert to
$/MWh
 Based on monthly avg.
heat rate/hub and
starting month
 Hub correlations
 Mean reversion
parameters
 Jump parameters
 Heat rate:
(Ppower/Pnatgas)
Time
horizon
 To cover the
default horizon
Generate normal
randoms and
correlate them
using Cholesky
Correlated
random
normals
NYC-ECO00111-007
ERCOT Public
Combine base
price and jump
simulations
 Based on hub
correlations for nonjump time series
Generate
uncorrelated
uniform and
normal random
numbers
Jump event
parameters
 Max and mean jump size
 No jumps included
 Po = implied forward price for
first forward month
Uncorrelated
uniform and
normals
 Uniforms for
jump placement
 Frequency of multiple
day events
 Use mean reversion
parameters, historical
spot volatilities
Base forward
price
simulations
 Normals for
jump size
Determine jump
event probability
based on historic
jump parameters
Probable
number of
jumps
Simulate jump
events (start day,
number of days,
and jump amount)
Simulated
prices per
day per hub
Jump
simulations
 Uniform randoms
determine numbers of
common and unique
jumps and their timing
 Jump size is lognormal,
based on price cap and
historical norms
User input
Input
Process
Calculation Output
Credit Loss Model
Default module schematic
1-yr PDs
for each
QSE
 Produced by Credit
Scoring Model
Time
horizon of
the
simulation
 Set by user (in days)
List of
QSE type
for each
QSE
 Each QSE is
segmented by type
Default
correlation
between
QSE types
Calculation of
time adjusted
PDs
Time
adjusted PDs
for each
QSE
Normalization
of time
adjusted PDs
 PDs are scaled
exponentially (based
on Poisson default
arrival rate)
 PDs for each
QSE scaled to
length
of simulation
 Time adjusted PDs
are transformed to
critical values using
Inverse Normal CDF
Creation of
default
correlation
matrix and
decomposition
 Correlation matrix
for each QSE is
produced based on
QSE type and
decomposed using a
Cholesky
decomposition
Generate and
normalize a set
of random
numbers
Critical values
 Critical values
represent the number
of standard deviations
from the mean that
correspond (in
likelihood) to the PD
of each QSE
Cholesky
matrix
 Correlation matrix
after decomposition
Normalized
random
numbers
 Random numbers are
generated (same # as # of
QSEs) and normalized using
Inverse Normal CDF
Correlate
random
numbers
using
Cholesky
matrix
 Matrix multiplication
using set of random
normals and the
Cholesky matrix
 For each QSE
compare critical
value to its
corresponding
correlated random; if
critical value is less
than correlated
random default has
occurred
Correlated
random
numbers
ERCOT Public
List of
defaulted
QSEs by
simulation
 For each simulation
record which QSEs
default (if any)
 The set of random
normals are now
correlated based on
the correlations
between the QSE
types
User input
NYC-ECO00111-007
For each QSE
determine if
default occurred
Input
Process
Calculation Output
Credit Loss Model
Price module schematic
Forward
ERCOT
price inputs
Monthly
natural gas
futures
price
Convert natural
gas monthly time
series to daily
time series
 Dependent on
time horizon
Store daily heat
rate by hub
Historical
ERCOT
price
parameters
Calculate
implied
ERCOT
forward prices
by hub
 Alternatively, these
can be used directly
once available
ERCOT
log forward
prices by hub
 Nat gas price *
heat rate
 Convert to log price
Generate daily
log price series
and convert to
$/MWh
 Based on monthly avg.
heat rate/hub and
starting month
 Hub correlations
 Mean reversion
parameters
 Jump parameters
 Heat rate:
(Ppower/Pnatgas)
Time
horizon
 To cover the
default horizon
Generate normal
randoms and
correlate them
using Cholesky
Correlated
random
normals
NYC-ECO00111-007
ERCOT Public
Combine base
price and jump
simulations
 Based on hub
correlations for nonjump time series
Generate
uncorrelated
uniform and
normal random
numbers
Jump event
parameters
 Max and mean jump size
 No jumps included
 Po = implied forward price for
first forward month
Uncorrelated
uniform and
normals
 Uniforms for
jump placement
 Frequency of multiple
day events
 Use mean reversion
parameters, historical
spot volatilities
Base forward
price
simulations
 Normals for
jump size
Determine jump
event probability
based on historic
jump parameters
Probable
number of
jumps
Simulate jump
events (start day,
number of days,
and jump amount)
Simulated
prices per
day per hub
Jump
simulations
 Uniform randoms
determine numbers of
common and unique
jumps and their timing
 Jump size is lognormal,
based on price cap and
historical norms
User input
Input
Process
Calculation Output
Credit Loss Model
Exposure module schematic
List of
defaulted
QSEs by
simulation
 Defaulted QSEs output
from Default Module
Correlation
of default
and price
by QSE
type
 Identifies QSE types
whose defaults are likely
to be correlated with price
Simulated
prices per
day per
hub
Place defaults in
the simulation
time horizon and
determine default
scenario
 For QSE types whose defaults
have low market event
sensitivity, the day of default is
picked at random over the time
horizon
 For QSEs types whose defaults
have high market event
sensitivity, a high price day
(based on a four day rolling,
weighted average of prices in
hubs where the QSE has
exposure) is randomly picked
according to user defined
parameters
Day of
default and
associated
prices
 Day of default and price
as well as prices for
days before and after
Historic BES
exposure
and BES and
total vol. by
QSE by hub
Exposure
for each
defaulting
QSE
Exposure
calculation
BES
“escalation”
type by
QSE type
 Amount BES exposure
may increase in a default
 Prices per day per hub
over the specified time
horizon from Price Module
 Based on the number of days of exposure (from
the default type) and the prices on those days, as
well as accounting for any BES escalation, the
exposure is calculated as price times volume over
each day and the total exposure is the sum of the
daily exposures
Default
types and
dimensions
 Identifies the types of
default and the
associated number of
days of exposure
NYC-ECO00111-007
ERCOT Public
User input
Input
Process
Calculation Output
Credit Loss Model
Key Assumptions – Default Module
• Default probabilities from credit scoring model
• Default correlations among market segments
• QSE default sensitivity to market events
• Default occurrence is a stochastic variable
Key Assumptions – Price Module
• Mean reversion and jump parameters
• Implied ERCOT market forward prices (based on NG forwards)
• Shape factors reflecting differences between simple average and
load-weighted prices
ERCOT Public
Credit Loss Model
Key Assumptions – Exposure Module
• Distinguishes between market price events and non-market
(random) shocks
• Volume escalation, depending on nature of Counter-Party
• Volume escalation and likelihood of default near a high-priced
day are stochastic variables
• Mass transition days
Key Assumptions – Collateral Module
• No excess cash collateral in base case
• Illiquid collateral, eg guarantees, can be haircut
• Collateral recomputed based on simplified EAL
ERCOT Public
Credit Loss Model
Model Base Case
– Did not include current (excess) collateral held by ERCOT
– Represented what was enforceable by ERCOT under Protocols
Model “Current Case”
– Used actual levels and types of collateral
– Assumed some degree of excess collateral would be maintained
until a default event
ERCOT Public
Credit Loss Model
Historic outcomes – Base Case
Simulations using FYE-2009 and Q3-2009 Financials
Potential Credit Loss - Base Case
($Millions)
Horizon (in days)
Simulations
$180
$160
$140
$120
$100
$80
$60
$40
$20
$0
90% (1:10)
95% (1:20)
99% (1:100)
99.9% (1:1,000)
Total defaults
Simulations with defaults
Simulations without defaults
Default simulations with zero loss
Total simulations with zero loss
($Millions)
Expected Loss
Median (1:2)
FYE-2009
Q3-2009
90% (1:10)
95% (1:20)
99% (1:100)
99.9% (1:1,000)
Max (1:10,000)
FYE-2009
Base Case
214
10,000
FYE-2009
Base Case
365
10,000
Q3-2009
Base Case
365
10,000
58,849
9,741
259
2,218
2,477
59,362
9,775
225
2,087
2,312
44,782
9,546
454
3,670
4,124
$2.6
$0.2
$3.2
$0.3
$2.8
$0.0
$6.8
$11.7
$34.3
$86.5
$180.9
$8.4
$15.4
$43.6
$92.2
$153.3
$6.5
$12.9
$40.8
$152.8
$304.0
Note 1-year mean and median
losses are higher than actual
experience to date in nodal market.
ERCOT Public
Former ERCOT Potential Future Exposure Model
Historic outcomes – Current Case
Simulations using FYE-2009 and Q3-2009 Financials
Potential Credit Loss - Current Case
($Millions)
Horizon (in days)
Simulations
$100
$80
Total defaults
Simulations with defaults
Simulations without defaults
Default simulations with zero loss
Total simulations with zero loss
$60
$40
$20
FYE-2009
Current Case
214
10,000
FYE-2009
Current Case
365
10,000
Q3-2009
Current Case
365
10,000
58,979
9,732
268
2,082
2,350
58,845
9,721
279
1,938
2,217
44,014
9,539
461
3,195
3,656
$4.8
$0.3
$5.4
$0.4
$5.1
$0.2
$16.2
$20.0
$30.7
$79.1
$212.2
$17.7
$22.6
$36.3
$78.1
$133.0
$17.3
$21.3
$35.6
$92.6
$204.4
$0
90% (1:10)
95% (1:20)
99% (1:100)
FYE-2009
Q3-2009
99.9% (1:1,000)
($Millions)
Expected Loss
Median (1:2)
90% (1:10)
95% (1:20)
99% (1:100)
99.9% (1:1,000)
Max (1:10,000)
ERCOT Public
ERCOT Public
Credit Loss Model
A model for the nodal market was prototyped but apparently never
validated or run.
Key Assumptions – Nodal Model
• Price simulation in key locations only
• DAM price locations considered as additional hubs with different
parameters
• Forward values for CRRs based upon implied ERCOT market
forward prices
• CRRs envisioned as one-month obligations only, settling in DAM
• Exposure calculation similar to zonal, with additions of DALE,
DAM invoices and Average Invoice Liability (AIL)
ERCOT Public
Former ERCOT Potential Future Exposure Model
In 2011 ERCOT solicited bids to update the credit loss model for the
nodal market.
With a cost estimated at $400k - $800k, F&A elected to not update the
model.
The credit scoring model could be updated internally with Market input.
ERCOT Public
Credit Updates
Questions
ERCOT Public