Risk management

Deregulation and competition increase the

volatility of energy prices. The more volatile
an energy market is, the riskier it is for
firms doing business in that market.
Energy traders call this risk
market risk, and some now
quantify it using measures other
than, but based on, the
original one—Value at Risk

Val u e
at
Ris k:
Variations
on a theme

alue at Risk (VaR) is Since then, some trading during a given period of time

V a number that the

chief executive of
any energy trading
company should
know before leaving the office
each day. It indicates how
much money the company
could lose if market prices
move wildly before the com-
pany can close its position.
Originally used in finance,
VaR was adapted years ago
by energy trading compa-
nies to meet their needs.
12
firms have begun using what
they consider better mea-
sures of market risk, all of
which are based on VaR.
Three of these new measures
are Profit at Risk (PaR), Earn-
ings at Risk (EaR), and Cash
Flow at Risk (C-far). The mea-
sures are described in the
four sections that follow.
Specifically, VaR measures
the worst expected loss (for
a portfolio) to a specified con-
fidence level (usually 95%)
under normal market condi-
tions. The final VaR commu-
nicated to top management
at the end of each day is
aggregated from the individ-
ual VaRs of the company’s
different trading desks and
portfolios. In the first section,
Dr. Carlos Blanco describes
the three main techniques for
calculating VaR: analytic VaR,
Monte Carlo, and historical
simulation.
Just how useful is VaR for
Global Energy Business, May/June 2001

energy companies? In the
second section, Chal Barn-
well points out its weak-
nesses. He says that if ener-
gy companies use VaR the
way that banks use it, they
could grossly underestimate
their risk exposure. The bet-
ter measure of market risk
that his company has come
up with is a variation of VaR
called PaR.
Also recognizing the limi-
tations of VaR are Dr. Gary
Dorris and Andy Dunn. In
the third section, they intro-
duce another measure of
market risk—EaR—that they
say is more suitable for phys-
ical-asset-intensive energy
companies. The article
includes an example of how
managers of a hypothetical
gas-fired merchant power
plant might use EaR to limit
their company’s market risk
exposure.
Finally, in the fourth section,
Louis Guth and Kristina
Sepetys explain how their
company’s patent-pending
C-far model can provide an
answer to a question that
energy trading firms often
ask: “What is the probabil-
ity that this year’s cash flow
will be inadequate to fund
our strategic investments?”
The answer takes the form
of a risk profile, which the
model generates by taking
into account the different
types of risk to which a com-
pany is exposed.
—Anne Ku
Global Energy Business, May/June 2001
Risk management
Calculating and
using Value at Risk
o s t r i s k m e a s u r e m e n t level for the set.
M methodologies are based
on the analysis of a set of
There are three main methodologies
for calculating Value at Risk (VaR): vari-
scenarios that describe pos- ance-covariance (also known as analytic
sible future “states” of the VaR), Monte Carlo simulation,
world. Each of the method- B Y D R . C ARLOS and historical simulation. The
B LANCO
o l o g i e s m a k e s d i ff e r e n t three are complementary, but
assumptions about the possible evolution each offers a different view of portfolio
of markets, but they follow a similar risk. Ideally, a firm would use all three
approach towards measuring market methods to obtain the most accurate pic-
risk: First calculate a profit or rev- ture of the market risk it faces.
enue for each scenario, and then aggre-
gate those results to form a distribu- Variance-covariance
tion and extract the mean and variability
of the profit, portfolio value, or revenue The most commonly used of the VaR
1. Monte Carlo simulation
Current forward Volatility and
curves, other prices rrelation information
Scenario generation engine
4
1 2
New prices
for scenario 1 for scenario 2 scenario 10,000
Hypothetical MTM for each price scenario
......
–$645,334 –$243,000 $543,000
160
140 Monte Carlo simulation VaR
120
No. of occurrences
100
80
60
40
20
0 -1.258 -1.146 -1.034 -0.922 -0.810 -0.698 -0.586 -0.473 -0.381 -0.249 -0.137 -0.025 0.087 0.199 0.311 0.424 0.536 0.648 0.760 0.827 0.984
Portfolio profit or loss, $1 million
13
 Risk management
2. Historical simulation ical profits and losses, or P&L, of the
firm’s portfolio under each scenario
are converted into a histogram of
Current Data base with historical forward expected profits and losses, from which
market prices curves, exchange rates, interest rates VaR can be calculated (Figure 1).
One advantage of Monte Carlo simu-
lation is that it does not assume that
2
portfolio returns are distributed normal-
ly. Another is that it is a forward-look-
New set of prices New set of prices New set of prices
for scenario 1 for scenario 2 ...... for scenario 100 ing assessment of risk that takes into
1 account options and non-linear posi-
3 tions. However, the methodology
......
requires the use of a correlation and
volatility matrix to generate the random
scenarios, and that makes it computa-
tionally intensive. It also requires the
company to have pricing models for all
the instruments in its portfolio.
Estimated P&L Estimated P&L Estimated P&L
4 –$1,215,334 –$443,000 $643,000
Historical simulation
160
Historical simulation refers to the
140 Historical simulation VaR The estimated P&L for
each scenario is the process of calculating the hypotheti-
120 difference between the cal distribution of profit and losses
No. of occurrences

MTM of the portfolio

100 under each scenario
of a portfolio based on how it would
and the current MTM. If have behaved under several hundred
80
you group all possible scenarios in the past. Its advantages are
P&Ls, you can get
60
the P&L distribution
that it does not use estimated vari-
40 ances and covariances, and does not
assume anything about the distribution
20
of portfolio returns. However, the big
0 -1.258 -1.146 -1.034 -0.922 -0.810 -0.698 -0.586 -0.473 -0.381 -0.249 -0.137 -0.025 0.087 0.199 0.311 0.424 0.536 0.648 0.760 0.827 0.984 disadvantage of historical simulation
Portfolio profit or loss, $1 million is its assumption that future risks are
much like past risks, and that is less
calculation methodologies is based to derive a “synthetic” portfolio of assets frequently the case in today’s fast-
on an analysis of the volatilities of, and held. The synthetic portfolio is made up changing energy environment.
correlation among, the different risk of (cash flow) positions in the risk fac- To calculate VaR through historical
exposures of the firm’s portfolio. tors, or “vertices,” whose volatilities and simulation, one needs two things: a data
The calculation of analytic VaR is a correlations are known. base with historical prices for all the
two-step process: The purpose of cash flow mapping is risk factors to be included in the simu-
■ Select a set of market risk factors to find the “best” replication of a finan- lation, and pricing models to re-evalu-
and systematically measure actual price cial exposure for the purpose of measur- ate the portfolio for each price scenario
levels, volatilities, and correlations. ing its risk in conjunction with the firm’s (Figure 2). Historical simulation could
■ Put the firm’s exposures into a other exposures. The most difficult part be considered a special case of Monte
form that can be analyzed using risk of this step is defining a set of risk fac- Carlo simulation in which all scenarios
factor information. This is called cash tors that is small enough to be manage- are defined before the event according
flow mapping. able, but comprehensive enough to cap- to the past behavior of market prices. In
Market risk factors refer to anything ture all the firm’s risk exposures. Once the case of electricity and over-the-
that affects the value of the portfolio. the cash flow map has been created, one counter energy markets, it is quite diffi-
Prices are the most common market fac- need only perform basic matrix manipu- cult to calculate VaR using historical
tors, but you could also include in the lation to calculate the VaR of a portfolio. simulation because price histories are
analysis non-market-related risks—such hard to come by.
as volume and weather. To be compati- Monte Carlo simulation
ble with the available risk factor data, A new way to calculate VaR
every instrument in a portfolio needs to Monte Carlo simulation generates ran-
be reduced to a collection of cash flows dom pricing scenarios. The hypothet- Risk managers are primarily concerned
14 Global Energy Business, May/June 2001

with the risk of events that are very
unlikely to occur but could lead to
catastrophic losses. Yet traditional
VaR calculation methods tend to ignore
extreme events and focus on risk mea-
sures that accommodate the entire
empirical distribution of returns. This
is a problem, because it is extreme
events—like a large market move—that
produce the largest losses.
Using stress tests and scenario analy-
ses to simulate the changes in the value
of a portfolio under hypothetically
extreme market conditions is no solu-
tion, because they cannot explore all
possible scenarios. What’s more, such
analyses do not indicate how likely it is
that extreme events will occur.
The problems resulting from extreme
events are not unique to risk manage-
ment; they also arise in disciplines such
as hydrology and structural engineer-
ing, where extreme events can have
devastating consequences. Researchers
and practitioners in these fields handle
the problem by using extreme value
theory (EVT). EVT is a specialist
branch of statistics that derives general
properties of the tail, or extreme end, of
a distribution by making the best possi-
ble use of a limited set of its realized
extreme values. By focusing on the
extreme tail of a distribution, VaR can
be estimated with a confidence of
greater than 95%.
The difference EVT makes to VaR
estimates is illustrated in Figure 3,
which shows the tail of the West Texas
Intermediate (WTI) daily return distrib-
ution from 1983 to 1999. The dots indi-
cate the actual extreme return observa-
tions, the continuous line by the right
vertical axis represents the tail (assum-
ing that logarithmic returns follow a
normal distribution), and the other con-
tinuous line represents the tail of an
extreme value distribution fitted to
these data. The message this figure con-
veys is that the confidence level of
EVT-calculated VaRs is much higher
than that of VaRs calculated using tradi-
tional methodologies.
Toward risk management
Traditionally, the use of VaR within
Global Energy Business, May/June 2001
Risk management
3. Extreme value theory applied to value at risk
1.6
EV-VaR vs. normal VaR
1.4
Historical
Cumulative probabilities, %
1.2 simulation
results
1.0
0.8
VaR based on extreme
VaR based
value distribution
0.6 on normal
distribution
0.4
0.2
0.0
–35 –30 –25 –20 –15 –10 –5 0
Loss, %
companies has been limited to a pas- is more complicated—and interest-
sive role—for reporting rather than ing—than risk management in most
prescriptive purposes. Now, howev- other fields because electricity and gas
er, firms in dynamic industries— production and trading are physical
such as energy—are beginning to use processes.
it for more proactive purposes—for As energy markets become more
risk management rather than just risk complex, energy firms are embracing
measurement. Active VaR can iden- risk management systems and proce-
tify business activities that incur too dures, such as Value at Risk, to become
much risk relative to their level of more competitive. ■
return. It can also point out which
assets, business processes, and activ-
Dr. Carlos Blanco is manager of global
ities are increasing or decreasing the support and educational services at
corporation’s overall risk exposure. Financial Engineering Associates,
Most risk professionals agree that (www.fea.com) Berkeley, Calif.
risk management in the energy industry
Profit at Risk:
More realistic than
Value at Risk
ompanies that generate or Volume risk
C deliver electricity, or sell
load-following ancillary ser- The financial risks estimated by typ-
vices or retail load services, ical VaR calculations are wholly relat-
need a more accurate and
robust risk-measurement met- B Y C HAL
ed to market price movements.
However, the real and increas-
B ARNWELL
ric than the “pure” form of ingly complex energy industry
Value at Risk (VaR). Profit at Risk engenders additional financial risks
(PaRTM) is a more useful measure for a that VaR cannot measure. The most
variety of reasons. important of these is volume risk,
15