Title: Valuation and Hedging of PowerSensitive Contingent Claims for Power with Spikes: a NonMarkovian Appr
1Return to Risk Limited website
www.RiskLimited.com
Valuation and Hedging of Power-Sensitive
Contingent Claims for Power with Spikes a
Non-Markovian Approach
Valery A. Kholodnyi February 25, 2004 Houston,
Texas
2Introduction
- As the power markets are becoming deregulated
worldwide, the modeling of the dynamics of power
spot prices is becoming one of the key problems
in the risk management, physical assets
valuation, and derivative pricing. - One of the main difficulties in this modeling is
to combine the following features - To provide a mechanism that allows for the
absence of spikes in the prices of
power-sensitive contingent claims while the power
spot prices exhibit spikes, and - To keep the dynamics of the prices of
power-sensitive contingent claims consistent with
the dynamics of the power spot prices.
3Models for Power Spot Prices with Spikes
- Mean-Reverting Jump Diffusion Process (Ethier
and Dorris, 1999 Clewlow, Strickland and
Kaminski, 2000) - the same mechanism is responsible for both the
decay of spikes and the reversion of power prices
to their equilibrium mean - Mixture of Processes (Goldberg and Read, 2000
Ball and Torous, 1985) - spikes and the regular, that is, inter-spike
regime do not persist in time - relatively difficult to estimate parameters
- Regime Switching Process (Ethier, 1999 Duffie
and Gray 1995) - discreet time regime switching
- inconsistent short term option values
- relatively difficult to estimate parameters
4The Non-Markovian Process for Power Spot Prices
with Spikes
- Motivation
- Different mechanisms should be responsible for
- the reversion of power prices to their
equilibrium mean in the regular, that is,
inter-spike state - the reversion of power prices to their long term
mean in the spike state, that is, for the decay
of spikes -
- This is, in our opinion, due to the substantial
difference in the scales of the deviations of
power prices from their equilibrium mean in the
spike and inter-spike states - For example, power prices in the US Midwest in
June 1998 rose to 7,500 per megawatt hour (MWh)
compared with typical prices of around 30 per MWh
5The Non-Markovian Process for Power Spot Prices
with Spikes
- Main Features
- The spikes are modeled directly as
self-reversing jumps, either multiplicative or
additive, in continuous time - The parameters that characterize spikes are
frequency, - duration, and magnitude
- The spikes parameters are directly observable
from market data as well as admit structural
interpretation - The spike state and the regular, that is,
inter-spike state do persist in time
6The Non-Markovian Process for Power Spot Prices
with Spikes
- Formal Definition
- Define (Kholodnyi, 2000) the non-Markovian
process for the power spot prices with spikes by - ?tgt0 is the power spot price at time t ,
- is the multiplicative magnitude of
spikes at time t , - is the inter-spike power spot price at
time t. - Assume that the spike process and
inter-spike process are independent Markov
processes.
7The Non-Markovian Process for Power Spot Prices
with Spikes
- Underlying Two-State Markov Process
- Denote by Mt a two-state Markov process with
continuous time t ? 0. - Denote the 2?2 transition matrix for the
two-state Markov process Mt by - Pss(T,t) and Prs(T,t) are transition
probabilities from the spike state at time t to
the spike and regular states at time T, and - Psr(T,t) and Prr(T,t) are transition
probabilities from the regular state at time t to
the spike and regular states at time T.
8The Non-Markovian Process for Power Spot Prices
with Spikes
Generators of the Underlying Two-State Markov
Process The family of 2?2 matrices L L(t) t
? 0 defined by is said to generate the
two-state Markov process Mt, and the 2?2
matrix is called a generator. In terms of the
generators, P(T,t) is given by
9The Non-Markovian Process for Power Spot Prices
with Spikes
Decompositions of the Transition Probabilities of
the Underlying Two-State Markov Process It can
be shown that Moreover where
10The Non-Markovian Process for Power Spot Prices
with Spikes
Underlying Two-State Markov Process in the
Time-Homogeneous Case In the special case of a
time-homogeneous two-state Markov process Mt the
transition matrix P(T-t) and the generator L are
given by and
11The Non-Markovian Process for Power Spot Prices
with Spikes
Construction of the Spike Process
?t
?(t,?)
1
Time
Mt
Regular State
Spike State
Time
12The Non-Markovian Process for Power Spot Prices
with Spikes
Formal Definition of the Spike Process The
transition probability density function for the
spike process ?t as a Markov process is given
by where ?(x) is the Dirac delta function.
13The Non-Markovian Process for Power Spot Prices
with Spikes
Inter-Spike Process For example, can be a
diffusion process defined by where
is the drift, is the
volatility, and
Wt is the Wiener process. In
the practically important special case of a
geometric-mean reverting process we have where
is the mean-reversion rate,
is the equilibrium mean, and is the
volatility.
14The Non-Markovian Process for Power Spot Prices
with Spikes
The Expected Time for ?t to be in the Spike and
Inter-Spike States The expected time for ?t
to be in the spike state that starts at time t
is Similarly, the expected time for ?t to
be in the inter-spike state that starts at time t
is In the special case of a time-homogeneous
two-state Markov process Mt
15The Non-Markovian Process for Power Spot Prices
with Spikes
- Interpretation of the Spike State of ?t as Spikes
in Power Prices - If the expected time for the non-Markovian
process ?t to be in the spike state is small
relative to the characteristic time of change of
the process then the spike state of ?t can
be interpreted as spikes in power spot prices - ?t can exhibit sharp upward price movements
shortly followed by equally sharp downward prices
movements of approximately the same magnitude. - For example, if is a diffusion process
then - and
- In this case is the expected lifetime of a
spike and
is the expected lifetime
between two consecutive spikes.
16The Non-Markovian Process for Power Spot Prices
with Spikes
- Estimation of the Spike Parameters
- In the special case of a time-homogeneous
two-state Markov process the expected life-time
of a spike is given by - Similarly, the expected life-time between two
consecutive spikes is given by - The estimation of the probability density
function ?(t,?) for the spike magnitude can be
based on the standard parametric or nonparametric
statistical methods - Scaling and asymptotically scaling distributions
are of a particular interest in practice
17The Non-Markovian Process for Power Spot Prices
with Spikes
The Non-Markovian Process ?t as a Markov Process
with the Extended State Space The state of the
power market at any time t can be fully
characterized by a pair of the values of the
processes , and at time t. Moreover,
although the process ?t is non-Markovian it can
be, in fact, represented as a Markov process that
for any time t can be fully characterized by the
values of the processes and at time
t. Equivalently, the non-Markovian process ?t
can be represented as a Markov process with the
extended state space that at any time t consists
of all possible pairs with and
.
18European Contingent Claims on Power in the
Absence of Spikes
Valuing European Contingent Claims on Power as
the Discounted Risk-Neutral expected value of its
payoff Denote by the value of the European
contingent claim on power with inception time t,
expiration time T, and payoff g. The value of
this European contingent claim can be found as
the discounted risk-neutral expected value of its
payoff where is the
risk-neutral transition probability density
function.
19European Contingent Claims on Power in the
Absence of Spikes
Example Geometric Mean-Reverting Process It can
be shown (Kholodnyi 1995) that where
20European Contingent Claims on Power in the
Absence of Spikes
Example Geometric Mean-Reverting Process For
example (Kholodnyi 1995) where with
21European Contingent Claims on Power in the
Presence of Spikes
Notation Denote by the value of the European
contingent claim on power with inception time t,
expiration time T, and payoff The payoff g can
explicitly depend, in addition to the power price
at time T, on the state, spike or inter-spike
state, of the power price and the magnitude of
the related spike. If g depends only on the power
price at time T we have
22European Contingent Claims on Power in the
Presence of Spikes
General Case The value E(t,T,g) can be found as
the discounted risk-neutral expected value of the
payoff g where is the the transition
probability density function for ?t represented
as a Markov process.
23European Contingent Claims on Power in the
Presence of Spikes
The Case When ?(t,?) is Time-Independent The
value E(t,T,g) is given by
24European Contingent Claims on Power in the
Presence of Spikes
The Case of Spikes with Constant Magnitude
Consider a special case of spikes with constant
magnitude ? gt 1, that is, when ?(?) is the delta
function ?(?- ?). The value E(t,T,g) is given
by
25European Contingent Claims on Power in the
Presence of Spikes
Linear Evolution Equation for European Contingent
Claims on Power with Spikes It can be shown
(Kholodnyi 2000) that the value E(t,T,g) of a
European contingent claim on power with spikes is
the solution of the following linear evolution
equation where and are the
generators of and as Markov processes.
26European Contingent Claims on Power in the
Presence of Spikes
Linear Evolution Equation for European Contingent
Claims on Power with Spikes In a practically
important special case when is a geometric
mean-reverting process the generator is
given by The generator is a linear
integral operator with the kernel
27European Contingent Claims on Power in the
Presence of Spikes
Linear Evolution Equation for European Contingent
Claims on Power with Spikes In the special case
of spikes with constant magnitude the generator
?(t) can be represented as the 2?2 matrix L(t)
transposed to the generator L(t) of the Markov
process Mt. In turn, v and g can be represented
as two-dimensional vector functions Note that
?(t) represented as L(t) can also be expressed
in terms of the Pauli matrices. This gives rise
to an analogy between the linear evolution
equation for E(t,T,g) and the Schrodinger
equation for a nonrelativistic spin 1/2 particle.
28Why Prices of European Claims On Power Do Not
Spike
Ergodic Transition Probabilities for Mt Assume
that the spikes have constant magnitude ? and the
underlying two-state Markov process Mt is
time-homogeneous. The transition probabilities
for Mt can be represented as follows Pss(T,t)
?s O(e-(T - t)a), Psr(T,t) ?s O(e-(T -
t)a), Prs(T,t) ?r O(e-(T - t)a), Prr(T,t)
?r O(e-(T - t)a), where ?s b/(a b) and
?r a/(a b) are the ergodic transition
probabilities.
29Why Prices of European Claims On Power Do Not
Spike
Values of European Contingent Claims on Power Far
From Expiration The values E?t?(t,T,g) and
E?t1(t,T,g) of European contingent claims on
power coincide up to the terms of order O(e-(T -
t)a) and hence can be combined into a single
expression as follows (Kholodnyi 2000) When T
- t gtgt , E?t?(t,T,g) and
E?t1(t,T,g) differ only by an exponentially
small term. As a result, prices of European
contingent claims on power do not exhibit spikes
while the power spot prices do.
30Why Prices of European Claims On Power Do Not
Spike
Values of European Contingent Claims on Power Far
From Expiration For example, (Kholodnyi 2000) the
values of European call and put options with
inception time t, expiration time T, and strike X
are given by
31Why Prices of European Claims On Power Do Not
Spike
Example Geometric Mean-Reverting Inter-Spike
Process It can be shown (Kholodnyi 2000) that
the value E(t,T,g) of a European options with
inception time t , expiration time T, and payoff
g is given by where
32Why Prices of European Claims On Power Do Not
Spike
Example Geometric Mean-Reverting Inter-Spike
Process For example, (Kholodnyi 2000) the values
of European call and put options with inception
time t , expiration time T, and strike X are
given by where
33Why Prices of European Claims On Power Do Not
Spike
Short-Lived Spikes Consider the case of
short-lived spikes, that is .
Then for the ergodic transition
probabilities we have ?s tch o(tch) and
?r 1 - tch o(tch), where In turn, the value
E(t,T,g) can be expressed as a correction to the
value Ê(t,T,g)
34Why Prices of European Claims On Power Do Not
Spike
Example Geometric Mean-Reverting Inter-Spike
Process It can be shown (Kholodnyi 2000) that
the values of European call and put options with
strike X are given by where
35Power Forward Prices for Power Spot Prices
Without of Spikes
Power Forward Prices as Risk-Neutral Expected
Power Spot Prices Denote by the power forward
price at time t for the forward contract with
maturity time T. Power forward price
can be found as the risk-neutral expected value
of the power spot prices at time T
36Power Forward Prices for Power Spot Prices
Without of Spikes
Example Geometric Mean-Reverting Inter-Spike
Process It can be shown (Kholodnyi 1995) that
power forward prices are given by
the following analytical expression where
37Power Forward Prices for Power Spot Prices
Without of Spikes
Example Geometric Brownian Motion (GBM) for
Power Forward Prices The risk-neutral dynamics of
is described by a geometric
Brownian motion where
38Power Forward Prices for Power Spot Prices With
Spikes
General Case Denote by the power forward price
at time t for the forward contract with maturity
time T. Power forward price F(t,T) can be found
as the risk-neutral expected value of the power
spot prices ?T at time T where
is the risk-neutral average magnitudes of spikes
39Power Forward Prices for Power Spot Prices With
Spikes
The Case When ?(t,?) is Time-Independent The risk
neutral average magnitude of spikes is given
by where is the risk-neutral conditional
average magnitude of spikes given by For
example, if ?(?) is corresponds to a scaling
probability distribution, that is, ?(?) ? ?-1-
?, then
40Power Forward Prices for Power Spot Prices With
Spikes
The Case of Spikes with Constant Magnitude
Consider a special case of spikes with constant
magnitude ? gt 1, that is, when ?(?) is the delta
function ?(?- ?). The risk neutral average
magnitude of spikes is given by
41Why Power Forward Prices Do Not Spike
Ergodic Transition Probabilities for Mt Assume
again that the spikes have constant magnitude ?
and the underlying two-state Markov process Mt is
time-homogeneous. The transition probabilities
for Mt can be represented as follows Pss(T,t)
?s O(e-(T - t)a), Psr(T,t) ?s O(e-(T -
t)a), Prs(T,t) ?r O(e-(T - t)a), Prr(T,t)
?r O(e-(T - t)a), where ?s b/(a b) and
?r a/(a b) are the ergodic transition
probabilities.
42Why Power Forward Prices Do Not Spike
Ergodic Average Magnitude of Spikes The
risk-neutral average magnitudes of spikes
and coincide up to the
terms of order O(e-(T - t)a). Therefore,
and can be combined into
a single expression as follows where
is the risk-neutral ergodic average magnitude of
spikes given by
43Why Power Forward Prices Do Not Spike
Power Forward Prices far From Maturity The power
forward prices F?t?(t,T) and F?t1(t,T) coincide
up to the terms of order O(e-(T -
t)a). Therefore, F?t?(t,T) and F?t1(t,T) can be
combined into a single expression as
follows When T - t gtgt ,
F?t?(t,T) and F?t1(t,T) differ only by an
exponentially small term. As a result, power
forward prices do not exhibit spikes while the
power spot prices do.
44Why Power Forward Prices Do Not Spike
Short-Lived Spikes Consider the case of
short-lived spikes, that is .
Then for the ergodic transition
probabilities we have ?s tch o(tch) and
?r 1 - tch o(tch), where For the average
magnitude of spikes we have In turn, F(t,T) can
be expressed as a correction to
45Why Power Forward Prices Do Not Spike
- Example GBM for Power Forward Prices
- Assume that the power forward prices
follow a geometric Brownian motion. - this is, for example, the case when the power
spot prices - follow a geometric mean-reverting
process. - Then power forward prices F(t,T) far from
maturity also follow the same geometric Brownian
motion. - This, for example, can be used for
- the estimation of the volatility for the
geometric Brownian motion for , - the estimation of the volatility and the
mean-reversion rate for the geometric
mean-reverting process for , and - dynamic hedging of derivatives on forwards on
power.
46European Contingent Claims on Forwards on Power
with Spikes
Geometric Mean-Reverting Inter-Spike Process and
Spikes with Constant Magnitude It can be shown
(Kholodnyi 2000) that the value of a European
contingent claim (on forwards on power for power
with spikes) with inception time t, expiration
time T, and payoff g is given by
47European Contingent Claims on Forwards on Power
with Spikes
Geometric Mean-Reverting Inter-Spike Process and
Spikes with Constant Magnitude For example,
(Kholodnyi 2000) the values of European call and
put options (on forwards on power for power with
spikes) with inception time t, expiration time T,
and strike X are given by
48European Contingent Claims on Forwards on Power
with Spikes
Geometric Mean-Reverting Inter-Spike Process and
Short-Lived Spikes with Constant Magnitude It can
be shown (Kholodnyi 2000) that the value of a
European contingent claim (on forwards on power
for power with spikes) with inception time t,
expiration time T, and payoff g can be
represented as the following correction
49European Contingent Claims on Forwards on Power
with Spikes
Geometric Mean-Reverting Inter-Spike Process and
Short-Lived Spikes with Constant Magnitude For
example, (Kholodnyi 2000) the values of European
call and put options (on forwards on power for
power with spikes) with inception time t,
expiration time T, and strike X can be
represented as the following corrections
50Extensions of the Model
- Both positive and negative spikes as well as
spikes of more complex shapes can be considered - European contingent claims on power with spikes
and another commodity that does not exhibit
spikes can also be valued. Those include fuel and
weather sensitive derivatives such as spark
spread options and full requirements contracts - European options on power at two distinct points
on the grid with spikes in both power prices can
also be valued. Those include transmission
options - Contingent claims of a general type such as
universal contingent claims on power with spikes
can be valued with the help of the semilinear
evolution equation for universal contingent
claims (Kholodnyi, 1995). Those include Bermudan
and American options.
51Acknowledgements
I thank my friends and former colleagues from
Reliant Resources, TXU Energy Trading, and
Integrated Energy Services for their attention to
this work. I thank my friends and colleges from
the College of Basic and Applied Sciences, in
general, and the Department of Mathematical
Sciences and the Center for Quantitative Risk
Analysis, in particular, of Middle Tennessee
State University for their warm welcome and
attention to this presentation. I thank the
organizers of the Energy Finance and Credit
Summit 2004 for their kind invitation and support
of this presentation. I thank my wife Larisa and
my son Nikita for their love, patience and care.
52References
- R. Ethier and G. Dorris, Do not Ignore the
Spikes, EPRM, July-August, 1999, 31-33. - L. Clewlow, C. Strickland and V. Kaminski,
Jumping the Gaps, EPRM, December, 2000, 26-27. - R. Goldberg and J. Read, Dealing with a
Price-Spike World, EPRM, July-August, 2000,
39-41. - C. Ball and W. Torous, On Jumps in Common Stock
Prices and Their Impact on Call Option Pricing,
Journal of Finance, XL (1), March 1985, 155-173 - R. Ethier, Estimating the Volatility of Spot
Prices in restructured Electricity Markets and
the Implications for Option Values, Cornell
University, 1999. - D. Duffie and S. Gray, Volatility in Energy
Prices, In Managing Energy Price Risk, Risk
Publications, London, UK, 1995. - V. Kholodnyi, Introduction to the
Beliefs-Preferences Gauge Symmetry, Elsevier
Science, Amsterdam, Holland, 2003, To appear. - V. Kholodnyi and J. Price, Foreign Exchange
Option Symmetry, World Scientific, River Edge,
New Jersey, 1998. - V.A Kholodnyi and J.F. Price, Foundations of
Foreign Exchange Option Symmetry, IES Press,
Fairfield, Iowa, 1998. - V. Kholodnyi, Beliefs-Preferences Gauge Symmetry
Group and Dynamic Replication of Contingent
Claims in a General Market Environment, IES
Press, Research Triangle Park, North Carolina,
1998. - V. Kholodnyi, A Non-Markov Method, EPRM, March,
2001, 20-24. - V. Kholodnyi, Analytical Valuation in a
Mean-Reverting World, EPRM, August, 2001, 40-45. - V. Kholodnyi, Analytical Valuation of a Full
Requirements Contract as a Real Option by the
Method of Eigenclaims, In E. I. Ronn, Editor,
Real Options and Energy Management, Risk
Publications, 2002. - V. Kholodnyi, On the Linearity of Bermudan and
American Options in Partial Semimodules, IES
Preprint, 1995. - V. Kholodnyi, The Stochastic Process for Power
Prices with Spikes and Valuation of European
Contingent Claims on Power, TXU Preprint, 2000. - V. Kholodnyi, Modeling Power Forward Prices for
Power with Spikes, TXU Preprint, 2001.