Valuation of Swing Contracts by Least-Squares Monte Carlo Simulation
Valuation of Swing Contracts by Least-Squares Monte Carlo Simulation
Probability
Gas price
Expected price
Strike price
Periods
Probability
Gas price
Probability Potential price
Strike price
Period 4
u=upward movement
d=downward movement
u>d u 4x
Period 3
Period 3
u=upward movement u 4x
x=value at Period 0
d=downward movement
p=chance of upward movement
u>d
Period 2
Period 2
Φ=exercise value
x=value at Period 0
u 3dx
Period 1
Period 1
u 3dx
Period 0
Period 0
u 2d 2x
ux u 2d 2x
x
dx
u d 3x
u d 3x
d 4x
d 4x
Fig. 3—The development of a binomial tree. Fig. 4—Solving a binomial tree. In the optimization of the green
node, a choice is made either to exercise an option and real-
ize a value at Period 3 or to save the option with an expected
value in Period 4.
is an example of a recombining tree in which an up/down or a
down/up movement results in the same state.
The binominal tree is solved for its economic value by step- of a gas-storage facility. Willigers and Bratvold (2009) provided a
ping backward in time down through the tree (Fig. 4). Starting detailed description of the LSM method and applied it to determine
at the right side of the tree, a decision is made at each period as the value of flexibility associated with a gas asset if the asset owner
to which of the following two choices is greater: the actual value had the option to optimize gas-production rates.
that can be realized by exercising the option in this period, or the The LSM methodology allows for the asset optimization (i.e.,
expected value of saving the option for a later period. Using p to optimal exercise decision strategy) to be separated from the
represent the probability of an upward movement (making 1–p price-evolution model. The modeling flexibility created by this
the probability of a downward movement), the expected value in separation is used in this study to model the main characteristics
a subsequent period is given by pux + (1–p)dx. of real swing contracts. Numerous swing contracts do not permit
Hence, a general formulation of the optimization procedure large adjustments in the volume of the traded commodity between
is given by periods. Instead, they provide a maximum allowable increment for
production to be increased or decreased. In contrast to Dörr (2003),
max[ pux t + (1 − p)dx t , ]. . . . . . . . . . . . . . . . . . . . . . . . . . . (1) this study addresses the resulting complexity. In addition, the gas-
production-decline rate has been modeled as an uncertain variable.
The optimized values for all states at Period tn are captured and enter In contrast to the FoT approach, the LSM method also allows for
the optimization procedure at Period tn-1. The binomial real option value simple implementation of these complexities and for the develop-
is determined by collapsing all branches into a single node at t0. ment of stochastic processes for multiple uncertainties that affect
When using the FoT approach to value a swing contract, a series the value of a swing contract (using the FoT approach, a separate
of trees is developed, with each tree representing one state of the set of lattices has to be developed for each source of uncertainty).
contract in terms of the number of swings, q, left to be exercised. Typical binomial models simulate cash flow (e.g., Copeland and
In solving the FoT at each node for Period tN to t0, where N = Antikarov 2001) opposed to the separate uncertainties, as is easily
total number of periods, the optimal choice is made between two performed using the LSM methodology. Furthermore, previous
alternatives: (1) exercising one swing option and realizing its value studies valued swing contracts from the perspective of the gas
or (2) saving all swing options for a later period. The continuation trader, whereas this study also addresses the perspective of the
value (expected value) of the latter alternative is captured in Tree gas producer.
m (blue nodes in Fig. 5) while the continuation value of the first
alternative is captured in Tree m – 1 (red nodes in Fig. 5).
Although the FoT method is simple to implement for a simple Description of the LSM Model
problem, the implementation and calculation requirements grow Optimization Procedure. The LSM analysis applied here starts
exponentially with the number of factors accounted for in the by simulating a set of 5,000 “paths” for each of the uncertain
model, making this approach impractical (the computing-time variables: gas price and the decline rate. A “path” is one realization
requirements grow exponentially) for realistic problems that are of how the uncertain variable might evolve over a time horizon of
rich in complexities. 5 years, divided into discrete periods of 2 months each. Although
continuous stochastic processes are used in the model, a single
LSM Real Option Valuation value for the uncertain variables is determined for each period. The
The LSM approach, as described by Jacques (1996) and Longstaff variables used in the analysis are listed in Table 1.
and Schwartz (2001), was first applied to the valuation of a gas The value of a single swing option in the final period of the con-
swing contract by Dörr (2003). Haarbrücker and Kuhn (2009) tract, Period T, can be determined easily. Because the swing option
used the LSM method to value swing contracts in the electricity has no value once the contract has ended, the value is simply
market, and Boogert and de Jong (2006) used LSM in an invest-
ment problem closely related to swing contracts—the valuation max[ST ( w) − K , 0] ⋅ c, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2)
Pe
rio
2
d
Pe
io
1
r
d
Pe
0
io
d
r
Pe
rio
Pe
One-swing options
Hig
h ga
s pr
ice
Zero-swing options
Low
gas
pric
e
Fig. 5—A recombining FoT used for the valuation of a swing contract.
where ST (w) is the gas price in Period T for path w = 1, …, 5,000; is larger than the expected value of the option in Period T. More
K is the contract price of one gas unit; and c is the gas volume generally, the value of the swing option in Period t is given by
traded in Period T.
This valuation is made for all paths and all states of q, the num-
ber of remaining swing rights (in Period T the value of the swing max{[St(w) – K]∙c + [t,St(w),q–1], [t,St(w),q]}, . . . . . . . (3)
option is the same for all states of q because only a single option
can be exercised per period and the swing option rights have no where [t,St(w),q] is the continuation value (conditioned on the
value once the contract has ended). In Period T – 1, the swing option state at time t – 1), defined at Period t, with commodity price S(w),
will be exercised if the value of exercising the option in that period and where q is the number of remaining swings.
Symbol Description
t Period
∆t Length of a period
T Total number of periods
rf Risk-free rate
w Path or iteration
c Total gas volume traded in one period
cmin Minimal gas volume traded in one period
cmax Maximum gas volume traded in one period
∆c Difference in volume of traded gas between two subsequent periods
CC Total amount of contracted gas
Pt (w) Gas produced at Period t and Iteration w
K Strike (contract) price for 1 gas unit
St (w) Spot market gas price at Period t and Iteration w
q Swing option
Q Total number of swing options
q′ Swing option increments
Q′ Total number of swing option increments
ψ(t, St (w), q) Continuation value
ψ̂(t, St (w), q) Estimated continuation value
2 1 2 3 4
′=
1
′=
2 3
′= ′=
4 5
′=
q t= q t= q t= q t= q t=
Trading c4
volume
c3
c2
Trading periods
Cumulative
consumed swing
option increments
Trading periods
Fig. 6—The traded volume per period is represented by a series of discrete states bounded by a floor and ceiling. Moving from
one period to the next, the volume will either increase or decrease by one volume increment represented by one swing incre-
ment. The coloring illustrates what is consumed in the subsequent periods [green = uneven periods (1, 3, and 5), and yellow =
even periods (2 and 4)].
The continuation value, (•), equals the expected remaining This relationship effectively uses the value of the stochastic
value of the swing contract at Period t given the actions taken variables in Period t to estimate the future continuation value of
in this period. The optimization procedure, as shown in Eq. 3, is the Period t + 1. Given that our interest is in the continuation value
executed for each path and all states of q starting in Period T – 1 in Period t, an adjustment for the time value of money needs to be
looping backward in time until Period t = 0. A continuation value made (in this case, continuous discounting):
(•) for each state of q is stored in a matrix defined for each path
ˆ [t , St ( w), q] = ˆ [t + 1, St ( w), q ] ∗ e f , . . . . . . . . . . . . . . . (6)
− r t
and period. Upon exercising a swing option, the data transfer
between different matrices is analogous to moving between trees
in the FoT method (Fig. 4). where rf represents the risk-free rate and Δt the length of each
period.
Estimation of the Continuation Value
Changing Traded Volumes Between Periods
In the LSM methodology, the continuation value (•) in Period t and
Numerous swing contracts constrain the amount of change in
Swing q is estimated using a least-squares mean regression in which
commodity trading between periods. Modeling this requires that
one or a series of stochastic variables are regressed against the out-
the volume of commodity traded in a period be treated as a state
come modeled in the simulation. The regression formula is given by
variable. We assume that in any given period, the owner of the
W
contract will trade at either the maximum or minimum permis-
mint ,q = ∑ [Yw − ( + bSt ( w))]2 . . . . . . . . . . . . . . . . . . . . . . . . . . (4) sible volume (that is, take full advantage of market conditions or
w =1 make no change). Thus, c can be modeled as a discrete state vari-
able (upper part of Fig. 6). Given the floor and the cap on traded
where Yw is the simulated value in iteration w and and b are volumes per period and the constraints on trading adjustments
coefficients to be determined in the regression. between subsequent periods, one can define c [cmin, ...., cmax] and
The gas price, St(w), is modeled stochastically and included ∆c = |c – c |. This type of problem, in which the solution is one
i j
in the regression analysis. A regression analysis is made for of two extreme states, is sometimes referred to as a “bang-bang
each Period t and possible state of q and results in an estimated type” (Jaillet et al. 2004).
continuation value (whose notation we simplify to yw) in Period A consequence of the constraints on temporal changes of
t + 1 (in the binomial model, the continuation is known and given trading volumes is that the gas trader does not necessarily purchase
by [t , St ( w), q] = p ⋅ u + (1 − p) ⋅ d ) : ˆ [t + 1, St ( w), q] = yw = + bsw . c , the maximum allowable gas volume. In order to account for
max
this potentially lower volume bought on swing options, the variable
ˆ
+ (1 − p) ⋅ d ) : [t + 1, St ( w), q] = yw = + bsw . . . . . . . . . . . . . . . . . . . . . . . . . (5) q is introduced to indicate the number of swing increments. The
number of swing increments q purchased in each period equals from the long-run mean x, we would expect that in ln(2)/ years
c − cmin xt will have reverted half way to the long-run mean. By analogy
c . with physics, we denote the term ln(2)/ as the “half-life” of the
mean-reverting process.
Conversion of Total Volume Traded Into
Swing Options Gas-Price Model
Numerous contracts do not constrain the number of swings but The gas-price model is parameterized on the basis of a data set of his-
instead define an overall commodity volume traded during the torical Bacton gas prices from October 2000 to December 2007. The
contract and a floor and ceiling for each trading period. Given that
the problem is a bang-bang type, the conditions of such contracts parameters required in the price model are the gas-price volatility,
can be translated into a specific number of swing options, as reversion speed, reverting price level, and seasonal price fluctuation.
given by All parameters used in the stochastic models are listed in Table 2.
Q=
(CC − cmin ) ⋅ T , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7) Gas-Production Model
(cmax − cmin ) ⋅ T The uncertain production rate of a gas field is also modeled by
where CC is the total volume of contracted commodity and T the a stochastic process. The production rate follows an exponential-
decline function. Given the decline rate and the flow rate at the
number of trading periods. (cmax −swing
The number of swing increments Q =in the
cmin )
⋅contract
Q is start of a period, the remaining reserves can be calculated. As the
given by: c decline rate for each path changes over time, the expected value
of recoverable reserves will change accordingly.
Q =
(cmax − cmin ) ⋅ Q. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8)
c Value Distribution Between Producer and Trader
Although swing contracts can help gas producers to mitigate risk
Description of the Stochastic Models of adverse future conditions, gas producers would benefit by being
An Ornstein-Uhlenbeck mean-reverting stochastic process (Wil- more knowledgeable about what quantity is effectively paid when
ligers and Bratvold 2009) has been used to simulate the logarithm committing a gas stream to a swing contract.
of the gas spot price and the overall gas-production rate. Future The value of a swing option for the gas buyer is given in Eq. 3.
values are described by the stochastic differential equation The revenue generated from the gas stream for the gas producer
consists of three components: (1) payment by the gas-buying party,
dx t = ( x − x t )dt + s dWt , . . . . . . . . . . . . . . . . . . . . . . . . . . . (9) (2) gas sold on the spot market, and (3) a penalty if insufficient gas
is produced to meet the terms of the swing contract. The revenue
where x is the long-run mean to which the log of the variable for the gas producer is given by:
reverts, describes the strength of the mean reversion, s is the
volatility of the process, and dWt represents increments of a standard K ⋅ c + St ( w) • max[(Pt (w) – c), 0][St ( w) − K ]*
Brownian motion process. As shown by Dixit and Pindyck (1994), max{[c − Pt ( w)], 0}, . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. (10)
this implies that the log of the gas prices, xt, is normally distributed
−2 r f
with mean E[x(t)] = x + (x0 – x)e −2 rf and variance s (1 − e ).
2
where Pt(w) is the amount of produced gas in Period t and Path
2 w = 1, ..., 5,000.
This shows that deviations from the long-run mean will fol- Gas produced in excess of the amount dedicated to the swing
low an exponential decline. Given a current log value of x0 away contract is sold at the spot price to the market. It is assumed that
if the gas produced, Pt(w), is insufficient to cover the gas volumes,
10,000 c, the producer will purchase the shortfall at the spot price and
9,000
Value of swing contract
sell this purchased gas at the agreed rate, K, to the gas trader. In
8,000 scenarios where the spot price is less than the contract price, the
7,000 gas producer will generate positive revenue by buying gas at the
6,000 spot price and selling it to the gas trader at the contract price.
5,000
4,000 Results
3,000 The Value of an Incremental Swing Option. In the initial analy-
2,000 sis, a model with simplified assumptions was used to (1) investigate
1,000 the relationship between the number of swing options and the value
0 of a swing contract and (2) obtain insight into the minimum gas
0 0 10 15 20 25 30 35 prices that are required to justify the execution of a swing option.
Number of swing options In this basic model a swing contract is modeled with any number
of swing options, Q, and the traded gas volume in each period is
Fig. 7—The value to the producer of a swing contract as a func- either cmin = 0 or cmax.
tion of the number of swing options given that a constant total The results shown in Fig. 7 are for a fixed total volume of 3,000
volume traded in the swing contract remains fixed. MMscf to be traded through exercise of the options. The results
Gas price
Gas price
Strike price
Periods Periods
Gas price
Strike price
Periods Periods
Fig. 8—The marginal value of a swing option in a swing contract declines with the number of swing options because these
incremental swing options are exercised at progressively smaller price spikes. Periods are indicated by colored boxes. Green
boxes indicate where there is value in exercising a swing option, and red boxes mark those periods where there is not.
show how the value changes as a function of how that volume is from this one price spike that has the highest price (see Fig. 8 for
traded. If the contract specifies that only one option can be exer- more details).
cised, then all 3,000 MMscf (c_max = 3,000 MMscf) is traded in The minimum gas price that justifies the execution of a swing
one transaction and the value of the contract is EUR 9.141 mil- option has been determined as a function of time and the number
lion. At the other extreme, if the contract allows 30 swings, each of swing options available to the owner. The results shown in Fig. 9
a trade of 100 MMscf (c_max = 100 MMscf), the value falls to suggest that the minimum price point for a single-swing contract
EUR 2.333 million. Thus, the value of the contract decreases as remains constant at approximately EUR 8/Mscf for the first 30
the number of swings increases (equivalently, as the traded volume periods, but progressively declines to EUR 4/Mscf, the strike price,
per swing decreases). The reason that fewer options (with large in the final periods of the swing contract. The minimum gas price
permitted trade volumes) are more valuable is as follows. During required to justify the execution of a swing option decreases from
the contract period, one can expect one large price spike and a EUR 8/Mscf to 5.5/Mscf when the number of available swing
large number of smaller spikes—the more swing options that are options is increased from 1 to 20 (Fig. 9). Gas producers can expect
used to buy the same overall volume of gas, the less one can profit increased gas demands toward the end of the gas swing contract
10
9
Minimum spot price, EUR
8
7
1 swing
6
5 2 swings
4
4 swings
3
2 10 swings
1
20 swings
0
0 5 10 15 20 25 30 35 40 45 50 55
Period
Fig. 9—The minimum gas price, in EUR, to justify the execution of a swing option as a function of time. The relationship is shown
for a series of contracts with different number of swing options.
8,000 5,000
0
7,500
–5,000
7,000 –10,000
0 500 1,000 1,500 2,000 2,500 3,000 3,500 8,354 4,167 2,500 418
Delta c Volume of produced gas
Fig. 10—The value of a swing contract, in millions of EUR, from Fig. 11—The value, in millions of the contracted gas.
the perspective of the gas trader as a function of the total vol-
ume of contracted gas for different assumptions of Δc.
5,000 therefore have more to sell on the spot market, at a low price (Fig.
13). Although gas producers could optimize production rates with
0 respect to gas prices, such trading strategies are rarely implemented
Certain production volumes
–5,000
effectively (Willigers and Bratvold 2009).
Uncertain production volumes Swing contracts can be an effective tool to hedge against
–10,000 market price volatility but gas producers would benefit by better
0 2,000 4,000 6,000 8,000 10,000 understanding the cost of such insurance. The cost of the hedge
Volume of produced gas equals the profit made by the gas trader, given that the gas is ulti-
mately sold on the spot market and the producer has enough gas
Fig. 12—The impact of uncertainty on the value creation, in to supply the amount of gas as stipulated in the swing contract.
millions of EUR, for the gas producer. The cost of the insurance increases with market volatility and the
Cost of insurance
Gas price
Strike
price
Fig. 13—The distribution of value between the gas trader and the producer and the cost of insurance against market risk. Panel
A indicates those periods in which the gas trader exercises their swing options. The periodic revenues, as shown in Panel B,
are aggregated in Panel C to illustrate the cash flows of the gas trader and gas producer.
extent to which gas traders can benefit from price fluctuations. constraints on the minimum and maximum amounts of gas stored
In the extreme, when the gas price equals the strike price in all at any given time. The choice to store or extract gas will be
periods (i.e., no price volatility), the gross profit and the cost for determined by the gas price, and, analogous to the swing-contract
the gas trader would be equal and the cost of the insurance would problem, there will be greater value in one or the other option.
be zero. Gas producers can reduce the cost of the insurance by, for
example, introducing a cap on the maximum difference of volume Conclusions
of gas traded in consecutive periods because such measures will This study intended to raise awareness among the upstream oil and
reduce the extent to which a gas trader can benefit from price gas community of the factors that affect the value of flexibility.
volatility (see Fig. 10). Our analysis formally quantified the value of flexibility associated
The expected cost of the insurance paid by the gas producer is with swing contracts from the perspectives of both the gas producer
raised significantly if there is a risk that not enough gas is produced and the gas trader (buyer). Gas traders realize that the flexibility
to fulfill the gas contract (Fig. 12). In this study, the gas producer associated with swing contracts has value, as demonstrated by the
buys the gas shortfall at the spot market and subsequently sells the widespread use of swing contracts. At present, the parties buying
gas at the strike price to the gas trader. Although this arrangement gas from upstream exploration and production companies are the
could yield a profit for the gas producer if the spot price is lower main beneficiaries of the option to trade increased volumes of gas
than the strike price, given that the largest volumes will be traded during select periods over the lifetime of a contract. Gas produc-
on the swing contract when the gas price is high, the net result is ers would benefit by improving their understanding of the value
that the profitability of the gas producer is reduced. associated with the flexibility of swing options and thereby ensure
that they receive a fair share of the value created during periods
Expansion of the Model of increased demand and prices. By failing to do so, they risk
The model developed in this study demonstrates that the LSM transferring value to the traders at no cost.
approach can be used to value swing contracts and their common Recently, several techniques have been proposed to evaluate
characteristics. However, numerous variations of swing contracts the flexibility associated with swing contracts. However, only the
are traded and some aspects have not been addressed here or have LSM method seems to have the flexibility to address the real-life
been simplified. Numerous contracts allow for daily adjustments complexities associated with swing contracts.
of production volumes as opposed to bimonthly, and the penalties The analysis of a swing contract is a showcase for the applica-
for not meeting the contract terms are generally more complex tion of stochastic price models in exploration and production proj-
than assumed here. However, the model can be readily extended ect evaluation. The holder of a swing option aims to capitalize on
to incorporate these complexities. price spikes that are expected to occur during the contract period,
Although we focused on swing contracts, the analysis of a gas- and the value of the option is determined by price extremities,
storage facility is remarkably similar (Boogert and de Jong 2006). rather than by the expected price. Hence, the use of an expected
The operator of a gas-storage facility has the option in each period price forecast will attribute no value to a swing option. To correctly
to store gas or extract gas. There will be operational constraints estimate the option value requires simulating the price volatility
on the rate at which gas can be extracted and stored and capacity with a probabilistic model.