In this paper, by means of the Lie-Trotter operator splitting method, we have presented a new unified approach not only to rigorously derive Kirk’s approximation but also to obtain a generalisation for multi-asset spread options in a straightforward manner. The derived price formula for the multi-asset spread option bears a great resemblance to Kirk’s approximation in the two-asset case. More importantly, our approach is able to provide a new perspective on Kirk’s approximation and the generalization; that is, they are simply equivalent to the Lie-Trotter operator splitting approximation to the Black-Scholes equation.
Spread options, whose payoff is contingent upon the price difference of two underlying assets form the simplest type of multi-asset options, are very popular in markets as diverse as interest rate markets, currency and foreign exchange markets, commodity markets, and energy markets nowadays [1] . Despite the rapid development of spread options, pricing spread options is a very challenging task and receives much attention in the literature. The major difficulty lies in the lack of knowledge about the distribution of the spread of two correlated
lognormal random variables. The simplest approach is to evaluate the expectation of the final payoff over the joint probability distribution of the two correlated lognormal underlyings by means of numerical integration. However, practitioners often prefer to use analytical approximations rather than numerical methods because of their computational ease. Among various analytical approximations, e.g. Carmona and Durrleman [1] , Deng et al. [2] , Bjerksund and Stensland [3] , Venkatramana and Alexander [4] , Kirk’s approximation seems to be the most widely used and is the current market standard, especially in the energy markets [5] . It is well known that Kirk’s approximation extends from Margrabe’s exchange option formula with no rigorous derivation [6] . Recently, Lo [7] applied the idea of WKB method to provide a derivation of Kirk’s approximation and discuss its validity. Nevertheless, it is not straightforward to provide a generalisation of Kirk’s approximation for the case of multi-asset spread option via this approach.
Accordingly, it is the aim of this paper to present a simple unified approach, namely the Lie-Trotter operator splitting method [8] , not only to rigorously derive Kirk’s approximation but also to obtain a generalisation for the case of multi-asset spread option in a straightforward manner. The derived price formula for the multi-asset spread option bears a great resemblance to Kirk’s approximation in the two-asset case. More importantly, the proposed approach is able to provide a new perspective on Kirk’s approximation and the generalisation; that is, they are simply equivalent to the Lie-Trotter operator splitting approximation to the Black-Scholes equation. Illustrative numerical examples for the three-asset spread options are also shown to demonstrate both the accuracy and efficiency of the extended Kirk approximation. Furthermore, it should be emphasized that our approach is completely different from the extended Kirk approximation proposed by Li et al. [9] . Li et al. suggested to approximate the sum of lognormal assets as a single lognormal variable and then apply Kirk’s approximation directly to this single lognormal variable and the remaining asset to price the -asset spread option.
2. Two-Asset Spread Options
The price of a European call spread option obeys the two-dimensional Black-Scholes equation
with the final payoff condition
Here and are the future prices of the two lognormal underlying assets, and are the vola- tilities, is the correlation between the two assets, is the strike price, is the risk-free interest rate, and denotes the time-to-maturity. It is well known that no analytical solution is available in closed form and one needs to resort to numerical methods. In the following, we propose a simple derivation of the well-known price formula of Kirk’s approximation by means of the Lie-Trotter operator splitting method [8] .
Proposition 1:
The price of the two-asset spread option can be approximated by
where
Proof:
In terms of the two new variables
Equation (1) can be rewritten as follows:
where
The final payoff condition now becomes
Accordingly, the formal solution of Equation (9) is given by
Since the exponential operator is difficult to evaluate, the Lie-Trotter operator splitting method [8] can be applied to approximate the operator by (see the Appendix)
and obtain an approximation to the formal solution, namely
for
It is not difficult to recognise that satisfies the partial differential equation
with the initial condition:, and that the admissible solution is given by
where is the cumulative normal distribution function and
As a result, we obtain
which is exactly the approximate price formula given in Equation (3). (Q.E.D.)
In terms of the spot asset prices, namely, the price formula given in Equation (3) turns out to be identical to the price formula of Kirk’s approximation
where
It should be noted that for the Lie-Trotter operator splitting approximation to be valid, needs to be sufficiently small, namely. In accordance with Equation (25), this implies that Kirk’s approximation is more favourable for positive correlation between the two assets. Furthermore, for, the operators and commute so that the Lie-Trotter splitting approximation becomes exact and Margrabe’s formula is recovered.
3. Multi-Asset Spread Options
To price a European -asset call spread option, we need to solve the -dimensional Black-Scholes equation
subject to the final payoff condition
where is the future price of the lognormal underlying asset with the volatility, is the correlation between the assets and, is the strike price, is the risk-free interest rate, and is the time- to-maturity. Unfortunately, this is a very formidable task. In the following, we apply the Lie-Trotter operator splitting method to derive a closed-form approximate price formula for the multi-asset spread option, which bears a great resemblance to the price formula of Kirk’s approximation in the two-asset case.
Proposition 2:
The price of the -asset spread option can be approximated by
where
Proof:
Introducing the new variables: for and, Equation (27) can be cast in the form
where
The formal solution of Equation (36) is given by
Then we apply the Lie-Trotter operator splitting method [10] to obtain an approximation to the formal solution, namely (see the Appendix)
where the relation
is utilized.
Next, in terms of the two new variables
we rewrite as follows:
where
It is clear that the exponential operator is difficult to evaluate, so we need to apply the
Lie-Trotter operator splitting method [10] again to approximate the operator by. As a
result, we obtain
for
Since is independent of, the standard workhorse of the Black-Scholes model can be used to evaluate such that
where
It is obvious that this approximate solution is identical to the approximate price formula given in Equation (29). (Q.E.D.)
In terms of the spot asset prices, namely, the price formula given in Equation (29) becomes
where
Obviously, this approximate price formula resembles the price formula of Kirk’s approximation in the two-asset case very closely. In fact, by setting we can recover Kirk’s approximation readily. Moreover, for the Lie-Trotter splitting approximation to be valid, we need to require to be sufficiently small, namely. In accordance with Equation (58), this implies that the extended Kirk approximation is more favourable for those cases with positive effective correlation.
4. Illustrative Numerical Examples for the Three-Asset Spread Options
In this section, illustrative numerical examples are presented to demonstrate the accuracy of the extended Kirk approximation for the three-asset spread options. Although most spread options involve two assets only, yet there is a growing demand for three-asset spread options which can be found in the models for power plants or their financial equivalents—tolling contracts. We examine a simple three-asset spread option with the final payoff. Table 1 tabulates the approximate option prices estimated by the extended Kirk approximation for different values of the strike price and time-to-maturity. Other input model parameters are set as follows:, , , , , , and. Monte Carlo estimates and the corresponding standard deviations are also presented for comparison. It is observed that the computed errors of the approximate option prices are capped at (in magnitude). In fact, most of them are less than. Then, in Table 2 the effect of increasing the three vol- atilities (from to) upon the approximate estimation of the option prices is investigated. Obviously only a slight increase occurs in the computed errors, and these errors are still less than (in magnitude). Finally, we study a case in which all the three volatilities are different, namely, and, while the other parameters remain the same. According to Table 3, the computed errors generally increase a little bit in this case but they do not exceed (in magnitude). As a result, it can be concluded that the extended Kirk approximation for the three-asset spread option is found to be very accurate and efficient.
5. Conclusion
In this paper, we have proved that Kirk’s approximation for two-asset spread options can be rigorously derived
Table 1
. Prices of a European three-asset call spread option. Other input parameters are: r = 0.05, s1 = s2 = s3 = 0.3, r12 = 0.4, r23 = 0.2, r13 = 0.8, S1 = 50, S2 = 60 and S3 = 150. Here “EK” refers to the extended Kirk approximation while “MC” denotes the Monte Carlo estimates with 900,000,000 replications. The relative errors of the “EK” option prices with respect to the “MC” estimates are also presented
K\T
0.25
0.5
1
2
30
13.5410
16.4210
20.7761
27.1974
EK
-0.3%
-0.3%
-0.3%
-0.3%
error
13.5763 ± 0.0089
16.4735 ± 0.0142
20.8471 ± 0.0185
27.2841 ± 0.0264
MC
35
10.3383
13.5024
18.1191
24.8196
EK
-0.2%
-0.2%
-0.2%
-0.2%
error
10.3576 ± 0.0086
13.5286 ± 0.0124
18.1541 ± 0.0176
24.8573 ± 0.0276
MC
40
7.6613
10.9586
15.7231
22.6176
EK
0.0%
0.0%
0.0%
0.1%
error
7.6608 ± 0.0068
10.9572 ± 0.0109
15.7197 ± 0.0161
22.6072 ± 0.0244
MC
45
5.5097
8.7824
13.5805
20.5856
EK
0.4%
0.3%
0.3%
0.3%
error
5.4903 ± 0.0051
8.7565 ± 0.0106
13.5421 ± 0.0141
20.5302 ± 0.0260
MC
50
3.8470
6.9540
11.6795
18.7162
EK
0.9%
0.7%
0.6%
0.5%
error
3.8140 ± 0.0040
6.9058 ± 0.0086
11.6109 ± 0.0136
18.6195 ± 0.0245
MC
Table 2
. Prices of a European three-asset call spread option. Other input parameters are: r = 0.05, s1 = s2 = s3 = 0.6, r12 = 0.4, r23 = 0.2, r13 = 0.8, S1 = 50, S2 = 60 and S3 = 150. Here “EK” refers to the extended Kirk approximation while “MC” denotes the Monte Carlo estimates with 900,000,000 replications. The relative errors of the “EK” option prices with respect to the “MC” estimates are also presented
K\T
0.25
0.5
1
2
30
20.1436
26.0640
34.5186
46.3820
EK
-0.3%
-0.3%
-0.1%
0.3%
error
20.2066 ± 0.0168
26.1269 ± 0.0278
34.5402 ± 0.0425
46.2242 ± 0.0716
MC
35
17.4529
23.5976
32.2944
44.4495
EK
-0.1%
0.0%
0.1%
0.5%
error
17.4778 ± 0.0172
23.6076 ± 0.0268
32.2508 ± 0.0390
44.2275 ± 0.0787
MC
40
15.0417
21.3320
30.2150
42.6221
EK
0.1%
0.2%
0.4%
0.7%
error
15.0290 ± 0.0171
21.2938 ± 0.0266
30.1079 ± 0.0362
42.3425 ± 0.0656
MC
45
12.9002
19.2587
28.2733
40.8938
EK
0.4%
0.4%
0.6%
0.9%
error
12.8527 ± 0.0182
19.1753 ± 0.0234
28.1113 ± 0.0381
40.5466 ± 0.0628
MC
50
11.0139
17.3676
26.4620
39.2590
EK
0.7%
0.7%
0.8%
1.0%
error
10.9347 ± 0.0151
17.2410 ± 0.0236
26.2513 ± 0.0345
38.8658 ±0.0637
MC
Table 3
. Prices of a European three-asset call spread option. Other input parameters are: r = 0.05, s1 = 0.5, s2 = 0.4, s3 = 0.3, r12 = 0.4, r23 = 0.2, r13 = 0.8, S1 = 50, S2 = 60 and S3 = 150. Here “EK” refers to the extended Kirk approximation while “MC” denotes the Monte Carlo estimates with 900,000,000 replications. The relative errors of the “EK” option prices with respect to the “MC” estimates are also presented
K\T
0.25
0.5
1
2
30
13.8987
16.9731
21.6091
28.4620
EK
-0.5%
-0.6%
-0.8%
-1.2%
error
13.9635 ± 0.0092
17.0801 ± 0.0119
21.7922 ± 0.0191
28.8014 ± 0.0242
MC
35
10.6503
13.9613
18.8011
25.8630
EK
-0.4%
-0.5%
-0.7%
-1.1%
error
10.6905 ± 0.0078
14.0301 ± 0.0118
18.9316 ± 0.0163
26.1380 ± 0.0238
MC
40
7.9011
11.3085
16.2480
23.4424
EK
-0.1%
-0.2%
-0.4%
-0.9%
error
7.9101 ± 0.0074
11.3342 ± 0.0104
16.3206 ± 0.0172
23.6452 ± 0.0223
MC
45
5.6664
9.0191
13.9501
21.1989
EK
0.4%
0.2%
-0.1%
-0.6%
error
5.6426 ± 0.0064
9.0004 ± 0.0092
13.9627 ± 0.0145
21.3286 ± 0.0210
MC
50
3.9253
7.0838
11.9023
19.1291
EK
1.3%
0.8%
0.4%
-0.3%
error
3.8758 ± 0.0058
7.0242 ± 0.0081
11.8587 ± 0.0147
19.1882 ± 0.0216
MC
by applying the Lie-Trotter operator splitting approximation to the Black-Scholes equation, and that for cases with vanishing strike prices, the Lie-Trotter splitting approximation becomes exact and Margrabe’s formula is recovered. Our derivation also shows that Kirk’s approximation is more favourable for those cases with two positively correlated assets. Moreover, we apply the Lie-Trotter operator splitting method to obtain a genera- lisation of Kirk’s approximation for multi-asset spread options in a straightforward manner. The derived price formula for the multi-asset spread option closely resembles Kirk’s approximation in the two-asset case. By setting the total number of assets to be two, we can recover Kirk’s approximation readily. Thus, the ge- neralisation possesses the same nice features as Kirk’s approximation. For instance, as shown by the illustrative examples, the generalization is found to be very accurate and efficient in pricing the multi-asset spread options. All in all, our approach is able to provide a new perspective on Kirk’s approximation and the generalisation; that is, they are simply equivalent to the Lie-Trotter operator splitting approximation to the Black-Scholes equation.
ReferencesCARMONA R. , DURRLEMAN V. ,et al. (2003)PRICING AND HEDGING SPREAD OPTIONS 45, 627-685.HTTP://DX.DOI.ORG/10.1137/S0036144503424798DENG S.J., LI M. , ZHOU J. ,et al. (2008)CLOSED-FORM APPROXIMATION FOR SPREAD OPTION PRICES AND GREEKS 15, 58-80.HTTP://DX.DOI.ORG/10.3905/JOD.2008.702506BJERKSUND, P. AND STENSLAND, G. (2011) CLOSED FORM SPREAD OPTION VALUATION. QUANTITATIVE FINANCE, IFIRST, 1-10.VENKATRAMANA A. , ALEXER, C. ,et al. (2011)CLOSED FORM APPROXIMATION FOR SPREAD OPTIONS 18, 447-472.HTTP://DX.DOI.ORG/10.1080/1350486X.2011.567120KIRK, E. (1995) CORRELATION IN THE ENERGY MARKETS. MANAGING ENERGY PRICE RISK. RISK PUBLICATIONS AND ENRON, LONDON, 71-78.MARGRABE W. ,et al. (1978)THE VALUE OF AN OPTION TO EXCHANGE ONE ASSET FOR ANOTHER 33, 177-186.HTTP://DX.DOI.ORG/10.1111/J.1540-6261.1978.TB03397.XLO C.F. ,et al. (2013)A SIMPLE DERIVATION OF KIRK’S APPROXIMATION FOR SPREAD OPTIONS 26, 904-907.HTTP://DX.DOI.ORG/10.1016/J.AML.2013.04.004TROTTER H.F. ,et al. (1958)APPROXIMATION OF SEMI-GROUPS OF OPERATORS 8, 887-919.HTTP://DX.DOI.ORG/10.2140/PJM.1958.8.887LI M., ZHOU J. , DENG S.J. ,et al. (2010)MULTI-ASSET SPREAD OPTION PRICING AND HEDGING 10, 305-324.HTTP://DX.DOI.ORG/10.1080/14697680802626323TROTTER H.F. ,et al. (1959)ON THE PRODUCT OF SEMI-GROUPS OF OPERATORS 10, 545-551.HTTP://DX.DOI.ORG/10.1090/S0002-9939-1959-0108732-6SUZUKI M. ,et al. (1985)DECOMPOSITION FORMULAS OF EXPONENTIAL OPERATORS AND LIE EXPONENTIALS WITH SOME APPLICATIONS TO QUANTUM MECHANICS AND STATISTICAL PHYSICS 26, 601-612.HTTP://DX.DOI.ORG/10.1063/1.526596DROZDOV A.N. , BREY J.J. ,et al. (1998)OPERATOR EXPANSIONS IN STOCHASTIC DYNAMICS 57, 1284-1289.HTTP://DX.DOI.ORG/10.1103/PHYSREVE.57.1284HATANO N. , SUZUKI M. ,et al. (2005)HATANO, N. AND SUZUKI, M. FINDING EXPONENTIAL PRODUCT FORMULAS OF HIGHER ORDERS 679, 37-68.BLANES S., CASAS F., CHARTIER P. , MURUA A. ,et al. (2013)OPTIMIZED HIGHER-ORDER SPLITTING METHODS FOR SOME CLASSES OF PARABOLIC EQUATIONS 82, 1559-1576.HTTP://DX.DOI.ORG/10.1090/S0025-5718-2012-02657-3