Andrea Roncoroni

Energy and Commodity Asian-Style Options under
Seasonal Data and Stochastic Volatility

Andrea Roncoroni
ESSEC Business School, Paris - Singapore
Practical Quantitative Analysis in Commodities
June 17-18, 2010
London, UK
Andrea Roncoroni Commodity Asian-Sytle Options
Commodity Price Modelling
Model construction focuses on:
1
Primitives = Input state variables
should be quantities with:
Reliable observations;
Economic signicance.
2
Structural elements = Form of drift, volatility, jump, if any
should be identied using statistical analysis of historical data and
then tted to observed prices.
3
Driving noise terms = Number (&nature) of noise terms
should be assessed based on historical price analysis (e g , exam
of the trajectorial properties of price paths, Principal Components
Analysis, jump ltering).
Modelling Frameworks
Frameworks: We identify four classes of arbitrage free models for
commodity prices according to selection of primitives:
1
[SC] Spot Price-Convenience Yield Models (Gibson-Schwartz (1990))
primitives = spot price + instantaneous spot convenience yield;
2
[FD] Forward Curve Models (Reisman (1991), Jamshidian (1991))
primitive = forward price curve;
3
[FC] Forward Convenience Yield Models (Cortazar-Schwartz (1994))
primitives = spot price + instantaneous fwd convenience yield;
4
[SP] Spot Price Models (Black (1976))
primitive = spot price (deterministic convenience yield)
Roncoroni, A., Commodity Price Models, in: Cont et al., Encyclopedia of
Quantitative Finance, Wiley (forthcoming).
Stylized Facts and Market Price Information
Principle Commodity derivatives should be priced using models
reproducing:
1
Stylized facts about underlying price dynamics:
Mean reversion characterizing spot price dynamics,
Time and stochastic patterns aecting historical price volatility,
Jump-like price behavior and non normal returns.
2
Market price information available at the valuation time:
Market quoted forward and futures prices,
Volatility surfaces (liquid option quotes).
Stylized Facts I: Empirical Evidence
2 4 6 8 10 12
0.16
0.18
0.2
0.22
0.24
0.26
Month
S
t
d
.

D
e
v
.
Corn Historical Volatility, 1980-2009
2 4 6 8 10 12
0.18
0.2
0.22
0.24
0.26
Month
S
t
d
.

D
e
v
.
Soybean Historical Volatility, 1980-2009
2 4 6 8 10 12
0.2
0.21
0.22
0.23
0.24
Month
S
t
d
.

D
e
v
.
Wheat Historical Volatility, 1980-2009
2 4 6 8 10 12
0.4
0.5
0.6
0.7
Month
S
t
d
.

D
e
v
.
HH Gas Historical Volatility, 1990-2007
Stylized Facts II: Empirical Evidence
Market Price Information I: Empirical Evidence
0 0.5 1
62
64
66
68
Light,Sweet CrudeOil (Nymex 1-3-2007)
Maturity (years)
0 0.5 1
7
8
9
10
HH Natural Gas (Nymex 1-3-2007)
Maturity (years)
0 0.5 1
1.7
1.8
1.9
2
Heating Oil (Nymex 1-3-2007)
Maturity (years)
0 1 2 3 4
3200
3400
3600
3800
4000
Corn (CBOT 1-12-2006)
Maturity (years)
Market Price Information II: Empirical Evidence
0.2
0.4
0.6
0.8
1
1.2
0.8
1
1.2
0.35
0.4
0.45
0.5
0.55
0.6
0.65
Moneyness
Smile curve implied by Crude Oil Futures Options on July 7, 2009
Maturity
I
m
p
l
i
e
d

V
o
l
Commodity Asian-Style Options
Discrete monitoring = Prices are monitored every time units.
Underlying variable = Price average

n
i =0

i
S
i
.
Name Weight
j
Average Avg
n
Standard arithmetic 1/ (n + 1) (n + 1)
1
n
i =0
S
i
Volume weighed V
j
/
i
V
i
(
k
V
k
)
1
n
i =0
V
i
S
i
Cash ows:
Fixed strike Floating strike
max Avg
n
K, 0 max Avg
n
S
n
, 0
Articles
Analytical Pricing of Commodity Asian-Style Options under Discrete
Monitoring (with G.Fusai,M.Marena). JBF32(10), 2033-2045, 2008
Analytical pricing (up to FT) of arithmetic average options on:
dS
t
=
t
S
t
dt +
t
_
S
t
dW
t
+ dJ
t
(VC-SQRT-J)
(Variants: CC-SQRT:const.coe.+dJ=0, CV-SQRT:const.vol.+dJ=0; SC-SQRT:seas.coe.+dJ=0, C-SQRT-J:const.par.)
Control Variates for Asian-Style Options under Seasonality, Stochastic
Volatility and Jumps (with G.Fusai, M.Marena). WP, ESSEC, 2009
Analytical pricing (up to FT) of geometric average options on:
d lg S
t
= (
t
m
t
v
t
/2) dt +
_
v
t
dW
1
t
+ dJ
1
t
(SV-JJ)
dv
t
= (
t
v
t
) dt +
_
v
t
dW
2
t
+ dJ
2
t
.
and use as control variable for pricing arithmetic average options. (Variants: JD:v=const., SV=dJ1=dJ2=0, SV-J:dJ2=0)
Transform-Based Option Pricing (Carr-Madan (1999))
Pay-o (including call and xed-strike Asian):
max 0, Y
T
k , k = constant, Y _ 0.
1
Vanilla call Y
T
= S
n
, = 1, k = K;
2
Fixed-strike Asian Y
T
=

S
i
, = 1/ (n + 1), k = (n + 1) K.
Laplace transform of the option price wrt strike k:
/:
call price=funct.of strike k
..
C
0,Y
0
(T, k)
Laplace transf.=funct.of
..
/[C] () ,
_
+
0
e
k
C
0,Y
0
(T, k) dk .
Transform-Based Option Pricing (Carr-Madan (1999))
Laplace transform:
/[C] ()
AF price
= e
rT
_
_
_
_
E
0
_
e
Y
T
_
2
+
E
0
[Y
T
]
2
_
_
_
_
.
Laplace inversion Option price:
C
0,Y
0
(T, k) = e
rT
_
/
1
_
/[Y
T
] ()
()
2
_
(k) +E
0
[Y
T
] k
_
.
Arithmetic Average Options under a CV-SQRT Model
Constant volatility square-root dynamics:
dS
u
= (r
u
c
u
) S
u
du +
_
S
u
dW
u
, starting at: S
0
= x.
Curve tting: Set r
t
c
t
=
t
;
Input Fwd prices observed for maturities up to option expiration.
Problem Find
t
such that the spot model ts fwd prices.
Solution F
0,T
= E
0
(S
T
) = x exp
_
T
0

s
ds i:
T
=
T
ln F
0,T
.
Main Result (Fusai-Marena-Roncoroni (2008))
Theorem: Mgf S
t
Mgf (nal price, arithm.avg.price):
v
0,x
(n, ; , ) , E
0
_
e
[
S
n
+
j
S
j
]
_
= e
0
(;,)x
,
where
j
(; , ) satises the recursive equation:
j
(; , ) = A
_
;
j +1
(; , )
_
+
j
, for j = n 1 0,
n
(, , ) = +
n
(starting condition),
with A(; ) = e
(r c)
/
_
1 +
2
_
e
(r c)
1
_
/2 (r c)
_
.
Pricing Formula
Price Fixed-strike Arithmetic Asian-style option price:
V = e
rt
_
1
2
_
1
_
a
l
+
_
1
a
l
_
1
e

n+1
K(n+1)
v
0,x
(n, ; 0, )
2
d
+
e
(r c)(n+1)
1
(e
r
1) (n + 1)
x K (n + 1)
_
.
Extensions:
1
Mean reversion + Time-varying volatility;
2
Time-varying drift + Jumps.
Geometric Average Options under a CC-SV-JJ Model
Constant coe. stoch.vol. double jump model:
d lg S
t
= (r c m
t
v
t
/2) dt +
_
v
t
dW
1
t
+ dJ
1
t
(SV-JJ)
dv
t
= ( v
t
) dt +
_
v
t
dW
2
t
+ dJ
2
t
dt = Cov
_
dW
(1)
, dW
(2)
_
,
1
=
2
, J
i
i .i .d.
~ A
Pay-o (Geometric Asian-style call):
C
g
(T, K) = max
_
_
0,
_
n
k=0
S
k
_ 1
n+1
geometric avg.=:Y
T
K
_
_
.
Main Result (Fusai-Marena-Roncoroni (2009))
Theorem: Cf (lg S
t
, v
t
) Cf log(geom.avg.price Y
T
):
0,x,v
(n, ; u) = E
0
_
e
iuY
T
_
= e
iux+
1
(u;n,)v+
1
(u;n,)
,
where
j
and
j
satisfy the recursive equations (j : n 1 1):
j
(u; n, ) = D
_
(n j + 1) / (n + 1) , i
j +1
(u; n, ) ;
_
,
starting at:
n
(u; n, ) = D (u/ (n + 1) , 0; ) ;
j
(u; n, ) =
j +1
(u; n, )
+C
_
(n j + 1) / (n + 1) u, i
j +1
(u; n, ) ;
_
+J
_
(n j + 1) / (n + 1) u, i
j +1
(u; n, ) ;
_
,
starting at:
n
(u; n, ) = C (u/ (n + 1) , 0; ) + J (u/ (n + 1) , 0; ),
and D, C, J are given in analytic form.
Test 1: Discrete vs. Continuous Monitoring Rules
Model = Square-root with constant volatility Analytical formula
of joint mgf
_
S (T) , T
1
_
T
0
S (u) du
_
.
n=12 n=50 n=250 cts
k Price %Di Price %Di Price %Di Price
-0.05 0.16897 -1.58 0.17099 -0.40 0.17153 -0.08 0.17167
0 0.14102 -1.91 0.14306 -0.48 0.14362 -0.10 0.14376
0.05 0.11540 -2.33 0.11745 -0.59 0.11800 -0.12 0.11814
1
Price dierences | as the number of monitoring dates .
2
The speed of convergence is almost linear in n (slow).
(Conv.speed for barrier options is even slower, - 1/
_
n.)
Test 2: Comparison to Methods for the Market Model
Methods for geometric Brownian motion dynamics
- Geman and Yor (1993): Laplace trans.inv. with cont.monitoring
- Turnbull and Wakeman (1991) approximation of the lognormal price distribution
Comparative model
SQRT
:Call
SRmod.
(
SQRT
) = Call
BSmod.
(
GBM
)
K
GBM
Option Prices in the GBM case
SR
Option price (SQRT)
Inv.Lap. Logn,
90 0.1 15.39763 15.39906 0.97411 15.39890
110 0.1 1.41362 1.41080 1.02356 1.41070
90 0.5 19.30572 19.55391 4.86178 19.37724
110 0.5 10.07128 10.18997 5.11247 10.06599
1
SQRT accurately approx.GBM quotes, yet SQRT real time val.
2
SQRT lies between the two approx.quotes (but for deep OTM)
Test 3: Including a Quoted Forward Curve
Methods Compute prices with at and market fwd curves
n=5
K Flat Non Flat %Di
-0.05 0.14 0.15 4.04
0 0.13 0.13 3.99
0.05 0.12 0.12 3.95
n=250
K Flat Non Flat %Di
-0.05 0.15 0.17 9.73
0 0.14 0.15 9.78
0.05 0.13 0.14 9.82
1
Monitoring frequency ; price discrepancy between considering
and discarding the quoted forward curve |
2
This is important in commodity/energy markets where fwd curves
often display seasonality
Test 4: Including Seasonal Volatility
Step I Compute historical avg.vol.
GBM
for each month
Step II Conv.
GBM

SR
:
GBM
Spot=
SR
_
Spot
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
GBM
0.17 0.14 0.16 0.17 0.20 0.22 0.27 0.22 0.19 0.18 0.16 0.13
SR
10.2 8.59 9.74 10.2 12.4 13.8 16.3 13.7 11.4 11.2 9.80 8.40
Step III Building 3 vols
Flat average vol.
(a)
:
2
(a)
_
T
0
1
F
0,s
ds =
_
T
0

2
s
F
0,s
ds
Flat implied vol.
(b)
matching a benchmark option (ATM Asian
with a 5-period monitoring)
Time varying historical market volatility structure
Test 4: Including Seasonal Volatility
Results
K/S(t) n
a
=11.21
%Di
V(
a
)-V(nf)

b
=10.84
%Di
V(
b
)-V(nf)
Non Flat Vol.
0.9 12 460.72 0.59 458.62 0.13 458.03
1 12 196.08 3.02 191.37 0.54 190.34
1.1 12 54.71 9.03 50.80 1.23 50.18
0.9 1000 472.95 0.66 470.81 0.20 469.85
1 1000 206.46 3.35 201.67 0.96 199.76
1.1 1000 60.28 10.08 56.16 2.55 54.76
1
Important price dierences
2
This eect is rather signicant for OTM options
3
Method 2) ~ method 1), but requires option price observation
Test 5: Variance Reduction
Description: Evaluate an arithmetic average option using:
Naive Monte Carlo,
Geometric control variate,
Normal antithetic variables.
Control variate:
CV
=
arith avg
g (X) +
estim.by simulation
Cov (g (X) , f (X))
Var (f (X))
_
_
_
geo avg
f (X)
geo opt.priceanalytic
..
E(f (X))
_
_
_
.
Antithetic:
AV
:=
1
n
n
i =1
g
_
X
_
A
1
i
, A
2
i
__
+ g
_
X
_
A
1
i
, A
2
i
__
2
.
Results: SV model Prices and standard errors:
|option+method / monitoring 12 25 50 100 250
Geo Analytic (our method)
Price (x0.01)
5.05161 5.06461 5.07075 5.07386 5.07574
Geo Naive Monte Carlo
Price (x0.01) + Std Error (x0.00001)
4.97484
(5.55)
5.02985
(5.58)
5.05107
(5.61)
5.06507
(5.62)
5.07065
(5.64)
Arithmetic Naive Monte Carlo 5.10154
(5.60)
5.15143
(5.66)
5.16970
(5.69)
5.18261
(5.71)
5.18737
(5.72)
Arithmetic Control Variate 5.17961
(0.12)
5.18672
(0.12)
5.18966
(0.12)
5.19152
(0.12)
5.19253
(0.12)
Arithmetic Antithetic Variable 5.10398
(1.96)
5.14992
(1.99)
5.17290
(2.00)
5.18120
(2.01)
5.18971
(2.01)
Arithmetic Antithetic+Control 5.17913
(0.10)
5.18641
(0.10)
5.18959
(0.10)
5.19140
(0.10)
5.19255
(0.10)
1
Control variate dramatically reduces standard errors, while
preserving the computational time (analytical method is almost in
real-time);
2
Antithetic variates reduces standard errors from both naive MC and
control variate, whereas computational time almost doubles.
Results: SV Model Convergence and std.errors:
|ratio or method/ n.simul. 100,000 200,000 300,000 400,000 500,000
MC/Antithetic Variable 2.84707 2.84326 2.84170 2.84381 2.84391
MC/Control Variate 46.39385 46.08531 45.77696 45.77019 45.88352
MC/Antithetic+Control V 59.69837 59.20087 58.81804 58.83800 58.94951
Arithmetic Naive Monte Carlo
(x 0.01)
5.16737 5.17290 5.18414 5.18125 5.18159
Arithmetic Control Variate 5.19153 5.19143 5.19156 5.19147 5.19137
Arithmetic Antithetic Variable 5.17596 5.17491 5.17556 5.17965 5.18035
Arithmetic Antithetic+Control 5.19130 5.19129 5.19138 5.19130 5.19126
1
Variance reduction is dramatic with control variate;
2
Control variate leads to fast convergence.
Conclusion
We price arithmetic Asian-style options under realistic
assumptions:
1
Averages are discretely monitored Real-world practice
2
SQRT Model Analytic pricing formulae
3
SV-JJ Model Eective control variate
4
The underlying dynamics exhibit stylized behavioral features:
Time varying volatility Seasonal price vol.
Jumps Spikes and non-normal returns
5
Market information is accounted for using:
Time varying drift Fitting the quoted fwd curve/price trend
Stochastic volatility + jumps smile tting (to be conducted)
The Author
Andrea Roncoroni is Professor of Finance at ESSEC Business School (Paris - Singapore) and regular Lecturer at Bocconi University
(Milan), He holds a BS in Economics from Bocconi University (Italy), an MS in Mathematics from the Courant Institute of Mathematical
Sciences (New York) and PhD's in Applied Mathematics and Finance from the University of Trieste (Italy) and University Paris Dauphine
(France), respectively. His research interests cover Energy Finance, Financial Econometrics and Derivative Structuring. He has consulted
for private companies (e.g., Gaz de France, Edison Trading, EGL, Dong Energy) and lectured for public institutions (e.g., International
Energy Agency, Central Bank of France, Italian Stock Exchange). He regularly writes on academic journals and has recently published
"Implementing Models in Quantitative Finance: Methods and Cases" (with G.Fusai), edited by Springer-Verlag in 2008.
E-mail: andrea.roncoroni@gmail.com
Web page: http://www45.essec.edu/faculty/andrea-roncoroni

Andrea Roncoroni

Uploaded by

Copyright:

Available Formats

Andrea Roncoroni

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Andrea Roncoroni

Uploaded by

Copyright:

Available Formats

Energy and Commodity Asian-Style Options under

Seasonal Data and Stochastic Volatility

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.