QUANTITATIVE FINANCE RESEARCH CENTRE QUANTITATIVE F

INANCE RESEARCH CENTRE

QUANTITATIVE FINANCE RESEARCH CENTRE

Research Paper 376 September 2016

Empirical hedging performance

on long-dated crude oil derivatives

Benjamin Cheng, Christina Sklibosios Nikitopoulos and Erik Schlögl

ISSN 1441-8010 www.qfrc.uts.edu.au

 Empirical hedging performance on long-dated crude oil derivatives
Benjamin Chenga , Christina Sklibosios Nikitopoulosa,∗, Erik Schlögla
a
University of Technology Sydney,
Finance Discipline Group, UTS Business School,
PO Box 123 Broadway NSW 2007, Australia

Abstract
This paper presents an empirical study on hedging long-dated crude oil futures options
with forward price models incorporating stochastic interest rates and stochastic volatility.
Several hedging schemes are considered including delta, gamma, vega and interest rate hedge.
Factor hedging is applied to the proposed multi-dimensional models and the corresponding
hedge ratios are estimated by using historical crude oil futures prices, crude oil option prices
and Treasury yields. Hedge ratios from stochastic interest rate models consistently improve
hedging performance over hedge ratios from deterministic interest rate models, an improve-
ment that becomes more pronounced over periods with high interest rate volatility, such as
during the GFC. An interest rate hedge consistently improves hedging beyond delta, gamma
and vega hedging, especially when shorter maturity contracts are used to roll the hedge
forward. Furthermore, when the market experiences high interest rate volatility and the
hedge is subject to high basis risk, adding interest rate hedge to delta hedge provides an
improvement, while adding gamma and/or vega to the delta hedge worsens performance.
Keywords: Stochastic interest rates; Delta hedge; Interest rate hedge; Long-dated crude
oil options;
JEL: C13, C60, G13, Q40

∗
Corresponding author
Email addresses: [email protected] (Benjamin Cheng),
[email protected] (Christina Sklibosios Nikitopoulos), [email protected]
(Erik Schlögl)
1. Introduction
Motivated by the debacle of the German company Metallgesellschaft (MG) at the end
of 1993, several research papers have investigated the methods and risks in hedging long-
dated over-the-counter forward commodity contracts by using short-dated contracts, typi-
cally short-dated futures. The investigation conducted in Edwards and Canter (1995) and
Brennan and Crew (1997) conclude that MG’s stack-and-roll hedging strategy was flawed
and exposed the company to significant basis risk.
Other papers in the literature that consider using multiple futures contracts to hedge
long-dated commodity commitments include Veld-Merkoulova and De Roon (2003), Bühler,
Korn, and Schöbel (2004) and Shiraya and Takahashi (2012). Veld-Merkoulova and De Roon
(2003) by proposing a term structure model of futures convenience yields, develop a strategy
that minimises both spot price risk and basis risk by using two futures contracts with differ-
ent maturities. Empirical results show that this two-futures strategy outperforms the simple
stack-and-roll hedge substantially in hedging performance. Bühler et al. (2004) empirically
demonstrate that the oil market is characterised by two pricing regimes - when the spot oil
price is high (low), the sensitivity of the futures price is low (high). They then propose a
continuous-time, partial equilibrium, two-regime model and show that their model implies
a relatively high (low) hedge ratios when oil prices are low (high). Shiraya and Takahashi
(2012) propose a mean reversion Gaussian model of commodity spot prices and futures
prices are derived endogenously. In their hedging analysis, by using 3 futures contracts with
different maturities to hedge the long-dated forward contract, they calculate the hedging
positions by matching their sensitivities to the different sources of uncertainty. They empir-
ically demonstrate that their Gaussian model outperforms the stack and roll model used by
Metallgesellschaft.
Trolle and Schwartz (2009) propose a multi-dimensional stochastic volatility model for
commodity derivatives featuring unspanned stochastic volatility. They show that adding op-
tions to the set of hedging instruments significantly improves hedging of volatility trades, such
as straddles, compared to using futures only as hedging instruments. Dempster, Medova, and
Tang (2008) propose a four-factor model for two spot prices and their convenience yields and
develop closed-form pricing and hedging formulae for options on spot and futures spreads
of commodity. Chiarella, Kang, Nikitopoulos, and Tô (2013) empirically demonstrate that
hump-shaped volatility specifications reduce the hedging error of crude oil volatility trades
(straddles) compared to the exponential volatility specification counterpart. However, all
these papers assume deterministic interest rates and it is unclear to what extent these mod-
els can provide adequate hedges for long-dated options positions. The research literature
on hedging long-dated commodity option positions is rather limited and this paper aims to
make a contribution by empirically investigating the hedging of crude oil futures options
with maturities up to six years.
We use the Light Sweet Crude Oil (WTI) futures and option dataset from the NYMEX1
spanning a 6-year period from January 2006 to December 2011. A call futures option matur-
ing on December 2011 is hedged from 3rd July, 2006 to 31st October, 2011. During this 6-year
period, a number of major events happened, such as the Global Financial Crisis (GFC) and

1
The database has been provided by CME.

2
the Arab Spring and Libyan revolution that had significant impact on spot crude oil prices,
crude oil futures and its options, as well as interest rates. The gross domestic product growth
rates of China and India have increased exponentially during the decade of the 2000s. To
support the growth of the economy, the consumption of energy also increased. In particular,
China’s demand of crude oil grew at a 7.2% annual logarithmic rate between 1991 and 2006.
This phenomenal growth rate, among other factors, increased the demand in crude oil. On
the supply side, Saudi Arabia, the biggest oil exporter in the world, reduced its crude oil
production in 2007. Consequently, crude oil prices increased sharply, sending the price to a
high of US$145 per barrel on 3rd July, 2008, which was immediately followed by a spectacular
collapse in prices and by the end of 2008, the spot crude oil price was below US$40 per barrel
(see Hamilton (2009) and Hamilton (2008)). The US Treasury yields were above 4.5% before
the July 2007 and had decreased steadily to nearly zero by the end of 2008, after the GFC.
The empirical analysis in this paper considers a position in a long-dated call futures
option with a maturity in December 2011. This option is hedged over five years using
several hedging schemes such as delta hedge, delta-vega hedge and delta-gamma hedge.
For the purpose of comparison, two models are used to compute the suitable hedge ratios;
one with stochastic interest rates as modelled in Cheng, Nikitopoulos, and Schlögl (2015)
and one with deterministic interest rates fitted to a Nelson and Siegel (1987) curve. These
models are estimated from historical crude oil futures and option prices and Treasury yields
by using the extended Kalman filter, see Cheng, Nikitopoulos, and Schlögl (2016a) for a
similar estimation application. The hedge ratios for delta-interest rate (delta-IR), delta-
vega-interest-rate (delta-vega-IR), and delta-gamma and delta-gamma-interest-rate (delta-
gamma-IR) hedges are also computed but they are limited to the stochastic interest rate
model. The hedge ratios are derived using the factor-hedging methodology, the effectiveness
of which has been analysed in Cheng, Nikitopoulos, and Schlögl (2016b).
From this analysis, several results have emerged. Firstly, when an interest rate hedge is
added to the delta, gamma and vega hedge, there is a consistent improvement to the hedging
performance, especially when shorter maturity contracts are used to roll the hedge forward
(thus more basis risk is present). Secondly, because of the high interest rate volatility, in-
terest rate hedging was more important during the GFC than in recent years. Over periods
of high interest rate volatility (for instance, pre-GFC and during GFC) and when shorter
maturity hedging contracts are used, the delta-IR hedge consistently improves hedging per-
formance compared to delta hedge, while adding a gamma or vega hedge to the delta hedge
worsens hedging performance. Thirdly, due to lower basis risks, using hedging instruments
with maturities closer to the maturity of the option to be hedged reduces the hedging error.
Fourthly, the hedging performance from the stochastic interest rate model is consistently
better than the hedging performance from the deterministic interest rate model, with the
effect being more pronounced during the GFC. However, there is only marginal improve-
ment over the deterministic interest rate model during pre-crisis period and no noticeable
improvement after 2010.
The remainder of the paper is structured as follows. Section 2 presents the application
of the factor hedging methodology on the three-dimensional stochastic volatility–stochastic
interest rate forward price model developed in Cheng et al. (2015). It also derives the hedge
ratios for a variety of hedging schemes including delta, delta-IR, delta-vega, delta-vega-IR,
delta-gamma and delta-gamma-IR. Section 3 describes the methodology to assess hedging

3
performance on long-dated crude oil futures options, including the details of the dataset
used. Section 4 presents the empirical results and discusses their implications. Section 5
concludes.

2. Factor hedging for a stochastic volatility–stochastic interest rate model

Factor hedging is a broad hedging method that allows one to hedge simultaneously mul-
tiple factors and multiple dimensions impacting the forward curve of commodities, the in-
stantaneous volatility component and the interest rate variation and subsequently the value
of commodity derivatives portfolios. By considering the 𝑛-dimensional stochastic volatility
and 𝑁-dimensional stochastic interest rate model developed in Cheng et al. (2015), to hedge
the 𝑛-dimensional forward rate risks (that is 𝑊𝑖 (𝑡) for 𝑖 = 1, 2, . . . , 𝑛) it is necessary to use 𝑛
number of hedging instruments such as futures contracts. To further hedge the 𝑛 number of
volatility risks (that is 𝑊𝑖𝜎 (𝑡) for 𝑖 = 1, 2, . . . , 𝑛), it is required to use an additional 𝑛 number
of volatility-sensitive hedging instruments such as futures options. Since the proposed model
considers stochastic interest rates, the 𝑁-dimensional interest rate risks (that is 𝑊𝑖𝑟 (𝑡) for
𝑖 = 1, 2, . . . , 𝑁) should be hedged by using 𝑁 number of interest-rate-sensitive contracts
such as bonds. For completeness, we present next the model used to compute the required
hedge ratios.

2.1. Model
The model assumes that the time 𝑡-futures price 𝐹 (𝑡, 𝑇, 𝜎𝑡 ) of a commodity, for delivery
at time 𝑇 , evolves as follows:
𝑛
𝑑𝐹 (𝑡, 𝑇, 𝜎𝑡 ) ∑ 𝐹
= 𝜎𝑖 (𝑡, 𝑇, 𝜎𝑡 )𝑑𝑊𝑖𝑥 (𝑡), (1)
𝐹 (𝑡, 𝑇, 𝜎𝑡 ) 𝑖=1

where 𝜎𝑡 = {𝜎𝑖 (𝑡), . . . , 𝜎𝑛 (𝑡)} and for 𝑖 = 1, 2, . . . , 𝑛,

𝑑𝜎𝑖 (𝑡) = 𝜅𝑖 (𝜎 𝑖 − 𝜎𝑖 (𝑡))𝑑𝑡 + 𝛾𝑖 𝑑𝑊𝑖𝜎 (𝑡).

Also,
𝑁
∑
𝑟(𝑡) = 𝑟(𝑡) + 𝑟𝑗 (𝑡)
𝑗=1

where, for 𝑗 = 1, 2, . . . , 𝑁,

𝑑𝑟𝑗 (𝑡) = −𝜆𝑗 (𝑡)𝑟𝑗 (𝑡)𝑑𝑡 + 𝜃𝑗 𝑑𝑊𝑗𝑟 (𝑡).

The functional form of the volatility term structure 𝜎𝑖𝐹 (𝑡, 𝑇, 𝜎𝑡 ) is specified as follows:

𝜎𝑖𝐹 (𝑡, 𝑇, 𝜎𝑡 ) = (𝜉0𝑖 + 𝜉𝑖 (𝑇 − 𝑡))𝑒−𝜂𝑖 (𝑇 −𝑡) 𝜎𝑖 (𝑡)

4
with 𝜉0𝑖 , 𝜉𝑖 , and 𝜂𝑖 ∈ ℝ for all 𝑖 ∈ {1, 2, . . . , 𝑛}. The correlation structure of the associated
Wiener processes is given by
{
𝜌𝑥𝜎
𝑖 𝑑𝑡, if 𝑖 = 𝑗,
𝑑𝑊𝑖𝑥 (𝑡)𝑑𝑊𝑗𝜎 (𝑡) =
0, otherwise
{
𝜌𝑥𝑟 ˆ
𝑖 𝑑𝑡, if 𝑗 = 1,
𝑑𝑊𝑖𝑥 (𝑡)𝑑𝑊ˆ𝑗𝑟 (𝑡) = (2)
0, otherwise
{
𝜌𝑟𝜎 ˆ
𝑖 𝑑𝑡, if 𝑗 = 1,
𝑑𝑊𝑖𝜎 (𝑡)𝑑𝑊ˆ𝑗𝑟 (𝑡) =
0, otherwise

for 𝑖 ∈ {1, . . . , 𝑛}, 𝑗 ∈ {1, . . . , 𝑛}, ˆ𝑖 ∈ {1, . . . , 𝑁} and ˆ𝑗 ∈ {1, . . . , 𝑁}.

2.2. Delta hedging

For an 𝑛-dimensional model, factor delta hedging requires 𝑛 hedging instruments such
as futures contracts with different maturities, denoted by 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 ), 𝑗 = 1, . . . , 𝑛. Let
Υ(𝑡, 𝑇𝑀 , 𝜎𝑡 ) be the value of the futures option to be hedged and 𝑇𝑀 be its maturity. Let
ΔΥ𝛿𝐻,𝑖 and ΔΥ𝛿𝑖 denote the change in price of the portfolio and the change in the price of
the option on futures (which is the target to be hedged) by the 𝑖th shock of the uncertainty
in the futures curve (that is 𝑑𝑊𝑖𝑥 (𝑡)) respectively and 𝛿𝑗 be the position corresponding to
the 𝑗 th hedging instrument, we have:
△
ΔΥ𝛿𝐻,𝑖 = ΔΥ𝛿𝑖 + 𝛿1 Δ𝐹𝑖 (𝑡, 𝑇1 , 𝜎𝑡 ) + 𝛿2 Δ𝐹𝑖 (𝑡, 𝑇2 , 𝜎𝑡 ) + . . . + 𝛿𝑛 Δ𝐹𝑖 (𝑡, 𝑇𝑛 , 𝜎𝑡 ). (3)
For an 𝑛-dimensional model with 𝑛 amount of hedging instruments the set of equations in
equation (3) forms a system of 𝑛 linear equations which can be solved exactly by matrix
inversion. However, from numerical results, some of the values of 𝛿𝑖 produced by using this
method may be unnecessarily large and consequently lead to very large profit and loss of the
hedging portfolio. We use an alternative and more general method by simply minimising the
sum of the squared hedging errors, ΔΥ𝛿𝐻,𝑖 :
{ }
minimise (ΔΥ𝛿𝐻,1 )2 + (ΔΥ𝛿𝐻,2 )2 + . . . + (ΔΥ𝛿𝐻,𝑛 )2 (4)
𝛿1 ,...,𝛿𝑛

with a constraint that 𝛿𝑖 ≤ ℓ.2 The changes in the value of the hedging instruments
Δ𝐹𝑖 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) can be approximated by the discretisation of the stochastic differential equa-
tion, as follows:3
( ) ( ( ))
Δ𝐹𝑖 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) = 𝐹𝑖 (𝑡, 𝑇𝑗 , 𝜎𝑡 )𝜎𝑖 (𝑡, 𝑇𝑗 )Δ𝑊𝑖 (𝑡) − 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 )𝜎𝑖 (𝑡, 𝑇𝑗 ) − Δ𝑊𝑖 (𝑡)
= 2𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 )𝜎𝑖 (𝑡, 𝑇𝑗 )Δ𝑊𝑖 (𝑡).

2
ℓ = 1 seems to be a good choice to give low hedging errors and the hedging results are stable by varying
ℓ slightly.
3
This first order approximation is sufficient for the study at hand, because any additional accuracy on
the calculation of the hedge would be drowned out by the fact that the model is only an approximation of
the empirical reality.

5
Lastly we can calculate the change in the option on futures ΔΥ𝛿𝑖 by the 𝑖th shock as:
( ) ( )
ΔΥ𝛿𝑖 (𝑡, 𝑇𝑗 ) = Υ 𝐹𝑖,𝑈 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) − Υ 𝐹𝑖,𝐷 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) , (5)

where 𝐹𝑖,𝑈 and 𝐹𝑖,𝐷 denote the ‘up’ and ‘down’ moves of the price of the underlying hedging
instrument (e.g. futures price):

𝐹𝑖,𝑈 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) = 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) + 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 )𝜎𝑖 (𝑡, 𝑇𝑗 )Δ𝑊𝑖 (𝑡) (6)

𝐹𝑖,𝐷 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) = 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 ) − 𝐹 (𝑡, 𝑇𝑗 , 𝜎𝑡 )𝜎𝑖 (𝑡, 𝑇𝑗 )Δ𝑊𝑖 (𝑡). (7)

We note that the change in 𝑊𝑖 (𝑡) is normally distributed, so an ‘up’ or ‘down’ move is just
a notation. An ‘up’ move does not necessarily mean Δ𝑊𝑖 (𝑡) > 0.

2.3. Delta-Vega hedging

The condition in equation (4) only immunises small risks generated from the uncertainty
that directly impacts the underlying asset, for instance, the futures curve. It cannot mitigate
risks originating from an instantaneous volatility which may be stochastic. In order to
account for the risks of a stochastic volatility process, an additional number of 𝑛 futures
options may be used as hedging instruments to simultaneously immunise both volatility and
futures price risks.4 Let 𝑣𝑖 be the number of futures options that have values of Ψ(𝑡, 𝑇𝑗 , 𝜎𝑡 )
for 𝑗 = 1, . . . , 𝑛. The number of futures options 𝑣1 , 𝑣2 , . . . , 𝑣𝑛 are determined similarly to
equation (4) such that the overall changes in the value of the hedged portfolio, due to
volatility risk, are minimised. Setting up the hedged portfolio, we have:
△
ΔΥ𝜎𝐻,𝑖 = ΔΥ𝜎𝑖 + 𝑣1 ΔΨ𝜎𝑖 𝐹 (𝑡, 𝑇1 ), 𝑡, 𝑇1 + 𝑣2 ΔΨ𝜎𝑖 𝐹 (𝑡, 𝑇2 ), 𝑡, 𝑇2 +
( ) ( )

. . . + 𝑣𝑛 ΔΨ𝜎𝑖 𝐹 (𝑡, 𝑇𝑛 ), 𝑡, 𝑇𝑛 ,
( )
(8)
( ) ( ) ( )
where ΔΨ𝜎𝑖 𝐹 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 = Ψ 𝐹 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 , 𝜎𝑡𝑖𝑈 − Ψ 𝐹 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 , 𝜎𝑡𝑖𝐷 . 𝜎𝑡𝑖𝑈 represents
the vector of stochastic volatility processes where the 𝑖th element 𝜎𝑖 (𝑡) has been shocked by
an ‘up’ movement in 𝑊𝑖𝜎 (𝑡), denoted by 𝜎𝑖,𝑈 and it can be obtained as follows:

𝜎𝑖,𝑈 = 𝜎𝑖 (𝑡) + 𝜅𝑖 𝜎 𝑖 − 𝜎𝑖 (𝑡) Δ𝑡 + 𝛾𝑖 Δ𝑊𝑖𝜎 (𝑡).

( )

𝜎𝑡𝑖𝐷 is interpreted in a similar way:

𝜎𝑖,𝐷 = 𝜎𝑖 (𝑡) + 𝜅𝑖 𝜎 𝑖 − 𝜎𝑖 (𝑡) Δ𝑡 − 𝛾𝑖 Δ𝑊𝑖𝜎 (𝑡).

( )

The hedging portfolio consisting of one target option to be hedged and 𝑛 additional options,
with number of contracts 𝑣𝑖 , which are determined by a similar minimisation procedure to
the one outlined in equation (4). This portfolio, however, is not delta-neutral. To make this
hedging portfolio also delta-neutral, an additional 𝑛 number of futures contracts with number
of positions 𝛿𝑖 are included in the vega-neutral portfolio and then the 𝛿𝑖 such that the portfolio

4
Note that we are not using a futures option as a hedging instrument that matches exactly both the
maturity and strike of the target futures option that we are hedging because if we can use an exactly
matching option to hedge then the hedged portfolio would have a value of zero at all times.

6
is both delta- and vega-neutral is determined. Thus a portfolio that is simultaneously delta-
and vega-neutral should satisfy:
△
Υ𝛿𝜎
𝐻 = Υ + 𝛿1 𝐹 (𝑡, 𝑇1 ) + 𝛿2 𝐹 (𝑡, 𝑇2 ) + . . . + 𝛿𝑛 𝐹 (𝑡, 𝑇𝑛 )+ (9)
( ) ( ) ( )
𝑣1 Ψ 𝐹 (𝑡, 𝑇1 ), 𝑡, 𝑇1 + 𝑣2 Ψ 𝐹 (𝑡, 𝑇2 ), 𝑡, 𝑇2 + . . . + 𝑣𝑛 Ψ 𝐹 (𝑡, 𝑇𝑛 ), 𝑡, 𝑇𝑛 .

Since the ‘up’ and ‘down’ shocks of the stochastic volatility process 𝜎𝑡 have no impact on
futures prices, we can determine 𝑣1 , . . . , 𝑣𝑛 by applying 𝑛 volatility shocks to equation (8)
to calculate ΔΥ𝜎𝐻,𝑖 and minimise its squared hedging errors as shown in equation (4). The
next step is to determine 𝛿𝑖 , . . . , 𝛿𝑛 by applying 𝑛 shocks to equation (9) (with 𝑣𝑖 determined
previous and are held as constants) and then minimising its squared errors.

2.4. Delta-Gamma hedging

Delta hedging can provide sufficient protection against small price changes, but not
against larger price changes. To hedge larger price changes, a second order hedging is
required to take into account the curvature of option prices, in other words, gamma hedging.
Gamma measures the rate of change of the option’s delta with respect to the underlying,
which is equivalent to the second derivative of the option price with respect to the underlying.
If gamma is low, re-balancing of the portfolio infrequently may be sufficient because the delta
of the option does not vary much when the underlying moves. However, if gamma is high, it
is necessary to re-balance the portfolio frequently because when the underlying moves, the
delta of the option is not accurate anymore. Hence, it cannot be used as an efficient hedge
against market risk. The hedging portfolio is the same as the portfolio in equation (9) but
we use a different method to calculate the positions 𝑣1 , 𝑣2 , . . . , 𝑣𝑛 . To determine the amount
of positions, we construct the change of the hedging portfolio given by:
△
ΔΥ𝛾𝐻,𝑖 = ΔΥ𝛾𝑖 + 𝑣1 ΔΨ𝛾𝑖 𝐹 (𝑡, 𝑇1 ), 𝑡, 𝑇1 + 𝑣2 ΔΨ𝛾𝑖 𝐹 (𝑡, 𝑇2 ), 𝑡, 𝑇2 +
( ) ( )

. . . + 𝑣𝑛 ΔΨ𝛾𝑖 𝐹 (𝑡, 𝑇𝑛 ), 𝑡, 𝑇𝑛 ,
( )
(10)

where

ΔΨ𝛾𝑖 𝐹 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 = Ψ 𝐹𝑖,𝑈 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 , 𝜎𝑡 − 2Ψ 𝐹 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 , 𝜎𝑡

( ) ( ) ( )
( )
+ Ψ 𝐹𝑖,𝐷 (𝑡, 𝑇𝑗 ), 𝑡, 𝑇𝑗 , 𝜎𝑡 .

𝐹𝑖,𝑈 and 𝐹𝑖,𝐷 are the ‘up’ and ‘down’ moves of the futures price as defined in equation (6)
and equation (7). )The change of the target futures option ΔΥ𝛾𝑖 is calculated similarly to
ΔΨ𝛾𝑖 𝐹 (𝑡, 𝑇1 ), 𝑡, 𝑇1 . The rest of the steps to determine 𝑣1 , . . . , 𝑣𝑛 are exactly the same as
(

in subsection 2.2.

2.5. Delta-IR, Delta-Vega-IR and Delta-Gamma-IR hedging

To immunise 𝑁 risks from the interest rate shocks, 𝑁 additional bond contracts with
different maturities are required. All together, to apply a delta-vega-IR factor hedge , we
need 𝑛 number of futures with different maturities, 𝑛 number of options on futures with
different maturities and 𝑁 number of bond contracts with different maturities. Since the
shocks from the stochastic volatility process have no impact on bonds and futures, the first

7
step is to determine 𝑣1 , . . . , 𝑣𝑛 , which are exactly the same as in equation (8). The shocks
from the interest rate process have an impact on futures options as well as on bonds, so
the next step is to determine the number of bond contracts by minimising the sum of the
squared errors of the hedging portfolio consisting of options and bonds. The construction
of a delta-gamma-IR factor hedge and the method to determine the number of the bond
contracts follow the same procedure as the delta-vega-IR factor hedge and the detail is
therefore omitted. To construct a delta-IR hedge we need 𝑛 number of futures with different
maturities and 𝑁 number of bond contracts with different maturities. Options are not used
as hedging instruments to construct this delta-IR hedge.

3. Hedging futures options

Crude oil futures options with maturities beyond five years are hedged by using the
above mentioned hedging schemes. The three-dimensional version of the stochastic volatility–
stochastic interest rate forward price model (1), (i.e., 𝑛 = 𝑁 = 3) is employed to compute the
required hedging ratios. For a comparison of the hedging performance between the stochastic
and the deterministic interest rate model, we also consider the deterministic interest rate
counterpart of this three-dimensional model, as fitted to a Nelson and Siegel (1987) curve,
see Cheng et al. (2016a) for details. To capture the impact of stochastic interest rates
in the hedging performance, hedging under the deterministic interest rate specifications is
compared to the stochastic interest rate specifications. Furthermore, to assess the hedging
performance of these schemes in terms of the choice of the hedging instruments, several
scenarios are considered for varying maturities of the hedging instruments.

3.1. Methodology
A 6-year crude oil dataset of futures and options is used in the investigations starting from
January 2006 till December 2011. Crude oil futures options that mature in December are
considered as the target option to hedge because December contracts are the only contracts
with maturities over five years in our dataset. June contracts are also liquid but they have
a maximum maturity of only four years in the dataset. There are only a few futures options
with a maturity of December 2011 and with non-zero open interest that persist throughout
the whole 5-year period. Thus the futures option with a strike of $62 and a maturity of
December 2011 is used as the target option to be hedged.
The 6-year crude oil dataset of futures and options is sub-divided into 12 half-year periods.
The first sub-period is between January 2006 and June 2006, the second is between July 2006
and December 2006, the third sub-period is between January 2007 and June 2007 and so
on. The last sub-period, which is the 12th , is between July 2011 and October 2011. Since a
crude oil option on futures contract ceases trading around the 17th of a month prior to the
maturity month of the futures option, a December contract would ceases trading around the
17th of November,5 and in the dataset we exclude options with maturities under 14 calendar
days, the hedging performance analysis is terminated at the end of October 2011. For each
of the 12 sub-periods, a set of model parameters is estimated for the stochastic interest rate

5
See CME’s Crude Oil Futures Contract Specs and Crude Oil Options Contract Specs.

8
model and for the deterministic interest rate model using Kalman filter, as outlined in Cheng
et al. (2016a).
We use model parameters from previous sub-period together with the current state vari-
ables to estimate the hedge ratio for the current sub-period. For example, during the hedging
analysis between July 2006 and December 2006, the model parameters estimated in the first
sub-period are used together with the up-to-date state variables to derive the hedging ratios.
This procedure is followed until October 2011. The idea of this is that out-of-sample model
parameters are used when we derive the hedge ratios, i.e., the hedge ratios only use market
information which is available at the time for which they are constructed.
To compare differences in the hedging performance of a long-dated option betweenus-
ing longer maturity hedging instruments and shorter maturity hedging instruments, four
scenarios are investigated where the hedging instruments have varying maturities. For the
three-dimensional model used, as illustrated in Section 2, to construct a delta-hedged port-
folio, it is necessary to use three hedging instruments, thus futures contracts with three
different maturities. In theory any three different maturities would be fine. But in practice,
the liquidity of futures contracts with a maturity of more than a year decreases significantly
with an exception of futures maturing in June or December. Only futures contracts maturing
in December are available with a maturity of more than 3 years. So in this empirical analysis,
June and December futures contracts with maturities of less than 3 years are considered,
and only December contracts with a maturity longer than 3 years.
In the first scenario, hedging instruments with the three most adjacent June or December
contracts are used. For instance, if the trading date is 15th of July, 2006, then hedging
instruments with maturities of December 2006, June 2007 and December 2007 are used.
By the end of October 2006, the December 2006 contract gets rolled over to June 2008
(and keeping the other two contracts), and so on. The second scenario is similar to the
first scenario except that in the beginning of the hedging period the first hedging contract
matures in December 2007, followed by June 2008 and December 2008. As the trading date
gets to the end of October 2007, the December 2007 contract gets rolled over to June 2009,
and so on. The third and fourth scenarios follow this idea. Table 1 shows the maturities
of contracts used in these four scenarios. To construct a delta-vega or delta-gamma hedged
portfolio, three additional options on futures are required. The maturities of these options
are selected according to Table 1. Their strikes are selected based on a combination of
liquidity and moneyness. We first filter out near-the-money options with strikes that are
±15% away from the at-the-money strike and then out of these strikes, we select the strike
with the highest open interest.

3.2. Monte Carlo simulation

It is unclear what is the best approach to choose the size of the ‘up’ and ‘down’ moves
(Δ𝑊𝑖 (𝑡), Δ𝑊𝑖𝜎 (𝑡) and Δ𝑊𝑗𝑟 (𝑡)) required in the factor hedging, as discussed in Section 2.
Along the lines of Chiarella et al. (2013), the size of the moves can be determined by using
a Monte Carlo simulation approach.
We first consider the delta hedging scheme outlined in Section 2.2. Let 𝑘 be the index
of the Monte Carlo simulation with 1000 iterations (that is, 𝑘 = 1, 2, . . . , 1000). Let 𝑡𝑑 be

9
 Scenario 1 Scenario 2 Scenario 3 Scenario 4
Dec-2006 Dec-2007 Dec-2008 Dec-2009
Jun-2007 Jun-2008 Jun-2009 Dec-2010
Dec-2007 Dec-2008 Dec-2009 Dec-2011
Jun-2008 Jun-2009 Jun-2010 Jun-2012
Dec-2008 Dec-2009 Dec-2010 Dec-2012
Jun-2009 Jun-2010 Jun-2011 Jun-2013
Dec-2009 Dec-2010 Dec-2011
Jun-2010 Jun-2011 Jun-2012
Dec-2010 Dec-2011 Dec-2012
Jun-2011 Jun-2012 Jun-2013
Dec-2011 Dec-2012
Jun-2012 Jun-2013
Dec-2012
Jun-2013

Table 1: Different maturities of contracts used in the four scenarios. The table
displays different maturities of futures / options contracts used in these four scenarios. For instance,
in Scenario 4 and assuming delta-hedging, a call option on futures is hedged in 3rd July 2006 by
using 3 futures contracts with maturities of Dec 2009, Dec 2010 and Dec 2011. June contracts are
not included here because on 3rd July 2006, futures with a maturity of June 2010 are not available.
As the call option approaches the end of October 2009, the futures with a maturity of Dec 2009
gets rolled over to June 2012 while the other two futures contracts with maturities of Dec 2010 and
Dec 2011 remain.

the trading day and 𝑑 = 1, 2, . . . , 1333 be the index of the trading day.6 At 𝑡1 , we generate
𝑛 number of independent Δ𝑊𝑖 by drawing 𝑛 random samples from a normally distributed
1
random number generator with mean of 0 and variance of 252 . Using these 𝑛 samples and
following the steps outlined in Section 2.2, we can construct a hedging portfolio consisting the
option to be hedged and 𝑛 number of futures with different maturities and with hedge ratios
𝛿1𝑘 , . . . , 𝛿𝑛𝑘 . We add the index 𝑘 in these hedge ratios to explicitly show their dependencies on
the 𝑛 random samples realised in iteration 𝑘. The profit and loss (P/L) of the delta-hedged
portfolio on trading day 𝑡2 is defined as:
△
P/L𝛿,𝑘 𝛿 𝑘 𝑘 𝑘
2 = ΔΥ2 + 𝛿1 Δ𝐹 (𝑡2 , 𝑇1 ) + 𝛿2 Δ𝐹 (𝑡2 , 𝑇2 ) + . . . + 𝛿𝑛 Δ𝐹 (𝑡2 , 𝑇𝑛 ), (11)

where Δ𝐹 (𝑡2 , 𝑇1 ) = 𝐹 (𝑡2 , 𝑇1 ) − Δ𝐹 (𝑡1 , 𝑇1 ) represents the difference of market quoted futures
prices between 𝑡1 and 𝑡2 7 and ΔΥ𝛿2 represents the difference of the market quoted option

6
Since we start the hedging scheme from July 2006, we have 𝑡1 representing the trading day of the 3rd of
July, 2006, which is a Monday. We note that the 4th of July, 2006 is a public holiday (Independence Day) in
the United States, so 𝑡2 represents the trading day of the 5th of July, 2006. 𝑡1333 represents the last trading
day under consideration which is on the 31st of October, 2011.
7
We drop the dependency of 𝜎𝑡 in these futures prices because these are quoted prices rather than model
prices.

10
prices from 𝑡1 to 𝑡2 . In each of the 1000 simulations, we have a total of 1332 P/Ls (P/L𝛿,𝑘
2
to P/L𝛿,𝑘
1333 ) and we define the root-mean-square errors (RMSEs) as follow:
v
u
u 1 ∑ 1333
)2
𝛿,𝑘
P/L𝛿,𝑘
(
RMSEtotal = ⎷
𝑑
1332
𝑑=2
v
u
u1 ∑ 51
)2
𝛿,𝑘
P/L𝛿,𝑘
(
RMSE1 = ⎷ 𝑑
50 𝑑=2
v
u
u1 ∑ 101
)2
𝛿,𝑘
P/L𝛿,𝑘
(
RMSE2 = ⎷ 𝑑
50 𝑑=52
..
.
v
u
u1 ∑ 1333
)2
RMSE𝛿,𝑘 P/L𝛿,𝑘
(
27 = ⎷
𝑑 .
32 𝑑=1302

To calculate the total average of the RMSEs, we simply take the average over 1000 simula-
tions. That is:
1000
1 ∑
RMSE𝛿total = RMSE𝛿,𝑘
total
1000 𝑘=1
1000
1 ∑
RMSE𝛿1 = RMSE𝛿,𝑘
1
1000 𝑘=1
1000
1 ∑
RMSE𝛿2 = RMSE𝛿,𝑘
2
1000 𝑘=1
..
.
1000
1 ∑
RMSE𝛿27 = RMSE𝛿,𝑘
27 .
1000 𝑘=1

In each path, seven hedging schemes 8 are considered, namely: unhedged, delta, delta-IR,
delta-vega, delta-vega-IR, delta-gamma and delta-gamma-IR. The results are the RMSEs of
the profits and losses of the hedging portfolios.
So in each of the 1000 simulations, seven RMSEs represent the variability of the hedging
portfolios under different hedging schemes for the whole six-year period.9 To illustrate the

8
Strictly speaking, one unhedged portfolio with just the option itself and six hedging portfolios are
considered.
9
Note that the hedging schemes are all run on the same underlying empirical data set covering 1333 days
of daily price history, and as explained above, the Monte Carlo simulations only serve to provide the 𝑑𝑊
“bumps” used to calculate the positions in each hedge instrument in the factor hedge.

11
variability of the hedging portfolios at various times over the six-year period, the whole
period from July 2006 to the end of 2011 is sub-divided into 27 sub-periods. Each of the
sub-periods consists 50 trading days, except the last one. So now, given a scenario, the
RMSE represents the RMSE of the profits and losses of the hedging portfolio per path per
sub-period per hedging scheme. Then the average of the RMSE over 1000 simulations for
each hedging scheme is computed, for each of the scenarios. Thus, we have 27 by 7 RMSEs,
as shown in Figure 3.

4. Empirical results
The results of our empirical hedging application are depicted in Figure 3 and Figure 4,
where the RMSE of the unhedged portfolio are compared with the RMSE of the hedged port-
folios for the six different hedging schemes,10 under the proposed three-dimensional stochastic
volatility–stochastic interest rate model. Figure 3 considers the hedging instruments of Sce-
nario 1 and 2, where short-dated futures contracts are considered. Figure 4 considers the
hedging instruments of Scenario 3 and 4, where longer-dated contracts are included (see
Table 1 for exact contract maturities). Thus, the results are separated into 27 groups of 7
bars. The first group, labelled ‘Aug06’, represents the RMSE over a 50-trading day period
from 5th of July, 2006 to 13th of September, 2006 inclusive. The second group, without
label, represents the RMSE over a 50-trading day period from 14th of September, 2006 to
24th of November, 2006 inclusive. We continue similarly, so the last group, labelled ‘Oct11’,
represents the RMSE over a 32-trading day period from 15th of September, 2011 to 31st of
October, 2011. Firstly, we compare the overall hedging error between hedged positions and
unhedged positions. Then we evaluate the contribution of stochastic interest rate models on
hedging performance compared to deterministic interest rate models. Finally, we discuss the
performance of each hedging scheme over the course of these six years.

4.1. Unhedged vs. hedged positions

Table 2 reports the improvement in the overall hedging error for a range of hedging
schemes compared to the unhedged futures option position. Several conclusions can be
drawn. Firstly, an interest rate hedge incrementally but consistently improves hedging per-
formance when it is added to delta, delta-vega and delta-gamma hedging. When longer
maturity futures contracts are used as hedging instruments (thus lower basis risk is present),
such as in Scenario 4, then the improvement is about 1%, while the improvement nearly
doubles when shorter maturity futures contracts are used as hedging instruments. When
short-dated contracts are used as hedging instruments, it is necessary to more frequently
rollover the hedge and as a result, a higher basis risk is present. Consequently, part of
the basis risk is essentially managed by hedging the interest rate risk. This is in line with
what would be expected theoretically, since the difference between futures prices for different
maturities is in part due to interest rates (though convenience yields would also play a role).
Secondly, longer maturity hedging instruments (Scenario 4) consistently provide a better
hedging performance across all hedging schemes compared to the shorter maturity hedging

10
The six hedging schemes are: delta, delta-IR, delta-vega, delta-vega-IR, delta-gamma and delta-gamma-
IR.

12
 Three-dimensional stochastic volatility–stochastic interest rate model
Scenario delta delta-IR delta-vega delta-vega-IR delta-gamma delta-gamma-IR
1 74.16% 75.08% 73.95% 74.40% 74.68% 75.56%
2 74.88% 75.66% 74.78% 75.68% 75.42% 76.42%
3 76.68% 77.33% 76.52% 77.46% 76.98% 77.64%
4 82.87% 83.02% 84.18% 84.67% 85.15% 85.77%

Table 2: Improvement of hedging schemes over unhedged positions. This table shows
the percentage improvement of the RMSE of the hedged position over the unhedged portfolio for
a range of hedging schemes.

Three-dimensional model with Stochastic IR (s) and Deterministic IR (d)

Scenario delta (s) delta (d) delta-vega (s) delta-vega (d) delta-gamma (s) delta-gamma (d)
1 74.16% 72.95% 73.95% 72.87% 74.68% 73.85%
2 74.88% 73.55% 74.78% 73.73% 75.42% 74.68%
3 76.68% 75.33% 76.52% 75.43% 76.98% 75.58%
4 82.87% 81.23% 84.18% 82.83% 85.15% 83.80%

Table 3: Comparison between stochastic interest rate model and deterministic

interest rate model. This table shows the percentage improvement of the RMSE over the
unhedged portfolio between the stochastic interest rate model and deterministic interest rate model.
In delta(s) and delta(d) columns, only futures are used as hedging instruments with hedge ratios
derived from the stochastic interest rate model and deterministic interest rate model, respectively.
Interest-rate hedging is not included (i.e., no bonds are used to hedge). Similarly apply for delta-
vega and delta-gamma hedges.

instruments (e.g. Scenario 1). With the shorter maturity futures contracts, the improve-
ment across the different schemes is ranging between 73.95% and 75.56%. With the longer
maturity futures contracts, the improvement across the different schemes is more sound,
ranging between 82.87% and 85.77%, with the delta-gamma-IR hedge outperforming all
other schemes. Thirdly, delta-gamma hedge tends to provide a better hedge compared to
delta-vega hedge. Thus when hedging long-dated options positions, gamma hedge seems to
be more efficient compared to vega hedge, with delta hedging being the least efficient.

4.2. Deterministic vs. stochastic interest rates

Table 3 compares the hedging error improvement over the unhedged positions for a range
of hedging schemes by using models with deterministic and stochastic interest rates. When
interest rate hedge is not considered additional to the delta, vega and gamma hedges, then
the option position is hedged by using hedge ratios from the deterministic interest rate and
the stochastic interest rate model specifications. Overall, the stochastic interest rate model
improves incrementally the hedging performance compared to the deterministic interest rate
model. The effect is more pronounced when the shorter maturity hedging instruments are
used, predominantly due to the additional basis risk.

13
4.3. Comparison of hedging schemes
Figure 5 and Figure 6 presents the results of Figure 3 and Figure 4 by omitting the errors
from the unhedged portfolio. A thorough inspection of the hedging performance of the six
different hedging schemes reveals different patterns over the course of these six years.
Figure 1 depicts the annualised monthly standard deviations of the US Treasury yields
and Figure 2 plots futures prices for four December contracts and the annualised standard
deviation of these log- futures returns, between 2005 and 2013. These figures reveal that
there is a period of high volatility of interest rates, high volatility of crude oil futures prices
till the beginning of 2008, then there is a period of extreme variation in yields and futures
prices between 2008 and 2009 associated with the GFC and then after that a period of
relatively low variation in interest rates, especially the 1-year and 2-year yields. Thus the
pre-GFC period, the GFC period and the post-GFC period, as specified above, have been
used to discuss the results.
As we have already discussed in Section 4.1, an interest rate hedge consistently provides
an improvement in the hedging performance, when combined with the other hedging schemes
including delta, delta-gamma and delta-vega hedging. This improvement is more sound when
the futures option is hedged with shorter maturity contracts, and weakens as the maturity
of the hedging instruments increases (thus lower basis risk is involved). Additionally, during
the GFC period, the improvement of including IR hedge to the other hedging schemes such
as delta, delta-vega and delta-gamma is substantial. In the pre-GFC period, yields were
relatively more volatile compared to the post-GFC period. Accordingly, the inclusion of IR
hedge has a marginal improvement in the pre-GFC period, and no perceptible improvement
in the post-GFC period. These observations are consistent with the numerical findings
in Cheng et al. (2016b), where it is demonstrated that during periods of low interest rate
volatility, stochastic interest rates marginally improve hedging performance. However, during
the high interest rate volatility period of the GFC (see Figure 1), adding IR hedge over delta,
delta-vega and delta-gamma hedging, considerably reduces the hedging errors.
Furthermore, delta-IR hedge outperforms consistently and substantially all other hedging
schemes, when the market experiences high interest rate volatility and the hedge is subject
to high basis risk (e.g. in scenario 1, where shorter maturity hedging instruments are con-
sidered, see top panel of Figure 5). In particular, during GFC period, adding IR hedge to
delta hedging typically improves hedging performance, while adding gamma or vega to delta
hedging worsens hedging performance. This effect is more evident, when more basis risk is
present. Note also that, over periods of extremely high volatility such as during the GFC,
the hedge is subject to model risk, since models would not be able to capture well these
market conditions. Another reason for this could be that the delta-vega or the delta-gamma
hedging schemes are more sensitive to model misspecification, i.e., to a mismatch between
the assumed stochastic dynamics and the “true” process generating empirical data, since
three short-dated options and three short-dated futures are used to hedge the long-dated
option. Consequently, for hedging applications that are subject to model risk or basis risk,
IR hedge tends to improve further delta hedging, something that other hedging schemes,
such as gamma and vega, do not attain.11

11
Even though a stochastic volatility model is used, delta-vega hedge does not provide any noticeable

14
Figure 1: Annualised monthly standard deviations of the US Treasury yields.
This plot presents annualised monthly standard deviations are calculated based on computing
√ the
standard deviations of the US Treasury yields over 20 trading days and then multiplied by 12.

On the other hand, under Scenario 4, that provides a hedge with the least basis risk
compared to the other scenarios, the delta-vega-IR hedge tends to perform better during the
pre-GFC, and the GFC period, while the delta-gamma-IR hedge typically performs the best
at the post-GFC period, see bottom panel of Figure 6. During the post-GFC period, the
volatility of the futures contracts is low which makes volatility hedge less necessary (or even
counterproductive). Finally, during post-GFC period, where the interest rate volatilities and
levels have been unusually low, the IR hedges do not typically provide further improvement
compared to the other hedges. These latter results do not generally depend on the basis risk
present in the hedging applications.

5. Conclusion
This paper employs the stochastic volatility–stochastic interest rate forward price model
developed in Cheng et al. (2015) to hedge long-dated crude oil options over six years. The
Light Sweet Crude Oil (WTI) futures and option dataset from the NYMEX spanning a
6-year period is used in the empirical analysis. Several hedging schemes, including delta,
gamma, vega and interest rate risk hedges, are applied to hedge a call option on futures, from
the beginning of 2006 to the end of 2011. Factor hedging is used to simultaneously hedge
multi–dimensional risks impacting the forward curve, the stochastic volatility and stochastic
interest rates.
Several conclusions are drawn from this empirical analysis on hedging long-dated crude
oil options. Firstly, using hedging instruments with maturities that are closer to the ma-

improvement, especially when high basis risk is present such as in Scenario 1. A volatility trade such as
a strangle or straddle would have probably been more sensitive to vega hedging compared to an outright
options positions.

15
turity of the option reduces the hedging error. This is predominantly due to the fact that
hedging instruments with longer maturities require less frequently to roll forward the hedge;
hence, leading to lower basis risks. Secondly, delta-gamma hedging is overall more effec-
tive than delta-vega hedging, especially when the maturities of the hedging instruments are
getting shorter. For hedging instruments with longer maturities, delta-vega performs better
compared to delta-gamma in the pre-GFC period.
Thirdly, interest rate hedging consistently improves hedging performance when it is added
to delta, delta-vega and delta-gamma hedging with the improvement being more evident
when the shorter maturity contracts are used as hedging instruments (thus when basis risk
is higher). Consequently, part of the basis risk is essentially managed via the hedging of
the interest rate risk. Furthermore, interest rate hedging was more effective during the GFC
when interest rate volatility was very high, while interest rate hedging does not improve
the performance of the hedge in the post-GFC period, where interest rates volatility was
extremely low.
Fourthly, comparing the hedging performance between a stochastic interest rate model
and a deterministic interest rate model, during the GFC, there is a significant improvement
from the stochastic interest rate model over the deterministic interest rate model. However,
there is only marginal hedging improvement from the stochastic interest rate model during
the pre-crisis period and no noticeable improvement at all after 2010.
Lastly, delta-IR hedge often outperforms delta-vega and delta-gamma hedges. When
hedging is carried out with shorter maturity hedging instruments and over periods of high
interest rates volatility, adding IR hedge to delta hedge improves hedging performance,
while adding gamma or vega hedge to the delta hedge deteriorates the hedge. This is a key
conclusion: When we have more exposure to model risk (due to turbulent market conditions)
and basis risk (due to a mismatch between the maturity of the option to be hedged and the
hedge instruments), IR hedge beyond delta hedge can consistently outperform all other
hedging schemes.

References
Brennan, M., Crew, N., 1997. Hedging long maturity commodity commitments with short-
dated futures contracts. Mathematics of Derivatives Securities, 165–190.
Bühler, W., Korn, O., Schöbel, R., 2004. Hedging long-term forwards with short-term fu-
tures: a two-regime approach. Review of Derivatives Research 7 (3), 185–212.
Cheng, B., Nikitopoulos, C. S., Schlögl, E., 2015. Pricing of long-dated commodity derivatives
with stochastic volatility and stochastic interest rates. Tech. rep., Quantitative Finance
Research Centre, UTS, Research paper 366.
Cheng, B., Nikitopoulos, C. S., Schlögl, E., 2016a. Empirical pricing performance on long-
dated crude oil derivatives: Do models with stochastic interest rates matter? Tech. rep.,
Quantitative Finance Research Centre, UTS, Research paper 367.
Cheng, B., Nikitopoulos, C. S., Schlögl, E., 2016b. Hedging futures options with stochastic
interest rates. Tech. rep., Quantitative Finance Research Centre, UTS, Research paper
375.

16
Chiarella, C., Kang, B., Nikitopoulos, C. S., Tô, T.-D., 2013. Humps in the volatility struc-
ture of the crude oil futures market: New evidence. Energy Economics 40, 989–1000.

Dempster, M., Medova, E., Tang, K., 2008. Long term spread option valuation and hedging.
Journal of Banking & Finance 32 (12), 2530–2540.

Edwards, F. R., Canter, M. S., 1995. The collapse of Metallgesellschaft: Unhedgeable risks,
poor hedging strategy, or just bad luck? Journal of Futures Markets 15 (3), 211–264.

Hamilton, J. D., 2008. Understanding crude oil prices. Tech. rep., National Bureau of Eco-
nomic Research.

Hamilton, J. D., 2009. Causes and consequences of the oil shock of 2007-08. Tech. rep.,
National Bureau of Economic Research.

Nelson, C. R., Siegel, A. F., 1987. Parsimonious modeling of yield curves. Journal of Business
60 (4), 473–489.

Shiraya, K., Takahashi, A., 2012. Pricing and hedging of long-term futures and forward
contracts by a three-factor model. Quantitative Finance 12 (12), 1811–1826.

Trolle, A. B., Schwartz, E. S., 2009. Unspanned stochastic volatility and the pricing of
commodity derivatives. The Review of Financial Studies 22 (11), 4423–4461.

Veld-Merkoulova, Y. V., De Roon, F. A., 2003. Hedging long-term commodity risk. Journal
of Futures Markets 23 (2), 109–133.

17
Figure 2: Futures prices and annualised standard deviations of the log-return of
futures prices for four December contracts. These plots present the futures prices of four
December contracts (top) and the annualised standard deviations of the log-return of futures prices
of four December contracts (bottom) over eight years.

18
Figure 3: RMSEs for Scenario 1 and 2. This plot compares the average RMSEs over 1,000
simulation paths of various hedging schemes under Scenario 1 (top) and Scenario 2 (bottom) with
the unhedged position.

19
Figure 4: RMSEs for Scenario 3 and 4. These plots compare the average RMSEs over 1000
simulation paths of various hedging schemes under Scenario 3 (top) and Scenario 4 (bottom) with
the unhedged position.

20
Figure 5: RMSEs for Scenario 1 and 2. These plots display the average RMSEs over 1000
simulation paths of various hedging schemes under Scenario 1 (top) and Scenario 2 (bottom).

21
Figure 6: RMSEs for Scenario 3 and 4. These plots display the average RMSEs over 1000
simulation paths of various hedging schemes under Scenario 3 (top) and Scenario 4 (bottom).

22
Figure 7: RMSEs for Scenario 1 and 2. These plots compare the average RMSEs of
portfolios assuming stochastic interest rates and deterministic interest rates, under Scenario 1
(top) and Scenario 2 (bottom).

23
Figure 8: RMSEs for Scenario 3 and 4. These plots compare the average RMSEs of
portfolios assuming stochastic interest rates and deterministic interest rates, under Scenario 3
(top) and Scenario 4 (bottom).

24