Deux

ontributions en gestion des risques de matieres premieres

Steve OHANA
19 septembre 2006

L'es alier de la s ien e est l'e helle de Ja ob, il ne s'a heve qu'aux pieds de Dieu

Albert Einstein

ii

Remer iements

Ce travail doit beau oup tout d'abord a ma dire tri e de these, le professeur Helyette Geman, qui,
tout au long de es quatre annees, m'a prodigue ave bienveillan e ses onseils et ses intuitions.
C'est a son onta t que j'ai appris e qu'etait veritablement le metier de la re her he. Je la remer ie inniment pour m'avoir appris (parfois dans la douleur...) a onstruire et a rediger un arti le
s ientique. Ce manus rit n'aurait sans doute pas ete le m^eme sans son regard attentif et exigeant.
Cette these est egalement le fruit d'un partenariat de trois ans ave la Dire tion de la Re her he
de Gaz de Fran e, qui a permis de donner a mon travail une dimension appliquee. Mes pensees
vont en parti ulier a Olivier Bardou, a Isabelle Garreau, a David Game, et a Guillaume Leroy, que
je remer ie haleureusement pour leur soutien. Je remer ie Celine Jerusalem pour son aide ainsi
que Christian De-Laorest, Damien Reboul-Salze, Florent Bergeret, Christophe Barrerra-Esteve,
Gregory Benmenzer, Solenne Gueydan, Marion La ombe, Pierre-Laurent Lu ille, et Jeanne Rey
pour leur a ueil et leur en adrement s ientique.
Je salue le professeur Ni ole El Karoui, qui m'a fourni l'essentiel de ma formation theorique en
nan e, le professeur Dann Lanneuville, qui m'a aide a former mon projet de these, ainsi que les
professeurs Guy Cohen, Pierre Carpentier, Ce ile Kharoubi, Steven Shreve, Marija Ili , Abraham
Lioui, Paul Kleindorfer, Mi hel Crouhy, et Frederi Bonnans, qui m'ont genereusement onsa re
iii

leur temps et m'ont apporte des onnaissan es pre ieuses.

Ces quatre annees de re her he ont ete une experien e humaine passionnante mais egalement di ile, ou l'euphorie laissait pla e parfois au doute et au de ouragement. Je veux i i rendre hommage
aux personnes dont la onan e et l'amitie m'ont aide a surmonter les epreuves qui jalonnent la vie
d'un do torant.
Je remer ie en premier lieu ma femme Nathalie pour son amour et son soutien quotidien ; la onan e tranquille, la joie de vivre ommuni ative, et l'energie bienveillante qu'elle degage ont ete des
moteurs indispensables pour mener a bien ette these. Ce travail lui est personnellement dedie.
Je voudrais remer ier egalement mes parents pour les valeurs d'humilite, d'honn^etete, et de questionnement qu'ils m'ont transmises; la re her he etant avant tout une forme d'ouverture d'esprit,
de sens ritique, et de qu^ete spirituelle, ils ont ete sans au un doute mes premiers guides dans mon
par ours de her heur.
Je remer ie haleureusement mon frere Jean-Ja ques Ohana, pour l'enthousiasme qu'il a toujours
manifeste pour mes "trouvailles", aussi modestes soient-elles, pour sa uriosite insatiable et sa
generosite, qui font qu'il a toujours ete pour moi un modele et un guide.
Je remer ie mes beaux-parents Marie-Claire et Fabien Belhassen pour leur bienveillan e et la onan e qu'ils ont toujours montree pour mon travail.
Je voudrais egalement saluer Mauri e Petrover, eminent professeur de mede ine, qui s'est interesse
tres t^ot a mon travail, et m'a a ompagne ave e oute, ae tion, pedagogie, et bienveillan e dans
ma de ouverte de la re her he.
Il est impossible de ne pas evoquer i i mes amis et freres Emmanuel Farhi, Emmanuel Goldsztejn,
Benjamin Petrover, Arie Bibas et Joseph Amar, dont l'importan e dans ma vie est si grande.
Enn, je salue mon ousin Mi hael Ohana et mes amis Emmanuel-Juste Duits, Muriel Darmon,
iv

Alexandre Espinoza, Harold Hauzy, Emilie Brunel, Aurelia Crouhy, Pas al Levy-Garboua, Patri k
et Emmanuelle Hayat, Fran ois-Charles S apula, Benjamin Kunstler, et Ygal Levy, qui ont ha un
ontribue, par leur soutien et leur inter^et pour mes re her hes, a la on retisation de e travail.

vi

Contents

Remer iements

iii

Introdu tion

xi

1 Time- onsisten y in managing a ommodity portfolio : a dynami risk measure

approa h

1.1 Introdu tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2 A omparison of dynami risk obje tives . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.1 Stati risk measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.2 Risk measures asso iated to a stream of ash ows . . . . . . . . . . . . . . .

1.2.3 Time onsisten y . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2.4 Risk and substitution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.3 The retailer's portfolio problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 De omposition results in two parti ular ases . . . . . . . . . . . . . . . . . . 18
1.3.3 The retailer problem in an in omplete/illiquid market . . . . . . . . . . . . . 20
1.3.4 A on avity property for Ji . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
vii

1.3.5 Ji as the arbitrage pri e of the portfolio in omplete markets . . . . . . . . . 22

1.3.6 A model for the evolution of the forward urve and demand . . . . . . . . . . 24
1.4 Numeri al results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.4.1 Expression of the retailer's problem on an event tree . . . . . . . . . . . . . . 26
1.4.2 Building the event tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4.3 The setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.4.4 Ee t of optimal strategies on the nal and minimal wealths . . . . . . . . . 32
1.4.5 Portfolio value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.5 Con lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.6 Annex: proof of the onvexity result . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1.7 Referen es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2 A new dependen e model for ommodity forward urves; appli ation to the US
natural gas and oil markets

47

2.1 Introdu tion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.2 The e onomi relations between oil and natural gas in the US . . . . . . . . . . . . . 52
2.2.1 Dependen e through the demand . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2.2 Dependen e through the supply . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.3 Empiri al observation of the dependen e between oil and gas forward urves in the US 60
2.3.1 Data des ription . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.3.2 De omposition of daily forward urve moves into short term and long-term
sho ks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.3.3 Slope and level: two state variables for the shape of the forward urve . . . . 71
2.3.4 Denition of lo al and global dependan e stru tures . . . . . . . . . . . . . . 81
viii

2.3.5 Analysis of lo al dependen e stru ture . . . . . . . . . . . . . . . . . . . . . . 81

2.3.6 Analysis of global dependen e stru ture . . . . . . . . . . . . . . . . . . . . . 90
2.4 A new dependen e model for ommodity forward urves . . . . . . . . . . . . . . . . 97
2.4.1 Formulation of the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.4.2 Calibration of the model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
2.4.3 Stability of the orre tion me hanisms . . . . . . . . . . . . . . . . . . . . . . 125
2.4.4 Evolution of the orrelations . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
2.5 Con lusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
2.6 Referen es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

ix

Introdu tion

La gestion des risques nan iers asso ies aux portefeuilles de matieres premieres est un probleme
d'une grande a tualite, dans un ontexte de forte volatilite des mar hes de ommodites internationaux et de deregulation des mar hes europeens de l'ele tri ite et du gaz.
Cette these apporte deux ontributions independantes dans e domaine.

Le premier hapitre a trait a la gestion et a la valorisation dynamique des portefeuilles de

ommodites. Les produ teurs, a heteurs et nego iants de matieres premieres sont exposes a la fois
a un risque de mar he, de par la volatilite des prix des matieres premieres, et a des risques "non
trades" sur le mar he, dont les plus importants sont le risque volumique, lie a l'in ertitude sur
leur propre onsommation ou sur elles de leurs lients1 , les risques de ontrepartie2 , les risques
physiques3, et les risques politiques4. Ces risques non trades et la liquidite restreinte et in ertaine
des mar hes de ommodites reent un ontexte de mar he in omplet qui distingue les mar hes de
1 Les derivees limatiques ont notamment pour vo ation la ouverture nan iere des risques volumiques, mais etant

donnee leur tres faible liquidite, on peut onsiderer le risque volumique omme un risque non trade sur le mar he
2 La faillite du geant des mar hes de l'energie qu'etait Enron a a ru la sensibilite des a teurs a e type de risques;
il est a noter que l'apparition des mar hes de futures de matieres premieres a ete notamment motivee par e risque
3 Un exemple est l'in endie qui a rendu indisponible le sto kage anglais de Rough en fevrier 2006
4 Les re entes mena es de rupture de l'approvisionnement europeen en gaz reees par la rise diplomatique russoukrainienne en sont une bonne illustration
xi

matieres premieres des mar hes taux et a tions. Ainsi, m^eme si les a tifs physiques omposant
les portefeuilles de matieres premieres ( ontrats d'approvisionnement exible ("swing options"),
apa ites de sto kage, unites de produ tion/transformation...) peuvent ^etre e onomiquement assimiles a des options "reelles" e rites sur des sous-ja ents spot ou futures de ommodites, ils ne
peuvent ^etre valorises a l'aide des prin ipes d'arbitrage, omme le sont les options portant sur des
sous-ja ents a tions ou taux d'inter^et. De plus, l'univers de mar he in omplet ree des synergies
entre les a tifs d'un portefeuille de matieres premieres qui interdisent la de omposition de la gestion
et de la valorisation d'un tel portefeuille en somme d'options reelles independantes. Les mouvements re ents de rappro hement entre dierentes ompagnies europeennes de gaz et d'ele tri ite5
ont d'ailleurs ete en partie motives par les synergies existant entre leurs dierents portefeuilles
d'a tifs physiques.
Pour prendre en ompte es synergies dans la valorisation et la gestion d'un portefeuille de matieres
premieres, il est ne essaire d'adopter une appro he globale modelisant l'intera tion entre les dierents
a tifs physiques et ontrats omposant e portefeuille.
Il existe un nombre important de travaux de re her he operationnelle adoptant ette demar he
globale. Cependant, la plupart d'entre eux developpent une vision statique de l'evaluation du
portefeuille: seule la valeur initiale du portefeuille est onsideree et optimisee. Or, la gestion d'un
portefeuille est un pro essus dynamique qui implique des de ision sequentielles (utilisation des
sto ks, nominations sur les ontrats d'approvisionnement, interventions sur les mar hes a terme...),
lors desquelles le gestionnaire doit remettre a jour son ritere d'evaluation, en tenant ompte du
5 Une

inquantaine de fusion/a quisitions entre ompagnies d'energie ont eu lieu en Europe entre 1998 et 2003,
parmi lesquelles Gas Natural-Endesa, EON-Ruhrgas, EDF-Edison, et le mouvement ontinue aujourd'hui ave
l'annon e des intentions d'OPA hostiles de EON sur Endesa, d'Enel sur Suez puis l'annon e du possible rappro hement
Suez-Gaz de Fran e
xii

nouvel etat du portefeuille et de l'information qui est devenue disponible. Dans le adre d'une
evaluation statique, le planning de de ision futures qui est de ide a la date t ne tient pas ompte
de la remise a jour du ritere d'evaluation qui sera ne essairement operee a la date t + 1. On omprend des lors que l'utilisation d'un tel ritere d'evaluation dans un probleme de gestion dynamique
peut onduire a mal anti iper la valeur et le risque futurs du portefeuille et a regretter ex-post des
de isions passees, e que les e onomistes appellent l'in onsistan e dynamique.
L'obje tif du premier hapitre est d'introduire et d'experimenter sur un exemple on ret un nouveau ritere d'evaluation dynamique pour les portefeuilles de matieres premieres, dans le ontexte
d'un mar he partiellement liquide et en presen e d'un risque volumique. Ce ritere, onstruit de
maniere re ursive a partir du futur, permet la onsistan e inter-temporelle des de isions de gestion.
D'autre part, par e qu'il depend de deux parametres fa ilement interpretables, l'un ontr^olant la
regularite temporelle des ash- ows, l'autre leur dispersion aleatoire, e ritere permet au gestionnaire de trouver le ompromis ideal entre ri hesse nale esperee, risque sur la ri hesse nale et
risque de tresorerie au ours de l'horizon6. Enn, notre appro he est a la fois un outil de gestion et
de valorisation d'un portefeuille; elle rend notamment possible l'evaluation des a tifs au sein d'un
portefeuille, ave des appli ations potentielles importantes en terme de sele tion de portefeuille

( ession/a quisition/renego iation d'a tifs...).

Le deuxieme hapitre de ette these a pour obje tif de proposer un modele d'evolution onjointe de deux ourbes a terme de matieres premieres. De nombreux portefeuilles de matieres

6 Pour denir le risque sur la ri hesse nale ou sur la tresorerie, on fera appel au on ept de Conditional Value at

Risk
xiii

premieres sont en eet a l'interfa e entre plusieurs mar hes de ommodites7 . La valeur nan iere
et la gestion de es portefeuilles "multi- ommodites" ne dependent pas seulement des prix spot mais
de l'ensemble des ourbes a terme des dierentes matieres premieres impliquees. Par exemple, la
valeur d'une entrale de produ tion d'ele tri ite a partir de gaz depend du spread entre les ourbes
a terme du gaz et de l'ele tri ite sur la duree de vie de la entrale: le pri ing et la ouverture
d'un tel a tif reposeront don sur un modele d'evolution onjointe des ourbes a terme du gaz
et de l'ele tri ite. Un deuxieme exemple est elui du detenteur d'un ontrat de livraison de gaz
indexe sur le prix spot du petrole qui, pour se uriser sa marge, souhaitera au moment "opportun"
prendre une position "short" sur le mar he a terme du gaz et une position "long" sur le mar he a
terme du petrole. Un troisieme as mettant en jeu les orrelations entre ourbes a terme est elui
d'un hedge-fund desirant onstruire une strategie "long/short" sur les futures de deux matieres
premieres, en pariant sur le retour a des relations e onomiques de "long terme" liant historiquement les prix a terme8 . Sur es trois exemples, on onstate que la onnaissan e des lois d'evolution
onjointe de plusieurs ourbes a terme des matieres premieres inter-dependantes est essentielle. Or,
s'il existe une abondante litterature sur la modelisation des orrelations entre les prix de matieres
premieres, elle- i on erne prin ipalement l'intera tion entre quelques points parti uliers sur une
m^eme ourbe a terme (par exemple, le prix spot et le prix rst-month) ou sur deux ourbes a
terme (par exemple, les deux prix rst-month). Jusqu'a present, le probleme des dependan es en7 On peut penser par exemple a des portefeuilles ontenant des ontrats de gaz indexes sur le prix du petrole, des

usines de transformation de matieres premieres omme des raneries, des entrales de produ tion de gaz fon tionnant
au harbon, au oul, ou au gaz, des usines de fabri ation de metaux ou de papier onsommatri es d'energie...
8 Il est important de souligner i i l'instabilite temporelle de ertaines relations de long terme entre matieres
premieres: par exemple, la relation entre les prix du rude oil et les prix des produits ranes (aussi appelee " ra k
spread") s'est montree parti ulierement instable es dernieres annees
xiv

tre l'ensemble de deux ourbes a terme n'a pas ete en ore traite dans la litterature. Le deuxieme
hapitre de ette these se propose d'elaborer un modele d'evolution onjointe des ourbes a terme de
deux matieres premieres inter-dependantes. Notre modele integre a la fois les dependan es globales
et lo ales entre deux ourbes a terme de matieres premieres. Les mouvements journaliers d'une
ourbe a terme sont de omposes en un ho ourt terme ae tant les premieres maturites et un
ho long terme, representant une translation globale des prix a terme. Cette de omposition des
deformations se traduit par une de omposition de la forme d'une ourbe a terme omme somme
d'une omposante deterministe saisonniere, d'une omposante aleatoire de ourt terme (la "pente"),
et d'une omposante aleatoire de long terme ("le niveau"). Les dependan es globales on ernent
les relations de long terme existant entre pentes et niveaux des deux ourbes a terme, tandis que les
dependan es lo ales de rivent les relations entre les ho s ourt et long terme journaliers des deux
ourbes. Comme dans un modele de ointegration lassique, les relations de long terme apparaissent dans notre modele a travers une prime de risque dans l'evolution des prix a terme: l'originalite
par rapport a un modele de ointegration lassique est la stru ture par terme des primes de risque,
qui, dans notre modele, est ompatible ave l'absen e d'arbitrage et est la somme d'une partie
" ourt terme", dependant de l'e art a la relation d'equilibre sur les pentes, et d'une partie long
terme, dependant de l'e art a la relation de long terme sur les niveaux. Le modele de dependan e
est applique aux mar hes ameri ains du gaz et du petrole ( rude oil et heating oil) de Janvier
1999 a O tobre 2004. Con ernant la stru ture de dependan e lo ale, nous mettons en eviden e des
relations de ausalite entre les ho s journaliers des ourbes a terme gaz et petrole, une volatilite
sto hastique pour l'ensemble des ho s, une volatilite saisonniere pour les ho s ourt terme du gaz
et du heating oil, des orrelations positives entre les o-mouvements journaliers des ourbes a terme
gaz et petrole. Con ernant la stru ture de dependan e globale, nous mettons en lumiere l'existen e
xv

d'une forte relation de long terme forte entre les niveaux des ourbes a terme gaz et petrole (ave
deux ruptures intervenant au debut de l'annee 2000 et au milieu de l'annee 2003), et une relation
de long terme moins signi ative entre les pentes des ourbes a terme. Enn, une etude portant sur
la stabilite temporelle du modele de dependan e revele que les me anismes de orre tion d'erreur
relatifs aux dependan es globales se sont renfor es depuis 2002 et que les orrelations entre les
mouvements journaliers des ourbes a terme gaz et petrole ont onnu un trend as endant sur la
periode 1999-2004.

xvi

Chapter 1
Time- onsisten y in managing a
ommodity portfolio : a dynami risk
measure approa h1

We onsider the problem of the manager of a storable ommodity (e.g. hydro, natural gas, oal)
portfolio fa ing demand risk while having a ess to storage fa ilities and illiquid spot and forward
markets. In this setting, we emphasize that a dynami ally onsistent way of managing risk over
time must be introdu ed. In parti ular, we demonstrate the temporal in onsisten y of stati risk
obje tives based on nal wealth and advo ate the use of a new lass of re ursive risk measures
su h as those suggested by Epstein and Zin (1989) and Wang (2000) for portfolio optimization and
valuation. This type of risk measures not only provide time- onsistent de ision plannings but allow
1 This

hapter is a slightly dierent version of an arti le with the same title written with Pr Geman; I thank
Guillaume Leroy, David Game, Olivier Bardou and Jean-Ja que Ohana for their support and helpful suggestions; I
a knowledge also Stanley Zin and Paul Kleindorfer for their omments whi h helped me to improve this paper
1

the portfolio manager to ontrol independently the o urren e of ash- ows a ross time and a ross
random states of nature. We illustrate the dis ussion in an empiri al se tion where the trade-o
between nal wealth risk and bankrupt y risk at an intermediate date is analyzed and the synergy
between the physi al assets omposing a ommodity portfolio is assessed.

1.1

Introdu tion

We onsider the situation of a retailer, who is engaged in long-term sale ontra ts, owns storage
fa ilities and an trade the ommodity in illiquid spot and forward markets. The retailer is fa ing
a portfolio optimization problem, that translates into de iding at ea h time step whi h quantity to
inje t in or withdraw from her storage fa ilities and trade in the spot and forward market, and a
portfolio valuation problem, that onsists in assessing the value of the global portfolio and of ea h

asset omposing it. The optimization and the valuation take pla e in the ontext of two types of
risk: the volume risk that arises from the random demand of long-term ustomers and is related to
exogenous non traded variables su h as weather, and the pri e risk that is linked to the volatility
of the ommodity pri e.
In this in omplete market setting, the value of the retailer's portfolio is not uniquely determined
by arbitrage onsiderations and an integrated portfolio approa h is needed to handle liquidity onstraints.
The sto hasti programming literature, on the one hand, has essentially treated situations where
portfolio management is analyzed through a mean-varian e riterion applied to nal or intermediate wealths, and fully dened at the rst de ision date. In parti ular, the risk re-evaluations arising
at intermediate de ision dates are not taken into a ount, leading to possible on i ts between
de isions taken over time. Examples of this approa h are found in Unger (2002), where a CVaR
onstraint on the nal wealth is addressed through a Monte-Carlo approa h, in Martinez-de-Albeniz
and Sim hi-Levi (2005), where mean-varian e trade-os are onsidered and yield expli it solutions
in a one-step framework, and in Li and Kleindorfer (2004), where the ase of a multi-period VaR
onstraint on ash ows is examined.
The literature on de ision theory, on the other hand, has paid a deserved attention to the prob3

lem of dynami hoi e under un ertainty. Originally, it was the problem of dynami onsumption
planning that was analyzed by e onomists. In a seminal paper, Epstein and Zin (1989) introdu e
a set of dynami utilities, dened re ursively in a dis rete time setting, and allowing one to separately a ount for the issue of substitution - ontrolling onsumption over time- and risk aversion
- ontrolling onsumption a ross random states of nature. In nan e, dynami risk measures were
re ently introdu ed to a ount for the o urren e of a stream of random ash- ows over time. A
general requirement for these risk measures is their time- onsisten y (see e.g., Artzner et al. (2002))
be ause, as emphasized by Wang (2000), multi-period risks are reevaluated as new information be omes available, whi h raises the issue of the ompatibility between onse utive de isions implied
by the risk measure.
Our arti le, to our knowledge, is the third attempt after Chen et al. (2004) and Ei hhorn and
Romis h (2005) to use dynami risk obje tives in inventory and ontra ts portfolio problems. Ei hhorn and Romis h (2005) use a restri tion of the set of oherent dynami risk measures dened by
Artzner et al. (2002) to solve an ele tri ity portfolio optimization problem but do not raise the
problem of time onsisten y of optimal strategies. Chen et al. (2004) dene their obje tive fun tion
as an additive inter-temporal utility of the onsumption pro ess of the portfolio manager. Instead,
we hoose the Epstein and Zin (1989) non additive inter-temporal utility obje tive and apply it
dire tly to the ash ow pro ess. The impa t of this hange is signi ant : in our setting, the initial
wealth is not a state variable, the only state variables being the inventory level, and the umulative
positions in the forward market for ea h future delivery period; in addition, the retailer's problem
appears as a ash- ow stream management one rather than a onsumption planning one; lastly,
the exibility of the non additive inter-temporal utility allows the portfolio manager to separately
ontrol the distribution of ash ows a ross time periods and a ross states of nature, whi h is not
4

allowed by an additive utility obje tive on the onsumption pro ess2 .

The ontribution of this paper is twofold: i) on the methodologi al side, we dene the on ept of
time- onsisten y of optimal strategies, show that the lassi ally used stati risk measures on nal
wealth are not time- onsistent and advo ate the use of re ursive utilities as a time- onsistent and
exible measure for portfolio risk management and valuation; ii) on the operational side, we provide
a tra table framework to dynami ally manage physi al assets under random demand and evolution
of spot and forward ommodity pri es, and show on a numeri al example how the use of re ursive
utilities an help strike a trade-o between nal and intermediate wealth risk management and
assess the synergy between the physi al assets omposing a ommodity portfolio.
The remainder of the paper is organized as follows. In se tion 2, we dene the time- onsisten y of
optimal strategies and ompare two obje tives with respe t to the issues of time- onsisten y, and
risk/substitution preferen es. In se tion 3, we present the retailer's portfolio management problem
and provide a pri ing formula and bid/ask pri es for physi al ommodity assets. Se tion 4 presents
a numeri al illustration of the main ndings. Se tion 5 ontains on luding omments.

1.2

A omparison of dynami risk obje tives

The obje tive of this se tion is to present two examples dynami risk preferen es and assess their
time- onsisten y properties, whi h we view as an original ontribution of the paper.
2 Note that our framework redu es to the one of Chen et al.(2004) when substitution preferen es are ignored and

when CARA utility fun tions are used

5

1.2.1 Stati risk measures

In the ase of one period settings, a number of stati risk measures have been dened to express
preferen es of risk averse agents (see e.g, Artzner et al. (2000) and Frittelli and Rosazza Gianin
(2004)). Mathemati ally, a (stati ) risk measure is a fun tion, here denoted  , asso iating to a
ontingent laim X a real number  (X ).  (X ) represents the pri e that it is a eptable to pay in
order to pur hase X and  ( X ) represents the apital that must be provisioned in order to make
a short position in X a eptable.

1.2.2 Risk measures asso iated to a stream of ash ows

Possible riteria for ash ow streams assessment
Dened on a ltered probability spa e (
; F ; P; (Ft )), the dis rete-time sto hasti pro ess G =
(Gi )i=1;:::;T , represents a sequen e of random ash ows o urring at times (i )i=1;:::;T .
set of all

G is the

Fi -adapted ash ow pro esses from i = 1 to i = T . We hoose F1 = f;;

g (G1 is

deterministi ), and FT = F , so that full information is revealed at date T .

A dynami value measure V = (Vi )i=1;:::;T onsists of mappings Vi : G
! R that asso iate to ea h
ash ow pro ess G 2 G and to ea h ! 2
a real number Vi (G; !). The resulting sto hasti pro ess
(Vi ) is Fi -adapted. Finan ially, it represents the value of the sequen e of ash ows (Gk )k=1;:::;T
or the apital requirement to over the liabilities ( Gk )k=1;:::;T at dates (k ).
Let us now propose two ategories of dynami values measures for streams of ash ows:

1. The rst ategory onsists of extensions of stati riteria depending on the wealth a umulated
6

between date i and date T :

Wi;T :=

T
X

G

=
Vi (G; !) = (Wi;T jFi )
 i

(1.1)

In the above equation,  is a one-step risk measure and the notation (:jFi ) refers to onditioning on the information available at date i .
2. A se ond ategory of riteria (proposed by Epstein and Zin (1989) and Wang (2000)) are
re ursively onstru ted from the end of the time period by dening:

VT (G; !) = GT
Vi (G; !) = W (Gi ; (Vi+1 jFi ))

8i  T 1

(1.2)

In the above equation,  is a one-step ertainty equivalent 3 and the mapping W : R2

R

is alled an aggregator. In this framework, the date i value is assessed re ursively by

aggregation of the urrent ash ow Gi and ertainty equivalent of Vi+1 seen from date i .
An important observation is that the pro ess (Vi ) is Fi -adapted.

1.2.3 Time onsisten y

Time- onsisten y is a property whi h guarantees that preferen es implied by a dynami value
measure do not on i t over time.
3 We adopt Wang's denition of the ertainty equivalent, i.e., a stati measure  verifying the monotoni ity property

(whi h insures that if a random variable X is larger than Y in every state of the world, then  (X )   (Y )) and
redu ed to the identity on the spa e of onstant random variables.
7

Examples of time-in onsisten ies

Consider the two ash ow streams A and B , where all transition probabilities are supposed to
equal 0:5:
(state uu)

7(state uu)

1(stateu)

HHH

H ud)
1(state

4

2(stateu)
HHH
HH1(state ud)




6(state du)




HH
0(state d) HH
H dd)
1(state



3(state du)



 
HH
1(state d)HH 1(state dd)
H
B

Let us evaluate stream A using the dynami value measure (1.1) with (X ) = u 1 (E [ u(X )),

u(x) = ln(x):

V2 (A; u) = exp(E (ln(W2A;3 ju))) = exp(0:5(ln(8) + ln(2))) = 4; V2 (A; d) = exp(E (ln(W2A;3 jd))) = 6
1

V1 (A) = exp(E (ln(W1;3 ))) = exp(0:25(ln(11) + ln(5) + ln(9) + ln(4))) = (55  36) 4
Now evaluate stream B :

V2 (B; u) = exp(E (ln(W2B;3 ju))) = exp(0:5(ln(6)+ln(3))) = 18; V2 (B; d) = exp(E (ln(W2B;3 jd))) = 8
1

V1 (B ) = exp(E (ln(W1B;3 ))) = exp(0:25(ln(9) + ln(6) + ln(7) + ln(5))) = (54  35) 4

We thus have simultaneously the following inequalities:

V2 (A; u) < V2 (B; u); V2 (A; d) < V2 (B; d); V1 (A) > V1 (B )
8

As a result, the dynami value measure V dened in (1.1) qualies B as preferable to A in all states
of the world at time 2 and A preferable to B at time 1, hen e its time in onsisten y.
Time onsisten y does not hold either if  is a mean-varian e instead of an expe ted utility riterion
in equation (1.1). To see this, onsider the two following ash ow streams A (left) and B (right),
with transition probabilities being written on top of ea h ar :

1 (state uu)
3
4

0 (state
Hu) 1
HH4H
1
2
H ud)
0 (state

 12


0 (state d)

0 (state u)
1
2

0.5

 12


0 (state d)

Let us evaluate stream A using the dynami value measure (1.1) with (X ) = E (X )

ju)) V ar(W2A;3ju)) = 34 ( 34 169 ) = 169

V2 (A; d) = E (W2A;3 jd)) V ar(W2A;3 jd)) = 0

V2 (A; u) =

V1 (A) =

E (W2A;3

E (W1A;3 ))

V ar(W1A;3 )) =

1 3

2 4

3
8

9
9
)=
64
64

Now evaluate stream B :

V2 (B; u) =

E (W2B;3

ju)) V ar(W2B;3ju)) = 12

V2 (B; d) =

E (W2B;3

jd)) V ar(W2B;3jd)) = 0

V1 (B ) =

E (W1B;3 ))

V ar(W1B;3 )) =
9

1 1

2 2

1 1

2 4

1
3 12
)= =
16
16 64

V ar(X ):

We thus have simultaneously the following inequalities:

V2 (A; u) > V2 (B; u); V2 (A; d)  V2 (B; d); V1 (A) < V1 (B )

Let us now formally dene the time- onsisten y property:

Denition 1.2.1: The dynami value measure V is intrinsi ly time onsistent if

for
8
>
>
<
>
>
:

A; B 2 G , for t 2 T = f1; :::; T g, for ! 2

;
A(t; !)  B (t; !)
Vt+1 (A; !0 )  Vt+1 (B; !0 ) 8!0 2 Ht (!)

In the above denition,

) Vt (A; !)  Vt (B; !)

Ht(!) denotes the set of events !0 2

having the same history as !

up to time t: intuitively, it is the set of all possible subsequent events after time t bran hing from
a given s enario !.

Property 1.2.2: If the aggregator W is monotoni , then the dynami value measures of the re ursive type (1.2) are intrinsi ly time- onsistent

Proof : for any t 2 T , for any ! 2

, if A(t; !)  B (t; !) and Vt+1 (A; !0 )  Vt+1 (B; !0 ) 8!0 2

Ht(!),
then, by monotoni ity of ertainty equivalents, (Vt+1 (A; :)jFt )(!)  (Vt+1 (B; :)jFt )(!).
In turn, by monotoni ity of the aggregator W ,

Vt (A; !) = W (A(t; !); (Vt+1 (A; :)jFt )(!))  W (B (t; !); (Vt+1 (B; :)jFt )(!)) = Vt (B; !).

Denition of time onsisten y of optimal strategies and omparison of the two riteria
In the previous se tion, we dened an intrinsi time- onsisten y property, related to the evaluation
of exogenous streams of random ash- ows. In this se tion, we assume instead that the ash
ows depend on de isions that are made at ea h date i , using the information available at this

date. De ision at date i is the result of the optimization of a dynami value measure of the type
10

des ribed above. This optimization not only yields the rst de ision at that date, but a whole
de ision planning for all subsequent stages. The question we pose in this se tion is the following:

are optimal plannings onsistent over time?

Let us dene the problem formally: onsider a ash ow sequen e (Gi )1iT , o urring at dates
(i )i1 , depending on de isions (qi )1iT and on a multi-dimensional random pro ess (i )1iT :

Gi := f (qi; i ). (i ) is assumed to be of the type i+1 = g(i ; i+1 ) for some reasonably behaved
fun tion g, and a white noise ve tor pro ess (i ).
We introdu e the state variables xi on whi h depend de isions at time i and denote A(xi ) the set
of admissible strategies (qk )ikT at time i . We suppose that, after de ision qi is made at time i ,
the state xi leads to xi+1 = h(xi ; qi ; i+1 ; i+1 ), where h is a deterministi fun tion and (i ) a white
noise ve tor pro ess possibly orrelated with (i ). We denote (Fi ) the ltration generated by the
pro esses (i ; i ); (qi ) is supposed to be an (Fi )-adapted pro ess.
Lastly, we onsider the following optimization problem, related to a dynami value measure V :

Ji (xi ) :=

Max V (G)
(qk )kt 2A(xi ) i

(1.3)

We denote (qki (xi ))ki the resulting (Fi )-adapted optimal strategy de ided at date i 4 . The
question of onsisten y of optimal strategies an be formulated in the following way:
(i+1)
i (x ; 
i
Is qi+1
i i+1 ; i+1 ) equal to (qi+1 (xi+1 )), where xi+1 = h(xi ; q (xi ); i+1 ; i+1 )?

We now turn to the time onsisten y of optimal strategies derived from the two dynami value
measures dened above.
- First, let us onsider the nal wealth obje tive dened in equation (1.1) with (X ) = u 1 (E [ u(X )),i.e,
4 We suppose throughout this se tion that all en ountered optimization problems have a unique solution

11

Vi (G; !) = u 1 (E (u(Gi + Gi+1 + ::: + GT )jFi )))5 :

Ji (xi ) : =

Max V (G)
(qk )ki 2A(xi ) i


1
= u
Max
Max E  (E i+1 (u(Gi + Gi+1 + ::: + GT )))
qi (qk )ki+1 i


1
= u
Max
E i (
Max
E
(u(Gi + Gi+1 + ::: + GT )))
qi
(qk )ki+1 2A(xi+1 ) i+1

The date i+1 implied problem Max E i+1 (u(Gi + Gi+1 + ::: + GT ))) diers from the one derived
(qk )ki+1
from the dynami value measure (Vi ), i.e., Max Vi+1 = E i+1 (u(Gi+1 + Gi+2 + ::: + GT )). As a
(qk )ki+1
result, the optimal strategy de ided at time i diers from the optimal strategy exhibited at time

i + 1.
Time in onsisten y remains if we use a mean-varian e obje tive instead of an expe ted utility. In
order to further investigate this issue, let us onsider a sequen e of three ash ows (G1 ; G2 ; G3 ),
depending on the (Fi )-adapted pro ess (i )i=1;2;3 and

Fi -measurable de isions (qi)i=1;2;3 , and

let us de ompose the varian e of the sum of these ash ows. As usual, we denote V ari (X ) :=

V ar(X jFi ).
V ar1 (G1 + G2 + G3 ) = V ar1 (G2 + G3 ) = E 1 [(G2 + G3 )2 [E 1 (G2 + G3)2
= E 1 [E 2 ((G2 + G3)2 ) [E 1 (E 2 (G2 + G3 ))2
= E 1 [E 2 ((G2 + G3)2 )

E 1 ([E 2 (G2 + G3 )2 ) + E 1 ([E 2 (G2 + G3 )2 )

[E 1 (E 2 (G2 + G3 ))2

= E 1 [V ar2 (G2 + G3 ) + V ar1 (E 2 (G2 + G3 )) = E 1 [V ar2 (G3 ) + V ar1 (G2 + E 2 (G3 ))

The last equality illuminates why total varian e is time in onsistent: the

F1 -measurable term

V ar1 (G2 + E 2 (G3 )) is ontrolled by both de isions q1 and q2 , in ontrast to the term G1 , whi h
depends only on the de ision q1 . This fa t ompromises the existen e of any dynami programming
5 From now on, we will denote E (X jF

) = E i (X )
12

equation linking optimal strategies at dates 1 and 2 :

J1 (x1 ) : =
=

Max
fE (G + G2 + G3) V ar1 (G1 + G2 + G3)g
(qk )k=1;2;3 2A(x1 ) 1 1
Max fG (q ) V ar1 (G2 + E 2 (G3 )) + E 1 (E 2 (G2 + G3 ) V ar2 (G3 ))g
(qk )k=1;2;3 1 1


6= Max
G1 (q1 ) V ar1 (G2 + E 2 (G3 )) + E 1 ( Max E 2 (G2 + G3 ) V ar2 (G3 ))
q1
(qk )k=2;3 2A(x2 )
- We now turn to the dynami value measures des ribed in equation (1.2).
As a rst observation, let us onsider the ase of a linear aggregator W (x; y) = x + y. The date i
obje tive derived from the value measure Vi dened by equation (1.2) is then:

Ji (xi ) : =

Max V (G)
(qk )ki 2A(xi ) i

= Max fGi (qi ) + i (Vi+1 )g

(qk )ki


= Max
Gi (qi ) +
q
i

Max

(qk )ki+1 2A(xi+1 )

i (Vi+1 )

The question at this stage is to know whether permuting the operators Max and operator  is
legitimate in the last equality, i.e., if the following property holds:

Max  (V ) =? i ( Max Vi+1 )

(qk )ki+1 i i+1
(qk )ki+1

(1.4)

If the permutation is valid, then the optimal strategies will be time- onsistent sin e the date i+1
implied problem Max Vi+1 will oin ide with the optimization problem at stage i + 1; otherwise,
(qk )ki+1
they will not.
13

Let us try the aggregator W (x; y) =  1 ((x) + (y)) and ertainty equivalent (X ) =

u 1 (E [ u(X )), where u and  are in reasing fun tions and  is a positive dis ounting fa tor6 :
Max V (G) = Max  1 ((Gi (qi ) + (i (Vi+1 )))
(qk )ki 2A(xi ) i
(qk )ki 2A(xi )


1
= 
Max f(Gi (qi )) + (i (Vi+1 ))g
(qk )ki 2A(xi )



1
= 
Max
(Gi (qi )) + ( Max i (Vi+1 ))
qi
(qk )ki+1

Ji (xi ) : =

The inversion between operators Max and  in the last equality is permitted as


Max  (V ) = Max u 1 (E i (u(Vi+1 ))) = u 1 E i (

Max
u(V ))
(qk )ki+1 i i+1
(qk )ki+1
(qk )ki+1 2A(xi+1 ) i+1


1
= u
E i (u(
Max
V )) = i (
Max
V )
(qk )ki+1 2A(xi+1 ) i+1
(qk )ki+1 2A(xi+1 ) i+1

We an now present a general su ient ondition of time onsisten y for optimal strategies:

Property 1.2.3: If there exist non de reasing fun tions a b, , and d and positive numbers t su h
that

Vi (G) = a hfb(Gi (qi )) + i [E i (d (Vi+1 (G))gi

(1.5)

then the dynami value measure (Vi ) leads to time- onsistent optimal strategies.

For the re ursive value pro ess dened by utility fun tions  and u, equation (1.5) holds with

a =  1 , b = , =  u 1 , and d = u. In the ase of lassi al expe tation maximization

(risk-neutrality), equation (1.5) holds with a = b = = d = Id.

1.2.4 Risk and substitution

We have mentioned earlier that the problem of dynami optimization under un ertainty involves
two dimensions, one with respe t to the distribution of ash ows a ross states of nature, the other
6 This

parti ular hoi e for the aggregator and the ertainty equivalent was rst suggested by Epstein and Zin
(1989) and later on extended by Wang (2000) to in orporate ambiguity aversion
14

over onse utive time periods. The rst dimension has an ee t on the nal wealth distribution
while the se ond one impa ts the likelihood of bankrupt y within the time period.
Dynami value measures dened in equations (1.1) are not appropriate to apture the risk atta hed
to intermediate ash ows sin e they are based on nal wealth. By ontrast, re ursive dynami
value measures allows one to disentangle randomness and time omponents, via the ertainty equivalent  and the aggregator W (respe tively a ounting for the risk aversion and the substitution
preferen es of the de ision maker). For instan e, in the ase of re ursive dynami value measures
based on utility fun tions, the on avity of the fun tions u and  leads to the smoothing of ash
ows distributions in both dimensions and in turn to a joint ontrol of the nal wealth risk and
bankrupt y risk.

Remark: The hoi e u =  in re ursive value measures derived from utility fun tions u and
P
 leads to the lassi al obje tive: Vi (G) = u 1 (E i ( Tk=i  k

i

u(Gk ))), whi h has been widely

used in onsumption and portfolio hoi e problems in nan e (e.g., onsumption-based CAPM).
Of ourse, this obje tive is time onsistent and aptures both risk aversion and substitution; its
drawba k is that it does not oer as mu h exibility as a more general re ursive value measure
sin e risk aversion and substitution are represented by the same fun tion u.

As a on lusion of this se tion, we an state that re ursive dynami value measures with utility
type aggregator and ertainty equivalent are satisfa tory in regard to time onsisten y of optimal
strategies and inter-temporal risk management.

15

1.3

The retailer's portfolio problem

1.3.1 The model

We adopt a dis rete time setting, with a nite horizon. The de ision periods are denoted (pi ),

i = 1; :::; T (typi ally months or quarters). The dates (i ) are dening the periods (pi ).
date 1
period 1 1

...

date 2
period 2 2

...

date T
period T T

We assume from now on that the retailer's portfolio is omposed of one sale ontra t and one
storage reservoir. In addition, the ommodity is supposed to be traded, stored, and onsumed
in the same lo ation (in order to avoid transmission osts and onstraints). The problem an be
represented in a stylized diagram:

storage

retailer

?6

lient

market

Lmax is the maximal level of storage, the minimal level of storage (at any date) being 0, Linit is
the initial storage level, Lend is the minimal storage level at the end of the horizon. Li represents the
storage level at the end of period pi . Qinj
denotes maximal inje tion in period pi , Qdraw
maximal
i
i
withdrawal; we suppose there are no inje tion/withdrawal osts nor holding ost. di denotes the
lient's random demand in period pi , Kis is the xed selling pri e of the ommodity for period i.
Only forward ontra ts are onsidered; ash ows due to forward ontra ting are settled at
16

maturity of the ontra t and ounter-party risk ignored. We denote by F (i; j ) the forward pri e
of the ommodity quoted during pi for delivery in period pj 7 (j

 i) and Si the spot pri e of the

ommodity, where Si := F (i; i).

Remarks:
1. In our model, trading is only authorized at de ision dates
2. Even in the ase of illiquid markets, the retailer is assumed to be a pri e-taker, meaning that
her trading de isions will have no impa t on market pri es
Storage de ision variables orresponding to period pi are subje t to the following onstraints:
draw
0  qiinj  Qinj
 Qdraw
i
i ; 0  qi

L0 := Linit ; Li+1 = Li + qiinj

0  Li  Lmax

i1

(1.6)

qidraw 0  i  T

(1.7)

8i = 1; :::; T ;

LT

 Lend

(1.8)

n(i; j ) denotes the net number of forward ontra ts bought during period pi for delivery in period
pj (j  i), the ase i = j being a spot transa tion. N (i; j ) represents the total forward position at
the end of period pi for delivery in period pj and satises the onditions:

N (0; j ) := 0

8j  1; N (i; j ) = N (i 1; j ) + n(i; j ) 8 1  i  j

(1.9)

We model the sequen e of events and de isions in the following way: during period pi , the retailer
dis overs the lient's demand and de ides on date i whi h quantities n(i; j ) to buy on the spot
and forward market and qiinj or qidraw to inje t in or withdraw from storage, respe ting the physi al
balan e of ommodity ows during period pi i.e.,
7 Here, F (i; j )

N (i; i) + qidraw

qiinj = di

8 1iT

(1.10)

an be onsidered as the average pri e over all the quotation dates belonging to period pi of all
forward ontra ts for delivery in period pj
17

Equation (1.10) expresses that market and storage are the two ways to serve demand at period pi .
We dene the dis rete set of states of nature
. Ea h ! 2
represents a realization of the pro ess

i = (di ; F (i; j )j i ), i = 1:::T . We denote by (Fi ) the ltration generated by (i ). Throughout
the paper, we assume the absen e of arbitrage opportunities in the ommodity spot and forward
markets. On (
; F ; Fi ), we dene a risk-neutral probability measure P, under whi h forward pri es
are martingales8 .
We dene the set A of admissible strategies as:
n

A := (qi)i1 = (qidraw ; qiinj ; n(i; j )ji )i1 Fi measurable and verifying onstraints (1.6) to (1.10)
1.3.2 De omposition results in two parti ular ases
In this se tion, it is assumed that there are neither onstraints nor osts asso iated to trading in
the forward market. The risk-free interest rate r is supposed onstant. The goal here is to present
two ases where the pri ing issues and management of the portfolio are parti ularly simple:
- the rst ase is the one of a liquid market and deterministi demand
- the se ond ase in ludes un ertain demand but assumes risk-neutrality of the retailer, hen e the
use of a riterion of expe ted prot maximization
In both ases, a full de omposition of the portfolio value and management is possible.
The total ash ow during period pi is denoted as Gi and may be written as:

Gi = di Kis

T
X

e r(j

i

) F (i; j )n(i; j )

(1.11)

j i

8 We hoose here to work under a risk-neutral probability measure P to rule out a spe ulative use of the spot and

forward markets; indeed, if forward pri es were not martingales under P, the trading de isions implied by our model
ould be in uen ed by possible spreads between forward pri es and P-expe ted values of spot pri es, a feature whi h
is not relevant in the retailer's ontext
18

Remark: Cash ows due to forward trading are in this paper registered at transa tion date and
dis ounted from delivery date at the risk free interest rate r. We adopt this unusual rule be ause
we want ash ows at dates i to depend only on date i de isions and not on previous ones9 ,
as would be the ase if ash ows from forward transa tion had been registered at delivery date.
Sin e interest rates are onsidered deterministi , this representation has no onsequen es on the
nal wealth but may have some on intermediate wealths10 .
Assuming liquid spot markets, the oupling onstraint (1.10) an be treated as an impli it one and
we fa e a fully de omposable problem, with onstraints only on individual assets.
Deriving from (1.9) and (1.10) the volume n(i; i) of spot transa tions, equation (1.11) be omes:
T
X

Gi = di Kis n(i; i)Si

e r(j

= +1

i

) n(i; j )F (i; j )

j i

= q

q S + di (K
inj
i
i

draw
i
i

s
i

Si ) + N (i 1; i)Si

T
X

= +1

e r(j

i

) n(i; j )F (i; j )

j i

In this form, Gi appears like the sum of three omponents:

1. qidraw Si

qiinj Si = period pi payo from the storage fa ility. Storage de isions taken over

time are inter-dependent due to the apa ity onstraints expressed in equation (1.7)
2. di (Kis Si ) = period pi payo from the sale ontra t devoided of any optionality, whi h is in
fa t a strip of swaps ex hanging the sale ontra t pri e Kis for the spot pri e Si . The volume
involved at period pi is either xed (deterministi demand) or random (unknown demand)
3. N (i 1; i)Si

PT

= +1 e

j i

r j i

) n(i; j )F (i; j ) = period pi ash ow from forward ontra ts

9 in a ordan e with the setting dened in se tion 1.2.3

10 we thus assume here that the retailer provisions in advan e all the future gains or liabilities at the signature of

a forward ontra t
19

Under this form, the portfolio appears as a ombination of various options written on the ommodity
spot pri e while the forward market appears as a way to hedge the spot pri e risk. The above
splitting of ash ows suggests a de omposition of the portfolio's value. In fa t, the latter will only
be possible in two parti ular ases:

 Portfolio de omposition in a omplete market setting: here, we assume that the demand
pro ess (di ) is deterministi (e.g., the ontra t sets a xed volume to be delivered in all future
periods). Then, the arbitrage pri e of the portfolio is the sum of maximal expe ted ash ows
under the (unique) risk-neutral probability measure; this value is the sum of the arbitrage
pri es of storage and sale ontra t. In this framework, the obvious strategy for the portfolio
manager onsists in optimizing independently the storage fa ility against the spot market
under the risk-neutral measure, and hedging spot pri e risk using the forward market.

 Portfolio de omposition for a risk-neutral retailer in a liquid market: we assume here that the
retailer fa es both demand and pri e risks but is risk-neutral, i.e., she only tries to maximize
her expe ted prot. Under the assumption that the physi al measure is a risk-neutral measure,
the optimal strategy for the risk-neutral retailer onsists again in optimizing independently the
storage fa ility against the spot market and doing no trade in the forward market. Moreover,
under deterministi demand, the optimum of the risk-neutral retailer's obje tive orresponds
to the arbitrage pri e of the portfolio.

1.3.3 The retailer problem in an in omplete/illiquid market

Illiquidity is modeled by deterministi volume onstraints on spot and forward trading, of the form:

nb(i; i +  )  nmax
(i;  ); ns(i; i +  )  nmax
(i;  )
s
b
20

(1.12)

where nb (i; j ) and ns (i; j ) stand for the number of bought and sold forward ontra ts during period

pi for delivery in period pj (with n(i; j ) = nb (i; j ) ns(i; j )).

We dene the set of admissible strategies from state xi = (Li ; N (i; :); i ):
n

A(xi) := (qk )ki = (qkdraw ; qkinj ; n(k; j )jk )ki Fk measurable verifying admissibility onstraints
(1.13)
and the analogous set of illiquid market admissible strategies

Aliq (xi ). The restri tions of the

previous de ision sets to date i , dening the admissibility sets for de isions qi only, will be denoted
by Ai (xi ) and Aliq
i (xi ).
We an now formulate the retailer's optimization problem as:

Ji (xi ) :=

Max V (G)
(qk )ki 2Aliq (xi ) i

(1.14)

where G is dened by (1.11) and Vi (G) by the re ursive equation (1.2), with aggregator W and
ertainty equivalent  derived from on ave in reasing fun tions  and u and positive dis ount
fa tors (i ):

W (x; y) =  1 ((x) + i (y)); (X ) = u 1 (E [ u(X ))

We denote su h a dynami value measure as Vt;u (G).
The optimal value Ji (xi ) satises the dynami programming equation:


1
Ji (xi ) =  1 ( Max

(
G
i (qi )) + i  u (E i (u(Ji+1 (xi+1 )))) )
qi 2Aliq
i (xi )

where the state xi+1 is given by the transition equation xi+1 = (Li + qiinj

(1.15)

qidraw ; N (i; :) +

n(i; :); g(i ; i+1 )). The existen e of equation (1.15) guarantees the time onsisten y of optimal
strategies, as shown in the previous se tion.
21

1.3.4 A on avity property for Ji

Proposition 1.3.1:
Choosing CARA type utilities (x) = e

x

and u(x) = e

x

su h that 0 <   , for all dates

t, and all states xi su h that Aliq

t (xi ) 6= ;, the maximization problem :


1
Max

(
G
i (qi )) + i  u (E i (u(Ji+1 (xi+1 ))))
qi 2Aliq
i (xi )

is on ave with respe t to de isions qi . Moreover, the de ision set

Aliq
i (xi ) is onvex.

The result

also holds for  = Id and u of CARA type.

The proof is provided in the annex.

1.3.5 Ji as the arbitrage pri e of the portfolio in omplete markets

In this se tion, we show that, in omplete markets, Ji is the arbitrage pri e of the portfolio under the two onditions: (x) = x (no preferen e for smooth versus irregular ash ows in time
dimension) and i = e r(i+1

i

) (one period dis ount fa tor). These two assumptions will hold

throughout se tion 3.5.

Property 1.3.2:
For a on ave fun tion u, Ji (xi ) =
obje tive Jirn (xi ) =

Max V Id;u (G) is never greater than the risk-neutral

(qk )ki 2Aliq (xi ) i
ViId;Id (G)

Max
(qk )ki 2Aliq (xi )
Proof : The on avity of u implies that for all random variables X :
u 1 (E [ u(X ))  E (X )

(1.16)

It results, by a simple re ursion, that:

Id;Id
Id;Id
8G 2 G ; 8i 2 T ; ViId;u(G) = Gi + i u 1(E i (u(VtId;u
+1 )))  Gt + i E i (Vi+1 ) = Vi (G)

22

and the property holds.

Property 1.3.3: When onditional values Vk+1 omputed at stages k (k = i; ::; T

1) are non

sto hasti , then ViId;u is the sum of dis ounted ash ows from stage i to stage T
Id;u
Proof : In this ase, u 1 (E i (u(VkId;u
+1 ))) = Vk+1 for all k = i; :::; T

Gi + i ViId;u
+1 =

PT

r (
=e k

k i

i

1, and, therefore, ViId;u (G) =

) Gk , by a simple re ursion.

The onsequen e is that, in a omplete market setting (i.e., deterministi demand and no liquidity
onstraints), Ji is at least equal to the arbitrage pri e of the portfolio.

Property 1.3.4: In a situation of market ompleteness, Ji (xi ) is equal to the arbitrage pri e
of the portfolio Jiap (xi ) =

Max

(qk )ki 2A(xi )

PT

EQ
i (

= e r(k

k i

i

) Gk ), where Q is the (unique) risk-neutral

measure

Proof : This property is derived from the following observations:

- Ji (xi ) 

Max V Id;Id (G), as exhibited in property 1.3.2

(qk )ki 2A(xi ) i
- Max ViId;Id (G) = Jiap (xi ), be ause the optimal value of the risk-neutral retailer's problem
(qk )ki 2A(xi )
is equal to the portfolio's arbitrage pri e
- Ji (xi )  Jiap (xi ), as shown in property 1.3.3.

Property 1.3.5: If markets are omplete and u stri tly on ave, then the risk of the optimal
strategy (qk )ki is null.

Proof : The equality between Ji (xi ) and Jirn (xi ) implies an equality in equation (1.16) for ea h
X = Vi+1 , and, be ause the fon tion u is stri ly on ave, the equality is possible only if un ertainty
on all Vt is null.
23

Consequently, we obtain the satisfa tory property that the optimization programme also provides
a hedging strategy.
To on lude this paragraph, we give a way of estimating the ask and bid pri es of a physi al asset
or nan ial ontra t in in omplete markets: as often done in the literature , we dene the ask (bid)
pri e as the dieren e of the values of Jt , with and without the bought (sold) asset. Under this
denition, the bid and ask pri es of an asset depend not only on the risk aversion of the manager
but also on her initial portfolio, a lassi al property in a situation of in ompleteness.

1.3.6 A model for the evolution of the forward urve and demand
We assume a lassi al one-fa tor evolution model for the market forward urve F (i; j ):

F (i; j ) = F (i 1; j )Mi;j exp(e

ki j i

) Xi ) 8j  i 8i  2

(1.17)

where (Xi )i2 is a dis rete-time sto hasti pro ess omposed of independent variables with law

N (0; (iX )2 ), (ki ) are positive parameters, and (Mi;j )j i are positive onstants ensuring that F (i; j )ij
are martingale pro esses. In this model, only one type of sho k is allowed for the forward urve,
namely translations, with an amplitude vanishing with time to delivery.
Regarding the demand pro ess (di )i2 , we assume that it is driven by a dis rete-time sto hasti
pro ess (Yi ) (typi ally the temperature), omposed of independent variables with law N (0; (iY )2 )
positively orrelated with the pri e pro ess with orrelation oe ients (i ):

di = max(fi ; di + Yi )

(1.18)

where (fi ) are positive oors ensuring that the demand pro ess is positive, and (di ) are the average
demands at ea h period.
As a on lusion, to simulate the joint evolution of forward urve and demand at periods (pi ), we
24

only need to jointly simulate the random variables (Xi ) and (Yi ) for i = 1; :::; T and then use
formulas (1.17) and (1.18).

1.4

Numeri al results

Three dierent methods to numeri ally solve sto hasti ontrol problems
Due to the nite horizon and the omplexity of the onstraints, the problem must be solved numeri ally. Three dierent numeri al methods are possible:
1. The rst approa h onsists in using the dynami programming equation (1.15) to solve the
problem re ursively from stage i = T to stage i = 1. The advantage of this approa h is that
the al ulation time is linear with respe t to the number of time steps. The drawba k is
that omplexity explodes exponentially with the dimension of the state xi and the number
of de ision variables. In the retailer's ase, the dimension of the state spa e at stage i 1 is
(T

i + 1) (number of forward positions with delivery after period i 1) + 1 (forward pri e

risk) + 1 (demand risk) + 1 (storage level). We easily understand that this dimension an be
very large, and an be even larger if we onsider several storages and/or ontra ts. Longsta
and S hwartz (2001) proposed a numeri al method, ombining Monte-Carlo simulations and
dynami programming; this method allows to handle large dimension in the sto hasti pro ess

 but not in the de ision spa e, and therefore is not appropriate in our setting.
2. The se ond method that an potentially be used here is the one used by Unger (2002) for
hydro power risk management. It onsists in using Monte-Carlo simulations to adjust a priori
feedba k rules (i.e. rules relating de isions to realizations of  ), determined by a nite set of
parameters. This method has the advantage to potentially apture any type of dynami s for
25

the pro ess  (jumps, sto hasti volatility...) in a very easy way, but is not appropriate to
ompute onditional expe tations and impose to settle forms of feedba k rules a priori, whi h
is a di ult task in our ase, due to the number of parameters intervening in de isions
3. The third method whi h an be used is sto hasti programming, where a nite number of
s enarios are represented on an event tree, and where de isions have to be taken at every
node of the tree. The main drawba k of this method is the exponential growth of al ulation
time with the number of time steps. In spite of this major drawba k, this method has three
advantages that make it the most adapted to our problem: rst, the al ulation time does not
explode with the dimension of de ision spa e; se ond, the onditional expe tations that are
the ore of our optimization obje tive an be assessed very easily on an event tree; thirdly,
we do need to provide any form of feedba k laws but optimize dire tly our obje tive on the
set of all possible de isions at ea h node.

1.4.1 Expression of the retailer's problem on an event tree

Re ursive evaluation of obje tive on the tree: Let Nt denote the set of time t nodes of the
tree. Mathemati ally, a node n 2 Nt is an event representing a realization of the pro ess ( ) from
time 1 to time t. We all

N the set of all nodes belonging to the tree. The root node n = 1

represents the departing point.

Note that Nt+1 is a renement of Nt and therefore, every node n of time t  2 has a unique an estor
(or father) node a(n) at time t 1, dened as the unique member of Nt 1 whi h in ludes node n.
Nodes n belonging to the set NT are alled leaves. A s enario orresponds to a path from the root
to a leaf. The su essors (or sons) to node n form the set S (n); NT = fn : S (n) = ;g.
With the given transition probabilities nm from node n to node m 2 S (n), we dene a probability
26

n for ea h node n by the re ursion n := a(n) a(n)n and by 1 = 1. Clearly,

2Nt n = 1 holds

for ea h t = 1; :::; T .

Vn = Gn 8n 2 NT

(1.19)



Vn =  1 (Gn ) +  u 1 (E n (u(Vm )))
8
<

=  1 (Gn ) +  u 1 (
:

2S (n)

9
=

nm u(Vm ))

(1.20)

8n 2 Nt; t  T 1

(1.21)

Expression of s enario onstraints on the tree:

Stati onstraints bounding de isions of stage i only have straightforward translations on orresponding de isions on the tree. However, s enario onstraints (1.6), (1.9) and (1.10) will be expressed via the variables Ln , and Nn (p), representing respe tively the storage level and the forward
position for delivery in period p at node n; these variables are dened by the following relations
(remember that n = 1 is the root node of the tree):

L1 = Linit + q1inj

q1draw

(1.22)

Ln = La(n) + qninj

qndraw 8n 2 Nt ; t  2

(1.23)

N1 (p) = nb1 (p) ns1 (p)

(1.24)

Nm (p) = Na(m) (p) + nbm (p) nsm (p) 8t  2; 8m 2 Nt ; 8p  t

(1.25)

The orresponding onstraints are:

0

Ln

 Ln  Lmax 8n 2 Nt; 1  t  T

(1.26)

 Lend 8n 2 NT

(1.27)

draw
dm = Nm (t) + qm

inj
qm
8m 2 Nt; 1  t  T

(1.28)

We dene the set Atree of admissible tree strategies as de isions

draw ; q inj ; nb (p); ns (p))
(qm )m2N = (qm
m
m2N whi h respe t stati onstraints spe ied in 1.3.1 and
m
m

27

dynami onditions (1.22) to (1.28).

Optimization problem of the retailer

Max V
(qn )n2N 2Atree 1

(1.29)

We are left with a large-s ale ontinuous non linear optimization programme on variables (qn ) in
every node of the tree. We solve this problem numeri ally using a large-s ale non linear solver.

1.4.2 Building the event tree

To build the event tree, we use a two-dimensional latti e (see Webber (1997)), repli ating exa tly
the rst two moments of the pro ess (X; Y ) at ea h time step.
The four vertexes of the unit square rst provide the equiprobable joint realizations of a ve tor
~ Y~ ) of two un orrelated zero mean unit varian e random variables:
Z~ = (X;

6
( 1; 1)

(1; 1)

( 1; 1)

(1; 1)

Figure 1.1: S enarios for two un orrelated random variables

The extension to two orrelated variables is straightforward: onsidering a ve tor of two un or~ Y~ ), the ve tor of random variables Z = (X; Y ) = AZ~ with
related unit varian e variables Z~ = (X;
28

0
B

A=B


x

C
C
A

have zero mean and ovarian e matrix  = B



(x )2 x y

1
C
C.
A

p
y
1 2 y
x y (y )2
Therefore, we pro eed in the following
way to build1the event tree on the pri e/demand pro ess:
0
B

- rst, using the matrix M = B



C
C,
A

whose olumns represent the four joint re-

1 1 1
1
~ Y~ ) of two un orrelated zero mean, unit varian e variables, we form the
alizations of a ve tor (X;
2  4 matrix N = AM , whose olumns are the realizations of the ve tor (X1 ; Y1 ), representing the
pri e/demand nodes at time 1
- then, we atta h to ea h node of period 1 the son nodes given by the matrix N = AM , and so on,
until the last period
- nally, we apply formulas (1.17) and (1.18) to get the forward urve and the demand at ea h
node, the term Mi;j being determined by the martingale ondition at node n:

Fn (i 1; j ) = E n (Fm (i; j )) =

1
F (i; j )
4 m
m2S (n)
X

(1.30)

whi h gives:

Mi;j =

1
1 exp(e
m2S (n) 4

ki j i

) Xim )

(1.31)

It is important to point out here that the term M depends only on i and j and not of node n
be ause the variables (Xi ; Yi ) are independent of (Xi 1 ; Yi 1 ), hen e the sets fXim ; m 2 S (n)g are
the same for every node n of date i 1 .
We obtain 4T 1 dierent s enarios from period 1 to period T .

29

(a) Realizations of the forward urve (e/MWh)

(b) Realizations of demand (TWh)

( ) Two-dimensional representation of the pri e and demand pro esses (X; Y ) at ea h time
step: the realizations of the pri e pro ess X an be read on the x-axis
Figure 1.2: Event tree

30

1.4.3 The setting

We assume the following setting:
- the retailer is trading an energy produ t, whose pri e is expressed in e/MWh
- there are ve periods of one quarter ea h: during the rst quarter, the retailer fa es no demand
and replenishes her storage fa ility using the spot market in order to meet the unknown lient's
demand in the following year
- the storage has an initial level at 20 TWh, a maximal inje tion/withdrawal per period of 10 TWh,
a maximal (resp. minimal) storage level of 50 TWh (resp. 0), and a minimal end level of 20 TWh
- the forward pri e dynami s are represented by the model des ribed in equation (1.17) with parameters ki = 2 years 1 and volatility iX = 0:2 8i  2; the initial forward urve is supposed to
be at at the level 20 e/MWh; in parti ular, the initial spot pri e equals 20 e/MWh
- the maximal allowed traded volume in the market de reases with time-to-delivery: it equals 30
TWh for ontra ts delivering in the present quarter ("spot" transa tion), 10 TWh for ontra ts
delivering in the next quarter, 5 TWh for ontra ts delivering in two quarters, and 0 TWh for
ontra ts delivering in the following periods
- the selling pri e on the sale ontra t is 21 e/MWh (hen e a margin of 5% with respe t to the
average market forward pri e); regarding the demand hara teristi s, we suppose that d1 = 0, and
8i  2: iY = 10 TWh, di = 20 TWh, fi = d3i , and i = 0:5. The realizations of (X; Y ) at ea h
time step are represented on gure (1.2( )): we note that there are four dierent realizations for
the demand pro ess and two only for the pri e pro ess
- we adopt CARA utility fun tions u(x) = e

x

and (x) = e

x

to represent risk aversion and

substitution preferen es, with varying risk aversion and substitution parameters  and ; interest
rates are set to 0.
31

Figures (1.2(a)) and (2.24(b)) show the forward urve and demand s enarios. The mean-reverting
nature of the spot pri e is visible.

1.4.4 Ee t of optimal strategies on the nal and minimal wealths

Figure (1.3(a)) shows the mean varian e trade-o in the nal wealth obtained when risk aversion
varies and the fun tion  remains equal to identity. When the risk is dened as the Conditional
Value at Risk11 on the nal wealth WT 12 :

CV aRq (W ) = E ( WT j WT > V aRq (W ))

(1.32)

the expe ted mean is an in reasing fun tion of risk, as shown in gure (1.3(a)). For example, a
de rease of the 0.5% (resp. 5%) CVaR on nal wealth from 611 (resp. 505) to 371 (resp. 291)
Me implies a de rease of the expe ted nal wealth from 67 to 15 Me. Figure (1.3(b)) represents
the trade-o between the risks of the nal wealth and temporal minimal wealth13 . Figure (1.3(b))
shows that it is possible to ex hange bankrupt y risk for nal wealth risk by de reasing the ratio of
parameter  to parameter . For example, to ut the 0.5% (resp. 5%) CVaR on temporal minimal
wealth from 1059 to 545 (resp. 473) Me, one has to a ept a rise of the 0.5% (resp. 5%) CVaR
on nal wealth from 365 (resp. 296) to 516 (resp. 458) Me. However, the ex hange of bankrupt y
risk for nal wealth risk has limits: Figure (1.3(b)) shows in parti ular that it is not possible to
bring down the 0.5% (resp. 5%) CVaR on temporal minimal wealth below a ertain threshold,
orresponding to the pair ( = 0:1;  = 0:001) (resp. ( = 0:01;  = 0:0005)).
Figures (1.4(a)) shows the umulative fun tion of the nal wealth over the 256 tree s enarios used
11 V aRq (W ) is the well-known Value-at-Risk asso iated to quantile q
12 the wealth Wi at the end of period pi is dened as the umulative sum of ash ows from period p1 to period pi
13 Temporal minimal wealth is dened as mini2f1;2;3;4;5g Wi ; the temporal minimal wealth distribution is thus

dire tly linked to bankrupt y risk

32

(a) Expe ted nal wealth in terms of CVaR (in Me); ea h urve orresponds to a dierent
CVaR quantile and is onstru ted with  taking the values f0; 0:001; 0:005; 0:01; 0:02g

(b) CVaR of the temporal minimal wealth in terms of CVaR of the nal wealth (in Me);
ea h urve orresponds to a dierent CVaR quantile and is onstru ted with (; ) taking
the values (0:1; 0); (0:05; 0:0001); (0:02; 0:0001); (0:01; 0:0001); (0:1; 0:001); (0:01; 0:005);
(0:01; 0:001); (0:001; 0:0001)
Figure 1.3: Trade-os between expe ted wealth/nal wealth risk and nal wealth risk/bankrupt y
risk

33

in the optimization pro edure under dierent values of risk aversion. In gure (1.4(a)), we observe
that a risk aversion of 0:02 allows to signi antly redu e the left tail up to 5% of the distribution
obtained under a risk-neutral strategy. The ost of a higher risk aversion is that the main part
of the nal wealth distribution (to the right of the 10% quantile) is signi antly moved upright.
Figure (1.4(b)) shows the distribution of the minimal wealth over time: we see that a more on ave
fun tion  signi antly redu es the likelihood of a very negative minimal temporal wealth, whi h is
a onsequen e of the smoothing of ash ows in the time dimension. However, as shown by gure
(1.4(a)), if the ratio




be omes too high (e.g.( = 0:01;  = 0:0005)), the nal wealth distribution

exhibits a large left tail. If the portfolio manager seeks to strike a balan e between nal wealth
and bankrupt y risk management, she may hoose ( = 0:1;  = 0:001) or ( = 0:01;  = 0:0001).
Figure (1.5) represents the intermediate wealths obtained at the dierent nodes of the event tree
for dierent ouples of (; ) and onrms the above on lusions: hoosing ( = 0:01;  = 0:0005)
allows one to ontrol the intermediate wealth risk but implies a great dispersion of the nal wealth;
onversely, hoosing ( = 0:02;  = 0) oers a very narrow range of nal wealths but with a high
bankrupt y risk at the end of the se ond period; the hoi e ( = 0:01;  = 0:0001) represents a
trade-o between and nal and intermediate wealth risks.

34

(a) Final wealth umulative fun tion (in Me); the ase  = 0 (resp.  = 0)
orresponds to a fun tion u (resp. ) equal to identity

(b) Temporal minimal wealth (in Me) umulative fun tion in in omplete markets; the ase  = 0 (resp.  = 0) orresponds to a fun tion u (resp. ) equal
to identity
Figure 1.4: Final and temporal minimal wealth umulative fun tions for dierent risk aversion and
substitution parameters

35

(a) Wealth prole in the ase (0,0)

(b) Wealth prole in the ase (0.02,0)

( ) Wealth prole in the ase (0.01,0.0001)

(d) Wealth prole in the ase (0.01,0.0005)

Figure 1.5: Cumulative wealths (in Me) in the dierent nodes of the event tree for dierent pairs
(; )

1.4.5 Portfolio value

Figure (1.6(a)) represents the portfolio value dened in se tion 1.3.5 for dierent risk aversion
parameters. The portfolio value is a de reasing fun tion of the risk aversion parameter. The spread
between the risk-neutral and positive risk aversion values an be interpreted as a risk premium,
whose value in reases logi ally with the risk aversion parameter.
The value of the sale ontra t, obtained by setting the storage exibility to zero in the original
portfolio14, behaves similarly. The storage value, obtained by setting the lient's demand to zero in
14 Setting the storage exibility to zero may ause the problem to be infeasible in the ase of illiquid markets and

non-interruptible lients; estimating the sale ontra t value may thus require in some situations the introdu tion of
arti ial interruption/emergen y supply osts to relax the possibly too restri tive volume onstraints; in our example,
36

the retailer's portfolio, does not depend on the risk aversion parameter: this is due to the fa t that,
under the liquidity assumptions made in se tion 1.4.3, the storage fa ility has a unique arbitrage
value (here 55.26 Me) whi h an be se ured by appropriate forward transa tions; in this ontext,
the optimum J1 of the storage management problem redu es to the storage arbitrage value, as
explained in se tion 1.3.5. The synergy value, dened as the spread between the storage portfolio
value dened in se tion 3.5 and the storage arbitrage value, is null for a risk-neutral retailer and
in reases with the risk aversion parameter, whi h expresses the fa t that the synergy between sale
ontra t and storage fa ility is in term of risk management rather than in term of expe ted return.
Figure (1.6(b)) represents the synergy value in term of the risk aversion parameter under dierent
demand volatilities. It is observed that the synergy value in reases with demand volatility, whi h
means that the storage fa ility's value-added in the retailer's portfolio in reases with the volume
un ertainty. Figure (1.7) shows that the storage's value-added be omes null in a ontext of high
forward market liquidity, even in the presen e of volume un ertainty: the synergy ee t arises
only under an illiquid forward market. In addition, the portfolio value varies from 89 to 37 Me,
depending on the forward market liquidity, whi h points out the importan e of liquidity assumption
for portfolio valuation.

the lients' demand ould be met in every s enario only with the illiquid market
37

(a) De omposition of portfolio value for dierent risk aversion parameters

(b) Synergy value in term of risk aversion parameter for dierent demand
volatilities
Figure 1.6: De omposition of J1 (x1 ) =
Maxliq V1Id;u (G) (in Me) and synergy value for
(qk )k1 2A (x1 )
dierent risk aversion parameters and dierent demand volatilities

38

Figure 1.7: Portfolio and synergy values (in Me) for the dierent settings of forward market
liquidity des ribed in table (1.1) (with  = 0:01 and demand volatility  = 10 TWh)

Q0 Q1 Q2 Q3 Q4
low liquidity setting

30

10

medium liquidity setting 30

10

10

10

10

30

30

30

30

high liquidity setting

30

Table 1.1: Des ription of the three liquidity settings: Q0 represents the maximal volume of "spot"
transa tions, Q1 the maximal volume for delivery in the next quarter, Q2 the maximal volume for
delivery in the next following quarter...

1.5

Con lusion

We have developed in this paper a tra table model to introdu e time- onsisten y and inter-temporal
wealth management in optimizing a ommodity portfolio. In this order, we ompared stati risk
measures expressed on nal wealth with utility-type dynami risk measures: only the latter lead
to time- onsistent optimal strategies and disentangle the omponents of temporal substitution and
39

risk a ross states of nature. These properties are illustrated on a numeri al example. The use
of the model signi antly redu es the left tail in the nal wealth distribution, and leads to a
satisfa tory trade-o between nal wealth risk and expe ted wealth when risk is represented by
Conditional Value at Risk. In addition, the model allows one to dene an optimal strategy between
de reasing the risk of the nal wealth and redu ing the likelihood of a bankrupt y within the
time horizon. Lastly, our approa h allows one to assess the synergy value between the dierent
physi al assets omposing a portfolio, with important appli ations in term of ommodity portfolio
stru turing. Our urrent areas of investigation on ern the improvement of the omputing time
of the time- onsistent strategies and the omparison through simulations of strategies based on
re ursive utilities with strategies based on stati risk measures. Regarding the rst point, the
numeri al te hnique that is urrently used ex ludes for the moment a number of de ision steps
higher than 5, due to the non-linearity of the obje tive fun tion and the explosion of the number of
de ision variables with the number of time steps; approa hing the obje tive fun tion by a pie e-wise
linear on ave fun tion (whi h is theoreti ally possible thanks to the on avity property presented
in se tion 1.3.4) ould be a way of redu ing the problem to a linear programming one, whi h would
permit the in orporation of more than 10 de ision steps (the event tree would then ontain several
million nodes, implying a number of de ision variables whi h is ertainly ompatible with urrent
linear programming te hniques). Con erning the se ond area of resear h, the rst step onsists in
building a simulator apable of dynami ally reprodu ing strategies based on dierent risk measures
(e.g, re ursive utilities, expe ted nal wealth, expe ted nal wealth/CVaR on the nal wealth...)
under pri e/demand s enarios whi h are independent of the ones used for the optimization. Then,
the obje tive will be to assess the value-added of a temporally onsistent measure with respe t
to a stati risk measure in term of de ision planning robustness, of expe ted return, and of inter40

temporal risk management.

1.6

Annex: proof of the onvexity result

We use here the three following lemmas:

Lemma 1.6.1 Let f : Rn  Rm ! R be a on ave fun tion. Let A 2 Rmn and b 2 Rm . Dene

P (b) = fx 2 Rn ; Ax  bg and
g(b) = max f (x; b)
s:t:x2P (b)

(1.33)

Let Q = fb 2 Rm ; P (b) 6= ;g. Then Q is a onvex set and g : Q ! R is a on ave fun tion.

The proof an be found in Martinez-de-Albeniz and Sim hi-Levi (2005).

Lemma 1.6.2 On a probability spa e (

; F ; P), denote by L(
) the set of random variables with
values in

Rm .

Let f :

Rn

 Rm ! R ?

be a on ave fun tion with respe t to its rst argument

and X be a random variable in L(

), su h that 8  1; 8x 2 Rn ; E (( f (x; X )) ) < 1. Then,
for CARA utilities u(x) =

g(x) =  u 1

x

and (x) =

x

with 0 < 

 ,

the fun tion dened by



u  1 (f (x; X )) , is on ave.


Proof : Straightforward al ulations lead to g(x) = [E (( f (x; X ))  ) 

1
So the problem is equivalent to showing that g~(x) = [E (f~(x; X ) )  is onvex for f~ onvex with

respe t to x and   1.
To prove this property, we shall show that:

8(x1; x2 ) 2 Rn ; 8 2 [0; 1; g~(x1 + (1 )x2 )  g~(x1) + (1 )~g(x2 )

(1.34)

First, by the onvexity of f~:

1
g~(x1 + (1 )x2 )  [E ((f~(x1 ; X ) + (1 )f~(x2 ; X )) ) 

41

(1.35)

Then, from Minkowski's inequality, whi h is valid for   1, we get:

1
[E ((f~(x1 ) + (1 )f~(x2 ; X )) ) 

 [E ((f~(x1 ; X )) )  + [E (((1 )f~(x2; X )) ) 

1
1
= [E ((f~(x1 ; X )) )  + (1 )[E ((f~(x2 ; X )) ) 

= g~(x1 ) + (1 )~g (x2 )

(1.36)

The ombination of (1.35) and (1.36) leads to (1.34).

Lemma 1.6.3 On a probability spa e (

; F ; P), denote by L(
) the set of random variables with
values in

Rm .

Let f : Rn  Rm

!R

be a on ave fun tion with respe t to its rst argument and

X be a random variable in L(
), su h that 8
for CARA utility u(x) = e

x

2 R?+ ; 8x 2 Rn ; E (exp( f (x; X ))) < 1.

Then,

( > 0), the fun tion dened by g(x) = u 1 (E [ u(f (x; X )), is

on ave.

The proof an be found in Chen et al. (2004).

Remark: Lemma 1.6.3 an be interpreted as the limit ase of lemma 1.6.2 when  onverges to zero.


To prove the on avity of (Gi (qi )) +  u 1 (E i (u(Ji+1 (xi+1 )))) 15 with respe t to de isions

qi , we pro eed by ba kward re ursion on i:

- Let us begin with the ase i = T
We denote by kT the number of de ision variables at time T and dene:
n

BTliq (dT ) = (LT ; N (T 1; :)); Aliq

1; :); dT ) 6= ;
T (LT ; N (T
15 The proof that follows is valid for both ases:

- (x) =
-  = Id

x

with   
42

(1.37)

Dene fun tion J~T by:

BTliq (dT ) 7!
(LT ; N (T

1; :))

7! J~T (LT ; N (T 1; :)) =

qT

Max

2Aliq
T (LT ;N (T 1;:);dT )

(GT (qT ))

The set of onstraints an be put under the equivalent form:

qT

8
>
>
<

2 A (xT ) () >
liq
T

>
:

0  LT + qTinj

N (T

qTsout  Lmax

1; T ) + nb (T; T ) ns (T; T ) + qTsout

qTinj = dT

Here, we omitted all bounding onstraints on individual de isions qTinj ; qTsout ; nb (T; T ); ns (T; T ).
1; T ); dT ) an thus be put under the linear form:
Aliq
T (LT ; N (T

AqT

 B (LT ; N (T 1; T )) + C

A, B , and C being appropriate matri es and ve tor. Hen e, using lemma 1.6.1, we know that J~T is
on ave with respe t to B (LT ; N (T

1; T )) + C , and thus also with respe t to ve tor (LT ; N (T

1; T )) on BTliq (dT ).
In addition, we see that:

(LT ; N (T

Hen e, 8dT ;
8
>
>
>
>
>
>
<
>
>
>
>
>
>
:

1; :)) 2 BTliq (dT ) ()

8
>
>
>
>
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
>
>
>
>
:

Lend  LT

 Lmax

N (T

1; T )  dT

N (T

1; T )  dT + nmax
(T; T ) + Qinj
s
T

nmax
(T; T ) Qdraw
T
b

N (T

1; T ) + LT

 dT + nmax
(T; T ) + Lmax
s

N (T

1; T ) + LT

 dT nmax
(T; T ) + Lend
b

BTliq (dT ) has the form:

LT

N (T

2 [0; Lmax

max
1; T ) 2 [
min
(dT ); N (T
T (dT );
T

43

1; T ) + LT

2[

min
T

(dT );

max
T

(dT )

max
where
min
is a onstant, and
min
,
T
T ,
T

min
T

max
T

are fun tions of the nal demand.

- Now, let us denote by ki the number of de ision variables at stage i and dene:
n

Biliq (di) = (Li; N (i 1; :)); Aliq

1; :); di ) 6= ;
i (Li ; N (i
Suppose that 8 di+1 ;
8
>
>
>
>
>
>
>
>
>
>
<

(1.38)

Biliq+1(di+1 ) is of the form:

Li+1 2 [0; Lmax
max
N (i; p) 2 [
min
i+1 (di+1 ; p);
i+1 (di+1 ; p) 8p  i + 1;

>
>
>
>
>
>
>
>
>
>
:

N (i; i + 1) + Li+1 2 [ +1 (di+1 ); +1 (di+1 )

min
i

(1.39)

max
i

min
max
where
min
i+1 is a onstant,
i+1 and
i+1 are fun tions of i + 1 demand and delivery period p,

+1 and max
i+1 are fun tions of period pi+1 demand. Assume also that, for every realization of

min
i

i+1 , the fun tion J~i+1 (Li+1 ; N (i; :)) = (Ji+1 (Li+1 ; N (t; :); i+1 )) is on ave with respe t to ve tor
(Li+1 ; N (i; :)). Dene fun tion J~i by:

Biliq (di ) 7!
(Li ; N (i 1; :))
(with xi+1 = (Li + qiinj

7!

Max
(Gi (qi )) +  u 1 (E i (u(Ji+1 (xi+1 ))))
qi 2A (
( 1;:);di )
liq
i Li ;N i

qisout ; N (i 1; :) + nb(i; :) ns (i; :); i+1 ))

Remark that J~i (Li ; N (i 1; :)) = (Ji (Li ; N (i 1; :); i )).

Lemma 1.6.2 (resp 1.6.3) ensures that the fun tion  u 1

u 1

E i

i

E i

i

u  1 (J~i+1 (xi+1 ))

(resp.

u(J~i+1 (xi+1 ) ) is on ave with respe t to variables Li+1 , and N (i; p)pi+1 .

As (Li+1 ; N (i; :)) = (Li + qiinj

qisout ; N (i

1; :) + nb (i

1; :)

ns (i

1; :)), this fun tion is

jointly on ave with respe t to de isions qi and state (Li ; N (i 1; :)) on Rki  Biliq (di ). Hen e, the
date i obje tive fun tion (Gi (qi )) +  u 1 (E i (u(Ji+1 (xi+1 )))) = (Gi (qi )) +  u 1 (E i (u

 1 (J~i+1 (xi+1 ))) is the sum of the on ave fun tion (Gi (qi )) and of the on ave fun tion 
44

u 1

E i

i

u  1 (J~i+1 (xi+1 )) . It is thus jointly on ave with respe t to de ision variables qi and

ve tor (Li ; N (i

1; :)) on Rki

 Biliq (di ). Let us now examine the set of onstraints. Denoting by

Hi(di ) the (nite) set of date i+1 demands bran hing from date i demand di, we have:
8
>
>
<

qi 2 Aliq
1; :); di ) ()
i (Li ; N (i
(Li + qiinj

qisout ; (N (i 1; p) + nb (i; p) ns(i; p))pi+1 ) 2 Biliq+1 (di+1 ) 8di+1 2 Hi (di )

>
>
:

N (i 1; i) + nb (i; i) ns (i; i) + q

sout
i

inj
i

= di

(1.40)

Here again, we omitted bounding onstraints on de isions qi .

First, the ompatibility between (1.40) and the bounding onstraints on (qi ) implies that Biliq (di )
is of the form (1.39). The form (1.39) hen e passes from i + 1 to i.
In addition, as 8di+1 ;

Biliq+1(di+1 ) is of the form (1.39), the set of onstraints Aliq

1; :); di )
i (Li ; N (i

an be put under the linear form:

Aqi  B (Li ; N (i 1; p)ipT ) + C

A, B , and C being appropriate matri es and ve tor.
Using lemma 1.6.1, we know that fun tion J~i is on ave with respe t to variables B (Li ; N (i
1; p)ipT ) + C , and thus also on ave with respe t to variables (Li ; N (i 1; p)pi ), whi h ends the
proof.

1.7


Referen es

Artzner P.,Delbaen F., Eber J.M., Heath D. (1999), Coherent Measures of Risk, Mathemati al
Finan e 9: 203-228
Artzner P., Delbaen F., Eber J.M., Heath D. , Ku H. (2002), Coherent Multiperiod Risk Measurement, Working Paper

45

Chen X., Sim M., Sim hi-Levi D., Sun P. (2004), Risk Aversion in Inventory Management, a epted
by Operations Resear h
Ei hhorn A., Romis h W. (2005), Polyhedral Risk Measures in Sto hasti Programming, SIAM J.
Optim. 16 , pp. 69-95
Epstein G., Zin S. (1989), Substitution, Risk Aversion, and the Temporal Behavior of Consumption
and Asset Returns: A theoreti al framework, E onometri a, Vol. 57, No. 4, pp. 937-969

Frittelli M., Rosazza Gianin M.(2002), Putting order in risk measures, Journal of Banking and
Finan e, Vol. 26 pp. 1473-1486
Frittelli M., Rosazza Gianin M.(2004), Dynami onvex risk measures, New Risk Measures for the
21th Century, G. Szego ed., John Wiley & Sons, pp. 227-248
Lide Li, Paul R. Kleindorfer (2004), Multi-Period VaR-Constrained Portfolio Optimization with
Appli ations to the Ele tri Power Se tor, Energy Journal

Longsta F., S hwartz E.S (2001), Valuing Ameri an Options by Simulation: A Simple Least
Squares Approa h, The Review of Finan ial Studies, Vol. 14, pp. 113-147

Martinez-de-Albeniz V., D. Sim hi-Levi (2003), Mean-Varian e Trade-os in Supply Contra ts,
Working paper, MIT
Martnez de Albniz, V. Sim hi-Levi, D. (2005), A portfolio approa h to pro urement ontra ts, Produ tion and Operations Management, Vol. 14, No 1, pages 90-114
Unger G. (2002), Hedging Strategy and Ele tri ity Contra t Engineering, PhD Thesis, ETH Zri h
Wang, T. (2000), A Class of Dynami Risk Measures, under revision
Webber N., M Carthy L. (1997), An I osahedral Latti e Method for Three Fa tor Models, Working
Paper

46

Chapter 2

A new dependen e model for

ommodity forward urves;
appli ation to the US natural gas and
oil markets1

The goal of this paper is to present and alibrate a model for the joint evolution of orrelated
ommodity forward urves. The main originality of the model is that it aptures both the lo al
and global dependen e stru tures of two forward urves, through an error- orre ting term in the

risk-premia of the forward pri e returns. The model is applied here to the US oil and gas forward
markets, whi h have strong e onomi relations, from the demand and supply sides.
1 I thank Gaz de Fran e for providing me with the data, Celine Jerusalem for helping me to treat them, Gregory

Benmenzer, Olivier Bardou, and Jean-Ja ques Ohana for stimulating dis ussions
47

2.1

Introdu tion

Surprisingly, the modeling of the o-movements of ommodity forward urves has re eived very little
attention in the nan ial literature. Yet, this is a subje t of onsiderable importan e for the pri ing,
risk management, and optimization of portfolios omposed of multi- ommodity assets su h as gasred power plants, oil-indexed natural gas ontra ts, or oil reneries. Indeed, the nan ial value of
a multi- ommodity asset is a fun tion of the entire forward urves and the hedging strategies for
multi- ommodity portfolios are based on futures ontra ts rather than spot transa tions. As a onsequen e, a model des ribing the evolution of ommodity spot pri es only, provides a partial view
of the risks/value entailed in su h portfolios and of the possible a tions of the portfolio manager.
A model des ribing the joint evolution of two ommodity forward urves should apture at the
same time their global and lo al dependen e stru tures. The lo al dependen e stru ture des ribes
the volatilities, the marginal densities and the orrelations of the daily forward urve moves. A
framework of analysis for this type of dependen e was des ribed in Clewlow and Stri kland (2000),
who propose to extend the lassi al PCA on one ommodity forward urve to a PCA on the returns
of two ommodity forward urves, thus obtaining several types of o-movements of the two forward
urves. By ontrast with the lo al dependen e stru ture, the global dependen e stru ture des ribes
the long-term relations existing between ommodity pri es2. Mu h attention has been devoted
to the study of ointegration between series of dierent spot/futures ommodity pri es3, with a
2 two

frequent examples of long-term intera tions between ommodity markets are the possibility to use a given
ommodity to produ t another one (natural gas to produ e power, rude oil to produ e heating oil...) or to use a
given ommodity as a substitute to another one (e.g. heating oil instead of natural gas for heating, oal instead of
natural gas to produ e power)
3 see e.g. Alexander (1999) for a study of the ointegration between gas/oil spot and futures pri es on the NYMEX,
Ates and Wang (2005) for an analysis of the relations between spot and rst-near by natural gas pri es in the US,
48

view to des ribing the intera tion between several parti ular points in the same forward urve or
in dierent forward urves (for example the relations between the front-month pri es of a pair of
ommodities or the relations between the spot and front-month pri es of the same ommodity).
There is extensive work also on the evolution of a single interest rate or ommodity forward urve,
either for fore asting (see Diebold and Li (2003)) or VaR al ulation (see e.g. Brooks (2001)). But
no work, to our knowledge, has ever proposed a framework to simulate the evolution of two entire
ommodity forward urves, des ribing the way the two urves "revert to ea h other". The retained

approa h for this problem follows Pilipovi (1997), Manoliu and Tompaidis (2002), S hwartz and
Smith (2000), and Geman and N'Guyen (2005), who de ompose the daily deformations of a forward
urve into a short-term sho k, ae ting only the rst maturities, and a long-term sho k, onsisting of an overall translation of the forward urve. Regarding the lo al dependen e stru ture, the
model aptures, on the one hand, the ausal relations between the daily short-term and long-term
sho ks of the two ommodities, and on the other hand, the time-dependent volatilities of the four
omovements (see e.g. Geman and Nguyen (2005), Ri hter and Sorensen (2000), and Due (2002),
for eviden e of sto hasti ity of the volatility of ommodity pri es, and Blix (2003) for eviden e of
seasonality of natural gas impli it volatility), and their possibly non Gaussian dependen e stru ture
(see e.g. Eydeland (2003) for eviden e of the non-normality of energy (log) pri es). The approa h
to apture the long-term relations between two forward urves an be viewed as an extension of the
on ept of ointegration to forward urves. The de omposition of the forward urve daily moves

Siliverstovs et al. (2005) for an analysis of ointegration between Japanese, European, and North Ameri an gas
pri es, Nguyen (2002) for the analysis of the ointegration between the futures pri es of metals on the London Metal
Ex hange, Pekka and Antti (2005) for the study of ointegration between spot and futures ele tri ity pri es on the
NordPool
49

translates into a de omposition of the shape of the forward urve into a seasonal term, slope4 and
level5 . The long-term relationships between the two ommodity forward urve slopes and levels
are looked for and the deviations to these equilibriums be ome predi tive variables for the future
relative evolution of the two urves. The model is applied here to the US natural gas, rude oil
and heating oil markets during the period 99-2004. These three markets, in spite of their dieren es, are intertwined by e onomi relations, from the onsumption side and the produ tion side.
Regarding the lo al dependen e stru ture, we nd eviden e of ausal relations between natural
gas and oil sho ks, sto hasti volatility for the dierent sho ks, seasonal volatility for natural gas
and heating oil short-term sho ks only, and positive orrelations between the o-movements of oil
and gas forward urves. Regarding the global dependen e stru ture, our analysis highlights the
existen e of a strong long-term relationship between the levels of natural gas and oil (with two
break points o urring in the beginning of year 2000 and in the middle of year 2003), and of a
weaker long-term relationship between natural gas and oil slopes. The analysis of the temporal
stability of the model parameters reveals that the error- orre tion me hanisms have been stronger
sin e 2002, that the orrelations between the daily o-movements of oil and gas forward urves
have in reased signi antly throughout the period 1999-2004 and that the orrelation between the
short-term sho ks of natural gas and heating oil peaked during the tight market winters 2000-2001,
2002-2003, and 2003-2004.
I view the ontribution of this paper as threefold: from an e onomi standpoint, the presented
forward urve model sheds light on the relations between the natural gas and oil markets in the
US, spe ifying at the same time the short-term and long-term relations between the three energies;
4 depending on the sign of the slope, the urve will be said to be in ontango or in ba kwardation
5 from an e onomi standpoint, the level is linked to the stru tural pri e of the ommodity, as observed

quotation date, and the slope to the short-term situation of inventory, produ tion, and demand
50

in the

from a statisti al standpoint, the model proposed here opens a new avenue for the modeling of
the joint evolution of several orrelated forward urves, giving a simple way to apture in a single
arbitrage-free model the long-term relations between the shapes of dierent forward urves and the
lo al statisti al relations between their daily o-movements. Lastly, from a nan ial standpoint,
the model developed here has important appli ations in terms of multi- ommodity asset pri ing,
of ommodity portfolio risk management and of ommodity portfolio optimization; as far as asset
pri ing is on erned, Duan and Pliska (2004) have shown that the ombination of ointegration and
sto hasti volatility has an impa t on asset pri es: thus the model would lead to dierent pri ing
results than standard lo al dependen e models without risk-premia. Regarding risk management,
the model, be ause it aptures the long-term relations between two urves, allows one to realisti ally simulate the evolution of two forward urves on long time horizons, whi h is important for
the estimation of portfolios' Earning-at-Risk on a long-term perspe tive. With respe t to portfolio optimization, our error- orre tion model allows the portfolio manager to fore ast the relative
evolutions of the two onsidered forward urves given their initial slopes and levels, a property
whi h has numerous impli ations; hedge funds will be provided with dire tional strategies based on
long/short positions on the two urves while physi al portfolio managers will have a way to hoose
the best moments to lo k in the margin of their assets with futures ontra ts.
The rest of this paper is organized as follows. In se tion 2, we des ribe the e onomi relations
between oil and natural gas markets in the US, from the demand side and the oer side. In se tion
3, we present the two-fa tor model and des ribe the global and lo al dependen e stru tures between
oil and gas forward pri es in the US. In se tion 4, the model is pre isely alibrated and the temporal
stability of the model parameters is studied. Se tion 5 ontains on luding omments.

51

2.2

The e onomi relations between oil and natural gas in the US

Even though there are strong dieren es between the natural gas and oil markets in the US (the
natural gas market is a ompetitive and lo al market whereas the rude oil market is an oligopolisti
and world market), the natural gas and oil pri es are intertwined by strong e onomi relations, both
from the demand and supply sides.

2.2.1 Dependen e through the demand

Figure (2.2( )) shows the repartition of natural gas and petroleum onsumption in the US by enduse. It appears that industry is the se tor where the overlapping of the two energies is the most
signi ant, the se ond one being the segment of residential/ ommer ial ustomers. As shown in
gure (2.1(b)), industry represents approximately 1/3 of the global US gas onsumption. Around
30% of the natural gas onsumption of this se tor, where we nd petroleum, hemi als, paper, and
metal industries, are represented by ustomers with dual-fuel apa ity (essentially old power plants
and boilers), who are able to swit h from natural gas to oil (generally distillate or residual fuel oil)
depending on the market pri es of the two energies6 . Figure (2.2(a)), taken from the Ameri an
Gas Foundation (2003), shows that the natural gas demand urve is omposed of in exible and
exible parts. When natural gas pri es are higher than distillate pri es, all the industrial and
power generation ustomers with dual-fuel apability are o-gas and the natural gas demand, due
to residential and ommer ial ustomers, shows very little pri e elasti ity. In this ontext, the
demand is mainly impa ted by the weather as residential and ommer ial ustomers use natural
gas for heating purposes essentially (indu ing the seasonal natural gas onsumption prole shown in
6 Note that the re ent environmental regulations, imposing air pollutant emission onstraints to industrials, tend

to prevent them from using distillate fuel or oal as a substitute to natural gas
52

gure (2.1(a))); this demand is slow in rea ting to higher gas pri es be ause it is mostly represented
by non-interruptible lients without dual-fuel apa ity and be ause the utility rates for this type
of lients respond to market pri es with a lag. When gas pri es are ompetitive with residual fuel
oil and/or distillate fuel oil, natural gas demand is mu h more pri e elasti , as the industrial and
power generation ustomers with dual-fuel apability hoose the less expensive fuel to run their
a tivities. Lastly, when natural gas pri es are below the point at whi h most dual-red apa ity
has swit hed from oil to natural gas, the demand is again insensitive to pri e variations. The
middle graph in gure (2.2(b)) shows that, in a situation of "stable pri es", a demand sho k,
resulting in a right translation of the demand urve, leads to a moderate natural gas pri e in rease.
Natural gas ustomers with dual-fuel apa ity (8-10 % of the global natural gas onsumption),
who have swit hed from natural gas to oil have ompensated for the in reased onsumption of
in exible natural gas onsumers. However, starting from the middle situation ("tighter natural
gas market"), when most of the fuel swit hable apa ity has already swit hed away from natural
gas, a further rise of the natural gas demand, whi h ould be provoked for instan e by a older
weather, an lead to a pri e spike. Of ourse storage, whi h has an ee t on the supply urve in
the winter (see gures (2.1(a)) and (2.2(a))) is a ru ial adjustment variable in this ontext. These
observations a ount for the variability of natural gas pri e volatility observed in the US market,
a phenomenon whi h will be studied in details in the next se tions. Figure (2.3) shows that an
oil pri e rise results in an upward translation of the demand urve, whi h in turn, leads to an
in rease of the natural gas pri e. This gure a ounts for the positive orrelations between the
US natural gas and oil pri es. Note that the ee t illustrated in gure (2.3) an happen both in
the very short term (for industrial onsumers already having exible swit hing apa ities), ausing
positive orrelations/ ausalities in the daily movements of natural gas and oil pri es, and in the
53

medium-long-term (for industrial onsumers who progressively instal a fuel swit hing-te hnology to
adapt to a ontext of a heap/expensive gas/oil pri es in the long run), ausing long-term relations
between the pri es of the two energies.
In addition, we expe t a spe i orrelation between natural gas and heating oil pri es as the demand
of these two energies is in uen ed by the weather and the US heating oil market an dis onne t from
the rude oil world market in the short-term due to ongestion/disruption problems in the rening
system; gure (2.2.1) shows that the apa ity surplus in the US rening industry is progressively
disappearing, a phenomenon responsible for the frequent ongestions in the US rening system and
instable ra k spreads between rude oil and the rened produ ts observed in the re ent years.
The EIA (2001) provides a detailed analysis of the orrelation between natural gas and heating
oil markets during winter peaks; in parti ular, in January 2000, where the market was tight for
natural gas and heating oil in the US, interruptible natural gas onsumers whose deliveries were
interrupted due to severe weather onditions, pur hased and burned fuel oil instead of natural gas
for heating, whi h led to a surge in the heating oil pri es in the US (even though the orresponding
volumes were very low).
In addition, be ause industrials often lo k in their margins using the forward markets, we expe t
positive orrelations not only between oil and gas short-term pri es but between oil and gas forward
pri es as well; for example, if industrials with dual-fuel apa ity observe that gas forward pri es are
above oil substitutes for a given maturity, they will pur hase oil forward ontra ts instead of gas
forward ontra ts for that maturity, leading to a orre tion of the spread between the two forward
pri es. This onvergen e between gas and oil forward pri es is reinfor ed by the urrent behaviour
of hedge funds and nan ial investors, who tend more and more to onsider the dierent ommodity
markets as a unied asset lass (see Geman (2005)).
54

(a) Monthly natural gas onsumption, produ tion, and net im- (b) Repartition of natural gas onsumption in the US
ports in the US (Sour e: EIA)
by end-use (Sour e: EIA)
Figure 2.1: Natural gas onsumption, produ tion, and imports in the US; in the left graph, the
gap between the seasonal onsumption and the at produ tion/import urves is lled by storage
movements

55

(a) demand and supply urves for natural gas (Sour e: (b) volatility of natural gas pri es in dierent market onAGA)
ditions (Sour e: AGA)

( ) Natural gas and oil onsumptions by end-use (Sour e: EIA)

Figure 2.2: E onomi relationships between oil and gas: demand side

56

Figure 2.3: Impa t of an oil pri e rise on natural gas pri e

2.2.2 Dependen e through the supply

The dependen e between oil and natural gas pri es in the US is also originating in the supply side.
The Gulf of Mexi o on entrates indeed major gas and oil elds, gas pro essing plants, and oil
reneries (see gure (2.5(a))). When the hurri anes Katrina, Rita, and Wilma stru k this region,
they ae ted at the same time the produ tion of natural gas, rude oil, and rened produ ts in the
US, thus ausing a rise in pri es of all these energies (see gures (2.5(b)), (2.5( )), and (2.5(d))).
The supply of rened produ ts in the US was impa ted be ause the Gulf of Mexi o on entrates
a large part of the US rening apa ity and be ause the import of rened produ ts from abroad
represents only 3% of the total US demand. The same applied to the US natural gas market, whi h
relies at more than 80% on the US produ tion (18% of whi h omes from the Gulf of Mexi o) to
meet the domesti demand, and where gas pro essing plants play a ru ial role in the treatment of
extra ted natural gas. The rude oil pri es were also ae ted, due to the tightness of the rude oil
world market at the time of Katrina's landfall.

57

Figure 2.4: Rening apa ity in the US (Sour e: EIA)

58

(a) Natural gas elds and pro essing plants in the US (Sour e: (b) Natural gas and oil produ tion interrupEIA)
tions after the hurri anes (Sour e: EIA)

( ) Impa ts of the hurri anes on natural gas and (d) Impa ts of the hurri anes on gasoline and
rude oil pri es (Sour e: EIA)
heating oil pri es (Sour e: EIA)
Figure 2.5: E onomi relations between oil and gas: supply side

59

2.3

Empiri al observation of the dependen e between oil and gas

forward urves in the US

2.3.1 Data des ription

The data used here are the NYMEX daily futures pri es of natural gas, rude oil, and heating
oil, from January 1999 to the end of O tober 2004. For the three energies, the pri es are the 1st
month, 2nd month,...,15th month futures pri es. Con erning natural gas, the pri e is based on
delivery at the Henry Hub in Louisiana, the rossing of 16 intra- and interstate natural gas pipeline
systems that draw supplies from the region's proli gas deposits. The pipelines serve markets
throughout the U.S. East Coast, the Gulf Coast, the Midwest, and up to the Canadian border.
The futures pri es are expressed in dollars per Million British Thermal Units (MMBtu). For rude
oil, the NYMEX futures ontra ts's delivery point is Cushing, Oklahoma, whi h is also a essible
to the international spot markets via pipelines and the pri es are expressed in dollars per barrel.
The NYMEX rude oil futures ontra t is the world's largest-volume futures ontra t trading on a
physi al ommodity. Be ause of its ex ellent liquidity and pri e transparen y, the ontra t is used
as an international pri ing ben hmark. Lastly, the Heating Oil futures ontra ts are based upon
delivery in New York harbor, the prin ipal ash market trading enter, and the pri es are expressed
in dollars per Gallon (1 Gal = 42 Barrel).

60

(a) Crude Oil futures pri es in dollars/Barrel

(b) Heating Oil futures pri es in dollars/Gal

( ) Natural Gas futures pri es in dollars/MMBtu

Figure 2.6: Pri e traje tories from January 99 to O tober 2004

61

Figure (2.6) represents the traje tories of 1st month and 15th month futures pri es for the three
energies:

 the traje tories of Crude Oil and Heating Oil 15th month futures pri es are quasi identi al
 there are dieren es in the short term between rude and heating oil
 the trends of natural gas and oil 15th month futures display a parallel dire tion
 even though the 1st month natural gas futures pri e exhibits mu h larger moves than oil
within the period, oil and gas approximately share the same ba kwardation and ontango
periods7

 the period 1999-2004 an be separated in several subperiods:

{ from January 1999 to end of 2001, the three traje tories display a "bump": they rst
follow an upward trend until the end of 2000, and then a de ay until the end of 2001

{ in the years 2002-2003, gas pri es start rising while oil pri es remain stable
{ from the beginning of 2004 to now, the three energies display a very lear surge, with
an exponential speed for oil and a linear speed for natural gas

2.3.2 De omposition of daily forward urve moves into short term and longterm sho ks
Justi ation and interpretation of the de omposition
In gure (2.7), it appears that forward urve moves de ompose into a long-term sho k, whi h
provokes a global upward or downward translation of the forward urve, and a short term sho k,
7 There

are a few notable ex eptions to this rule su h as the summer 2004, when the gas forward urve was in
ontango and the heating oil and rude oil urves were ba kwardated
62

whi h only impa ts the short term futures pri es, with an amplitude that de ays with time-tomaturity. In e onomi terms, the interpretation of the de omposition is the following:

 the short term sho k refers to events that are expe ted to ae t the market for a limited period
of time (temperature hange, transitory supply shortage or transportation ongestion...)8

 the long-term sho k relates to events or news that potentially impa t the long-term energy
pri e (news about the likelihood of a war or politi al instability in an oil produ ing ountry,
dis losure of lower than expe ted reserves...)

8 One ould wonder why events of weekly time s ale su h as a temperature drop or a bottlene k in the transportation

system should ae t the pri es of the ontra ts delivering in the following months; this link between spot and forward
markets is explained by the storability of the three onsidered energies. Indeed, tensions in the day-ahead market
prompt utilities and distribution ompanies to pump on their reserves in order to take advantage of high spot pri es
or be able to deliver their rm lients; this in turn reates a situation of s ar ity in the medium term, whi h, as
explained by the theory of storage, has a dire t impa t on the slope of the monthly forward urve
63

(a) Natural Gas futures pri es (in $/MMBtu) as a fun tion of time to maturity (in months)
from January 4th to January 19th, 1999

(b) Natural gas futures pri es returns as a fun tion

of time to maturity (in months) from
64
January 5th to January 19th, 1999
Figure 2.7: De omposition of returns into a short and long term sho ks

Mathemati al formulation of the de omposition

We denote Fe (t; T ) the futures pri e at time t of energy e for delivery at month T . For the sake of
simpli ity, we assume that delivery o urs the last trading day of the futures ontra t. We assume
the following arbitrage-free daily evolution model for the forward urve of energy e:

Fe (t; T )
=e
Fe (t; T )

)
X e +
Y e

ke T t

(2.1)

X;t e;X

Xte = X;t
e + e t
Y;t e;Y

Yte = Y;t
e + e t

where:
Y;t
 (X;t
e ) and (e ) are (Ft )-adapted pro esses representing the drifts

 (eX;t ) and (eY;t) are (Ft )-adapted pro esses representing the volatilities
 (te;X ) and (te;Y ) are orrelated pro esses formed of i.i.d variables
 k1e represents the hara teristi time of the short term sho k
Cal ulation of the short term and long-term sho ks
Assuming that the short term sho k does not ae t the 14th month return9, the short term and
long-term sho ks an be readily derived from the observed short term and long-term returns of
energy e:

X

e
t

Fe (t; T14 )

Fe (t; T14 ) obs

ke (T1 t)
Fe (t; T1 )
= e
Fe (t; T1 )

Yte =

9 This is equivalent to the assumption 3  1 < 14 months

ke

65

(2.2)

Fe (t; T14 )
Fe (t; T14 )


obs

where the variable Ti denotes the last trading day of the i-th month futures ontra t observed at
date t.

Estimation of the short term hara teristi time for the three energies
To estimate ke for the three energies, we minimize the root mean squared errors (RMSE) i.e., the
root of the mean squared dieren es between the observed returns and the model implied returns


Fe (t; Ti )
Fe (t; Ti )

Fe (t; T14 )

=
Fe (t; T14 )
model


obs

+e

ke Ti T1

)
Fe (t; T1 )
Fe (t; T1 )

Fe (t; T14 )

Fe (t; T14 )


obs

(2.3)

Therefore, we solve, for ea h energy e, the following programme:

M inRMSE
ke

v
u
u
=t

N X
14 
F (t; T ) 
X
e
i

N  14 t=1 i=1

Fe (t; Ti )

obs

Fe (t; Ti )
Fe (t; Ti )

2


model

(2.4)

where N is the number of observations.

rude oil heating oil natural gas

ke

2:52

3:34

3:33

1 (in months)

4:77

3:59

3:60

ke

Table 2.1: ke for the three energies

rude oil heating oil natural gas

ke

2:52

3:10

2:97

1 (in months)

4:77

3:87

4:04

ke

Table 2.2: ke for the three energies when ex eptional deformations are removed (the date t is
r
h

i2
P

Fe (t;Ti ) 

Fe (t;Ti ) 
removed if RMSEt = 141 14
> RMSR)
i=1
Fe (t;Ti ) obs
Fe (t;Ti ) model
66

Table (2.1) reports the short term hara teristi times of the three energies. As a rst observation, these hara teristi times are ompatible with the assumption 3  k1e < 14 months, whi h
helped us al ulate the short term and long-term sho ks. In addition, we observe that the short
term hara teristi times of natural gas and heating oil are similar and signi antly smaller than
the one of rude oil. The e onomi interpretation is that the short term sho ks in the heating oil
and natural gas lo al markets are linked to very short-lived events (e.g., sudden drop of temperature
in the US, bottlene k in the US rening system et ...) whereas the short term sho ks in the global
rude oil market orrespond to events with a longer time s ale (e.g. dis losure of a lower than
expe ted world inventory).

67

29

35

33
27
31

25

29

27
23
25

21

23

21
19
19

17

17
1

(a) RMSE/RMSR (in %) in terms of ke for rude oil

46

42

38

34

30

26
2

( ) RMSE/RMSR (in %) in terms of ke for natural gas

Figure 2.8: RMSE/RMSR (in %) in terms of ke

68

(b) RMSE/RMSR (in %) in terms of ke for heating oil

50

Figure (2.8) shows the behavior of the relative error fun tion (i.e., the square root of the mean
squared errors (RMSE) divided by the square root of the mean squared returns (RMSR)) in terms
of ke . We see that the performan e of the model in explaining the varian e of the observed returns
of the residuals = RM SE
2 of 7%) than for
is signi antly lower for natural gas (with a ratio varian e
total varian e
RM SR
of the residuals of 3%). A rst explanation is that the relative importan e
oil (with a ratio varian e
total varian e
of "twist" moves (whi h are not a ounted for in the two fa tor model) in the global forward
urve volatility is more pronoun ed for natural gas than for oil. This is onrmed by a Prin ipal
Component Analysis on the 14 series of forward urve returns, whose results are displayed in table
(2.3):
rude oil heating oil natural gas
1st fa tor

95:95%

95:13%

92:54%

2nd fa tor

2:94%

3:77%

4:95%

3rd fa tor

0:41%

0:81%

1:34%

Table 2.3: Proportion of overall varian e explained by the 1st (translation), 2nd (rotation), and
3rd fa tors (twists) for the three energies
A se ond explanation ould be that kgaz is more variable than k rude and koil . Figure (2.9)
represents the evolution of the optimal ke 10 for the three energies with a 6 months time step from
summer 99 to summer 2004. We onsider that summer months are April, May,..., September and
winter months are O tober, November,..., Mar h. We had to remove the dates when the mean
squared error of the two-fa tor model was greater than the mean squared returns over the whole
period, be ause otherwise, the optimal ke took abnormally high values at some point during the
10 the optimal ke

on period p is the one minimizing the RMSE of the two-fa tor model on period p
69

horizon. Thus, the plotted traje tories orrespond to those ke that t the "normal" movements
of the forward urve. We see in table (2.2) that the removal of the ex eptional moves leads to
signi antly smaller estimations of ke for heating oil and natural gas. The parameter ke is seasonal
(with higher values during the winter season than during the summer season) and volatile for
natural gas and heating oil and more stable for rude oil. The higher winter values of ke for heating
oil and natural gas are explained by the fa t that the winter demand is more weather-sensitive hen e
more volatile than the summer demand and that inventory withdrawals are needed to fulll the
demand for heating during the winter season, hen e the short term winter sho ks are more brutal
and short-lived than the short-term summer sho ks. The winter 2001-2002 ( orresponding to the
6th point in gure (2.9)) was ex eptionally mild and experien ed higher than normal inventory
levels, ausing an abnormally low ke for heating oil and natural gas. This is onrmed by the
observation of the natural gas and heating oil futures pri es during the winter 2001-2002, when the
two urves are in ontango and the 1st month futures pri es of heating oil and natural gas display
a relative stability. On the whole, it is the heating oil's parameter that is the most variable during
the period, with ample seasonal variations and higher average values in the period 99-2001 and
lower values in the period 2002-2004. As a onsequen e, the lower performan e of the two-fa tor
model for natural gas ompared to heating oil is not due to more variable and seasonal ke but to
the higher relative importan e of twist moves in the forward urve deformations. In what follows,
we will use the onstant values of ke given by table (2.1), minimizing the total RMSE over the
whole period.

70

Figure 2.9: Evolution of ke with a 6 months time step from summer 1999 to summer 2004

2.3.3 Slope and level: two state variables for the shape of the forward urve
The evolution model (2.1) implies a forward urve shape model. Indeed, if we negle t the se ondorder terms:

lnFe (t; T ) 

Fe (t; T )
=e
Fe (t; T )

)
X e +
Y e

ke T t

we obtain the following expression for the shape of the forward urve at date t:

lnFe (t; T ) = lnFe (0; T ) +

t
X

=0

)
X e + X
Y e
s
s
s=0

ke T s

(2.5)

Let us assume that the shape of the initial forward urve is of the type:

lnFe (0; T ) = Q(T ) + e

ke T

X 0e + Y0e

(2.6)

where T takes integer values representing months and Q is a fun tion of period one year and zero
mean. Then, equation (2.5) leads to:

lnFe(t; T ) = Q(T ) + e
71

) X e + Y e

ke T t

(2.7)

with:

X te = X 0 e
Yte = Y0 +

ke t

t
X

=1

( )
X e
s

ke t s

(2.8)

s
t
X

=1

Yse

(2.9)

Equation (2.7) shows that, under model (2.1), the shape of the forward urve at any date t is the
superposition of a seasonal fun tion Q(T ), a slope X t , and a level Yt . The slope and level an
be derived from the daily sho ks (
Xte ;
Yte ) via (2.8)-(2.9). The slope follows a mean-reverting
pro ess driven by the short term sho ks and the level a random walk driven by the long-term
sho ks:

X te = X te
t e

+
X e
t

ke t

(2.10)

Yte = Yte
t +
Yte

(2.11)

E onomi interpretation of the season, slope and level

The forward urve model (2.7) has very lassi al e onomi interpretations: the seasonality of the
forward urve is explained by a stru tural imbalan e between winter and summer onsumptions
and by the small number of market parti ipants having a ess to storage reservoirs; the level is
related to the long-term pri e of the ommodity and the slope to the benet ( lassi ally referred
to as the " onvenien e yield") of holding the physi al ommodity vs holding a ontra t for future
delivery. Fama and Fren h (1988) in parti ular use the slope of the forward urve as a proxy for
the inventory level.
72

0.05

0.10

0.04

0.08

0.03
0.06
0.02
0.04

0.01
0.00

0.02

-0.01

0.00

-0.02
-0.02
-0.03
-0.04

-0.04
-0.05

-0.06
1

13

17

21

25

(a) Heating Oil

13

17

21

25

(b) Natural Gas

Figure 2.10: Seasonal fun tions Q for heating oil and natural gas

Estimation of the seasonality for natural gas and heating oil

To estimate the seasonality of the natural gas and petroleum forward urves, I ompute, for ea h
ommodity, the average log futures pri es of the ontra ts with a time-to-maturity below one year
for delivery in January, February,..., De ember from the beginning of January 99 to the end of
De ember 2003. This way, I obtain, for ea h energy e, the average log forward urve fe (T );

T = 1; :::; 12. To obtain the zero mean fun tion Q, I subtra t the mean of fe to fe . As expe ted, I
nd no seasonality for the rude oil forward urve. The obtained seasonal fun tions for heating oil
and natural gas are displayed on gure (2.10). The winter peaks are mu h more pronoun ed for
natural gas than for heating oil.

Initial slope and level

In order to al ulate the slope and level at all dates t, we need the daily short and long-term sho ks
on the one hand and the initial slope and level on the other hand. On e the parameter ke is known,
73

the short term and long-term sho ks are obtained from equation (2.2). Regarding the initial slope
and level, they are obtained by "inversion" of formula (2.6) using the 1st month and 13th month
log futures pri es observed on January 4th 1999:

X 0e = eke (T1 0) ln(Fe (0; T1 )=Fe (0; T13 ))

Y0e = lnFe(0; T13 ) Q(2)
From the estimated slope and level, I an re onstru t an estimated log forward urve using formula
(2.6). Figure (2.11) displays the observed and estimated log forward urves on January 4, 1999. For
rude oil, the two-fa tor model fails to reprodu e orre tly the shape of the initial forward urve.

74

2.67

-0.86
-0.88

2.65

-0.90
2.63
-0.92
2.61

-0.94

2.59

-0.96
-0.98

2.57

-1.00
2.55
-1.02
2.53

-1.04

2.51

-1.06
1

11

13

15

(a) Crude Oil

11

(b) Heating Oil

0.91

0.87

0.83

0.79

0.75

0.71

0.67
1

11

13

15

( ) Natural Gas
Figure 2.11: Comparison of observed (bla k) and estimated (red) log forward urves

75

13

15

Comparison of estimated slope and level with dire t estimators

From the initial slope and level and the daily short and long-term sho ks, I an ompute the slope
and level at any date using formula (2.10)-(2.11). I ompare this (indire t) estimation with a dire t
estimation by "inversion" of formula (2.7) using the observed 1st month and 13th month log futures
pri es at date t:

X te = eke (T1 t) ln(Fe (t; T1 )=Fe (t; T13 ))

Yte = lnFe (t; T13 ) Q(t + 1 month)
The traje tories of dire t and indire t estimators of slope and level for the three energies are
displayed in gures (2.12) and (2.14). There are remarkable similarities between the slopes of the
three energies, whi h are most of the time of the same sign. The dieren es between dire t and
indire t slope estimators are due to the existen e of twists in the forward urve: these twists, whi h
lead to an inexa t estimation of the slope, have only a temporary ee t on the dire t estimator, as
the ee t disappears when the twist is no longer visible, and a longer term ee t on the indire t
estimator, due to its re ursive form (2.10). This is illustrated in gure (2.13): from August 20,
1999 to August 21, 1999, the dire t estimator jumps upward after the last trading day of the
September 1999 ontra t due to the bump observed in the August forward urve; however, the
indire t estimator is ontinuous as it is only impa ted by the very small variation of the 1st month
futures pri e, i.e., the O tober ontra t pri e. This is the ause of the rst break point between
the rude oil dire t and indire t slope estimators observed in gure (2.12(a)). Regarding the levels,
there are big dis repan ies between the dire t and indire t estimators, always in the same dire tion
for the three energies. This is be ause the two-fa tor model implies that the forward urve is
almost at above a maturity of k3e

 12 months, whereas the observed forward urves are often

76

ba kwardated at this time s ale; as a onsequen e, at ea h rolling date when ba kwardation is

observed in long-term futures pri es, the dire t estimator of the level jumps down whereas the
indire t one remains steady, and the errors a umulate over time. The ee ts of this bias however
seem to be very omparable for the three energies, whi h allows one to use indierently the dire t
and the indire t estimators for the analysis of the long-term relations between gas and oil levels.
From now on, the global dependen e stru ture will be analyzed using the dire t estimator for the
slopes and the indire t estimator for the levels. The reason for this hoi e is the following: for the
slopes, we saw that the dire t estimator, ontrary to the indire t one, is only temporarily ae ted
by the existen e of twists in the forward urve; for the levels, the indire t estimator is preferred
be ause it re e ts the gains of an investor entering a 15-th month futures ontra t and rolling over
his position at ea h last trading day11 .

11 the

gain of the onsidered investor would be Ptk=1
F (k; T14 ) and the indire t level estimator is Yt = Y0 +

P =1
((
t
k

F k;T14
F k;T14

77

0.4

0.4
0.3

0.2

0.2

0.0

0.1
0.0

0.2

0.1
1999

2000

2001

2002

2003

2004

2005

1999

2000

2002

2003

2004

2005

(b) Heating Oil

0.4

0.2

0.0

0.2

0.4

0.6

0.8

(a) Crude Oil

2001

1999

2000

2001

2002

2003

2004

2005

( ) Natural Gas
Figure 2.12: Comparison of dire t estimation (bla k) and indire t estimation (red) of the slopes

78

22.2

21.8

21.4

21.0

20.6

20.2

19.8

19.4

19.0

18.6
1

11

13

15

Figure 2.13: Crude oil forward urves on August 20 (bla k) and August 21 (red) 1999 (August 20
being the last trading day of the futures ontra t delivering in September 1999

79

0.5

4.5

0.0

4.0

0.5

3.5

1.0

3.0
2.5
1999

2000

2001

2002

2003

2004

2005

1999

2000

(a) Crude Oil

2001

2002

2003

2004

2005

1.0

1.5

2.0

2.5

(b) Heating Oil

1999

2000

2001

2002

2003

2004

2005

( ) Natural Gas
Figure 2.14: Comparison of dire t estimation (bla k) and indire t estimation (red) of the levels

80

2.3.4 Denition of lo al and global dependan e stru tures

From now on, we will refer to the relations between slopes and levels as the global dependen e
stru ture and to the orrelation between o-movements as the lo al dependen e stru ture. The
global dependen e stru ture des ribes the long-term relations between slopes and levels of the
three energies and the error orre tion me hanism that insures the reversion of slopes and levels
to the long-term equilibrium. It will translate into an error orre tion term in the drifts et . The
lo al dependen e stru ture des ribes the relations between the o-movements of the three energies
as well as the dependen e of date t returns on date t

t returns, whi h will add a term in the

drifts et .

2.3.5 Analysis of lo al dependen e stru ture

Figures (2.15) to (2.19) present the lo al dependen e stru ture between oil and gas futures pri es.

Dependen e on past sho ks

Regarding the ross-energy dependen e on past sho ks, we nd that the ausality generally runs
from oil to natural gas and is negative12 , ex ept for the natural gas and heating oil short-term
moves, where the ausality runs from gas to heating oil and is positive13 . Regarding the intertemporal dependen e on past sho ks, the ausality runs both ways between the short-term and the
long-term, but is positive in the dire tion long-term ,! short-term and negative the other way14 .
Lastly, regarding the auto- orrelation of sho ks, we nd that in the short-term, oil markets tend to
12 The e onomi intuition being that the natural gas pri e rst over-rea ts to the oil pri e and then is subje t to a

orre tion the next trading day

13 This nding is in line with the analysis of natural gas and heating oil markets led in EIA (2001)
14 The e onomi interpretation being that the long-term rst over-rea ts to the short-term moves and is subje t to
a orre tion the next trading day
81

amplify the previous move, whereas in the long-term, they are more likely to orre t it.

Dependen e between o-movements

Regarding the ross-energy o-movements, we obtain the expe ted positive orrelations between
the natural gas and oil markets. Regarding the inter-temporal o-movements, we an see that the
orrelations between the short-term and long-term sho ks are positive for the three energies (the
orrelations being of 40% for rude oil and natural gas and of 30 % for heating oil)

82

st_shocks_gas & st_shocks_crude

0.15
0.10
0.05

ACF

0.05
0.10

0.05

0.05

0.00

0.00

ACF

0.10

0.20

0.15

0.25

st_shocks_gas & lt_shocks_crude

10

10

10

Lag

10

Lag

(a) gas short-term sho ks/ rude short-term (b) gas short-term sho ks/ rude long-term
sho ks
sho ks
lt_shocks_gas & st_shocks_crude

0.15
ACF

0.05

0.05

0.00

0.00

0.05

0.10

0.10
0.05

ACF

0.15

0.20

0.25

0.20

lt_shocks_gas & lt_shocks_crude

10

10

10

Lag

10

Lag

( ) gas long-term sho ks/ rude short-term (d) gas long-term sho ks/ rude long-term
sho ks
sho ks
Figure 2.15: Cross orrelation fun tions between natural gas and rude oil daily sho ks with lags
of one to ten days; the ross orrelation fun tions with lag i (resp.
between rude oil at time t and gas at time t + i (resp. t i)

83

i) represent the orrelation

st_shocks_gas & st_shocks_heat

0.10

ACF

0.05

0.10
0.05

0.05 0.00

0.05

ACF

0.15

0.15

0.20

0.20

0.25

0.25

0.30

st_shocks_gas & lt_shocks_heat

10

10

10

Lag

10

Lag

(a) gas short-term sho ks/heating oil short- (b) gas short-term sho ks/heating oil longterm sho ks
term sho ks
lt_shocks_gas & st_shocks_heat

ACF

0.1

0.05
0.10

0.0

0.00

ACF

0.10

0.2

0.15

0.20

0.3

lt_shocks_gas & lt_shocks_heat

10

10

10

Lag

10

Lag

( ) gas long-term sho ks/heating oil short- (d) gas long-term sho ks/heating oil longterm sho ks
term sho ks
Figure 2.16: Cross orrelation fun tions between natural gas and heating oil daily sho ks with
lags of 1 to 10 days; the ross orrelation fun tions with lag i (resp.
between heating oil at time t and gas at time t + i (resp. t i)

84

i) represent the orrelation

Series st_shocks_gas

0.6
ACF

0.0

0.0

0.2

0.4

0.4
0.2

ACF

0.6

0.8

0.8

1.0

1.0

Series lt_shocks_gas

10

Lag

10

Lag

(a) auto orrelation of gas short-term (b) auto orrelation of gas long-term sho ks
sho ks

0.2
0.1

0.0

0.1

ACF

0.3

0.4

st_shocks_gas & lt_shocks_gas

10

10

Lag

( ) gas short-term sho ks/gas long-term

sho ks
Figure 2.17: Auto and ross- orrelation fun tions gas short-term sho ks/gas long-term sho ks with
lags of 1 to 10 days; the ross orrelation fun tion with lag i (resp.

i) represents the orrelation

between the long-term sho k at time t and the short-term sho k at time t + i (resp. t i)

85

Series st_shocks_crude

ACF

0.4
0.2

0.4
0.0

0.0

0.2

ACF

0.6

0.6

0.8

0.8

1.0

1.0

Series lt_shocks_crude

10

Lag

10

Lag

(a) auto orrelation of rude oil short-term (b) auto orrelation of rude oil long-term
sho ks
sho ks

0.1

0.0

0.1

ACF

0.2

0.3

0.4

st_shocks_crude & lt_shocks_crude

10

10

Lag

( ) rude short-term sho ks/ rude longterm sho ks

Figure 2.18: Auto and ross- orrelation fun tions rude oil short-term sho ks/ rude oil long-term
sho ks with lags of 1 to 10 days; the ross orrelation fun tion with lag i (resp. i) represents the
orrelation between the long-term sho k at time t and the short-term sho k at time t + i (resp.

t i)

86

Series st_shocks_heat

0.4

ACF

0.2

0.0

0.0

0.2

0.2

0.4

ACF

0.6

0.6

0.8

0.8

1.0

1.0

Series lt_shocks_heat

10

Lag

10

Lag

(a) auto orrelation of rude oil short-term (b) auto orrelation of heating oil long-term
sho ks
sho ks

0.1
0.1

0.0

ACF

0.2

0.3

st_shocks_heat & lt_shocks_heat

10

10

Lag

( ) heating oil short-term sho ks/heating

oil long-term sho ks
Figure 2.19: Auto and ross- orrelation fun tions heating oil short-term sho ks/heating oil longterm sho ks with lags of 1 to 10 days; the ross orrelation fun tion with lag i (resp. i) represents
the orrelation between the long-term sho k at time t and the short-term sho k at time t + i (resp.

t i)

87

Volatilities
For the three energies, the volatilities of the short-term and long-term sho ks are estimated by
the standard deviation of the sho ks within a 50-days sliding window. The obtained traje tories
are displayed on gure (2.20): all sho ks exhibit volatility lusters, jumps, and the natural gas
and heating oil short-term volatilities follow a seasonal pattern, with high values in winter (60
% in normal winters for natural gas, 25 % in normal winters for heating oil) and lower values
in summer (20 % for natural gas, 10% for heating oil). The phenomenon of sto hasti volatility,
observed in most ommodity markets, is linked to the temporal variations of some key indi ators of
the supply exibility, su h as the deviation to "normal" storage level, and the proportion of spare
produ tion/rening apa ity. Note also that the short-term volatility peaks orrespond to periods
of high positive forward urve slopes, an observation whi h is onsistent with the theory of storage
(Kaldor (1939)), and whi h was also observed by Ates and Wang (2005) in the US gas market.
The seasonal pattern of natural gas and heating oil short-term volatilities an be explained by the
fa t that the demand is more sensitive to the temperature during the heating season and that the
demand and produ tion sho ks have more impa t on the pri es during the winter, when storage is
part of the supply urve and the market is tight, than during the summer, when storage is part of
the demand urve and the market is loose. The seasonal behavior of gas impli it volatilities was
already observed by Blix (2003) in the US gas market.

88

2001

2002

2003

40
30
20

natural gas long term volatility in %

2000

10

100
80
60
40

natural gas short term volatility in %

20
1999

2004

1999

(a) natural gas short-term volatility

2000

2001

2002

2003

2004

35
30
25
20
15

crude oil long term volatility in %

35
30
25
20
15
10

crude oil long term volatility in %

40

(b) natural gas long-term volatility

1999

2000

2001

2002

2003

2004

1999

( ) rude oil short-term volatility

2000

2001

2002

2003

2004

2000

2001

2002

2003

2004

35
30
25
20

heating oil long term volatility in %

1999

15

50
40
30
20

heating oil short term volatility in %

60

(d) rude oil long-term volatility

1999

(e) heating oil short-term volatility

2000

2001

2002

2003

2004

(f) heating oil long-term volatility

Figure 2.20: short-term and long-term volatilities (in %) of the three energies estimated with a
50-days sliding window

89

2.3.6 Analysis of global dependen e stru ture

Stationarity properties of the slopes and levels
Table (2.4) reports the results of the Phillips-Perron unit root tests on the slopes and levels of the
three energies. Not surprisingly, the slopes are found to be mean-reverting while the levels display
a random walk behavior.

90

Crude oil slope

Di key-Fuller

Lag Parameter p-value

3:914

0:01319

Heating oil slope

Di key-Fuller

Lag Parameter p-value

3:5531

0:03911

Natural gas slope

Di key-Fuller

Lag Parameter p-value

3:4437

0:04771

Crude oil level

Di key-Fuller

Lag Parameter p-value

1:2474

0:897

Heating oil level

Di key-Fuller

Lag Parameter p-value

1:2974

0:8757

Natural gas level

Di key-Fuller

Lag Parameter p-value

1:7799

0:6715

Table 2.4: Philipps-Perron unit root tests on the slopes and levels of the three energies; the teststatisti s, trun ation lag parameters, and p-values of the tests are reported

91

Long-term relation between forward urve slopes

Figure (2.21) displays the relation between natural gas and rude oil/heating oil slopes. We see
that, when the natural gas forward urve is in ba kwardation (positive slope), the oil forward
urve is also in ba kwardation15 . However, a ba kwardated oil urve does not ne essarily imply a
ba kwardated natural gas forward urve. In parti ular, year 2002 experien ed a ba kwardated oil
urve and a natural gas forward urve in ontango. The results of the linear regression of natural
gas slope on rude oil and heating oil slopes are reported on table (2.5). Note that the regression
oe ients are not signi antly dierent from 1, the regression R2 being around 30% for rude and
40% for heating oil.
Estimate Std. Error t value

b rude
a rude

0:137
1:027

0:00710
0:0414

19:23
24:79

P r(> jtj)
< 2:10 16
< 2:10 16
R2 = 29:76%

bheat

0:0979

0:00489

aheat

0:987

0:0298

20:01
33:10

< 2:10 16
< 2:10 16
R2 = 43:04%

Table 2.5: Linear regression of natural gas slope on rude oil and heating oil slopes: a denotes the
regression oe ient and b the inter ept; the estimated oe ients, standard deviations, t-statisti s,
and two-sides p-values are reported

15 note however that there are outliers in the linear relation:

for instan e, during the winters 2000-2001 and 20022003, the natural gas slope was very high while the oil slope was mildly positive
92

0.9
+

0.9

+ +
+
+++ + +++
++ +++ +
++
++ +
+
+
+
+
+
+ ++
+ + ++
++
+
+
+
+ + + ++ + + + +
++
+ ++
+
+
+
++++++
++ ++++ ++++++++
++++ ++
++
+++ +
+++++ +
++++
++++
++++++++
+++++++++++++++
+
+ +
++++
++ +++ +++++++
+
++
+
+
++++++++++++ +++++
+
++
+
+
+
+
+
+
+
+
+
+++ +++++ + ++
+
+
+++++
+++++++++
++++++++++
++++++
+ +++
++ + + + + ++
++
++++++
+
++++
++
++
++++
+ +++++++++ +++
+++++++++++++++
++ +++
+++
+
++++
+++
++
+++++
++
++++++
++++
++
+++
+++
+++++
++++++++++
++++
+ +++++ ++++
++++++++
++ ++
+ +
++++ +++
+++
++
+
+++++++++
+
++
+++
++
+++++
++++
++++
++++ ++++
+
++ ++ + +++++++ + + + +
+ +
++
+++++
++
++
++
++++
++++
+ + ++ + + + ++++
+
+ +++++++++++++++
+
++++++++
+++
++
+
++
+++
++++
++++
++++++++++
++
++
+ ++
+
+
+
+
+
+
+
+ +
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++++++++++++++
+++
+++
+++++
++++
+++++
+++++
++ +
++
++
++++++++++ + + +
+++++++
++
+
++
+
++++
+++++
+
+
+
+
+
+
++ ++
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ + +
+ +++
+ +
++++
+++ +++++++++++
++++++
+++++++++
++++++ ++++ + +++++++ ++ + ++
++++ ++++++++++++
+++++++++++
++ +
+++
++
+
+
+
+++ + ++ +++++
++
++ +++ ++
+++++++++++ +++++++
+
+
+
+
+
+
+++
+
++ +++++
+
+
+++++++ ++ ++++ ++++++ +++++ ++ +++
+ + + ++++++++++++++++++++ + ++
++
+ +
++++++++++
+++ + +
+ ++
+ +++ + ++++++++++ ++ ++++
+
+
+
+
+ + ++ +
++
++++++++++++ +
++
++
++
+
+ ++ +++++
+ ++
+
+
+++++++++++++++
+++++
++ ++
+++++++
++ +++ ++
+++++++++
+++
+++++++
++
+
+++ ++++++++ + ++ ++++++
+
+
+
+
+++ +
+++++++
++
++ ++ + ++ +
++++++
+
+
++
+
+
+++
+++ +
+ + ++ +
++ ++
++
+
++ ++
+ ++ +++++
+
+

0.7

0.5

0.3

0.1

-0.1

-0.3

-0.5
-0.3

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

0.7

+ +++
+
+ + ++ +
++

0.5

0.3

0.1

-0.1

-0.3

-0.5
-0.3

0.5

(a) natural gas slope in terms of rude oil slope

+
+

+
+ +
++
+
+
+
+
+ ++
+
+
+
++
+
+
+ + +
+ + +
+
++
+
+
+
++
+
+
+ ++
+ ++++++++ +++++ +++
++ ++++++++++++
+ +++
++++++
++++++++ +++++
++++ + +++ +++ +
+
+
+
+
+ + ++ ++ +++
+ ++++++ +++
++
++++ + + +++
+
+
++++ +
++++++
++++++++
++++ + +
+++++++++ +++++
++++++
+++++++
+ +++
++
+ ++ + ++ ++ + + + ++
+++
++++
+++++++
++
+ ++
+
++++
++++++++++++
++
+
++++
++++
++++++++
+
++
+++++++++++
++++++
+++++
++
++++
+++
++ +
+++++
+++
++
++++
++++
++
++
+++
++ ++
+++++++ +++++++++
+++++
+
+++++++++ +++++
++++++
+ +++++++ +++
++
+
+++
+++++
++
++++
++++++++++
+++++
++
+++ ++++
+ + +
+++++++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
++
++++++
+
++ ++ ++++
+ +++++ + + ++++ +
++++++++++++
++
+++++
++
+
+
++++++
++ + + ++++
++++
++ ++++++ + ++++++++++++++++
++++
++
++ ++++++++++ +++ ++
+++++
++
++ + + ++++
++
+++
++ +
+++++++++
+ ++++
++ +++
+ +
+ + +++ +
+
+ ++++++++++
++
++
+++
+ +++ +++
+++ +++ ++++
+++
++++
+++++
+ +++++++++++
+ ++
++++++++++++++
++++ +
++++
+
+ ++++++++
+
++ + + +++ + ++ ++++
+++
++++++
+++++ ++
+ +++++++
+ +
++
++++
++
++++
+++
++++
++
+
+++++++++++++++
++
+
+
+
+++++ +++
+
+
++++ ++
+ ++++++++++++
+
+
+
+
+
++ +++ ++++++
+++
+
++
+++
+++++ + +
+++++ ++++++
+++ ++++++++++++++
+
+
++++++
++
++++ + ++ ++ ++++++++++
+ ++++++ ++ +
+
+
+
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
+ ++
+ ++++++ ++ ++ + +++ ++ ++
++ ++++ +++ + ++++++
+
+
+++ + ++++
+
+
+
+
+
+
+
+
+
+
+
+
++ ++
++ + + ++++++ ++ +++
+ +++
+ ++ ++++ ++++ ++
+ ++
++ +++ + +
+ ++++++++
++ + ++ ++ +++ ++ ++++++
++ ++ ++++ +
++
+
+++
++
+ + ++++++ ++
+
+
+++
++
+++
+++ + + +
++
+++++
+
++ ++
+ ++
+++
+
+

-0.2

-0.1

0.0

0.1

0.2

0.3

0.4

0.5

(b) natural gas slope in terms of heating oil slope

Figure 2.21: Natural gas slope in terms of rude oil (left) and heating oil (right) slopes; the linear
t is displayed in green

2.7

3.3
+++
++++
++++
+++
++++++++
+
++++++
++
++++++++
+++
++
+++++
+++++
+++
+++
++
+++
+
++
+++++++++
+++
++
+++
+++
++
+
++++
++
+++++++
++
++
++
+++++
++++++++++
+
++
++++++++++++
+
++
++
++
++
+
+++
++++
++
+++
+
+
+
+
+
+
+
+
+
+
++
+
+
+
+
+
+
+
+
+
++++
+++++
+++
+ ++ ++++
++++++
++
++
+ +++
+++++++
+
+ +++++
+++
+
++++
+++++++
++++
+++
++++++
+
+++++++
++
+++
++++
++ ++
+++++++ ++ ++
+++++
+++
++
+
++++++++++++++
+
+
+
+
+
+
+
+
++
+
+
+
+
+++++++++
+
++
++++
++
+
++
++
+
+
+
++
++
++
++++
++
+
+++++
++
+++
++
+
+++
++
++++ ++
++
++
+++
+
++++
+
+
+
++
+
+
+++
+++++
+ ++++++++++
+++++
++
+++++
+
+
+
++
++
++
++++++
+++
++
++
+++++
+++++++++
+++
+
+++
+++
+++
+
+++++++++
++++
+
+
+
+
++
+
+
+++
+++++
++++
++++
++++
++
+
+++
++++
+++
+
++++
++
++++
++++++
+
+
+
+
+
+
+++ +
++++
+
++++++++
++
+++
+
+++
++++++
+++++
++
++
++++
+
++
++
++++ +++
++
+++
++
+
++++
++
+++
+
+
++
+
+++++++++++
++
++
+ ++
+
+++++++
++++
++
++
+
++++
+
++
++
+++
++
++
+++++
+++++++
++
++
+++++
+
+

2.3

1.9

1.5

1.1

2.9

2.5

2.1

1.7

1.3

0.7

0.9

0.3
2.4

2.8

3.2

3.6

4.0

4.4

0.5
-1.1

4.8

(a) natural gas level in terms of rude oil level

+++
++
++
+++++
++++
++
++++
+
+
++++++
++
+
+++++++
++++
+++++
++
++++++
++
++++
++
+++++
+
+++++
+
+
+
+
+
++
+
+
+
+
++
+
+++
+++
+++++++
++++
+++
+
++
++++++
+
++++++ +++++
++
+
+
+++
+++
+
++
++
+++++
++
+++++++++
++
++++
+
++
++++
+++++
++++++++
++++++
++
+++++
+++
++
+ +++ +++
+++
++
+++
+++
+
+
+
+
+
+
+
+
++
+
+
+
+
++
++
+
+
+
+
+
+
+
+
+
++
+
+
+
+
++
+
+
++
+
++++
+++ ++++
++++
+++
+
++
++
++++
++
++
+++++++++
++++
+++
+
+++
++
+
++
+
++++++
+
+
+++
++
+
+++
++
+
++
+
+
+
++
++
++++
+
++
++
++
++
++
+
++++
+++
++
+
++
++ +
++++
+
+++
++
+
++
+
+
+
+++++ ++
+++
+
++
+++++
++
+
+
++
+
+
++
+
+
+
+
++
+
+++++
+++
++++
++
+++
++
+
++
++
+++++
++
++
+++
+
+
+
+ ++++
++++
+
++++
++
+
+++
+
+++
++
+++++
++
++++
+++++++++
+
++
+
++
+
+
+
+
++
++++ +
++++
+
+++
+++
++++
+++++++++++++
++
++
++++++++
+
++
+++++++++
++
++
++
+++
+ ++++
++++
++++
+
+++++
+++++++
++
+++
++
+
+++
+
+
++
+++++++++++++++
++
+++
+++
++++
++
+++
++++++
+++
+
+++
++
+++
++
++++++++
++
+++++
+
+++
+

-0.7

-0.3

0.1

0.5

0.9

1.3

(b) natural gas level in terms of heating oil level

Figure 2.22: Natural gas level in terms of rude oil (left) and heating oil (right) levels; the linear t
is displayed in green; the best three-lines t is displayed in red; in dark blue: year 99; in medium
blue: year 2000; in lear blue: year 2001; in pink: year 2002; in yellow: year 2003; in bla k: year
2004
93

Long-term relation between forward urve levels

Figure (2.22) displays the relation between natural gas and rude oil/heating oil levels: in both
ases, a pie ewise-linear relation appears, with two break points o urring at the beginning of year
2000 (where gas long-term futures pri es start rising at an in reased pa e, the pa e of the oil pri e
rise remaining stable), and in the middle of year 2003 (where oil long-term futures pri es start
rising sharply, while the gas pri es rise remains stable).
Tables (2.6) and (2.7) report the results of the lassi al linear regression and the pie ewise-linear
regression of gas level on rude oil and heating oil levels. First, we see that the R2 is mu h higher
than for the regression on the slopes: the long-term equilibrium between the levels is mu h stronger
than the long-term relation between the slopes. Se ond, the pie ewise-linear regression oe ients
are signi ant, whi h onrms the validity of the pie ewise linear model, and of the same negative
sign, ausing the gas long-term pri e to be less sensitive to the variations of oil long-term pri e
above the up-threshold Y and below the down-threshold Y. Lastly, table (2.8) shows that the unitroot hypothesis an be reje ted for the residuals of the pie ewise linear relation between gas and
oil levels but not for the residuals of the linear relation between gas and oil levels. As a on lusion,
only the pie ewise linear relation allows one to obtain the desired stationary residuals.

94

Estimate Std. Error t value

b rude
a rude

2:277
1:095

P r(> jtj)

0:0314

72:44

< 2:10 16

0:0089

123:04

< 2:10 16
R2 = 91:26%

bheat

1:774

0:00355

499:7

< 2:10 16

aheat

1:174

0:0086

136:0

< 2:10 16
R2 = 92:73%

Table 2.6: Linear regression of natural gas level on rude oil (up) and heating oil (down) levels: a
denotes the regression oe ient and b the inter ept; estimated oe ients, standard deviations,
t-statisti s, and two-sided p-values are reported

95

Estimate

b rude
a rude

4:323
1:679

Std. Error

t value

P r(> jtj)

0:0648

66:69

< 2:10 16

0:0185

90:89

< 2:10 16

a rude

1:323

0:0425

31:16

< 2:10 16

a+ rude

1:004

0:032

31:22

< 2:10 16

Y rude = 3:15

Y rude = 3:73 R2 = 95:12%

bheat

1:885

0:00412

457:71

< 2:10 16

aheat

1:736

0:0170

102:05

< 2:10 16

aheat

1:287

0:0432

29:80

< 2:10 16

a+heat

1:045

0:0297

35:22

< 2:10 16

Yheat = 0:53 Yheat = 0:031 R2 = 96:18%

Table 2.7: Pie ewise-linear regression of natural gas level on rude oil and heating oil levels; the
regression variables are Ye , (Ye

Ye ) = Min(0; Ye

Ye ), and (Ye

Ye)+ = Max(0; Ye

Ye),

with e= rude oil (up) or e=heating oil (down); a denotes the dierent regression oe ients and

b the inter epts; the thresholds Ye and Ye are determined by the minimization over the ouples
(Ye; Ye ) of the sum of squared residuals of the regression of Ygas on the variables Ye , (Ye

Ye ) ,

and (Ye Ye )+ ; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

96

Residuals of the rude-gas pie ewise linear relation

Di key-Fuller

Lag Parameter p-value

3:507

0:0417

Residuals of the rude-gas linear relation

Di key-Fuller

Lag Parameter p-value

2:428

0:397

Residuals of the heating oil-gas pie ewise linear relation

Di key-Fuller

Lag Parameter p-value

4:251

0:01

Residuals of the heating oil-gas linear relation

Di key-Fuller

Lag Parameter p-value

2:514

0:361

Table 2.8: Philipps-Perron unit root tests on the residuals of the pie ewise linear and linear relations
between gas and oil levels; the test-statisti s, trun ation lag parameters, and p-values of the test
are reported

2.4

A new dependen e model for ommodity forward urves

2.4.1 Formulation of the model

We want to introdu e an error- orre tion me hanism on the levels and on the slopes between the
energies e and e0 . Therefore, we postulate that the drifts are the sums of a onstant part, a term
97

expressing dependen e on past returns, and an error- orre tion term:

0
B
B
B
B
B
B
B
B
B
B


X

e
t

C B
C B
C B
0
B

Xte C
C B
C=B
C B
B
e

Yt C
C B
C B
A 

Yte0

X;e

C
C
C
X;e0 C
C
C+
C
Y;e C
C
C
A

B
B
B
B
B
B
B
B
B
B


Y;e0

X 1
e
t

Xte0 1

Y 1
e
t

Yte0 1

X;e R

X
t lX

C B
C B
C B
C B
C B
C+B
C B
C B
C B
C B
A 

X;e0 RtX lX
Y;eR

Y
t lY

Y;e0 RtY

lY

C B
C B
C B
C B
C B
C+B
C B
C B
C B
C B
A 

X;e
t

C
C
C
X;e0 C
t C
C
C
Y;e C
t C
C
0 A

(2.12)

Y;e
t

0
0
RtX = X te fXe;e (X te )
0

RtY = Yte fYe;e (Yte )

In the model (2.12):

 e stands for natural gas and e0 stands alternatively for rude oil and heating oil
  = (X;e; X;e0 ; Y;e; Y;e0 ) is the 1  4 ve tor omposed of the onstant part of the drifts
 is a 4  4 matrix
  = (X;e; X;e0 ; Y;e; Y;e0 ) is the 1  4 ve tor omposed of the error- orre tion speeds
 Xte and Yte denote the slope and level of the forward urve of the energy e
 lX and lY refer to the lags between an observed deviation to the long-term equilibrium and
the orre tion of this deviation

 x ! fXe;e0 (x) is the relation between the slopes of energy e and e0 (in the ase of gas and oil,
fX is a linear fun tion)

 x ! fYe;e0 (x) is the relation between the levels of energy e and e0 (in the ase of gas and oil,
fY is pie ewise linear fun tion)
98

 (RtX ) is the pro ess omposed of the deviations to the long-term relation between the slopes
 (RtY ) is the pro ess omposed of the deviations to the long-term relation between the levels
0
0
0
0
0
0
 the pro esses (X;e
= tX;e tX;e ; X;e
= tX;e tX;e ; Y;e
= tY;etY;e ; Y;e
= tY;e tY;e ) follow
t
t
t
t

independent GARCH pro esses ; we in lude a seasonal omponent in the GARCH pro ess

followed by natural gas and heating oil short-term sho ks

 the residual sho ks (tX;e ; tX;e0 ; tY;e; tY;e0 ) are assumed to be i.i.d
We use the 4  1 ve tor pro ess
Zt = (
Xte ;
Xte 0 ;
Yte ;
Yte 0 )0 . A few omments are required
here. First, keeping in mind equations (2.10) and (2.11), assuming linear fun tions fX and fY ,
and making abstra tion of the dependen e between
X and
Y indu ed by the term
Zt 1 ,
the model (2.12) implies a ve tor autoregressive model (VAR) for the slopes and a ve tor error orre tion model (VECM) for the levels, whi h makes sense from an e onomi standpoint. Se ond,
we believe that the model (2.12) is su iently general to a ount for the evolution of any pair of
0

related ommodity forward urves, with appropriate long-term relations fXe;e and fYe;e . However, as
0

we hoose to model the pro esses (tX;e tX;e ; tX;e tX;e ; tY;e tY;e ; tY;e tY;e ) as independent GARCH
pro esses, we ex lude from our s ope the relations slope/volatility (whi h are studied by Ates
and Wang (2005) in the US gas market) and the ee t of volatility transmission between the
two ommodity pri es, an ee t whi h was highlighted before in the litterature on erning gas
and oil markets (see Pindy k (2004) and Ewing et al. (2003)). Lastly, we assume a onstant
0

dependen e stru ture between the residuals (tX;e ; tX;e ; tY;e ; tY;e ), thus negle ting the possible
orrelation lustering (see Eydeland and Wolynie (2003)) and the potential relations between
orrelation and volatilities (see e.g. Goorbergh et al. (2005)).
99

2.4.2 Calibration of the model

To alibrate the model, we pro eed in three steps: rst, we nd the lags lX and lY and we estimate

 and by a linear regression of
Zt on
Zt 1 , RtX lX , and RtY

lY

; se ond, we apply independent

GARCH models to the residuals of this linear regression; third, we study the dependen e stru ture
between the standardized residuals of the independent GARCH models. The hoi e of this imperfe t
pro edure, whi h is also done by Ng and Pirrong (1994) and Ates and Wang (2005), was motivated
by the high number of parameters to be estimated.

Estimation of lX , lY ,  and
Figures (2.23) and (2.24) represent the ross- orrelations between the short-term (resp. long-term)
sho ks and the residuals of the long-term relations on the slopes (resp. levels). For the pair
rude oil-natural gas, we observe that rude oil and natural gas short-term sho ks both orre t
the deviations to the long-term relation on the slopes with a lag of one day and that only natural
gas long-term sho ks orre t the deviations to the long-term relation on the levels (with no lag).
We therefore hoose lX = 1, lY = 0 for the pair rude oil-natural gas and on lude that rude oil
plays the leading role in the long-term pri e dis overy. For the pair heating oil-natural gas, we
observe that only the natural gas short-term sho ks orre t the deviations to the long-term relation
on the slopes (with no lag), and that both heating oil and natural gas long-term sho ks orre t
the deviations to the long-term relation on the levels (with no lags). We therefore hoose lX = 0,

lY = 0 for the pair heating oil-natural gas and on lude that heating oil plays the leading role in
the short-term pri e dis overy. Tables (2.9) to (2.16) report the results of the 8 linear regressions:
100

we nd many signi ant dependen e relations on past returns16 and a onrmation of the error orre tion me hanism des ribed above; note also the positive trends on the long-term moves for
the three energies.

16 On the whole, the results onrm the results of se tion 2.3.5, ex ept that natural gas moves are now found to be

positively auto- orrelated

101

st_shocks_gas & resX_ngco

0.00
ACF
0.04

0.00
0.10

0.08

0.05

ACF

0.05

0.02

0.10

0.04

lt_shocks_gas & resY_ngco

30

20

10

10

20

30

30

20

10

Lag

10

20

30

Lag

(a) residuals of the long-term relation on (b) residuals of the long-term relation in the
the slopes/gas short-term sho ks
levels/gas long-term sho ks
st_shocks_crude & resX_ngco

0.00
ACF

0.00

0.08

0.06

0.04

0.02

ACF

0.02

0.02

0.04

0.04

0.06

lt_shocks_crude & resY_ngco

30

20

10

10

20

30

30

Lag

20

10

10

20

30

Lag

( ) residuals of the long-term relation in the (d) residuals of the long-term relation in the
slopes/ rude oil short-term sho ks
levels/ rude oil long-term sho ks
Figure 2.23: Cross orrelation fun tions between the sho ks and the residuals of the rude oilnatural gas long-term relations (in the slopes and levels) with lags of 1 to 30 days; the ross
orrelation fun tions with lag i (resp.

i) represent the orrelation between the residuals at time

t and the sho ks at time t + i (resp. t i)

102

st_shocks_gas & resX_ngho

0.00

ACF

0.04

0.00
0.10

0.08

0.05

ACF

0.05

0.02

0.04

0.10

0.06

lt_shocks_gas & resY_ngho

30

20

10

10

20

30

30

20

10

Lag

10

20

30

Lag

(a) residuals of the long-term relation in the (b) residuals of the long-term relation in the
slopes/gas short-term sho ks
levels/gas long-term sho ks
st_shocks_heat & resX_ngho

0.00

ACF

0.00

0.10

0.06

0.04

0.05

0.02

ACF

0.02

0.05

0.04

lt_shocks_heat & resY_ngho

30

20

10

10

20

30

30

Lag

20

10

10

20

30

Lag

( ) residuals of the long-term relation in the (d) residuals of the long-term relation in the
slopes/heating oil short-term sho ks
levels/heating oil long-term sho ks
Figure 2.24: Cross orrelation fun tions between the sho ks and the residuals of the heating oilnatural gas long-term relations in the slopes and levels with lags of 1 to 30 days; the ross orrelation
fun tions with lag i (resp.

i) represent the orrelation between the residuals at time t and the

sho ks at time t + i (resp. t i)

103

Estimate Std. Error t value P r(> jtj)

X;gas

0:000161

0:000812

0:199

0:843

1;1

0:104

0:0297

3:513

0:000457

***

1;2

0:169

0:0585

2:896

0:00384

**

1;3

0:167

0:0601

2:776

0:00558

**

1;4

0:103

0:0647

1:599

0:110

X;gas

0:0110

0:00509

2:165

0:0305

Table 2.9: Linear regression of the natural gas short-term sho ks on
Zt 1 and RtX lX : pair rude
oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

104

Estimate Std. Error t value P r(> jtj)

X; rude 0:000565

0:000397

1:424

0:155

2;1

0:0267

0:0145

1:839

0:0661

2;2

0:0917

0:0286

3:212

0:00135

**

0:00258

**

2;3

0:0886

0:0294

3:019

2;4

0:144

0:0316

4:570

5:3:10 6

***

X; rude

0:00543

0:00249

2:187

0:0289

Table 2.10: Linear regression of the rude oil short-term sho ks on
Zt 1 and RtX lX : pair rude
oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

Y;gas

Estimate Std. Error t value

P r(> jtj)

0:00117

0:004207

0:000408

2:867

**

3;1

0:0499

0:0149

3:348 0:0000835 ***

3;2

0:0588

0:0293

2:004

3;3

0:0547

0:0303

1:809

0:00453

0:0706

3;4

0:0229

0:0326

0:704

0:482

Y;gas

0:0119

0:00408

2:923

0:00352

**

Table 2.11: Linear regression of the natural gas long-term sho ks on
Zt 1 and RtY

lY

: pair rude

oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported
105

Estimate Std. Error t value P r(> jtj)

6:56:10 5 ***

Y; rude

0:00148

0:000367

4:003

4;1

0:00787

0:0135

0:584

4;2

0:0745

0:0265

2:810

0:00503

4;3

0:00647

0:0273

0:237

0:813

4;4

0:107

0:0294

3:640

0:000282

0:00582

0:00369

Y; rude

1:580

0:559
**

***

0:114

Table 2.12: Linear regression of the rude oil long-term sho ks on
Zt 1 and RtY

lY

: pair rude

oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

106

Estimate Std. Error t value P r(> jtj)

X;gas

0:000188

0:000812

0:231

1;1

0:126

0:0303

4:168

0:817
3:26:10 5 ***

1;2

0:111

0:0498

2:238

0:0254

1;3

0:172

0:0601

2:867

0:00420

**

1;4

0:142

0:0563

2:524

0:0117

X;gas

0:0152

0:00569

2:661

0:00787

**

Table 2.13: Linear regression of the natural gas short-term sho ks on
Zt 1 and RtX lX : pair
heating oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided
p-values are reported

Estimate Std. Error t value P r(> jtj)

X;heat 0:000730

0:000456

1:602

0:109

2;1

0:0538

0:0170

3:161

0:00160

**

2;2

0:0523

0:0279

1:873

0:0613

2;3

0:0456

0:0337

2;4

0:153

0:0316

4:845

0:00319

0:282

X;heat 0:000901

1:354

0:176
1:40:10 6 ***
0:778

Table 2.14: Linear regression of the heating oil short-term sho ks on
Zt 1 and RtX lX : pair heating
oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported
107

Estimate Std. Error t value P r(> jtj)

Y;gas

0:00122

0:000407

2:991

0:00283

**

3;1

0:0402

0:0151

2:652

0:00808

**

3;2

0:0778

0:0249

3:125

0:00181

**

0:0494

3;3

0:0594

0:0302

1:967

3;4

0:0432

0:0283

1:524

0:128

Y;gas

0:0131

0:00461

2:840

0:00458

**

Table 2.15: Linear regression of the natural gas long-term sho ks on
Zt 1 and RtY

lY

: pair heating

oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

108

Estimate Std. Error t value P r(> jtj)

Y;heat

0:00147

0:000402

3:649

0:000272

4;1

0:0160

0:0150

1:071

0:284

***

4;2

0:0692

0:0246

2:813

0:00497

4;3

0:0198

0:0298

0:663

0:507

4;4

0:170

0:0280

6:076 1:58:10 9 ***

0:0108

0:00455

Y;heat

2:370

0:0179

**

Table 2.16: Linear regression of the heating oil long-term sho ks on
Zt 1 and RtY

lY

: pair heating

oil-natural gas; the estimated oe ients, standard deviations, t-statisti s, and two-sided p-values
are reported

GARCH models for the volatilities

In this se tion, we model the volatility pro esses of the residuals of the eight previous regressions (i.e. the pro esses (t ) in model (2.12))17 . Ljung-Box tests on the pro esses (2t ) show the
heteros edasti ity of the dierent residuals:

17 Unsurprisingly, we obtain very similar volatility models for the natural gas residuals of the pair rude oil-natural

gas and for the the natural gas residuals of the pair heating oil-natural gas; in what follows, we report only the former
109

data

X2

df

p-value

(X;gas
)2
t

25:03

5:65:10 7

data

X2

df

p-value

(Y;gas
)2
t

20:24

6:827:10 6

data

X2

df

p-value

(X; rude
)2
t

9:668

0:00187

data

X2

df

p-value

(Y; rude
)2
t

36:178

1:801:10 9

data

X2

df

p-value

(X;heat
)2
t

324:488

< 2:2:10 16

data

X2

df

p-value

(Y;heat
)2
t

25:717

3:954:10 7

Table 2.17: Box-Ljung tests on the pro esses (X;gas

)2 ,(Y;gas
)2 ,(X; rude
)2 , (Y; rude
)2 , (X;heat
)2 ,
t
t
t
t
t
and (Y;heat
)2 ; the test-statisti s, degrees of freedom of the approximate hi-square distribution of
t
the test statisti s, and p-values of the tests are reported
Moreover, gure (2.20) exhibits a signi ant seasonal omponent in the natural gas and heating
oil short-term volatilities. The following seasonal GARCH model, proposed by Diebold (2003) for
the modeling of temperature series, a ounts for this phenomenon:

t = t t

(2.13)

t2+1 = t2 + 2t + (1   )(a0 + a1 os(2t=252) + b1 sin(2t=252))

(2.14)

(t )

(2.15)

i:i:d
110

Note that the volatility of volatility is itself seasonal sin e the volatility sho ks 2t = t2 t2 have
dierent average winter and summer values. This hara teristi is ompatible with the observation
of gas and heating oil short-term volatilities, whi h mostly luster during the winters (see gure
(2.20)). This model was alibrated by Quasi-Maximum Likelihood on natural gas and heating oil
short-term residuals (X;gas
) and (X;heat
). The log-likelihood of the model in the ase of Gaussian
t
t
residuals (t ) is:

LL =


2t
+ log(t )
2t2
t=1

N
X

Estimate

Std. Error t value

T
log(2)
2
P r(> jtj)

gas

0:795

0:0376

21:171

< 2:10 16

***

gas

0:127

0:0282

4:503

6:691:10 6

***

a0;gas

0:00100

0:000131

7:660

1:865:10 14 ***

a1;gas

0:000368

0:000102

3:621

2:930:10 4

***

3:750:10 4

***

b1;gas

0:000348

0:0000977

3:557

Table 2.18: Quasi-Maximum-Likelihood estimation of a seasonal GARCH model on (X;gas

); the
t
estimated oe ients, standard deviations, t-statisti s, and two-sided p-values are reported
Table (2.20) reports the results of the Jarque-Bera (resp. Ljung-Box) tests on the residuals
(resp. squared residuals) of the two seasonal GARCH models. The Jarque-Bera tests allow us
to reje t the hypothesis of Gaussian residuals. The Ljung-Box test indi ates that the squared
residuals on heating oil are still auto orrelated, whi h shows that the seasonal GARCH model is
not appropriate to lter out the heteros edasti ity of the heating oil returns. By ontrast, we
annot reje t the hypothesis of independen e for the squared residuals of natural gas, whi h is an
indi ation of the validity of the model for natural gas. We thus adopt an ARCH model rather than
111

Estimate

Std. Error t value

P r(> jtj)

heat

0:880

2:148:10 2

40:955

< 2:10 16

heat

9:337:10 2

1:771:10 2

5:271

1:357:10 7 ***

a0;heat

2:958:10 4

5:206:10 5

5:681

1:336:10 8 ***

a1;heat

1:488:10 4

5:175:10 5

2:876

4:034:10 3 ***

b1;heat

6:768:10 5 5:236:10 5

***

1:292 1:962:10 1

Table 2.19: Quasi-Maximum-Likelihood estimation of a seasonal GARCH model on (X;heat

); the
t
estimated oe ients, standard deviations, t-statisti s, and two-sided p-values are reported
a GARCH model for heating oil:

t = t t

(2.16)

t2+1 = 2t + (1  )(a0 + a1 os(2t=252) + b1 sin(2t=252))

(2.17)

(t )

(2.18)

i:i:d

The estimated parameters are reported on table (2.21) and the tests on the residuals are reported
on table (2.22): we observe now that we annot reje t anymore the hypothesis of independen e for



 X;heat 
the squared residuals. Figures (2.25) and (2.26) plot the traje tories of X;gas
 and t
,
t

together with the volatilities t predi ted by the seasonal GARCH models and the long-term
seasonal varian e fun tions a0 + a1 os(2t=252) + b1 sin(2t=252).

112

Jarque-Bera
data

X2

df

p-value

tX;gas

1853:361

2:2:10 16

data

X2

df

p-value

(tX;gas )2

0:0189

0:8907

data

X2

df

p-value

tX;heat

61:5038

4:4:10 14

data

X2

df

p-value

(tX;heat )2

4:5605

0:0327

Ljung-Box

Jarque-Bera

Ljung-Box

Table 2.20: Jarque-Bera and Box-Ljung tests on the residuals of the seasonal GARCH models for
natural gas and heating oil; the test statisti s, degrees of freedom, and p-values of the tests are
reported

113

Std. Error t value

P r(> jtj)

2:619:10 2

4:697

2:636:10 6 ***

a0;heat 2:734:10 4 1:235:10 5

22:172

< 2:10 16

***

a1;heat 1:460:10 4 1:597:10 5

9:143

< 2:10 16

***

b1;heat 3:890:10 5 1:391:10 5

2:795

5:191:10 3

**

Estimate

heat

0:262

Table 2.21: Quasi-Maximum-Likelihood estimation of a seasonal ARCH model on (X;heat

); the
t
estimated oe ients, standard deviations, t-statisti s, and tow-sided p-values are reported

Jarque-Bera
data

X2

df

p-value

tX;heat

147:947

< 2:2:10 16

data

X2

df

p-value

(tX;heat )2

0:992

0:319

Ljung-Box

Table 2.22: Jarque-Bera and Box-Ljung tests on the residuals of the seasonal ARCH model for
heating oil; the test statisti s, degrees of freedom, and p-values of the tests are reported

114

0.30
0.25
0.20
0.15
0.10
0.00

0.05

natural gas short term volatility

500

1000

1500


Figure 2.25: Traje tories of X;gas
 (bla k), volatility t predi ted by a seasonal GARCH model
t

(red), and square root of the long-term varian e a0 + a1 os(2t=252) + b1 sin(2t=252) (green)

115

0.12
0.10
0.08
0.06
0.04
0.00

0.02

heating oil short term volatility

500

1000

1500


Figure 2.26: Traje tories of X;heat
 (bla k), volatility t predi ted by a seasonal ARCH model
t

(red), and square root of the long-term varian e a0 + a1 os(2t=252) + b1 sin(2t=252) (green)

116

We model the other series by lassi al GARCH models. The results are reported on tables
(2.23)-(2.31). When the Ljung-Box test leads to reje t the independen e hypothesis on the squared
residuals of the GARCH(1,1) model, we use instead the ARCH(1) model. This happens here for
the rude oil long-term sho ks.
Estimate

Std. Error t value

a0 1:214:10 6 4:203:10 7

2:888

a1 6:712:10 2 7:712:10 3

8:704

P r(> jtj)
0:00388

**

< 2:10 16 ***

b1 9:305:10 1 7:113:10 3 130:817 < 2:10 16 ***

Table 2.23: Quasi-Maximum-Likelihood estimation of a GARCH model t2+1 = a0 + a1 2t + b1 t2
on natural gas long-term sho ks; the estimated oe ients, standard deviations, t-statisti s, and
tow-sided p-values are reported
Jarque-Bera
data

X2

df

p-value

tY;gas

502:0754

< 2:10 16

data

X2

df

p-value

(tY;gas )2

0:1007

0:751

Ljung-Box

Table 2.24: Jarque-Bera and Ljung-Box tests on the residuals of the GARCH model for natural
gas long-term sho ks; the test statisti s, degrees of freedom, and p-values of the tests are reported

117

Estimate

Std. Error t value P r(> jtj)

a0 8:655:10 6 2:611:10 6

3:315

0:000916

***

a1 7:928:10 2 1:248:10 2

6:355

2:09 10

***

b1 8:827:10 1 2:028:10 2 43:530 < 2:10 16

**

Table 2.25: Quasi-Maximum-Likelihood estimation of a GARCH model t2+1 = a0 + a1 2t + b1 t2

on rude oil short-term sho ks; the estimated oe ients, standard deviations, t-statisti s, and
tow-sided p-values are reported

Jarque-Bera
data

X2

df

p-value

tX; rude

62:5259

2:242:10 14

data

X2

df

p-value

(tX; rude )2

0:6365

0:425

Ljung-Box

Table 2.26: Jarque-Bera and Ljung-Box tests on the residuals of the GARCH model for rude oil
short-term sho ks; the test statisti s, degrees of freedom, and p-values of the tests are reported

118

Estimate

Std. Error t value P r(> jtj)

a0 1:677:10 5 5:692:10 6

2:946

0:00322

**

a1 7:387:10 2 1:329:10 2

5:557

2:74 8

***

b1 8:543:10 1 3:406:10 2 25:083 < 2:10 16 ***

Table 2.27: Quasi-Maximum-Likelihood estimation of a GARCH model t2+1 = a0 + a1 2t + b1 t2
on heating oil long-term sho ks; the estimated oe ients, standard deviations, t-statisti s, and
tow-sided p-values are reported
Jarque-Bera
data

X2

df

p-value

tY;heat

75:6266

< 2:2:10 16

data

X2

df

p-value

(tY;heat )2

0:196

0:658

Ljung-Box

Table 2.28: Jarque-Bera and Ljung-Box tests on the residuals of the GARCH model for heating oil
long-term sho ks; the test statisti s, degrees of freedom, and p-values of the tests are reported

Dependen e stru ture of the standardized o-movements

We model the dependen e stru ture of the residuals () using the opula representation:
P(tX;e

 z1 ; tX;e0  z2 ; tY;e  z3 ; tY;e0  z4 ) = C (F X;e(z1 ); F X;e0 (z2 ); F Y;e(z3 ); F Y;e0 (z4 )) (2.19)

where the opula fun tion C is dened in [0; 14 with values in [0; 1, and (F X;e ; F X;e0 ; F Y;e; F Y;e0 )
denote the marginal distributions of the residuals (X;e ; X;e0 ; Y;e ; Y;e0 ). We will use here the
119

Ljung-Box
data

X2

df p-value

(tY; rude)2 4:5107

0:0337

Table 2.29: Ljung-Box test on the residuals of the GARCH(1,1) model for rude oil long-term
sho ks; the test statisti s, degrees of freedom, and p-values of the tests are reported
Estimate

P r(> jtj)

Std. Error t value

a0 1:623:10 4 6:295:10 6

25:782

< 2:10 16

a1 1:590:10 1 2:593:10 2

6:132

8:67:10 10 ***

***

Table 2.30: Quasi-Maximum-Likelihood estimation of an ARCH(1) model t2+1 = a0 + a1 2t on

rude oil long-term sho ks; the estimated oe ients, standard deviations, t-statisti s, and towsided p-values are reported
Gaussian opula dened by:

C (u1 ; u2 ; u3 ; u4 ) = 4 ( 1 (u1 );  1 (u2 );  1 (u3 );  1 (u4 ))

(2.20)

where 4 is a 4-variate normal distribution of orrelation matrix , and  1 is the inverse of the
univariate standard normal distribution. The log-likelihood of the opula model (2.19)-(2.20) is:
0

LL(; F X;e ; ::; F Y;e ) =

T
X

=1

ln (F X;e (z1t ); :::; F Y;e (z4t )) +

N 
X

=1

i

ln f X;e (z1t ) + ::: + ln f Y;e (z4t )

(2.21)

where (f X;e ; f X;e0 ; f Y;e; f Y;e0 ) are the univariate densities, is the density of the 4-variate normal
opula (2.20):

(u1 ; u2 ; u3 ; u4 ) =
120

4C
u1 :::u4

Jarque-Bera
data

X2

df

p-value

tX;heat

62:5259

2:242:10 14

data

X2

df

p-value

(tY; rude )2

0:6365

0:425

Ljung-Box

Table 2.31: Jarque-Bera and Ljung-Box tests on the residuals of the ARCH(1) model for rude oil
long-term sho ks; the test statisti s, degrees of freedom, and p-values of the tests are reported
and (zit )1tT are the observations of the i-th variable. As explained in Joe and Xu (1996), the
alibration of the model (2.19) is done in two steps:
- we rst make a non parametri estimation of the marginal densities (f X;e; f X;e0 ; f Y;e; f Y;e0 ) of the
dierent sho ks, using the Gaussian kernel estimator (Silverman (1986)):
N
1 X
z zi
^
f (z ) =
K(
)
Nh i=1
h

2
where K (z ) = p12 exp( z2 ), (zi ) are the observations, and h is an appropriate bandwidth; the ob-

tained densities are plotted and ompared to the standard normal density in gures (2.27)-(2.28);
the univariate distributions (F X;e ; F X;e0 ; F Y;e; F Y;e0 ) are obtained by numeri al integration of the
densities.
- on e the marginal distributions are determined, the orrelation matrix  is estimated by maximization of the rst part of the log-likelihood (2.21), whi h ontains the information on the dependen e
stru ture.
The estimates of matrix orrelations  for the pairs gas/ rude oil and gas/heating oil are reported
on tables (2.32)-(2.33). We note that all orrelations are signi antly positive and that the depen121

den e stru ture is very similar for the two pairs, apart form the orrelation between the short-term
sho ks, whi h is higher for the pair gas-heating oil than for the pair gas- rude oil. This higher
orrelation was expe ted be ause weather is a ommon determinant of natural gas and heating oil
demand urves.

122

0.4
0.3
0.2
0.1
0.0

10

Figure 2.27: densities of natural gas short-term sho ks (bla k), natural gas long-term sho ks (red),
rude oil short-term sho ks (green), rude oil long-term sho ks (blue), and normal density ( lear
blue)

123

0.5
0.4
0.3
0.2
0.1
0.0

10

Figure 2.28: densities of natural gas short-term sho ks (bla k), natural gas long-term sho ks (red),
heating oil short-term sho ks (green), heating oil long-term sho ks (blue), and normal density ( lear
blue)

124

short-term gas short-term rude long-term gas long-term rude

short-term gas

0:258(0:024)

0:583(0:015)

0:278(0:024)

short-term rude

0:258(0:024)

0:280(0:024)

0:480(0:019)

long-term gas

0:583(0:015)

0:280(0:024))

0:329(0:023)

long-term rude

0:278(0:024)

0:480(0:019)

0:329(0:023)

Table 2.32: Maximum-likelihood estimation of the orrelation matrix of the Gaussian opula for
the pair gas- rude oil (standard errors in parenthesis)
short-term gas short-term heat long-term gas long-term heat
short-term gas

0:337(0:022)

0:584(0:015)

0:299(0:023)

short-term heat

0:337(0:022)

0:309(0:023)

0:428(0:020)

long-term gas

0:584(0:015)

0:309(0:023)

0:337(0:022)

long-term heat

0:299(0:023)

0:428(0:020)

0:337(0:022)

Table 2.33: Maximum-likelihood estimation of the orrelation matrix of the Gaussian opula for
the pair gas-heating oil (standard errors in parenthesis)

2.4.3 Stability of the orre tion me hanisms

The obje tive here is to study the temporal stability of the dierent orre tion me hanisms whi h
were found in se tion 3.4.2. As in Alexander (1999), gure (2.29) shows the temporal evolution
(with a two-year sliding window) of the t-statisti s on X;gas and X; rude for the pair gas- rude.
We observe rst that the oe ient X;gas has a positive trend during the period: as a onsequen e,
from 2000 to 2002, the orre tion is done through rude oil rather than through natural gas, and the
125

situation is inverted in the period 2002-2003. Se ond, we observe that the t-statisti s on X;gas and

X; rude often move in opposite dire tions, whi h implies that the strength of the global orre tion,
represented by the bla k traje tory, tends to remain lo ally stable. Yet, after a transitory period
in 2001-2002, the bla k urve takes higher values in the period 2002-2003 than in the period 20002002, indi ating that the long-term equilibrium was stronger in the latter period than in the former.
Figures (2.30) to (2.32) present similar results. Most on lusions remain the same, in parti ular
the fa t that all orre tion me hanisms have been stronger sin e 2002.

126

3.0
2.5
2.0
1.5
1.0
0.5
0.0
2000

2001

2002

2003

Figure 2.29: Stability of the orre tion me hanism on gas and rude oil slopes: traje tories (with
a two-year sliding window) of the t-statisti s of X;gas (gas short-term sho ks to deviations RtX )
(blue), of X; rude ( rude short-term sho ks to deviations RtX ) (red), mean of the two previous
t-statisti s (bla k), 95% signi an e level (green)

127

3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
2000

2001

2002

2003

Figure 2.30: Stability of the orre tion me hanism on gas and rude oil levels: traje tories (with
a two-year sliding window) of the t-statisti s of Y;gas (gas short-term sho ks to deviations RtY )
(blue), of Y; rude ( rude short-term sho ks to deviations RtY ) (red), mean of the two previous
t-statisti s (bla k), 95% signi an e level (green)

128

3
2
1
0
1
2
2000

2001

2002

2003

Figure 2.31: Stability of the orre tion me hanism on gas and heating oil slopes: traje tories (with
a two-year sliding window) of the t-statisti s of X;gas (gas long-term sho ks to the deviations

RtX ) (blue), of X;heat (heating oil long-term sho ks to the deviations RtX ) (red), mean of the two
previous t-statisti s (bla k), 95% signi an e level (green)

129

3.0
2.5
2.0
1.5
1.0
0.5
2000

2001

2002

2003

Figure 2.32: Stability of the orre tion me hanism on gas and heating oil levels: traje tories (with
a two-year sliding window) of the t-statisti s of Y;gas (gas long-term sho ks to the deviations

RtY ) (blue), of Y;heat (heating oil long-term sho ks to the deviations RtY ) (red), mean of the two
previous t-statisti s (bla k), 95% signi an e level (green)

130

2.4.4 Evolution of the orrelations

The obje tive here is to study the stability of the dependen e stru ture whi h was found between
the forward urves o-movements. Figure (2.33) represents the temporal evolution of Kendall's
orrelation18 (with a one-year sliding window) between dierent pairs of sho ks. All orrelations
display an upward trend on the period. A possible explanation for this observation is the fa t that,
in ommodity markets, in ontrast to equity markets, orrelation is generally bigger when pri es
are rising. This ould also a ount for the fa t the orre tion me hanisms between oil and gas have
been stronger sin e 2003, when gas and oil pri es started rising. Lastly, gure (2.34) represents
the temporal evolution of Kendall's orrelation (with a 50-days sliding window) between natural
gas and heating oil short-term sho ks. Not surprisingly, stronger orrelations prevailed during the
tight market winters 2000-2001, 2002-2003, and 2003-2004, when natural gas and heating oil urves
both were in ba kwardation.

18 The Kendall's orrelation 

is linked to the orrelation matrix of the Gaussian opula estimated above through
the relation  = 2 ar sin() (see Lindskog et al. (2001)) and is known to be more robust than the usual Pearson
linear orrelation
131

0.25

0.30

0.25

0.20

0.20

0.15

0.15
0.10

2000

2001

2002

2003

2004

2001

2002

2003

2004

(b) short-term gas/short-term heating oil

0.10

0.10

0.15

0.15

0.20

0.20

0.25

0.25

0.30

0.30

0.35

(a) short-term gas/short-term rude

2000

2000

2001

2002

2003

2004

2000

( ) long-term gas/long-term rude

2001

2002

2003

2004

(d) long-term gas/long-term heating oil

Figure 2.33: Kendall's orrelation (with a one-year sliding window)

132

0.5
0.4
0.3
0.2
0.1
0.0
1999

2000

2001

2002

2003

2004

Figure 2.34: Kendall's orrelation between gas and heating oil short-term sho ks (with a 50-days
sliding window)

133

2.5

Con lusion

This paper has presented a new dependen e model for ommodity forward urves. Like popular
models on single ommodity forward urves, it de omposes the forward urve moves into a shortterm and a long-term sho ks, with sto hasti and possibly seasonal volatilities. The orrelation
between the sho ks of the two urves is aptured through a non-Gaussian dependen e stru ture.
The originality of the model is that, in addition to this lo al dependen e stru ture, it a ounts for
the long-term relations between the ommodity forward pri es through an error- orre tion term
in the risk-premia of the forward pri e returns. The long-term relations are based on the state
variables des ribing the shape of a forward urve under the two-fa tor model, namely the slope
and level. Our urrent resear h on erns the extension of the model to three sto hasti fa tors for
ea h urve (as in Diebold and Li (2003)), the modeling of sto hasti dependen e stru ture, and the
impli ations of the model for multi- ommodity asset pri ing and portfolio optimization.

2.6

Referen es

Alexander C. (1999), Correlation and Cointegration in Energy Markets in Managing Energy Pri e
Risk (2. nd. Edition) V. Kaminsky (ed.).

Ates A., Wang G. (2005), Storage, Weather and Dynami s of Natural Gas Pri es in Futures and
Spot Markets, Working Paper

Ameri an Gas Foundation (2003), Natural Gas and Energy Pri e Volatility, Energy and Environmental Analysis, In .
Blix M. (2003), The volatility Stru ture of Gas Futures Contra ts, Working Paper, Sto kholm
S hool of E onomi s
134

Brooks R. (2001), Value at Risk Applied to Natural Gas Forward Contra ts, WP01-08-01, Departmentof E onomi s, Finan e and Legal Studies, University of Alabama
Campbell D., Diebold F. (2003), Weather Fore asting for Weather Derivatives, NBER Working
Papers 10141, National Bureau of E onomi Resear h, In .PIER Working Paper No. 01-031.
Clewlow L., Stri kland C. (2000), Energy Derivatives: Pri ing and Risk Management, La ima Publi ations (2000)
Diebold F., Li C. (2003) Fore asting the Term Stru ture of Government Bond Yields, NBER Working Papers 10048, National Bureau of E onomi Resear h, In
Duan J., Pliska S. (2004), Option Valuation with Co-integrated Asset Pri es, Journal of E onomi
Dynami s and Control, 28(4), 727-54
Due D. (1999), Volatility in Energy Pri es in Managing Energy Pri e Risk (2. nd. Edition) V.
Kaminsky (ed.).
Energy Information Administration (2001), Impa t of Interruptible Natural Gas Servi e on Northeast Heating Oil Demand

Ewing B., Malik F., Ozdan O. (2002), Volatility Transmission in the Oil and Natural Gas Markets,
Energy E onomi s 24 (2002) 525-538
Eydeland A., Wolynie K. (2003), Energy and Power Risk Management: New Developments in
Modeling, Pri ing, and Hedging, WileyEurope

Fama E., Fren h K. (1988), Business Cy les and the Behavior of Metals Pri es, Journal of Finan e,
Vol. 43, No. 5, pp. 1075-1093
Geman H. (2005), Commodities and Commodity Derivatives : Modelling and Pri ing for Agri ulturals, Metals and Energy, Wiley Finan e

Geman H., Nguyen V. (2005), Soybean Inventory and Forward Curves Dynami s, Management
135

S ien e, Fall 2005

Goorbergh R. van den, Genest C., Werker B. (2005), Bivariate option pri ing using dynami opula
models, Insuran e: Mathemati s and E onomi s, 37, 101-114.

Joe H., Xu J. (1996), The estimation method of inferen e fun tions for margins for multivariate
models, Te hni al Report no. 166., Van ouver: University of British Columbia, Department of

Statisti s.
Kaldor N. (1939), Spe ulation and E onomi Stability, Review of E onomi Studies, O tober 1939,
pp. 1-27
Lindskog F., M Neil A., S hmo k U. (2001), Kendall's Tau for Ellipti al Distributions, RiskLab
Resear h Paper, ETH Zuri h
Manoliu M., Tompaidis S. (2002), Energy Futures Pri es: Term Stru ture Models with Kalman
Filter Estimation Applied Mathemati al Finan e, vol. 9, issue 1, pages 21-43

Ng V., Pirrong C. (1994), Fundamentals and Volatility: Storage, Spreads and Dynami s of Metal
Pri es, Journal of Buisness, 67 (2) 203-230

N'guyen V. (2004), Etude des mar hes de matieres premieres : Valorisation et ouverture d'instruments
derives, PhD Thesis, University Paris Dauphine

Pekka M., Antti K. (2005), Evaluating Multivariate GARCH models in the Ele tri ity Markets,
Helsinki S hool of E onomi s, Working Paper, W-382
Pilipovi D. (1997), Energy Risk: Valuing and Managing Energy Derivatives, M Graw Hill
Pindy k R. (2004), Volatility in Natural Gas and Oil Markets, Journal of Energy and Development
Ri hter C., Sorensen C. (2000), Sto hasti Volatility and Seasonality in Commodity Futures and
Options: The Case of Soybeans, Department of Finan e, Copenhagen Business S hool, Working

Paper
136

S hwartz E., Smith J. (2000), Short-Term Variations and Long-Term Dynami s in Commodity
Pri es, Management S ien e

Siliverstovs B., Neumann A., LHegaret G., Hirs hhausen C. (2005), International Market Integration for Natural Gas? A Cointegration Analysis of Gas Pri es in Europe, North Ameri a and
Japan, Energy E onomi s, Vol. 27, No. 4, 603-615.

Silverman B. W. (1986) Density Estimation for Statisti s and Data Analysis, London: Chapman
Hall.

137