thèse: régulation du marché des matières premières
TRANSCRIPT
UNIVERSITE PARIS 1-PANTHEON SORBONNEECOLE DOCTORALE DE SCIENCES DE GESTION
Labex-ReFi
Thèse: Régulation du marché des matières premières
THESEPrésentée et soutenue publiquement le 11 Avril,2016 en vue de l’obtention du Doctorat en Sciences de
Gestion par
HUANG XiaoyingDoctorant contractuel du Labex ReFi porté par heSam Université, portant la référence ANR-10-LABX-0095
JURY
Directeurs de thèse Didier MarteauProfesseur, ESCP-EuropeSteve OhanaProfesseur associé, ESCP-Europe
Rapporteurs Michel AlbouyProfesseur, Conseiller scientifique Grenoble School of ManagementYves SimonProfesseur, Université Paris-Dauphine
Suffragants Gemei AochiDirectrice des risques, Groupe AxéréalFrançois-Gilles Le TheuleInspecteur général de l’agriculture, Ministère de l’agricultureDirecteur exécutif, Laboratoire d’excellence sur la régulation financièreProfesseur affilié, ESCP-EuropePierre-Charles PradierMaître de conférences & HDR,Département d’économie (UFR 02), Université Paris 1 Panthéon-Sorbonne
Ce travail a été réalisé dans le cadre du laboratoire d’excellence ReFi porté par heSam Université, portant la référenceANR-10-LABX-0095. Ce travail a bénéficié d’une aide de l’Etat gérée par l’Agence Nationale de la recherche au titre du projet
Investissements d’Avenir Paris Nouveaux Mondes portant la référence n° ANR-11-IDEX-0006-02.
Contents
Contents 2
List of Figures 4
List of Tables 5
Acknowledgement 7
I General introduction 8
1 General introduction 9
Bibliography 14
II Commodity prices and risk management 15
2 Behavior of French wheat prices 162.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.2 The features of commodity prices and related works . . . . . . . . . . . . . . . . . . 182.3 The wheat market and empirical analysis of prices . . . . . . . . . . . . . . . . . . . 192.4 Model application and estimation results . . . . . . . . . . . . . . . . . . . . . . . . . 222.5 Diagnostic tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
Bibliography 42
3 Market risk evaluation of agricultural cooperatives 473.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.2 Modelling price returns and VaR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.3 Comparison under different scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . 543.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Bibliography 68
2
III Financialization of commodity markets 70
4 Does the impact of index flows on commodities prices involve stockpiling as a signal? 714.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 714.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.3 Inventory and index funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 754.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.5 Robustness Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Bibliography 107
5 The impact of Central Banks’ announcements on commodities markets 1105.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1105.2 Literature on the interaction between commodity prices and monetary policy . . . . 1125.3 Preliminary observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1145.4 Response of the commodity market to the Federal Reserve monetary policy . . . . . 1175.5 Additional tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1205.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
Bibliography 145
IV General conclusion 147
6 General conclusion 148
List of Figures
2.1 Seasonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.2 One example of simulated prices and original prices . . . . . . . . . . . . . . . . . . . . 302.3 Comparison between real log return and simulated log return . . . . . . . . . . . . . . . 312.4 An example of prices gap between harvests . . . . . . . . . . . . . . . . . . . . . . . . . 342.5 Fundamentals of wheat market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352.6 Daily prices series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362.7 First-order difference of prices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372.8 More simulation series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1 Three harvests of French wheat prices from 2011 to 2014 . . . . . . . . . . . . . . . . . 603.2 Portfolio 1 - 10-Day VaR 99% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.3 Portfolio 2 - 10-Day VaR 99% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633.4 Portfolio 3 - 10-Day VaR 99% . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1 Relation model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 734.2 Relation model 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 744.3 Commodity Index Funds Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.1 Commodities prices,Interest rate and Dollar index (a) and (b) . . . . . . . . . . . . . . . 1255.2 Commodities prices,Interest rate and Dollar index (c) . . . . . . . . . . . . . . . . . . . 1265.3 GSCI and dollar index of all trading days and of unexpected monetary policy days . . . 1285.4 Delayed effect - Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1315.5 Delayed effect - Median . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
4
List of Tables
2.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Augmented-Dickey Fuller test and Hurst exponent . . . . . . . . . . . . . . . . . . . . . 22
2.3 Parameter k and µx . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.4 Parameter of volatility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 Parameters of Jump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.6 Likelihood ration tests(LR) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7 Moments comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.8 Estimated parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.1 Portfolio 1 - Number of exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2 Portfolio 1 - Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.3 Portfolio 2 - Number of exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4 Portfolio 2 - Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5 Portfolio 3 - Number of exceptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.6 Portfolio 3 - Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.7 Parameters of Black-Scholes model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.8 Parameters of Jump model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.9 Basel Committee’s criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.1 Inventory proxy and CITF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.2 Inventory proxy and ETF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.3 Inventory proxy and CITF Total . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.4 Futures returns and CITF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.5 Futures returns and ETF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.6 Futures returns and CITF Total . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.7 Bi-monthly regression results comparison (a) . . . . . . . . . . . . . . . . . . . . . . . . 93
4.8 Bi-monthly regression results comparison (b) . . . . . . . . . . . . . . . . . . . . . . . . 93
4.9 Monthly regression results comparison (a) . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.10 Monthly regression results comparison (b) . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.11 Specific period study: 06/2006-30/06/2008 (Commodities’ boom) a . . . . . . . . . . . 95
4.12 Specific period study: 06/2006-30/06/2008 (Commodities’ boom) b . . . . . . . . . . . 95
4.13 Specific period study: 07/2008-31/01/2009 (Commodities’ collapse) a . . . . . . . . . . 96
5
4.14 Specific period study: 07/2008-31/01/2009 (Commodities’ collapse) b . . . . . . . . . 964.15 Specific period study: 02/2009-31/03/2011 (Recovery) a . . . . . . . . . . . . . . . . . . 974.16 Specific period study: 02/2009-31/03/2011 (Recovery) b . . . . . . . . . . . . . . . . . 974.17 Specific period study: 04/2011-2014 (Fall in commodities prices) a . . . . . . . . . . . . 984.18 Specific period study: 04/2011-2014 (Fall in commodities prices) b . . . . . . . . . . . . 984.19 No-agricultural commodities (a) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 994.20 No-agricultural commodities (b) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 994.21 Correlation between Stock-to-use and different inventory proxy methods . . . . . . . . 1004.22 Relationship between CITF and inventory proxy using alternative calendar spreads a) . 1014.23 Relationship between CITF and inventory proxy using alternative calendar spreads b) . 1024.24 Results with bi-monthly and monthly data using M1/spot spread as inventory proxy . 1034.25 The impact of CITF using stock-to-use ratio . . . . . . . . . . . . . . . . . . . . . . . . . 1044.26 Relationship between prices returns and CITF using spot prices . . . . . . . . . . . . . 1054.27 Table of AIC for choosing lag number in the regression . . . . . . . . . . . . . . . . . . . 106
5.1 Correlation of major variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1275.2 Elementary event study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1295.3 Observation before 2008: 2005/2008 (21 unexpected events) . . . . . . . . . . . . . . . 1335.4 Observation before 2008: 2000/2005 (16 unexpected events) . . . . . . . . . . . . . . . 1335.5 Results based on unexpected event found from inflation swap . . . . . . . . . . . . . . . 1345.6 Transmission channels: Inventory and trading position . . . . . . . . . . . . . . . . . . . 1355.7 Transmission channels of inventory for five individual commodities . . . . . . . . . . . 1365.8 Role of dollar in the variation of commodity prices . . . . . . . . . . . . . . . . . . . . . 1375.9 Synthesis table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1385.10 List of unexpected monetary defined from dollar index events from 2008 to 2014 . . . 1395.11 List of unexpected monetary defined from inflation swaps from 2008 to 2014 . . . . . . 143
Acknowledgement
Above all, I thank my supervisors, Pr.Didier Marteau and Pr.Steve Ohana for their support, con-fidence and patience. I would like to thank professor Marteau for spending time with me evenif he has tough schecule. His advice on research has been invaluable. I would also like to thankprofesseur Ohana for encouraging my research and helped me come up with thesis topic. I amthankful for the excellent example both of my professors provided as a good researcher.
I thank Pr.Michel Albouy and Pr.Yves Simon for agreeing to be the reviewers and giving me feed-back to improve my thesis. I also thank Madame Gemei Aochi, Pr.François-Gilles Le Theule andPr.Pierre-Charles Pradier, giving me the honor of being the member of jury.
I thank Labex-Refi for having financed my thesis for four years that made my Ph.D work possible.Labex-ReFi has been a source of friendships as well as good advice and collaboration, which havecontributed to my personal and professional experience during my Ph.D research. I thank Pr. LeTheule, Pr.Douady and Pr. Bancel for their support in any phases of my thesis. A special thank tomy friends, for sharing the best moments together during the Ph.D thesis. I am grateful for timespent with the colleges, friends in Labex and their supporting,that made my research enjoyable.
I would like to thank my dad, mum and my sister for their continuous support and for staying inclose contact in spite of geographical distance.
Finally, I thank Kehan for supporting and encouraging during the final stage of Ph.D. Thank youfor everything.
Xiaoying Huang
Paris
7
Part I
General introduction
8
1General introduction
Marchés des matières premièresAu cours de la période 2006-2008, la plupart des prix des produits agricoles se caractérisent parune forte volatilité accompagnée d’une hausse historique. Ainsi, les prix des principaux produitsont augmenté de façon spectaculaire entre 2006 et 2007, atteignant des pics extrêmes à la fin del’année 2007 et le début 2008 pour connaître ensuite une période d’accalmie à partir de mi-2008.Puis, mi-2010, de nouveau, une phase de fortes hausses des prix est intervenue.La hausse des prix a provoqué des problèmes sociaux, surtout auprès des pays les moins dévelop-pés consommateurs de produits alimentaires. Les prix élevés forcent les populations les plusvulnérables à réduire leur consommation, entrainant ainsi une crise alimentaire. Les décisions deproduction ne peuvent pas être basées sur les évolutions des prix, ce qui cause une allocation desressources inefficaces. Par ailleurs, l’impact négatif concernant l’excès de la volatilité impliquesouvent l’incertitude des prix qu’elle provoque. Puis, l’excès de volatilité déstabilise le marchédes matières premières, qui peut provoquer un risque systémique étant donné que les marchésdes matières premières sont liés avec les autres marchés financiers tels que les marchés des ac-tions et des obligations.Une partie de cette variation des prix est liée aux caractéristiques spécifiques des matières pre-mières agricoles. En effet, ces produits sont fortement dépendants des conditions naturelles,telles que la météo, le climat mais aussi les parasites qui peuvent décimer les cultures. En outre,l’élasticité-prix de la demande est relativement faible. Après un choc d’offre, les prix vont varierfortement afin de rétablir un nouvel équilibre. Pour ce dernier, le niveau de stockage joue un rôleimportant. En 2007, le stockage des produits de céréales atteint un niveau historiquement bas auplan mondial, ce qui est considéré comme l’une des raisons de la hausse des prix de ces produitssur la période.Un ensemble de facteurs complexes issus de la modification de l’environnent économique et desmarchés des matières premières, a entrainé une crise sur la période de 2006 à 2008 et sur lesmarchés des matières premières. L’environnement macroéconomique aux Etats-Unis s’est carac-térisé par un niveau faible de taux d’intérêt et par une politique monétaire expansionniste. Lahausse des prix de l’énergie et de la dépréciation du dollar font grimper les prix des produitsagricoles. Par conséquent, certains (ex : Frankel 2008) soulignent que les matières premièresdeviennent l’un des investissements alternatifs pour les gestionnaires de portefeuille, ce qui faitaugmenter les prix. De plus, la hausse de la demande en provenance des pays émergents (laChine, l’Inde par exemple), contribue à la hausse de prix au début 2006.
9
Néanmoins, ces deux facteurs expliquent difficilement l’accélération extrême de hausse de prixet de volatilité. La spéculation sur les marchés dérivés des matières premières est le plus souventcitée comme une explication de la flambée des prix à court terme. Ce facteur est lié au thème «financialization » des marchés des matières premières. La « financialization » sur ces marchés estdécrite par le phénomène de co-variation entre prix des matières premières et des actifs financiers(ex : actions et obligations). L’investissement sur les matières premières permet de diversifier leportefeuille afin de réduire le risque global. Dans ce contexte, de nouveaux produits dérivés desmatières premières, en particulier les fonds indiciels, contribuent à l’essor rapide et importantdans le marché des commodités. Cela va causer un risque systémique et une menace pour la sta-bilité du système financier en vue de la dépendance entre les acteurs du marché et les produitsde différents marchés financiers. L’analyse des facteurs individuels ne peut pas fournir une ex-plication convaincante de l’évolution des prix. Il y a encore un désaccord sur les facteurs qui secachent derrière la crise entre 2006 et 2008 sur les marchés des matières premières.
La réponse des régulateursFace à la flambée des prix, à la volatilité et à l’instabilité des marchés, les régulateurs vont organ-iser des réponses portant d’une part sur la politique de stockage du marché physique et d’autrepart, sur la contrainte des opérations de dérivés sur le marché financier. Plusieurs approches ontété proposées après l’épisode 2006-2008 avec un accent particulier sur les pics et volatilité desprix.
Marché physique : renforcer la transparence des informationsDans le contexte des interventions sur le marché physique, la politique de stockage et la politiquecommerciale internationale sont deux outils principaux. La politique agricole commune (PAC)dans l’Union Européenne existe depuis 1960 pour la stabilisation des marchés agricoles. Son ob-jectif consiste à améliorer la productivité agricole et à protéger les bénéfices des pays membrescontre des chocs externes issus du commerce international, souvent à travers des subventions. Lesrèglements de la PAC en 2014 se concentrent sur le bon fonctionnement des systèmes de stockageprivé1.La transparence de l’information est l’autre volet de la régulation du marché physique. A ce titre,le thème du Système d’information sur le marché des produits agricoles (AMIS) a été proposé lorsdu sommet du G20 2011, dans le but de recueillir des données du marché des produits agricoleset de contribuer à la transparence du marché. Dans l’attente de sa mise en place, il est égalementimportant de gérer et d’atténuer les risques pour les acteurs de ces marchés physiques qui ont lemoins de connaissance en terme de finance. C’est pourquoi notre étude sur le marché physiquese centre sur la gestion des risques au sein des coopératives agricoles, acteurs les plus exposés dece point de vue.
1La contrainte est que la politique de stockage ne fonctionne pas très bien parce qu’il est difficile d’identifier les prixappropriés. De plus, les politiques mises en place peuvent décourager la spéculation excessive, et souffrent de la mêmeinefficience à cause de la complexité des contrats dérivés.
10
Marché financier des matières premières:Superviser et contrôler l’impact des activités spéculativesUne autre réponse face à la complexité des marchés des matières premières est l’intervention surles marchés financiers des dérivés. Le dispositif clé des régulateurs repose sur la surveillance desactivités de ces marchés afin de vérifier qu’ils assurent correctement leurs fonctions de couver-ture des risques et de formation des prix. Aux Etats-Unis, CFTC (Commodity Futures TradingCommission) est un régulateur spécifique sur les marchés américains des futures et options quipossède une longue expérience dans la lutte contre les manipulations sur ces marchés. Dans leursréponses et moyens d’actions, peuvent être cités en exemple : les limites de position (limiter lenombre de contrats ouverts détenus par opérateurs) ; la publication des rapports des activités destraders (CFTC publie régulièrement la situation des marchés et la répartition des positions decontrats parmi les différents acteurs du marché).Depuis 2008, les reformes de Dodd-Frank Act et Consumer Protection Act consistent à renforcerla surveillance des marchés. Cette fois-ci, CFTC propose quelques politiques concernant lesswaps qui sont des produits de gré-à-gré. La surveillance des activités des traders se fait à traversle rapport de COT (Commitments of Traders) qui inclut une nouvelle catégorie d’opérations,celles des investisseurs indiciels. Au niveau européen, la révision de la directive MiFID (Direc-tive marché d’instruments financiers), démarrée depuis 2012, contient des réglementations quiconcernent les marchés financiers. Et pour la première fois, les marchés dérivés des matièrespremières apparaissent dans la directive. L’enjeu de MiFID est d’assurer l’intégrité et le fonction-nement des marchés des matières premières pour la couverture de prix et la formation des prixqui sont souvent la référence pour les produits physiques. De plus, les autres décisions régulatri-ces liées aux marchés des matières premières contiennent la révision de Market Abuse Directivepour la transparence des opérations ; la révision de UCITS (Undertakings for the collective in-vestment in transferable securities) pour les produits indiciels des commodités ; et la réformed’évaluation des risques pour les banques qui tient compte des activités des matières premières.Par ailleurs, les institutions internationales ont aussi publié des rapports sur la régulation desmarchés des matières premières (par ex. : IOSCO, OECD etc..). Néanmoins, les activités de cesinstitutions internationales se limitent aux recherches. L’application des politiques régulatriceset coopération internationale sont encore limitées.
Problématiques et structure de thèseLes sujets portant sur la régulation des matières premières sont larges. Suite à la discussionrelative à la situation et à la régulation des marchés des matières premières, cette thèse va sestructurer en deux parties, le marché physique et le marché financier des matières premièresagricoles et prendre trois thèmes intéressants à étudier. (i) D’une part, si les interventions surles marchés des matières premières sont limitées, une autre option toujours ouverte est de tenterd’atténuer les impacts négatifs des comportements extrêmes des prix. Ce thème fait le point surla gestion efficace des risques. (ii) D’autre part, à ce jour, aucun consensus n’a pu être trouvéconcernant l’impact des nouveaux produits dérivés (ex : fonds indiciels). Cette question peuten fait être abordée directement dans l’impact des fonds indiciels sur le stockage des matières
11
premières. (iii) Enfin, parmi les régulateurs sur les marchés des matières premières, le rôle desbanques centrales est souvent ignoré. Pourtant, elles peuvent aussi être un régulateur efficace surces marchés. S’appuyant sur ces trois points de départ, cette thèse se structure comme suit.
La partie I est composée de deux chapitres qui se concentrent sur les marchés physiques et ges-tion des risques.Chapitre 1, Etude du comportement des prix du blé, est une étude de départ qui aide à comprendreles caractéristiques des prix des produits agricoles. L’attention est portée sur les sauts extrêmesdes séries des prix physiques de 2006-2014. Pour cela, un modèle stochastique de Double sautsexponentiel (Kou 2002) est appliqué. L’intérêt de ce modèle est de nous permettre de constaterles comportements de sauts et de la volatilité du blé. Les résultats justifient l’évidence de sautsfréquents pour l’épisode 2007-2008 et 2010-2011. Une volatilité forte des prix est observée lorsde la crise financière et lors de période comportant un choc climatique. Ce chapitre sur le com-portement des prix donne une base d’étude pour le chapitre suivant relatif à la gestion de risqueset prémisse à la recherche de régulation des marchés.Chapitre 2, Gestion des risques au sein des coopératives porte sur la prise en compte des variationsextrêmes des prix sur la gestion de risques. L’indicateur de risque VaR (Value-at-Risk) est simuléavec un modèle traditionnel et un modèle de sauts. La comparaison de cet indicateur sur troisportefeuilles similaires des activités des coopératives donne la conclusion suivante : la prise encompte de sauts sur les mesures de risques peut effectivement améliorer la prévision des risqueset éviter leur sous-estimation. La VaR avec prise en compte des valeurs extrêmes est un outilcomplémentaire dans la gestion des risques des coopératives et ne pose pas de problème de sur-coût de réservation ou de provision. Cependant, la prévision de survenance des sauts doit êtreprudemment étudiée.
La partie II comprend deux chapitres relatifs aux financialization du marché des matières pre-mièresChapitre 3, L’impact des fonds indiciels, analyse l’influence potentielle de fonds indiciels sur lemarché des prix agricoles. Parmi les études sur le sujet, cette étude se tourne vers la relationdirecte entre fonds indiciels et stockage. Nous commençons par le débat entre (Krugman, 2008)et (Pierru Babusiaux, 2010) concernant le rôle du stockage dans la relation entre spéculation etfonctionnement du marché. En vertu de régression de série temporaire, nous avons trouvé quel’impact de fonds indiciels sur le prix des matières premières est significatif à court terme maisl’impact sur le stockage n’est pas significatif. Cet impact n’a donc pas besoin de passer par lestockage à cause de l’inélasticité-prix offre et demande. Cette étude justifie la nécessité de régu-lation sur les produits indiciels des matières premières. Toutefois, le rôle des produits indicielsdans la volatilité extrême des prix physiques des matières premières reste limité. L’impact le plusimportant s’explique par les conditions fondamentales des marchés (climat, météo, . . . ).Chapitre 4, La réaction des marchés des matières premières aux annonces de la banque centrale, reprendla recherche de l’impact de l’environnent macro-économique sur les marchés des matières pre-mières. L’étude met l’accent sur la réaction des indices sectoriels des matières premières sur les
12
communications de la Réserve Fédérale. Nous avons prouvé que les prix des matières premièresdeviennent plus corrélés avec les activités de banque centrale. Et cette relation repose surtout surle taux de change et l’inflation. De l’autre coté, nous avons montré que la banque centrale peutaussi jouer un rôle dans la régulation des matières premières et ses propositions de politiquesmonétaires doivent aussi tenir compte des conséquences sur les marchés des matières premières.Ce sont des propositions de régulation des matières premières basées sur la compréhension desmarchés des matières premières qui constitue l’objet de cette thèse.
13
Bibliography
[1] Frankel A.Jeffrey. (2006), The effect of monetary policy on real commodity prices. NBERWorking papers series 12713.
[2] Kou S.G. (2002) A Jump-Diffusion Model for Option pricing. Management Science, Vol. 48,No. 8, pages1086-1101
[3] Krugman (2008), cite dans son bloc au Financial Times : The only way speculation can havea persistent effect on oil prices, then, is if it leads to physical hoarding – an increase inprivate inventories of black gunk. . . .”
[4] Pierru Babusiaux (2010), Speculation without oil stockpiling as a signature : a dyanmicperspective, OPEC Energy Review, Vol. 34, Issue 3-4, pages 131-148
14
Part II
Commodity prices and risk management
15
2Behavior of French wheat prices
Abstract
This paper uses a Double-exponential jump model (DEJM) including mean reversion and time-varyingvolatility to investigate the stochastic behavior of the daily price of wheat delivered in Rouen (France) from2004 to 2014. The parameters are estimated separately for 10 harvests using maximum likelihood estima-tion. The main empirical results can be summarized as follows: 1) The estimated parameters among the10 campaigns illustrate the development of wheat market and the influence from the new economic envi-ronment 2) The empirical results show that jump component and stochastic volatility do play an importantrole in the behavior of wheat spot prices, especially during 2007/2008 and 2010/2011. 3) by distinguishingthe upward jump and downward jump, the DEJM is effective to fit the data in wheat spot prices. Likelihoodratio test and model fitting indicate also that the DEJM model outperforms alternative models. Overall, itis appropriate to consider the prices’ discontinuity in the risk measurement and management.
Keywords: Double-Exponential jump model; wheat spot price; mean reversion; stochastic volatility; likeli-hood maximizationJEL Classification:G17 Q11
2.1 Introduction
Since the 2000s, most commodities markets have been characterized by high volatility and up-ward drift. Although some commodities’ prices decreased after 2008, the price level for thesecommodities remains high. A few studies focus on examining the features of evolution of differ-ent kinds of commodity prices. Askari and Noureddine (2008) discuss the dynamic of oil pricesthat are consistent with fundamentals of the markets and global economy. Trostle et al. (2011)reexamine the recent phenomenon of food commodity and summarize the factors explaining theprice surge. The relation of demand and supply is the main explanation of commodity prices,according to most studies. The objective of this paper is to analyze the variation of agriculturalcommodity prices during the last several years, taking here wheat as the example, and to investi-gate the price jump in french wheat market in order to discuss its risk management application.
Wheat is one of the three main cereals for human food consumption. Production and prices ofwheat have increased greatly with the increased demand from developing countries. The US Sen-ate Permanent Subcommittee (2009) produced a report about the excess volatility in the wheatmarket. The report indicates there is significant evidence that financial investment is one of themajor causes of ‘unwarranted changes’. Considering the French market, production reached 36.3
16
million tons per year during the period 2003-2007, which is the fourth-highest rate of produc-tion in the EU. Besides the fundamental market factors, some opinions rely on the argument thatthis high volatility is due to the speculation in commodity derivatives markets. Recent commod-ity prices display a discontinuity evolution, which may be a reaction to an event or information,such as unpredictable weather disaster or behavior from financial market.
An understanding of the dynamics of price series is of great interest as this is a premise of mar-ket regulation, and risk management requires a further understanding of commodity spot pricesmodeling. The focus of this paper is to estimate a model which mimics the stochastic volatility,jumps and seasonality of agricultural commodities. With the parameters estimated, we expectto understand the reasons and the characteristics of agricultural prices. Parameters will be es-timated by using different data sets (10 harvests in total). One harvest of wheat is from July ofone year to June of the next year. Wheat in different harvests would be considered as differentproducts, since the quality could be distinct. The price gap between two different harvests canbe observed from the price difference between the last opening day in June and the first openingday in July next year (Figure 2.4 of the appendix).With the objective of identifying the jump, the application of Double-Exponential Jump modeldeals with the problem that commodity physical market is less liquid and probably have less fre-quent jumps than other financial markets. Comparing to other traditional jump models whichusually suppose that jump size has normal distribution, this model is relevant to apply in dis-crete data. And Double-Exponential Jump model gives not only the information on jump sizeand frequency, but also the distinction of positive and negative jump, which is of interest in thestudying of jump behaviour.
The results show that modeling the wheat spot prices using Double Exponential Jump methodis flexible enough to allow to capture the special characteristics of wheat prices such as mean-reversion, time-varying volatility and jump. The French wheat prices exhibit mean reversion, asmost of the agricultural commodities and are volatile during several harvests. It is crucial forrisk managers and policymakers to take the jump into account, since the jump term is significantaccording to the results.
The rest of the paper is organized as follows. In section 2, we give a summary of the characteris-tics of commodity prices and the related literature. In section 3, a stochastic model is developedto capture most of the features of agricultural commodities. In section 4, the data and some pri-mary empirical analysis of the data will be introduced. In section 5, we analyze the behavior ofwheat prices by describing the estimated parameters and comment on the results. Finally, section6 gives the conclusions and applications.
17
2.2 The features of commodity prices and related works
A wide range of literature has developed time series models to simulate the commodity price dy-namics. The time series models permit us to capture the accurate commodity price behavior suchas skewness right, excess kurtosis and the volatility behavior. These features may be consideredin the stochastic process, including mean reversion, time-varying volatility, discontinuity, etc.
It is stylized empirical evidence that agricultural and other commodity prices exhibit mean re-version especially in the competitive markets framework. Under the reaction between demandand supply, commodity prices increase in the case of shortage. As a response, consumers reducetheir demand so that the prices revert to supply and demand balance. The Ornstein-Uhlenbeckmodel first discussed by Ornstein and Uhlenbeck (1930) describes the mean-reversion feature.Bessembinder et al. (1996) point out that mean reversion is more likely to occur in agriculturalcommodity markets than other commodity market. Its application in commodity price modelingcan be seen, for example, in Gibson and Schwartz (1990), Schwartz (1997), Schwartz and Smith(2000) and (Bernard et al.2008). They incorporate mean reversion in the drift term of the stochas-tic differential equation of spot prices.
Except at the prices level, volatility is another indicator which needs to be examined. A modelwithout stochastic volatility could give biased estimates of the process (Crain et al. 2000). Thevolatility models, Autoregressive conditional heteroskedasticity (ARCH) by Engle (1982) and theGeneralized Autoregressive conditional heteroskedasticity (GARCH) by Bollerslev and Ghysels(1996) are the most widely used. ARCH and GARCH models capture the excess kurtosis in com-modity price distribution, as the unconditional distribution may be not normal. Beck (2001)develops the GARCH model in the commodity market and surmises that a serially correlatedprice variance process exists only for storable commodities.
Concerning the distribution of price returns, empirical evidence indicates that price returns re-vealed excess kurtosis and are not normally distributed. The discontinuity of the price series(jumps/spikes)is one of the explanation of this phenomenon. Merton jump diffusion model (Mer-ton, 1976) used initially in the market is widely applied in finance, such as bonds and bond op-tions (Das & Foresi, 1996; Das, 2002) and commodity prices (Hilliard & Reis, 1998). Jumps in theprice series may be the result of the unpredictable shock to the financial market or a stock-out inthe commodity market. The traditional Jump-diffusion model uses the Poisson-Gaussian process:the continuous part of which is a Wiener process which measures the “normal” dynamic in pricesdue to relation between supply and demand; the jump component of which is a Poisson processand captures the non-continuous price dynamics. Other applications can be seen in Bernard etal. (2008, 2007)and Askari and Krichene (2008), etc. In contrast to the hypothesis of the Mertonjump diffusion model, the double exponential jump model used in this paper assumes that thejump size follows a double exponential distribution, which indicates that the upward and down-ward jumps have independently exponential distribution. The model is proposed and applied in
18
the option pricing by Kou (2002) and Ramezani and Zeng (1998). Some empirical works in thederivatives pricing models support that the double exponential diffusion model outperforms thenormal jump diffusion model (Ramezani & Zeng, 2004). Moreover, the jump size measures theprice changes in response to outside news; Kou (2002) argues that the double exponential jumpmodel captures the investors’ behavior in the overreaction and under-reaction to outside news.Based on this result, this research applies and develops the model of Kou (2002) and Ramezaniand Zeng (1998) on the modeling of wheat spot prices where it also includes the feature of time-varying volatility.
2.3 The wheat market and empirical analysis of prices
The market
The data that is used is the prices of wheat delivered to Rouen. Rouen is one of the biggest ports asa delivery place for agricultural commodities for the physical market and for the financial deriva-tives market as well. A great part of French wheat production is used for exportation. France isa net-exporter country. The level of inventory in France was relatively stable from 2005 to 2012.More about production and commercial balance of wheat delivered to Rouen is given in figure2.5 of the appendix.
By looking at the French wheat price series, it can be seen that the recent wheat market exhibitshigher-level price and volatility. Figure 2.6 in the appendix gives the price series from 2004 to2014. Prior to 2004, the wheat price is found at the lowest level, which corresponds to the re-cession in the USA. The prices swing up or down, but reverse to a mean level. The sharp pricespikes during the period 2003-2004 occur due to the low amount of inventory, which is causedby unfavorable weather conditions. The prices attain a peak of 292 Euro/ton in mid-2007, andincrease 100% compared to the price level in 2006. From January 2006 to April 2008, the priceof wheat increase about 159%, and then decrease abruptly to the long-term level in the secondsemester of 2008. Meanwhile, the prices in the period of 2007/2008 become volatile compared tothe previous harvest. The price changes between two business days may be 8 Euros intra-harvest,which is large in the physical market. The prices continue to fluctuate in the following two har-vests, although the average return to the level of before the financial crisis in 2008. Another peakis in 2010/2011 and compared to the summit in 2007/2008, prices are also volatile and a higherprice level is maintained in the next harvest of 2011/2012, when Russia banned the exportationof wheat due to an extremely dry climate, which produces a supply shock to the global wheatmarket.By also comparing between price series and the first-order difference series in figure 2.7, firstly,the first-order difference series show the existence of price jumps. First-order difference pricesoscillate around zero, and some extreme price difference between two days can be observed. Sec-ondly, the effect of “Volatility clustering” is found. The higher volatility period in Figure 2.7
19
coincides with the higher prices of Figure 2.6, which leads to a modeling with GARCH.
Primary empirical analysis
The price series properties are first analyzed in order to establish how to model the series. Thedaily spot prices of wheat delivered at Rouen from 2004 to 2014 are chosen, which incorporatehigh volatile period of 2007/2008 and the recent peak in 2011/2012. A few arrangements aremade: Firstly, the data is divided by 10 harvests, from July to June the next year, as the paperintends to study the prices harvest by harvest and considers that wheat in different harvests aredistinct products, although this approach leads to decrease the accuracy of parameter estimationin the small sample. Secondly, the physical market is less liquid than the financial market. Theprice may not change for several days. The days in which the price is the same as the previousday are deleted.
Table 2.1: Descriptive statistics
Log return Mean Std Deviation Variance Excess Kurtosis Skewness
2004/2005 -0.0011 0.0196 0.0004 9.7423 -0.41202005/2006 0.0004 0.0092 8.43E-05 26.6499 3.50992006/2007 0.0026 0.0174 0.0003 3.9229 0.92902007/2008 0.0004 0.0237 0.0006 0.6912 -0.36912008/2009 -0.0022 0.0226 0.0005 10.1772 1.58422009/2010 -0.0002 0.0144 0.0002 0.7993 -0.15662010/2011 0.0013 0.0249 0.0006 0.7192 -0.07712011/2012 0.0009 0.0162 0.0003 0.5689 0.39882012/2013 -0.0007 0.0168 0.0003 6.1398 -0.09392013/2014 -0.0011 0.0115 0.0001 0.2206 0.0211
Note: Descriptive statistic are calculated from log-return series for every harvests.
Descriptive statistics for the log return (defined as the difference of log prices of two openingdays) is given in Table 2.1. The return series are not normally distributed, as they have non-zeroskewness and positive excess kurtosis. It is justified that modeling by a simple Brownian processis ineffective.
Two years after the financial crisis of 2008, the mean returns are negative. All the returns seriesdisplay a ‘fat tailed’ characteristic, with positive excess kurtosis. This evidence motivates a jumpmodel and it is important when examining the high frequency (daily) data (Wilmot & Mason,2011). However, another explanation of leptokurtosis is time-varying volatility. For this reason,time-varying volatility will also be considered in the following, and the likelihood ratio test will
20
be taken to examine the effect of time-varying volatility on the model fitting.
Figure 2.1: Seasonality
Note: The monthly averaged prices are presented below to examine the seasonality. Thisgraph gives the monthly averaged price of the harvest: 2006/2007, 2007/2008, 2010/2011 and2011/2012. The conclusions of other 6 harvest are the same: There is not seasonality obviouslyobserved from the graph.
From Figure 2.1, where monthly wheat prices are presented separately, the seasonality of wheatcannot be found in the lines. According to the seasonality of agricultural products, the price ishigher when it is near the harvest due to the depletion of inventories. The non-seasonality ob-served in this graph may be explained by the fact that inside one harvest the inventory of wheatcan play a role of smoothing the prices. That is why in the following model estimation seasonalityis not added.1
For testing the stationarity of each harvest, Table 2.2 gives the results of Augmented Dickey-Fuller tests. Hypothesis of unit-root is rejected at 5% level for all the harvest series. All log pricesseries are stationary, which may suggest the mean-reversion in log prices. The mean-reversionproperty is also verified with the calculation of Hurst exponent.
1It should be noted that the seasonality is taken into account by a trigonometric function in the beginning of thiswork; however it is difficult to find the parameters with the optimization algorithm and, even if the estimated parameterswere found, the results were inconsistent and impossible to interpret.
21
The Hurst exponents obtained by R/S analysis with log prices series of every campaign are smallerthan 0.5 and significant at the 5% level according to the t-stat. On the basis of Hurst exponent,the wheat prices have a mean-reversion behavior.
Table 2.2: Augmented-Dickey Fuller test and Hurst exponent
ADF-test Test critical value: 5% level07/2004-06/2005 -6.7787 -1.942107/2005-06/2006 -5.6823 -1.942107/2006-06/2007 -17.1640 -1.942107/2007-06/2008 -8.2637 -1.942107/2008-06/2009 -17.0044 -1.942007/2009-06/2010 -4.4711 -1.942307/2010-06/2011 -9.3745 -1.947707/2011-06/2012 -6.2747 -1.973607/2012-06/2013 -4.1501 -1.942407/2013-06/2014 -1.3184 -1.9425
Hurst exponent t-stat07/2004-06/2005 0.4358 8.58E-2107/2005-06/2006 0.2077 3.44E-0707/2006-06/2007 0.3107 7.06E-3007/2007-06/2008 0.3350 3.23E-2307/2008-06/2009 0.4293 1.28E-3307/2009-06/2010 0.2319 1.43E-2207/2010-06/2011 0.2890 2.88E-3307/2011-06/2012 0.4053 2.41E-3207/2012-06/2013 0.3927 1.69E-1507/2013-06/2014 0.2681 1.14E-12
Note: These two tests are taken on the log prices series. Lag numbers of ADF-test are chosen byAIC/BIC criteria.
2.4 Model application and estimation results
Model application
In this model, the log price is assumed to follow a Brownian motion with a mean-reversion term,plus a compound Poisson process with jump sizes double exponentially distributed (Kou, 2002;Ramezani & Zeng, 1998). This model allows to study the positive and negative jumps respec-tively. Volatility is supposed to be stochastic (GARCH). With Xt = ln(St), log prices and Vt = σ2
t ,
22
volatility:
dXt = κ (µx −Xt)dt + σtdBt + JtdNt (2.1)
dVt = κv (µv −Vt)dt + σvVtdBvt (2.2)
Xt = ln(St), log prices can be stated by a Jump-diffusion process which is presented by an Ornstein-Uhlenbeck process plus a Poisson term.In equation (2.1) and (2.2), the Ornstein-Uhlenbeck process captures the mean-reversion behav-ior of prices. k denotes the rate that the log prices Xt return to a equilibrium or a mean value µx;Volatility
(Vt = σ2
t
)is supposed to be stochastic. The volatility with GARCH behavior is modeled
as a mean-reversion process. The Brownian motion Bt and Bvt , following Nt ∼ (0,dt) are assumedto be independent.
JumpThe discontinuous component of the price process is described by a Poisson counter Nt in theequation (2.1), with intensity λ and jump size Jt . The intensity λ is the average number of jumpsper unit of time, which can also be explained as the probability that jump occurs, P rob (∆Nt = 1) =λdt and probability of no jump, P rob (∆Nt = 0) = 1−λdt.When the jump occurs, the log price jumps from X(t−) to Xt = Jt +X(t−). We know that Xt = ln(St),so ln (St) = Jt + ln
(S(t−)
), or St = e(Jt)S(t−). We have
(eJt
)− 1, the percentage change of the prices.
The double exponential distribution of Jt is supposed to be independent of Bt and Bvt . It is the Jtthat illustrates the leptokurtic of the price returns. The double exponential distribution of Jt is:
fJt =
η1e−η1Jt P robability = p (2.3)
η2eη2Jt P robability = 1− p (2.4)
The probability p, is the probability that an upward jump occurs, and 1-p is the probability ofa downward jump. According to the exponential distribution, the means of exponential randomvariables are 1/η1 and −1/η2 for positive and negative jump sizes respectively. The mean of Jt isobtained easily, which is equal to p/η1 − (1− p)/η2. The jump size is not normally distributed buthas the leptokurtic feature. And the Double Exponent Jump model (DEJM for simplicity in therest of this paper) gives the feature of overreaction and under-reaction to outside news and alsothe reaction to good or bad news; “DEJM can be interpreted as an attempt to . . . incorporate in-vestors’ sentiment” (Kou 2002). And the hypothesis of double exponent jump is more practicallyapplied in illiquid market with discrete jumps. In this case, the two types of jump are supposedto be independent for simplification.2
2Further research may be conducted also on the correlation of positive and negative jump sizes.
23
VolatilityFrom equation (2.2), the specification of volatility is a continuous case of GARCH (Brockwell andChadraa 2006). For facilitating estimation, we can re-write Vt = ω + αε2
t−1 + βVt−1 with εt−1 theerror term at time t-1; ω > 0 et α ≥ 0,β ≥ 0; and α + β < 1 (stationary condition). Transforming tothe continuous form of GARCH, we have kv = 1−α − β and σv = α
√2 (Daniel B. 1990).
The equation of volatility, Vt , is the discrete form of GARCH, whereω
1−α − βmeasures the long-
run average variance per day by using the daily data here. Using the square-root-time rule, the
annualized volatility may be calculated by√
ω1−α − β
√252; α + β is the persistence of volatility.
Likelihood functionFor the likelihood function, the density of Xt is simplified as a Bernoulli weighted sum of normaland exponential density. The three sources of randomness, Bt ,Bvt and Nt , are assumed to be in-dependent. Price processes are divided into two regimes: Jumps happen with probability λ; Nojumps with probability 1−λ.Define a random variable Xt as a sum of independent normal (with mean µ and variance σ2) andexponential (with parameter δ) random variables. Its density function is:
g(X′ )
=δ2e
δ2
(2µ+δσ2−2X
′ )erf c
(−µ+ δσ2 −X ′√
2σ
), (2.5)
where erf c (x) the complementary error function =2√π
∫∞xe−s
2ds
Then density function can be written also as:
g(X′ )
= δe
δ2
(2µ+δσ2−2X
′ )Φ
X ′+µ−δσ2
σ
, with Φ (): Cumulative distribution CDF of standard normal
variables.
The density function of Xt combined with a Bernoulli distribution in our case can be calculated(see also Ramezani and Zeng (1998) and Kou (2002)):
f(Xt) = λ[pη1e
σ2η21
2 e−(X(t)−(1−κ)X(t−1)−κµx)η1Φ
(− (X (t)− (1−κ)X (t − 1) +κµx −Vtη1)
√Vt
)
+(1− p)η2e
σ2η22
2 e−(X(t)−(1−κ)X(t−1)−κµx)η2Φ
((X (t)− (1−κ)X (t − 1) +κµx −Vtη2)
√Vt
)]
+(1−λ)1√
2πe
− (X (t)− (1− k)X (t − 1)−κµx)2
2Vt
(2.6)
24
The log likelihood function to be maximized is: L (P ) =∑Tt=1 ln [f (Xt)] with T the last trading day
of the sample.
Parameters P = (κ,µx,p,η1,η2,λ,ω,α,β) can be estimated by maximization of the likelihood func-tion numerically3. Positive value constraints are put on all the parameters. The initial parametersare chosen carefully since results are highly dependent on the initial values.
Estimation results
Table 2.8 in the appendix gives the results from the maximum likelihood function. Parametersare given by different harvest periods (10 harvests in our sample). All the estimated parametersare significant, with the level at least 10%. The standard errors reported are calculated using theHessian matrix. The tables below, which provide explanation of different behavior, are extractedfrom Table 2.8 of the appendix.
Mean-reversion behavior
Table 2.3: Parameter k and µx
k 1/k µx eµx
Speed of returning to mean Days of returning to mean Equilibrium log price Equilibrium price2004/2005 0.0218 45.9144 4.5803 97.54352005/2006 0.0908 11.0189 4.6426 103.81872006/2007 0.0514 19.4561 4.9601 142.61302007/2008 0.0191 52.4271 5.4882 241.82832008/2009 0.0169 59.3109 4.7776 118.81262009/2010 0.0600 16.6652 4.7706 117.98782010/2011 0.0363 27.5373 5.5063 246.25112011/2012 0.0903 11.0761 5.2870 197.75592012/2013 0.0200 49.997 5.2874 197.8292013/2014 0.0725 27.5781 5.4492 232.5602
κ presents how quickly the log-price reverts to its long-run mean in the log-price process and1/κ is the days to return to the equilibrium level (The transformation of number of days is basedon the opening days). Prices in the harvest during 2007/2008 and 2008/2009 revert to meanmore slowly than in other years, where it takes more than 50 days (two months). Another period,2004/2005, also takes about 45 days to revert to the long-run level, which may be explained bythe lower storage pressure in the previous harvest and the market uncertainty. In both 2004/2005and 2008/2009, the prices do not return to the equilibrium level as quickly as other periods, oneof the potential explanations is that during these harvests inventory is relatively lower.
3Parameters are estimated by using Matlab. A brief Matlab code is given in the annex
25
Regarding the equilibrium level of prices, the mean price is coherent with price variation and re-flect the commodity inventory adjustment. Peaks are in 2007/2008, 2010/2011 and 2013/2014,where prices tend to be about 241.8 Euros/ton. The 2007/2008 harvest coincides with the crisisfrom the financial industry. As a result, the explanations of the 2007 peak consist of the increasingdemand from emerging countries and the financialization of the commodity markets accordingto several researches. Then, the price peak in 2010/2011 is probably due to Russia’s exportationban, which put pressure on the international wheat market. The difference between 2007/2008and 2010/2011 harvests is that we observe that, after 2010/2011, the equilibrium price remainsrelatively high after the peak of the previous year; however, prices drop down dramatically in theharvest following 2007/2008.
Volatility behavior
Table 2.4: Parameter of volatility
ω α β α + β√
ω1−α − β
√252
Persistence of volatility Mean volatility
2004/2005 0.000079 0.2625 0.0078 0.2703 0.16512005/2006 0.000011 0.6161 0.0427 0.6855 0.09152006/2007 0.000087 0.0921 0.4389 0.5309 0.21632007/2008 0.000155 0.2045 0.4439 0.6484 0.33312008/2009 0.000103 0.0205 0.6489 0.6695 0.27972009/2010 0.000088 0.0282 0.4277 0.4559 0.20172010/2011 0.000024 0.1917 0.7643 0.9560 0.37202011/2012 0.000003 0.0040 0.9740 0.9780 0.18912012/2013 0.000150 0.1464 0.0415 0.1879 0.21572013/2014 0.000095 0.10 0.1119 0.2119 0.1740
The volatility is a little bit smaller than the volatility calculated4 in the descriptive statistic tableand in the traditional lognormal model, since part of volatility is interpreted by the jump term inour model. The volatility obtained by the GARCH model measures the ‘normal’ price variation,which is caused by fundamental conditions. In spite of this, the evolution path of volatility isthe same, with highest yearly mean volatility of 33% in 2007/2008 and 37% in 2010/2011. Thehigh volatility implies uncertainty in the wheat market and the market is sensitive to exogenousinformation. It is surmised that the uncertainty in 2007/08 comes from the tremors in the finan-cial market and relatively small harvest. And the supply shock in the international market dueto Russia’s exportation ban may be a main cause of high volatility during 2010/2011.
4Volatility in the table of descriptive statistic can be transformed to be annual by multipliering√
252
26
Another highly volatile campaign is the 2004/2005 harvest, which may be a result of the histor-ically lowest inventory in 2003/2004. The tension of the supply side due to unfavorable climatein the previous year caused the uncertainty to remain in the 2004/2005 harvest. Conformingto the simple statistical analysis, the volatility in 2008/2009 remains higher although the pricedecreases from the peak. Moreover, after another summit in 2010/2011, the prices vary less vi-olently. Using the specification of GARCH, volatilities in all harvests are not persistent (α + β < 1).
Jump behavior
Table 2.5: Parameters of Jump
p1η1
− 1η2
1λ
Probability of upward jump Mean of upward jump Mean of downward jump Number of jumps2004/2005 0.5577 0.0716 -0.0835 7.10112005/2006 0.5354 0.0284 -0.0162 7.61262006/2007 0.7965 0.0376 -0.0347 15.80742007/2008 0.3013 0.0595 -0.0589 14.49272008/2009 0.4989 0.0667 -0.0478 14.28182009/2010 0.5013 0.0300 -0.0325 16.66182010/2011 0.4600 0.0589 -0.0578 16.59672011/2012 0.5798 0.0358 -0.0327 12.37312012/2013 0.400 0.0667 -0.0667 16.662013/2014 0.5872 0.0250 -0.0333 24.99
The jump is distinguished by upward and downward jump. The table above gives the results ofjump parameters. The mean of jump is defined as the averaged difference between ln (St) andln
(S(t−1)
), if jump occurs. Jumps have happened more frequently since 2006. 2005/2006 has the
smoothest trait with no-frequent and small jumps. Prices increase from 2006 and oscillate alonga high level during 2007/2008. The probability of an upward jump is larger than a downwardjump, with p = 0.79 in 2006/2007. The upward jumps are dominant in this period. Althoughprices remain higher in 2007/2008, the downward jumps occur more often.Otherwise, two kinds of cases can be distinguished according to the results: Firstly, the harvestof 2004/2005 is categorized by jump with large size and lower frequency. Secondly, in the years2006/2008 and 2010/2011, the jumps happen more frequently but have lower value. The resultsof the jumps are confirmed with those of mean reversion and volatility: In the higher volatileperiod, the jumps are more frequent and prices deviate to the market equilibrium much longerthan in other periods. In 2004/2005, the jumps may be caused by the fundamental conditions,which provoke a large change in the wheat price. According to the previous results, the first twoharvests in our sample are dominated by the fundamental information (inventory status), whilethe smaller and lower frequency jumps in 2007/2008 and 2010/2011 may be explained by supplyshock from domestic and international market. The surmise of impact from the financial marketshould refer to more rigorous analysis.
27
2.5 Diagnostic tests
In this section, robust tests will be taken to examine the validation of the model. The objectiveis to investigate if the mean reversion, jump behavior and time-varying volatility are appropriateto fit wheat spot prices. For answering this question, likelihood ratio tests are taken with threealternatives models under different hypotheses; and four moments calculated from the simulatedprices series are compared with the moments by original data.
Likelihood ratio tests
Table 2.6: Likelihood ration tests(LR)
Without jump Without mean reversion Constant volatility MC p-value
2004/2005 49.579 9.620 6.896 0.012005/2006 59.963 1.907 3.109 0.012006/2007 29.470 5.923 5.890 0.012007/2008 13.224 4.042 23.254 0.012008/2009 39.970 1.227 1.844 0.012009/2010 8.486 2.511 32.174 0.012010/2011 4.991 8.152 23.492 0.012011/2012 2.865 1.033 1.596 0.012012/2013 18.571 1.492 2.058 0.012013/2014 4.343 3.770 1.148 0.01
Note: Likelihood ratios are calculated for every harvest by comparing with the models without jump,models without mean reversion and models with constant volatility
The likelihood ratios are calculated by estimating three other models, which are ARCH modelwith stochastic volatility (without mean reversion); Mean-reversion model with stochastic volatil-ity (without jump term); and Jump diffusion model with constant volatility (without stochasticvolatility). We are interested only on the LR-test. The results of parameters are not given in thispaper.Simply comparing the likelihood values does not permit us to choose the fitted model, as themodel with the larger number of parameters may have a higher likelihood value. The threegroups of likelihood ratios are calculated in the table.Likelihood ratio is defined as LR = −2[L (p̂,X)−L (P ,X)], withp̂ the parameters estimated by re-stricted models, P the parameters estimated by the complete model, and X the log prices. LR isChi-square distributed with k degree of freedom (k is the number of restrictions).We can see that the hypothesis is rejected. Compared to all alternative models, the initial model(DJEM) is significant. It is effective to investigate the wheat prices by the complete model accord-
28
ing to the likelihood ratio test.
Moment fittingThe prices will be simulated using the estimated parameters. Then the moments are calculatedby the new simulated series. The comparison will be taken between the original moments andthe moment simulated. Firstly, the simulated series and the original series are taken in the samegraph, as shown in figure 2.2.
Figure 2.2: One example of simulated prices and original prices
Note:The original prices are presented by the solid curve and one of the simulated prices isshown by the point curve.
29
Figure 2.3: Comparison between real log return and simulated log return
Note: The log returns in the simulated log-return graphs are calculated from the simulatedprices.In the simulation we have simulated 1000 series, which constitute a surface aroundthe real prices line. Only the simulated series below 10% standard deviation and above 10%standard deviation are considered.
Figure 2.3 shows only one simulated series, which is considered the closest to the real data series.The estimation and simulation are produced by a discontinuity Poisson process and for every har-vest. The simulated prices and log returns fit the real data during 2005/2006, which is also themost stable period among the 10 harvests. The deviation between the simulated series and theoriginal is much higher in the high volatility period. It should be noted that the real data dependsalso on the economic conditions and outside information, and the simulated price series dependsonly on the assumptions of time series. As in the simulation process, a Poisson distribution sim-ulates the jump with a well-defined frequency. However, concerning the jump size, whether it isan upward or downward jump, the estimation is conducted by the mean of the jumps size. As a
30
result by simulation, the jumps happen randomly without considering the real information fromthe market.
Table 2.7: Moments comparison
Mean Variance Kurtosis SkewnessSimulated Original Simulated Original Simulated Original Simulated Original
2004/2005 -0.0005 -0.0011 0.0037 0.0004 1.8414 9.7423 -0.1168 -0.41202005/2006 0.0004 0.0004 0.0004 8.43E-05 1.0896 26.6499 1.3459 3.50992006/2007 0.0026 0.0026 0.0090 0.0003 4.2099 3.9229 -0.7021 0.92892007/2008 0.0009 0.0004 0.0192 0.0006 2.0985 0.6912 -0.3561 -0.36922008/2009 -0.0022 -0.0022 0.0215 0.0005 10.9567 10.1772 0.8724 1.58412009/2010 -0.0004 -0.0002 0.0026 0.0002 7.0331 0.7993 0.2725 -0.15672010/2011 0.0030 0.0013 0.0156 0.0006 50.7101 0.7192 -1.3379 -0.0772011/2012 -0.0005 0.0009 0.0033 0.0003 3.7895 0.5689 -0.366 0.39882012/2013 -0.0019 -0.0007 0.0002 0.0003 0.2688 6.1398 2.8415 -0.09432013/2014 -0.0006 -0.0011 0.0001 0.0001 0.1555 0.2206 3.1607 0.0216
Note: Four moments are calculated:mean, variance, kurtosis and skewness from original log-return series and simulatedlog-return series.
The four moments for 10 harvests are obtained in Table 2.7. The first two moments do not di-verge too much from the original data. In contrast, simulated skewness and kurtosis are muchsmaller than the original. The explanation of this fact can also be due to the lack of assumptionabout the economic condition in the simulation. Another 3 simulation series are given (Figure2.8) in the appendix. The deviation between real data and simulated data always exists for thehigh volatility period. And these simulated series are quite different from each other given therandom simulation. At least, the simulated series replicate the main features of real data, such asmean reversion, and higher volatility for certain harvests.A brief conclusion of this section: The model used in this article is to capture the key featuresof wheat prices. Using mean reversion, jump process and varying volatility is effective accordingto the simple likelihood ratio tests. Chan (2005) finds similar results about short-term interestrate: that models combining mean reversion, jump and time-varying volatility fit the data, andthe conclusion of Ramezani and Zeng (2004) for stock market is mixed. The simulation cannotproduce a similar series during the high volatility period as a result of complicated market con-ditions. The first two moments are coherent with the original moments. It should be mentionedthat the skewness and kurtosis are unstable according to different series. After taking all theabove into account the complete model (DEJM) used in this paper is effective to investigate thetrait of wheat spot prices and to be applied in the market with less and discrete jumps observed.
31
2.6 Conclusion
In this paper, a stochastic jump process is used to capture the French wheat spot prices from 2004to 2014. The stochastic process incorporates almost all the categories of price variation, includingmean reversion, jump behavior and varying volatility. The modelling of jump as an exponentialdistribution deals with the problem of less liquidity in physical market and allow to distinguishbetween positive and negative jump. The parameters are estimated by a numerical maximizationof the likelihood function. Estimation is taken by different harvests independently. In the end,there are 10 groups of parameters to interpret. Several robust tests are taken after estimation. Itis reasonable to incorporate the mean reversion and jump term into the process. The focal pointis to study the wheat spot prices features that can be found from the analysis.At first, the wheat prices exhibit the mean-reversion behavior: after a shock on the prices, theprices will return to an equilibrium level. However, the time taken to return depends on the mar-ket condition. During the volatile period (2007/2008 and 2010/2011), prices fluctuate far fromequilibrium level. Secondly, the period of 2007/2009 exhibits higher volatility for French wheat.The changes in inventory and influence from other countries do have an impact on the Frenchwheat market, since the market is never only domestically confined. As a result, fundamentalcondition is still the main explanation of the fluctuation of wheat prices. Thirdly, 2007/2008 dis-plays more frequent downward and small jumps. In addition, compared with the early years, thewheat market undergoes more shocks from the outside, with more jumps happening. Finally, thefundamentals seem still to be able to provide the major explanation of the volatility, as supplyshock provoke higher volatility in respective periods. However, harvests with frequently jumphappening coincide with the periods of financial crisis. This fact leads to the conjecture of impactfrom financial markets.The finding in this paper gives us some features of wheat prices and is relevant for the regula-tors. According to the results of our modeling, it is crucial that the agricultural cooperatives andthe policymakers take the jump into account in their risk management strategies. However, theresults of this paper are not enough to justify the impact of “financialization” or precise factorson the wheat market. Which are the factors behind the volatile prices? Are the fundamental con-ditions (supply/demand) dominant in the agricultural market or does the speculation from thecommodity derivatives market produce the recent price changes? All the questions not concludedin this paper will be discussed in future research.
32
Ap
pen
dix
Figu
re2.
4:A
nex
amp
leof
pri
ces
gap
betw
een
harv
ests
Not
e:T
hegr
aph
give
sth
ep
rice
sbe
twee
nha
rves
ts20
04/2
005
and
2005
/200
6.T
hela
stop
enin
gd
ayfo
r20
04/2
005
is27
thJu
nean
dth
efi
rst
open
ing
day
for
2005
/200
6is
1st
July
.The
blac
kcu
rve
link
edth
ese
two
poi
nts
inth
egr
aph.
The
rear
eap
par
ent
pri
cech
ange
son
thes
etw
od
ays.
Asi
mil
arp
heno
men
onm
ayal
sobe
fou
ndin
othe
rye
ars.
Inth
isp
aper
,the
esti
mat
ion
isca
rrie
dou
ton
10ha
rves
tsse
par
atel
yby
cons
ider
ing
whe
atfr
omd
iffer
ent
harv
ests
asd
isti
nct
pro
duct
s(d
iffer
ent
qual
itie
s).
33
Figu
re2.
5:Fu
ndam
enta
lsof
whe
atm
arke
t
Not
e:T
heex
por
tati
on,
imp
orta
tion
and
tota
lou
tpu
tof
the
Fren
chso
ftw
heat
mar
ket
from
2004
to20
14ar
egi
ven
inth
isgr
aph
(y-a
xis:
1000
tons
/uni
ty).
As
ap
rodu
cer
cou
ntry
,Fre
nch
imp
orta
tion
issm
all.
Abo
ut
half
ofou
tpu
tis
use
dfo
rex
por
t.T
hele
vels
ofp
rodu
ctio
nan
dex
por
tati
onof
2007
/200
8ar
ere
lati
vely
smal
lcom
par
edto
othe
rye
ars.
(Sou
rces
:Agr
este
)
34
Figu
re2.
6:D
aily
pri
ces
seri
es
Not
e:T
hegr
aph
give
sth
ed
aily
spot
pri
ces
(Eu
ro/t
on)
ofFr
ench
soft
whe
atan
dit
isal
soth
ed
ata
that
are
use
din
the
pap
er.
The
sam
ple
per
iod
isfr
om20
04to
2014
,10
harv
ests
into
tal.
Bef
ore
2006
,p
rice
sva
ried
smoo
thly
alon
g10
0Eu
ros/
ton
and
incr
ease
daf
ter
that
unt
ilre
achi
nga
pea
kin
2007
.T
hep
rice
sos
cill
ated
betw
een
300
Euro
/ton
and
200
Euro
/ton
duri
ng20
07an
d20
08.
The
nw
heat
pri
ces
dec
reas
edaf
ter
2008
but
mai
ntai
ned
vola
tili
ty.T
here
cent
high
pri
ces
bega
nin
2010
and
are
char
acte
rize
dal
soby
ahi
ghvo
lati
lity
.
35
Figu
re2.
7:Fi
rst-
ord
erd
iffer
ence
ofp
rice
s
Not
e:T
hegr
aph
give
sth
efi
rst
ord
erd
iffer
ence
ofp
rice
s.T
hehi
gher
vola
tili
typ
erio
dis
seen
toco
inci
de
wit
hth
ehi
gher
pri
ces
per
iod
whi
chbe
gins
onJu
ly20
06.T
hefa
ctju
stifi
esth
eex
iste
nce
of“V
olat
ilit
ycl
ust
erin
g”w
hich
isre
pre
sent
edby
aG
AR
CH
mod
el.
36
Tabl
e2.
8:E
stim
ated
par
amet
ers
Cam
pag
ne
kµx
pη 1
η 2λ
ωα
β-L
ogL
ikel
ihoo
d20
04/2
005
Val
ue
0.02
184.
5796
0.40
1128
.149
324
.540
40.
0489
8.10
E-0
50.
2867
0.00
59-4
52.7
std
erro
r(3
.66E
-03)
(1.0
5E-0
2)(0
.002
1)(0
.090
1)(0
.012
6)(0
.002
3)(1
.25E
-05)
(5.1
3E-0
3)2.
14E
-03
2005
/200
6V
alu
e0.
1003
4.64
780.
5398
63.8
271
87.1
363
0.05
078.
49E
-06
0.19
850.
5775
-453
.95
std
erro
r(8
.91E
-03)
4.24
E-0
30.
0266
0.98
951.
000.
0303
1.62
E-0
23.
38E
-02
5.45
E-0
220
06/2
007
Val
ue
0.04
034.
9832
0.59
9728
.184
136
.224
10.
0310
2.84
E-0
50.
1379
0.66
82-6
07.8
std
erro
r(6
.15E
-03)
(3.2
3E-0
2)(0
.051
5)(1
.03)
(1.0
3)(1
.2E
-02)
(7.2
4E-0
6)(3
.03E
-02)
(0.0
463)
2007
/200
8V
alu
e0.
0239
5.47
540.
451
23.6
0321
.166
20.
035
0.00
020.
2119
0.37
98-5
27.1
std
erro
r(0
.005
2)(0
.077
1)(0
.892
0)(1
.228
8)(1
.69)
(0.0
210)
(2.5
3E-0
5)0.
0564
0.05
0820
08/2
009
Val
ue
0.02
264.
8342
0.43
3917
.947
523
.658
70.
0634
1.04
E-0
40.
0212
0.64
63-4
98.2
std
erro
r(9
.92E
-03)
(0.0
579)
(0.0
412)
(1.0
9)(1
.42)
(0.0
406)
(1.1
6E-0
5)(0
.018
3)(0
.035
1)20
09/2
010
Val
ue
0.06
004.
7706
0.50
1333
.316
330
.764
70.
0600
8.79
E-0
50.
0282
0.42
77-5
70.0
5st
der
ror
(0.0
015)
(1.2
1E-0
2)(0
.374
6)(1
.004
8)(1
.01)
(2.2
1E-0
2)(4
.20E
-06)
(0.0
129)
(5.8
8E-0
2)20
10/2
011
Val
ue
0.06
065.
4719
0.61
6526
.819
619
.233
90.
0472
3.51
E-0
50.
2012
0.73
67-5
14.4
std
erro
r(0
.010
6)(2
.09E
-02)
(0.1
105)
(0.9
451)
(1.0
1)(0
.036
4)1.
15E
-05
(0.0
433)
(0.0
307)
2011
/201
2V
alu
e0.
0903
5.28
740.
5798
27.9
485
30.9
311
0.06
013.
19E
-06
0.00
400.
9743
-419
.7st
der
ror
(2.2
6E-0
2)(1
.29E
-02)
(0.3
540)
(4.7
725)
(0.0
331)
(0.0
573)
(7.0
1E-0
7)(2
.90E
-03)
(3.0
8E-0
3)20
12/2
013
Val
ue
0.02
005.
4491
0.40
015
.00
14.9
960.
0600
1.50
E-0
40.
1464
0.04
15-4
92.2
std
erro
r(6
.66E
-03)
4.15
E-0
2(0
.014
5)(0
.408
9)(0
.036
5)(0
.022
6)(2
.14E
-05)
(1.7
8E-0
2)(1
.26E
-02)
2013
/201
4V
alu
e0.
0725
5.21
100.
5872
39.9
9729
.999
80.
0300
9.47
E-0
50.
1000
0.11
19-4
89.4
std
erro
r(1
.34E
-02)
(5.2
8E-0
2)(0
.185
)(1
0.44
18)
(9.7
4)(0
.046
4)(1
.17E
-05)
(8.9
5E-0
2)(8
.9E
-02)
Not
e:T
his
tabl
egi
ves
all
esti
mat
edp
aram
eter
su
sing
max
imu
mli
keli
hood
func
tion
.St
and
ard
erro
rgi
ven
inp
aren
thes
esis
calc
ula
ted
byH
essi
anm
atri
x.A
llth
ep
aram
eter
sar
esi
gnifi
cant
–at
leas
t10
%le
vel.
Inth
ees
tim
atio
n,si
gnco
nstr
aint
sar
egi
ven
onal
lpar
amet
ers
for
inso
me
case
sne
gati
veva
lue
isfr
aud
.
37
Figu
re2.
8:M
ore
sim
ula
tion
seri
es
Not
e:T
his
grap
hgi
ves
3si
mu
late
dse
ries
.s1
,s2
,s3
rep
rese
ntth
e3
sim
ula
ted
seri
esof
log
pri
ces
and
the
blac
kth
ick
curv
eis
the
real
dat
ase
ries
.In
the
2010
/201
1ha
rves
t,it
seem
sth
atth
esi
mu
late
dse
ries
cann
otar
rive
toth
esa
me
high
leve
las
the
real
one.
The
pri
ces
are
low
erat
the
begi
nnin
gof
the
harv
est
com
par
edto
the
end
ofth
eha
rves
t.T
hep
rop
erty
ofm
ean
reve
rsio
np
rodu
ces
asi
mu
late
dcu
rve
vary
ing
alon
gan
equ
ilib
riu
mle
vel,
whi
chis
low
erth
anth
ehi
gher
pri
ces
ofth
em
idd
leha
rves
t.Y-
axis
isth
elo
gpr
ices
.
38
Annex: Estimation using Matlab
In the following, we give the estimation code using maximum likelihood function in Matlab en-vironment for the first year data of 2004/2005.
load(’Camp.mat’)
1) Optimization functionSt=Camprest(:,1);t = Camprest(:,2);options = optimset(′Algorithm′ ,′ Interior − point′ ,′MaxIter ′ ,500);
- Initial parameters should be chosen firstly:parametre=[0.02 4.6 0.45 30 28 0.03 0.0001 0.01 0.01]’;
- Constraint conditions on certain parameters and t-value for significance test is applied withHessian matrix obtained from the function:[parametre, likeli, exitf lage,output, lamdba,grad,hessian] = f mincon(@(parametre)like1(parametre,St, t),parametre,[000000− 100;0000000− 10;00000000− 1] , [0;0;−0.001], [], [], [], [], [], options);
stanerr=sqrt((diag(hessian)).∧− 1);tstat = parametre./stanerr;
p1=parametre(1);p2=parametre(2);p5=parametre(6);p10=parametre(3);p11=parametre(4);p12=parametre(5);p6=parametre(7);p7=parametre(8);p8=parametre(9);
2) Define the vt: volatility would be a vector with length of Stvt(1)=0.0004;vt(2)=0.0005;
for i=3:157;vt(i)=p6+p7*(St(i-1)-St(i-2)-p1*(p2-St(i-2)))2 + p8 ∗ vt(i − 1);
end
mj=p10/p11-(1-p10)/p12;vj=p10*(1-p10)*(1/p11+1/p12)2 + p10/(p112) + (1− p10)/(p122);
3) Define the likelihood function
fun = @(x) exp(-x.2/2);
39
for i=2:157;X(i)= -(St(i)-St(i-1)-p1*(p2-St(i-1))-vt(i)*p11)/(vt(i)0.5)Y (i) = (St(i)− St(i − 1)− p1 ∗ (p2− St(i − 1))− vt(i) ∗ p12)/(vt(i)0.5)q(i) = integral(f un,X(i), Inf )p(i) == integral(f un,Y (i), Inf )
Lt(i)=log((1-p5)*exp(-(St(i)-St(i-1)-p1*(p2-St(i-1)))2/(2 ∗ vt(i)))/(2 ∗ pi ∗ vt(i))0.5 + p5 ∗ (p10 ∗ p11 ∗exp(vt(i)∗p112/2)∗exp(−(St(i)−St(i−1)−p1∗ (p2−St(i−1)))∗p11)∗q(i)+(1−p10)∗p12∗exp(vt(i)∗p122/2) ∗ exp((St(i)− St(i − 1)− p1 ∗ (p2− St(i − 1))) ∗ p12) ∗ p(i));
end
40
Bibliography
[1] Alexander Carol and Emese Laxar,2005, On The Continuous limit of GARCH. Working paper.
[2] Angus Deaton and Laroque Guy,1996, Commodity storage and commodity price dynamics.Journal of Political Economy, Vol. 104, No.5, 896-923.
[3] Angus Deaton and Laroque Guy,1992, On the behavior of commodity prices. The Review ofEconomic Studies, Vol. 59, Issue 1, 1-23.
[4] Barry K. Goodwin, Schnepf Randy and Dohlman Erik,2005, Modelling soybean prices in achanging policy environment. Applied Economics, Vol. 37, Issue 3, 253-263.
[5] Feldman Barry and Till Hilary,2006, Backwardation and commodity futures performance:Evidence from evolving agricultural markets. The Journal of Alternative Investments, Vol.9,No.3, 24-39.
[6] Stacie Beck,2001, Autoregressive conditional heteroscedasticity in commodity spot prices.Journal of Applied Econometrics, Volume 16, Issue 2,115-132.
[7] Jean-Thomas Bernard; Lynda Khalaf; Maral Kichian; and Sebastien McMahon. 2008, Oilprices: Heavy tails, Mean Reversion and the Convenience yield. Working paper, Bank ofCanada & Quebec Ministry of Finance, 2008.
[8] Jean-Thomas Bernard; Lynda Khalaf; Maral Kichian; and Sebastien McMahon.2008, Fore-casting Commodity Prices: GARCH, Jumps, and Mean Reversion. Journal of Forecasting, 27,2008: 279-291.
[9] Hendrik Bessembinder;Jay F. Coughenour; Paul J.Seguin and Margaret Smoller.1996, Isthere a term structure of futures volatilities? Reevaluting the Samuelson hypothesis. TheJournal of Derivatives, Vol.4, No. 2, 45-58.
[10] Michael J. Brennan 1958, The supply of storage. The American of Economic Review Vol. 48,No.1, 50-72, 1958.
[11] Peter Brockwell ; Chadraa Erdenebaatar and Alexander Lindner. 2006, Continuous-timeGARCH processes. The Annals of Applied Probability Vol. 16, No.2,790-826.
[12] Sorensen Carsten.2002, Modeling seasonality in agricultural commodity futures. The Journalof futures markets Vol.22, Issue 5,393-426.
41
[13] Kam Fong Chan.2005, Modelling conditional heteroscedasticity and jumps in Australianshort-term interest rates. Accounting and Finance, 45, 537–551.
[14] Pirrong Craig. 1999, The organization of financial exchange markets: Theory and evidence.Journal of Financial Markets Vol.2, Issue 4, 329-357, 1999.
[15] Cyrus A. Ramezani, and Yong Zeng.2004, An empirical assessment of the Double Exponen-tial Jump-Diffusion Process. Working paper, California Politechnic.
[16] Cyrus A., Ramezani, and Yong Zeng.1998 Maximum likelihood estimation of asymmetricjump-diffusion processes: Application to security prices. Working paper, University of Wis-consin.
[17] Daniel B. Nelson.1990 ARCH models as diffusion approximations. Journal of Econometrics,Vol. 45, Issues 1-2, 7-38.
[18] Daniel B. Nelson.2002, Conditional Heteroskedasticity in asset returns: A new approach.Econometrica 59, 347-340.
[19] Sanjiv R. Das 2002, The surprise element: jumps in interest rate. Journal of Econometrics Vol.106, Issue 1, 27-65.
[20] Pilipovic Dragana. 1998, Energy risk: Valuing and managing energy derivatives. Mc-GrawHill.
[21] E.J. Hannan; R.D. Terrell and N.E. Tuckwell 1970, The seasonal adjustment of economictime series. International Economic Review Vol. 11, No.1, 24-52.
[22] Eduardo S. Schwartz.1997, The stochastic behavior of commodity prices: Implications forvaluation and hedging. The Journal of Finance Vol. 52, Issue 3, 923-973.
[23] Eduardo S. Schwartz and James E. Smith 2000, Short-term variation and long-term dynamicsin commodity prices. Management Science Vol.46, No.7, 893-911.
[24] Robert F. Engle 1982, Autoregressive conditional heteroscedasticity with estimates of thevariance of United Kingdom inflation. Journal of the Econometric Society Vol. 50, No.4, 987-1007.
[25] Eugene F. Fama and Kenneth R. French 1988, Business cycles and the behavior of metalsprices. The Journal of Finance Volume 43, Issue 5, 1075-1093.
[26] Feike C. Drost and Bas J.M. Werker 1996, Closing the GARCH gap: Continuous time GARCHmodeling. Journal of Econometrics Vol 74, Issue 1, 31-57.
[27] Fischer Black and Myron Scholes.1973, The pricing of options and corporate liabilities. TheJournal of Political Economy Vol.81,No.3, 637-654.
42
[28] G.E. Uhlenbeck and L.S.Ornstein 1930, On the theory of the Brownian motion. Physical Re-view Vol. 36, Issue 5, 823-841.
[29] Helyette Geman. 2007, Mean-reversion versus random walk in oil and natural gas prices.Advances in Mathematical Finance 219-228.
[30] Helyette Geman and Vu-Nhat Nguyen.2005, Soybean inventory and forward curve dynam-ics. Management Science Vol.51, No.7, 1076-1091.
[31] Holbrook Working.1949, The theory of price of storage. The American Economic Review Vol.39, No.6, 1254-1262.
[32] Hossein Askari and Krichene Noureddine.2008, Oil prices dynamics (2002-2006). Energyeconomics Volume 30, Issue 5, 2134-2153.
[33] Hyun J Jin and Darren L.Frechette 2004, A new t-test for the R/S analysis and long memoryin agricultural commodity prices. Appllied Economics Letters Vol. 11, Issue 11, 661-667.
[34] Jean-Michel Zakoian.1994, Threshold heteroskedastic models. Journal of Economics Vol. 18,Issue 5, 931-955.
[35] Jeffrey A. Frankel, and Rose Andrew K.2010, Determinants of agricultural and mineral com-modity prices. HKS Faculty Research Working Paper Series RWP 10-038.
[36] Jeffrey C. Williams and Brian D. Wright 1949, Storage and Commodity Markets.
[37] Jimmy E. Hilliard, and Jorge A.Reis 1999, Jump processes in commodity futures prices andoptions pricing. American Journal of Agricultural Economics Vol. 81, Issue 2, 273-286.
[38] Jimmy E. Hilliard and Jorge A. Reis 1998, Valuation of commodity futures and options un-der stochastic convenience yields, interest rates and jump diffusion in the spot. Journal ofFinancial and Quantitiative Analysis Vol. 33, Issue 1, 61-86.
[39] Komkrit Ovararin and Meade Nigel.2010, Mean reversion and seasonality in GARCH ofagricultural commodities. International Conference on Applied Economics.
[40] S.G.Kou 2002, A Jump-Diffusion Model for Option pricing. Management Science Vol. 48, No.8, 1086-1101.
[41] Mario J. Miranda and Xiongwen Rui.1995, Commodity storage and Interseasonal Price Dy-namics. Proceedings of the NCR-134 Conference on Applied Commodity Price Analysis, Forecast-ing and Market risk management, Chicago, IL.
[42] Richter Martin and Carsten Sorensen .2002, Stochastic volatility and seasonality in com-modity futures and options: the case of soybeans. Working paper, University of CopenhagenBusiness School.
43
[43] John F. Muth 1961, Rational expectations and the theory of prices movements.EconomicaVol. 29, No.3, 315-335.
[44] Nicholas Kaldor.1939, Speculaiton and economic stability. The Review of Economic StudiesVol. 7, No.1, 1-27.
[45] Paul A. Samuelson.1965, Proof that properly anticipated prices fluctuate randomly. Indus-trial Management Review Vol.6, No.2, 41-50.
[46] Paul C. Westcott, and Linwood A.Hoffman 1999, Price determination for corn and wheat:the role of market factors and government programs. AgEcon SEARCH, Technical Bulletin.
[47] Rajna Gibson, and Eduardo S.Schwartz 1990, Stochastic convenience yield and the pricingof oil contingent claims. The Journal of Finance Vol. 45, Issue 3, 959-976.
[48] Robert C. Merton.1976, Option pricing when underlying stock returns are discontinuous.Journal of Financial Economics Vol. 3, Issue 1-2, 125-144.
[49] Robert S. Pindyck.1999, The long-run evolution of energy prices. The Energy Journal Vol.20,No.1999.
[50] Robert S. Pindyck and Daniel L.Rubinfeld 1991, Econometric models. McGrawHill.
[51] S Ghosh, CL Gilbert and AJH Hallett 1987, Stabilizing speculative commodity markets. Ox-ford: Clarendon.
[52] Sanjiv Ranjan Das, and Silverio Foresi.1996, Exact solutions for bond and option prices withsystematic jump risk. Review of derivatives research 1,7-24.
[53] Shijie Deng.2000, Stochastic Models of Energy Commodity prices and their applications:Mean-reversion with jumps and spikes. Univesity of California Energy Institute.
[54] Steen Koekebakker, and Gudbrand Lien .2004, Volatility and price jumps in AgriculturalFutures prices : Evidence from wheat options. American Journal of Agricultural Economics86:1018-1031.
[55] Subcommittee, US Senate Permanent Permanent. June 24 2009, "Excessive Speculation inthe Wheat Market." Committe on Homeland Security and Government Affaires, US SenatePermanent Subcommittee Investigations.
[56] Svetlana Borovkova and Ferry Jaya Permana.2006, Chapter:Modelling electricity prices bythe potential jump-diffusion. Stochastic Finance,pp239-263.
[57] Tim Bollerslev and Eric Ghysels.1996, Periodic autoregressive conditional heteroscedastic-ity. Journal of Business and Economic Statistic Vol. 14, Issue 2, 139-151.
[58] Ronald Trostle.2011, Why another food commodity price spike. USDA.
44
[59] Victor K. Ng and Stephen Craig Pirrong.1994, Fundamentals and volatility: Storage, Spreadand the Dynamcis of metals prices. Journal of Business Vol. 67, No.2, 203-230.
[60] Yanjun Tian and P. Fackler 2000, A seasonal stochastic volatility model for futures priceterm structure. Working paper, North California State University.
45
3Market risk evaluation of agricultural cooperatives
Abstract
Normal distribution model is widely used for Value-at-Risk (VaR) estimation in the agricultural coopera-tives industry. In the framework of a new financial environment and agricultural prices dynamics, tradi-tional estimation method has been challenged. This paper applies two VaR estimations with Normal distri-bution and Jump distribution under three portfolios to mimic the cooperatives’ activities. The comparisonsshow that Normal distribution VaR can estimate market risk quite well in most of the cases; Jump-VaR doesnot overestimate risk and has an efficient capital allocation as Normal-VaR in the portfolios. And there isone occasion where Normal-VaR is located in the Yellow zone according to Basel criterion, which inducesrisk underestimation of Normal-VaR in the case of relative high volatile period.In briefly, it is suggested totake into account jump risk/extreme risk in risk management for agricultural cooperatives.
Keywords: Agricultural cooperatives, Market risk exposure, Normal distribution VaR, Jump DistributionVaRJEL Classification: C15, G18, Q12, Q14
3.1 Introduction
The agricultural industry is highly dependent on the climate, which may lead to un-forecastableprice variation. Nowadays, cooperatives often use financial markets to hedge this kind of risk.At the same time, by looking at the price variation, jumps and time-varying volatilities are foundto be significant in agricultural commodities price series. The failure of forecasting risk mightproduce systemic risk given the role of agricultural cooperatives in the market. Based on this,a good indicator showing and measuring market risk is needed. In the framework of agricul-tural cooperatives, this paper discusses Value-at-Risk (VaR), a measure of the potential loss givena probability within a period (one to ten days); VaR is actually a quantile of portfolio returns,which is a risk measurement widely used by financial institutions and the agricultural industry.
Cooperatives play an important intermediary role in the agricultural industry or agribusiness. AsDeclerck and Mauget (2008) indicates, “The job of agricultural cooperatives is to collect agricul-tural commodities from farmers and commercialize these commodities in the market.” The aimof cooperatives is to ensure that their members sell commodities at a reasonable price level. Thisis in the context of economic solidarity. The same case exists for French agricultural cooperatives.French cooperatives usually pay an "average price” to the members. This payment contract canbe explained as follows:
46
1. Cooperatives pay the members for an advance price when the contract is established be-tween them.
2. At the end of a harvest, two cases exist: cooperatives may pay for a complement price (dif-ference between average price and advance price) when the average price is higher than theadvance price; cooperatives do not need to pay for a complement price if average price islower than the advance price.
After collecting commodities from members, cooperatives commercialise these products to sellto their clients; most of them are industrial producers. In recent years, agricultural cooperativesface a multitude of difficulties, which are increasing costs, low commodities’ prices and higherprice volatility (Gray & Kraenzle, 2002). Price volatility challenges the risk management amongcooperatives. Higher volatility and improper hedging strategy may lead to cooperatives filing forbankruptcy.
Under the new environment with different dynamic nature of agricultural prices and volatilefinancial derivatives markets, cooperatives managers should revise risk management and mea-surement tools for confronting diverse market risks. On the one hand, commodity derivativecontracts are common hedging tools among the agricultural cooperatives providing diversifica-tion benefits. However, a price change in derivative contracts provoke devaluation of hedgingportfolio, which probably creates disaster for cooperatives. Rigorous risk management shouldtake this into account (Roussillon-Montfort, 2008). On the other hand, unlike the VaR that ap-plies in an equity or bond portfolio, VaR in commodity portfolio should capture the characteris-tics of commodities. Skewness and fat-Kurtosis phenomenon, which exist in recent agriculturalcommodities price returns, are not included in the normal distribution model. Finally, marketrisk is highlighted by financial regulators, for example, market risk framework of Basel III Capi-tal. In regulatory practices, VaR is used to define Minimum Capital Risk Requirement, which isrequested for the banks to set aside unforeseen loss. A proposal of the Basel committee defines aVaR measure using 99% confidence interval over 10 trading days, and the resulting VaR is multi-plied by a factor to arrive at the minimum capital requirement for regulatory purposes.
VaR is also suggested for firm’s risk management and as well in agricultural economics, whereVaR is recommended in the reporting of market risk. Given the immediacy role of cooperativesin the agricultural industry, bankruptcies of cooperatives are costly and will create systemic risk,which threatens the order of the whole market and the benefit of farmers. Requirements for therisk evaluation tools are discussed in Basel III perspective and the European Union’s Capital Ad-equacy Directive (CAD III) for financial institutions especially in order to control the tail risk.Considering the important role of cooperatives in the agricultural commodity industry, the issueof extreme risk should be also considered in agricultural cooperatives.
In this research, we focus on the calculation of VaR. An appropriate VaR applied in agriculturalcooperatives should deal well with the trade-off between sufficient risk prevision and not overus-
47
ing resources. We are going to compare two VaRs under different price processes: Black-Scholesmodel under normally distributed prices and Double-Exponential-Jump model. The comparisonis taken by using the VaR in different portfolios. Portfolios in this paper try to mime coopera-tives’ activities, including contract between farmers and cooperatives, and contract of hedging.The only product on which we concentrate is wheat. It is no doubt that cooperatives disposemultiplied kinds of agricultural products and it is more realistic to combine all these products.However, the interaction of different agricultural commodities makes the simulation more com-plicated and is out of the scope of this research.
In the three portfolios cases of this paper, one model does not always outperform the other one.Results are dependent on the periods and portfolios. However, from another perspective we canconclude that in terms of frequency and size of exception, Jump-VaR does not overestimate riskand has an efficient a capital allocation as Normal-VaR in most of the cases. On one occasion,Normal-VaR is located in the Yellow zone, according to Basel criterion, which induces risk under-estimation of Normal-VaR in the case of a period of relatively high volatility.
The contribution of this research is that: firstly, it contributes to the literature on risk evaluationin the agricultural business. VaR is examined often in the financial market assets, but little inthe physical market and among agricultural cooperatives practices, one of these researches is byManfredo & Leuthold (1998). Secondly, VaR estimation is based on the portfolio that is close tothe cooperatives’ operations and is practical for professionals. Former studies focus on the accu-rate estimation of correlation and volatility of agricultural portfolio, and this paper consists of theeffectiveness of two VaRs in simulated cooperatives’ activities. Finally, the results give anotherinsight for the regulators on the regulation in commodities physical markets in terms of marketrisk measurement and risk management.
The paper is organised as follows: section 2 summarises the two models of spot price modellingand presents VaR methods and backtesting; section 3 gives results of three kinds of portfolios andthe conclusion is presented in section 4.
3.2 Modelling price returns and VaR
Black-Scholes model and Jump model
Black-Scholes Model
The Black-Scholes model in this article is the Value-at-Risk under Black-Scholes framework wherethe commodities price returns St have a lognormal distribution with constant drift and volatility(Black & Scholes, 1973), where they apply for pricing options. More precisely, price St is sup-posed to be:
48
St = St−1exp([µ− 0.5σ2
] 1T
+ σ1TZt
)(3.1)
with Zt(0,1), standard normal distribution. There are two parameters in this model: µ denotesthe mean log-returns and σ the mean return volatility.
Criticism on this model is that more and more research has found that commodities price returnsare not normally distributed but have excess skewness and fat-tailor. The application under nor-mal distribution hypothesis may produce bias in the results. The objective of this paper is notto estimate the model, we give the estimated parameters directly in Table 3.7. In the following,we denote Normal-VaR as VaR simulated under Black-Scholes framework. Due to the evidence ofno-normal distribution of price returns, we have estimated a Double-Exponential-Jump Model,which includes all the characteristics of recent price evolution.
Double-Exponential Jump Model
Some preliminary statistical results have proved the argument that wheat prices are not normallydistributed. As we are particularly interested in jumps of the wheat prices, we applied a Double-Exponential-Jump model to the physical wheat prices (Kou, 2002), (Ramezani & Zeng, 1998). Themodel supposes that price returns have a double-exponential jump distribution with stochasticvolatility. For simplifying, we call this model ‘jump model’ in the rest of this paper.
This jump model assumes that:
dXt = κ (µx −Xt)dt + σtdBt + JtdNt (3.2)
dVt = κ (µv −Vt)dt + σvVtdBvt (3.3)
With, dBt ,dBvt Brownian process for price returns and volatility. dNt , Poisson counter with pa-rameter intensity is the average number of jumps per unit of time.
And the jump size has
fJt =
η1e−η1Jt P robability = p (3.4)
η2eη2Jt P robability = 1− p (3.5)
Different from the Normal model, the Jump model takes into account the fat-tailed and time-varying volatility in wheat prices, as in the previous paper, a special event may introduce jumpin the price series.
Both models are applied to the wheat prices for wheat delivered to Rouen for the harvests of2010/2012, 2012/2013 and 2013/2014. The parameters are estimated using maximisation oflikelihood function. The results are given in Table 3.8. The VaR estimated under this jump model
49
is noted as Jump-VaR.
Value-at-Risk(VaR)
VaR is primarily designed in the financial institutions. VaR measures the potential loss duringone period and is widely used as an indicator of market risk of financial portfolios. Marshall andSiegel (1996) state that “VaR has gained strong support from industry and regulatory bodies suchas G20, BIS and EU.” VaR is recommended in reporting of market risk associated with derivativescontracts. In agribusiness, Boehlje and Lins (1998) and Manfredo and Leuthold (2001) impliedVaR to quantify market risk for farms. Nevertheless, the uncertainty of VaR estimates may pro-duce a new risk, as VaR estimates might not be appropriate.
In general, there are two types of estimation method by which to obtain this indicator, non-parametric method and parametric method. Each of them has disadvantages when comparedwith the other.
The historical simulation method, as one of the most used non-parametric methods, assumes thathistory repeats itself. This method simulates asset returns based on historical prices and reads the1% worst loss. Historical simulation does not have any assumptions about return distributions. Itis easy to calculate and may capture the characteristics in the dataset. However, this method canbe inefficient for some complex assets and be biased in the case of higher volatility. For example,historical simulation VaR might not take into account the features of commodities, which prob-ably have seasonality or business cycles. Also, historical simulation VaR might be biased whenthere is an extreme event.
The parametric method: variance-covariance method or delta-normal method is based on thenormally distributed returns assumption. This method uses RiskMetrics methodology originallydeveloped by JP Morgan. The method supposes that portfolio return is a linear combination ofnormal variables, which may conduct problems in the case of nonlinear effect of options. Vari-ance and covariance of each asset in the portfolio are needed. It will be difficult in the caseof large positions in the portfolio. This method is widely used in cooperatives, where there isa variance-covariance matrix of different products, wheat, corn etc. As with historical simula-tion, the variance-covariance method may underestimate extreme outcomes. Another parametricmethod widely used in practice is Monte Carlo simulation, which will be applied in this paper.This method is flexible, based on different return distribution assumptions. It is easily applied inindustry and assumes that the underlying returns are normally distributed and simulated whenassets are some derivatives contracts that have a nonlinear and path-dependent pay-off function.Aussenegg and Miazhynskaia (2006) show that models’ accuracy depends on the type of distribu-tion and probability. The changes of risk report under different assumptions and methodologiesmay be dramatic. Risk estimation can be biased under these kinds of models, if market returns
50
do not follow respective distribution.
VaR tries to explain tail risk of a given distribution, and it will be distorted if the distribution ofdata is leptokurtosis under normal distributed assumption. This research is not the first try totake into account the jump risk or tail risk in the calculation of VaR. Duffie and Pan (1999) givean analytical VaR by considering jump risk in the derivatives portfolio. Tail risk can be explainedby an extreme risk. More rigorously, VaR based on extreme value theory is analysed by Brookset al. (2005) and Danielsson and Morimoto (2000). Moreover, other time-series models, such asARMA, GARCH, EGARCH, also take into account time-varying volatility and excess kurtosis infinancial assets returns. By fitting these time-series models, VaRs can also give more complete re-sults than VaR under normal distribution. In this research, for the purpose of applying the jumpmodel, we use only the jump model to compare the normal distributed model in simulating VaR,as we consider different wheat produced in different harvests are not the same products in termsof quality. Analyses are conducted on every year separately, with only about 200 data prices peryear.
VaR Backtesting
Qualitative or quantitative backtesting techniques are needed for testing the performance ofNormal-VaR and Jump-VaR. One of the backtesting techniques is the Basel approach, where theyallocate the banks to three zones based on the number of exceptions observed over 250 tradingdays. The exception is defined when Profit/Loss surpasses the estimated VaR, which means thatVaR fails to estimate the risk at this time. The VaR measure would be distinguished into oneof the three colored zones. VaR measures falling in the green zone have no particular concerns.Those in the yellow zone require monitoring with increased multiplier in calculating market riskscapital charges. The detail of the Basel criteria is given in Table 3.9.
The other backtesting technique is based also on the number of exceptions and is more rigorous.Kupiec’s test is to “determine whether the observed frequency of exceptions is consistent withthe frequency of expected exceptions according to the VaR model and chosen confidence”, and itis under the hypothesis that the number of exceptions has a Bernoulli distribution. The LR teststatistic is:
LR = −2ln(
(1− p̂)n−x p̂x
(1− p)n−x px
)v X2 (1) , (3.6)
With p̂ the VaR stated probability level; p the exception frequency; xn , n the total trading days andx the number of exceptions.The null hypothesis under this test is that the realised number of exceptions is equal to the theo-retical number of exceptions defined in the VaR model and model is efficient.
51
We conduct backtesting on the frequency of exceptions and also on the size of expected loss. Thethird backtesting is based on the magnitude of estimated VaR. A larger difference between VaRand real Profit/Loss in the case of exception will be a disaster for cooperatives. For this purpose,we calculate a Root-Mean-Square Deviation (RMSD), which is the difference between Expectedshortfall and the P/L in the case of exceptions.And,
RMSE =
√∑T
i=1 (Expected short f alli − P /Li)2
Tif exception accurs (3.7)
0 if no exception (3.8)
with T the total trading days.
RMSD is inspired from Angelidis and Degiannakis (2006), where they propose a two-stage VaRevaluation. They point out that, given an identical number of exceptions, one model underesti-mates the risks in the case of exception, which is costly as well.
The last backtesting VaR, Performance Criterion (VPC), used also in Fuss et al. (2010), and Baoand Saltoglu (2006), considers “the trade-off between efficient capital allocation and sufficient re-serves”. This measure includes the distance between returns and VaR, correlation between returnvolatility and VaR, and penalties when there is exception. VPC is calculated as follow:
VPC = α11n
X∑x=1
(Real returnx −V aRx)2 +α21n
√∑Kk=1 |V aRk −Real returnk |I (Real returnk < 0)
+α3cor(Real return2,V aR
)+α4|p̂ − p|
(3.9)
n is the number of observations; X the number of exceptions; K the numbers of times when thereis no exception and real return is negative. The first term and the last term are concerned with thepenalties of exceptions, which incorporate the size and frequency of exception. The second termmeasures the cost of reserves. It is also undesirable if VaR overestimates market risks. The thirdterm is the correlation between squared return and VaR, since a good VaR increases in volatileperiods for estimating the risk and recedes during tranquil time for reserving resources. p̂, p aredefined as in the Topiec’s test. α1, α2, α3, α4 are the weight of each component and sum to unity.They can be adjusted according to the risk managers’ preference. In our case, we use an equalweight, with αi = 0.25.
52
In general, a good VaR estimation should explain accurate risk level under both volatile marketconditions and a quiet market without unnecessary reserve costs. In the following section, we willestimate VaR in different scenarios and compare the results with three backtesting techniques.
3.3 Comparison under different scenarios
French wheat market conditions from 2011 to 2014
Before looking at the risk measure results, the wheat market conditions are briefly presentedfrom the price series and estimation results of pricing models. We consider that prices on dif-ferent harvests are prices of different products. Figure 3.1 shows wheat prices in three harvestsseparately. During the harvest 2012/2013, wheat prices have a much higher level than other twoharvest years. Cycles in the graphs show several periods where wheat prices display a decreasinginterval. These are also the periods where we observe most exceptions of VaR in the followingempirical results.
Parameters of the two pricing models for theses three harvests are given in Tables 3.7 and 3.8. Re-sults from the two pricing models, Black-Scholes model and Jump model, show that price returnvolatility is higher in the period 2012/2013 with a rate of 21%. However, according to the Jumpmodel, in the year of 2013/2014, jumps happen more frequently than in the other two periods.Harvest 2011/2012 may be considered as a quiet period, and the 2012/2013 and 2013/2014 cam-paigns are volatile in certain period intervals.
Empirical results
As it is mentioned before, cooperatives sign a contract with farmers to guarantee a selling price.Cooperatives buy products from farmers and pay for an “advance price” to members. At the endof the contract, if average price is larger than advance price (average price ≥ advance price), thecooperative should pay farmers a premium, which is the difference between average price andadvance price. This contract is equivalent to short an Asian call. Moreover, we assume that coop-eratives buy one unit of wheat for simplification.
According to two different price distribution frameworks, a Monte Carlo simulation is used toestimate the VaR of the portfolio. A comparison is taken between VaR simulated by Black-Scholes model and VaR simulated by Jump model. VaR estimation is conducted on three harvests,2011/2012, 2012/2013 and 2013/2014. The Monte-Carlo simulation process for VaRs (for bothBlack-Scholes and Jump model) is summarised:
1. Simulate 1000 times the underlying spot prices of both models and volatility in the jumpmodel using Euler approximation at t and iterate until time T (one year, the end of the
53
harvest in our paper). The time interval ∆t = 1/T . For t, generate random increment, ∆Zt ,Brownian motion for the Black-Scholes model; ∆Bt ∆Bvt , Brownian motion for spot pricesin the Jump model and for volatility equation; ∆Nt Poisson process for jump;
2. Approximate the spot prices according to the equations and the random variables generatedin step 2.
3. Simulate the Profit/Loss of the portfolio, which includes Asian Options. Daily VaR of the10-day horizon will be the 0.01-quantile of distribution of 1000 portfolio change values.
Portfolio 1: Market risk in spot pricesCooperatives only have wheat and hold it until the end of the harvest
We begin on the simplest case where cooperatives hold only wheat. They do not sign any contractswith farmers or sell products to clients. The objective of this portfolio is to study the market riskcreated purely by price fluctuations. Since we estimate VaR with a 10-days horizon profit andloss (noted P/L) of portfolio 1 is actually price difference of 10 days’ discard.
Portfolio 1:Portfolio = Long WheatProfit loss (P/L):t: P /L = (St+10 − St)t+1: P /L = (St+11 − St+1)...
Figure 3.2 gives the daily 10-day VaR for 99% confidence level and P/L of the portfolio with realprices. VaRs from the two models are a little different in the year of 2012/2013, where price seriesexhibit higher volatility than in the other two harvests. And for the other two years, the curves ofNormal-VaR and Jump-VaR are very close to each other. Combined with Table 3.1, which showsthe number of exceptions in the 2011/2012 harvests, there are 3 times for Black-Scholes and oncefor the Jump model where VaR does not approximate market risk. And in the last two harvests,Normal-VaR and Jump-VaR have the same exception times. If we look precisely at the violationdays, both models have violation on the date 07/11/2011, when wheat prices decreased brutallyby 14 Euro/ton within 10 days. Normal-VaR does not estimate the risk, while Jump-VaR estimatesthe risk on 14/09/2011 with a 21 Euro/ton drop, and this is also one of the highest price decreasesin the whole year. For the second harvest, 2012/2013, averaged wheat price is much higher thanin the harvest 2011/2012. The violations gather at the end of the second harvest, 2012/2013. Forexample, the two models have an exception on 28/05/2013, where wheat prices decrease by 27Euros/ton within 10 days. After that, wheat prices have decreased but still maintained volatilein 2013/2014. The violations of both VaR models take place in the period from 01/05/2014 to12/05/2014. Normal-VaR does not approximate market risk in the period of December 2013 verywell, while the cut down of wheat prices is estimated by Jump-VaR. The other indicator that we
54
calculate in Table 3.2 is 97.5% expected-shortfall, which defines the expected loss given 97.5%level of confidence1. Overall, Normal-VaR underestimates risk compared with Jump-VaR, sincethere are more times that the real loss exceeds 97.5% expected-shortfall in the case of Normal-VaR than in Jump-VaR.
For comparing the performance of the two VaR models, we use three backtesting techniques, asnoted above. Under the Basel framework, both VaRs are in the secure zone for all the periods.Additionally, Jump-VaR and Normal-VaR have the same results in Kupiec’s test for the last twoyears. According to Kupiec’s test, we accept both VaR models. The third backtesting, rootedsquare mean errors (RSME), is the average difference between Profit/Loss and Expected short-fall (97.5%) in the case of exceptions. Normal-VaR underestimates risk in the first two yearswith higher difference between expected shortfall and real loss. The larger distance between realloss and expected loss when VaR does not approximate market risk well will create disaster foragricultural cooperatives, as the loss is far more than the reserved capital. In the last harvest,2013/2014, Jump-VaR has RMSE slightly higher than Normal-VaR but the difference is not toolarge. And the final test on the combined performance shows that Jump-VaR is more efficientthan Normal-VaR in the first two harvests, but overestimates in the recent year 2013/2014.Overall, in the case of portfolio 1 with linear payoff, by combining all the backtesting results,although both models are in the secure zone of Basel and are accepted by Kupiec’s test, Normal-VaR underestimates market risk and is less inefficient in the first two harvests. And Jump-VaRunderestimates risk in the harvest 2013/2014.
Portfolio 2: Market risk without hedgingCooperatives buy wheat and short Asian call with their clients without hedging risk.Wheat is sold in the same proportion every day.
In this case, we assume that cooperatives buy wheat from farmers and sign a contract for guaran-teeing selling prices that are not smaller than the average prices of the year. Signing the contractwith farmers can be considered as short Asian Call. Moreover, cooperatives are assumed not tohedge risk; thus, they face the risk when price decreases. Meanwhile, we suppose that coopera-tives sell the same proportion of wheat every day. The contract has value 0 in the beginning. Forsimplifying, cooperatives are assumed to hold one unit of wheat. By including Asian call in theportfolio, the payoff of portfolio 2 becomes nonlinear. Under this case, profit and loss becomes:
Portfolio 2:Portfolio=Long wheat S + short Asian call YProfit and Loss (P/L):t : P /L = (St+10 − St) + (0−Yt+10)t+1: P /L = (St+11 − St+1) (1− 1/T )− (Yt+11 −Yt+1)
1Expected shortfall, ESα = E[X |X ≤ V aRα], with X, the loss of portfolio
55
...
The specification of Asian call Y (contract signed with farmers): Maturity T for one year/one har-vest; K strike price is supposed to be 190 Euros (the same value for the following cases), which isthe advanced prices to pay for farmers. Using this contract, cooperatives buy wheat from farmerswith the price at 190 Euros. At the end of year, if the average price during this year is higherthan 190 Euros, cooperatives should pay the farmers the difference between average price and190 Euros.
We have the results for Black-Scholes and Jump models, as shown in Figure 3.3. In the initialdays, the Normal-VaR is quite different from that computed by Jump-VaR. Profit/Loss of portfo-lio is volatile in the beginning due to the uncertain prices. When time is close to maturity, priceuncertainty decreases and wheat is sold out, value is near zero. By looking at Table 3.3 as well,Jump-VaR has less exception times than Normal-VaR except for the year 2012/2013, where Jump-VaR has one more violation. More specifically, in the 2011/2012 harvest, the same violation as thefirst portfolio has happened on the date of 09/09/2011 when wheat price has decreased by 21 Eu-ros/ton within 10 days. In the second harvest, one exception of Jump-VaR is on the 06/02/2012;however on this date, price does not exhibit as large a decrease as on other days. This leads to aquestion of whether the Jump model could not always estimate the occurrence of jumps. For theharvest 2012/2013, the exception period consists also of the period from 22/05/2013 to the endharvest, where wheat prices display a dropdown trend, as noted in Figure 3.1. The second indi-cator in Table 3.3 shows that expected loss measured by expected shortfall is higher in Jump-VaRthan in Normal-VaR, except in the 2013/2014 harvest.
The backtesting of the three techniques is concluded in Table 3.4. Both Normal-VaR and Jump-VaR are in the secure zone according to the Basel criterion. This is the same for the Kupiec’s test;both results have accepted the hypothesis that the VaR models are accurate. However, we findthe difference from the Root-Mean-Square error (RMSE). For the first harvest and in the case ofexception, Normal-VaR has underestimated much more risks than Jump-VaR, as the differencebetween real loss and expected shortfall is much higher. The conclusion of RMSE is inverse inthe 2013/2014 harvest. And, concerning the VPC test, the results of both VaRs are quite close interms of efficiency.
In portfolio 2, farmers begin to engage in a guaranteed contract with agricultural cooperatives.Without hedging risk, cooperatives confront more risks if wheat prices decrease. Estimation re-sults and backtesting results are quite close in both VaRs. In the first two harvests, Jump-VaRoutperforms Normal-VaR with less violations and more efficient resource allocation. Jump-VaRwill be more desirable for cooperatives in the first two harvests. In the 2013/2014 harvest, Jump-VaR overestimates risk. By using Jump-VaR, more capital is required, which is costly from thepoint of view of cooperative managers.
56
Portfolio 3: Market risk with hedgingCooperatives buy wheat and short call Asian call with their clients, and for hedging they longAsian Put. Same quantity of wheat (1/T) is sold every day until the end of maturity.
In this case, we assume that cooperatives long an Asian Put for hedging risk. The Asian call andput have the same maturity and strike price. Cooperatives also sell the same amount of wheatevery day. The portfolio can be presented as:
Portfolio 3:Portfolio=Long wheat (S) + short Asian Call (Y)+ long Asian Put (X)P/Lt : P /L = (St+10 − St) + (0−Yt+10)t+1: P /L = (1− 1/T ) (St+11 − St+1)− (Yt+11 −Yt+1) + (Xt+11 −Xt)...
The specification of Asian call Y (contract signed with farmers) is the same as in portfolio 2. AsianPut for hedging is specialised as: Maturity T for one year/one harvest; K strike price is supposedto be 190 Euros (same strike price as Asian Call Y). By long Asian Put, cooperatives do not hedgerisk perfectly. The holding wheat is decreasing as time passes, and we assume that cooperativessell 1/T portion of wheat every day. The P/L in portfolio 3 is slightly lower than that of portfolio 2.
Figure 3.4 gives the results of VaR of the two models by comparing with real data. Since we sup-pose that cooperatives sell the same portion of wheat every day, the Profit/Loss with real datadisplays the similar form as portfolio 2, except with smaller portfolio value due to the hedging.As in portfolio 2, the two models have larger difference in the beginning but converge in theend of harvest. Days of exceptions are observed in the similar dates as in the two other portfo-lios, for example, on 07/11/2011, when wheat prices decrease to 14 Euros/ton, the period from22/05/2013 to the end harvest, and from 30/04/2014 to 07/05/2014, when wheat prices reducecontinually. From Table 3.5, it can be seen that Jump-VaR has lower frequency of violation in allof the three periods. And, on average, Jump-VaR is larger than Normal-VaR and frequency of realloss exceeding expected shortfall is higher in the Normal model.
Concerning the backtesting of portfolio 3 and from Table 3.6, Normal-VaR has exceeded the realP/L in the harvest 2013/2014 5 times, which indicates that Normal-VaR is located in the Yellowzone according to Basel criterion. Normal-VaR and Jump-VaR are accepted to be correct accord-ing to Kupiec’s test. Performance of VaR is similar for the first two harvests. In the 2013/2014harvest, Normal-VaR has underestimated risk with expected shortfall lower than real loss in thecase of exceptions. Jump-VaR is also more efficient during this harvest on the basis of VPC.
In portfolio 3, cooperatives start to hedge risk with an Asian Put. Risk hedging is not perfectbut portfolio return volatility decreases. Normal-VaR and Jump-VaR work well in the first twoharvests; however, in the last year, Normal-VaR underestimates risk. A question will be raised
57
regarding the accuracy of the Normal-VaR model, but also regarding the accuracy of the Black-Scholes model’s estimated parameters. Jump-VaR in this case may be a complete market riskmeasure for cooperatives managers.
3.4 Conclusion
This paper simulated Value-at-Risk using Monte-Carlo methodology under different price returndistribution frameworks. We compared two price models, Black-Scholes model under normaldistribution and Double-Exponential-Jump model under no normal distribution. The compari-son is taken from the framework of simplified portfolios in cooperatives. As cooperatives have anintermediary role in the agricultural industry, accurate risk estimation is crucial.
The conclusion of the question of whether Value-at-Risk under the Jump model is superior to VaRunder Black-Scholes model depends on the studied periods. Firstly, as expected, the frequencythat real loss exceeds VaR is usually higher in Normal-VaR than in Jump-VaR. However, the dif-ference is small in some cases. Normal-VaR cannot be rejected, since both VaRs, Normal-VaR andJump-VaR, are in the security zone according to the Basel criterion in most of the cases. Sec-ondly, Jump-VaR does not have a cost reserve problem. Performance test of the two models forefficiency does not display a big cap. Thirdly, it should be noted that there is one occasion whenNormal-VaR is located in the Yellow zone of the Basel criterion. Normal-VaR might bring aboutunderestimation risk problems in the case of a highly volatile market or other extreme marketcondition. Finally, although the jump model takes into account jump risk in price modelling,this model cannot repeat the same jumps as real prices. That is why in Jump-VaR, exceptionsor underestimation can still happen when there is actually jump in the data. Overall, given thechanging market environment and uncertain price distributions, Jump-VaR can be complemen-tary to this kind of market condition, although jumps cannot forecast well and resources reservemay be costly in the Jump-VaR.
This paper gives three simple portfolios for imitating cooperatives’ activities. Wheat portfolios inthis research include the contract between cooperatives and farmers, and hedging contract. Theresults suggest that Jump-VaR may be a useful risk measure tool to complete in the reporting ofmarket risks for agricultural cooperatives. And this research uses only wheat in the portfolio.The Profit/Loss of cooperatives depends also on other commodities (unit-type commodity in thispaper). Future work may include other products, which will increase the calculating complicityand lead to computational burden with much more market risk factors.
58
Figure 3.1: Three harvests of French wheat prices from 2011 to 2014
(a) 2011/2012
(b) 2012/2013
(c) 2013/2014
Note: The graphs give the daily French wheat prices from 2011 to 2014. Three wheat harvests arepresented separately. Cycles give several main prices downward period. These periods coincidewith the time when VaR violations happen.
59
Figure 3.2: Portfolio 1 - 10-Day VaR 99%
(a) 2011/2012
(b) 2012/2013
(c) 2013/201460
Table 3.1: Portfolio 1 - Number of exceptions
VaR vs Realised P/L Expected shortfall vs Realised P/LNumber ofexceptions
Frequency ofexceptions
Number ofexceptions
Frequency ofexceptions
2011/2012VaR(Black-Scholes) 3 0.017 2 0.011
VaR(Jump) 1 0.006 2 0.011
2012/2013VaR(Black-Scholes) 4 0.023 4 0.011
VaR(Jump) 4 0.023 5 0.028
2013/2014VaR(Black-Scholes) 4 0.025 3 0.017
VaR(Jump) 4 0.025 4 0.025
Table 3.2: Portfolio 1 - Backtesting
Basel approach Kupiec’s test RMSE P/Land expectedshortfall
VPC(VaRperformancecriterion)
2011/2012VaR(Black-Scholes) Green zone 20.775 0.624 0.373
VaR(Jump) Green zone 30.143 0.086 0.330
2012/2013VaR(Black-Scholes) Green zone 17.306 1.605 0.796
VaR(Jump) Green zone 17.306 0.902 0.71
2013/2014VaR(Black-Scholes) Green zone 17.251 0.631 0.254
VaR(Jump) Green zone 17.251 0.801 0.583
61
Figure 3.3: Portfolio 2 - 10-Day VaR 99%
(a) 2011/2012
(b) 2012/2013
(c) 2013/201462
Table 3.3: Portfolio 2 - Number of exceptions
VaR vs Realised P/L Expected shortfall vs Realised P/LNumber ofexceptions
Frequency ofexceptions
Number ofexceptions
Frequency ofexceptions
2011/2012VaR(Black-Scholes) 3 0.017 3 0.017
VaR(Jump) 2 0.011 1 0.006
2012/2013VaR(Black-Scholes) 3 0.017 3 0.017
VaR(Jump) 4 0.023 4 0.023
2013/2014VaR(Black-Scholes) 4 0.025 3 0.019
VaR(Jump) 3 0.019 4 0.025
Table 3.4: Portfolio 2 - Backtesting
Basel approach Kupiec’s test RMSE P/Land expectedshortfall
VPC(VaRperformancecriterion)
2011/2012VaR(Black-Scholes) Green zone 20.775 0.541 0.188
VaR(Jump) Green zone 24.931 0.020 0.142
2012/2013VaR(Black-Scholes) Green zone 20.772 0.063 0.321
VaR(Jump) Green zone 17.306 0.089 0.332
2013/2014VaR(Black-Scholes) Green zone 17.245 0.414 0.023
VaR(Jump) Green zone 20.726 1.783 0.023
63
Figure 3.4: Portfolio 3 - 10-Day VaR 99%
(a) 2011/2012
(b) 2012/2013
(c) 2013/201464
Table 3.5: Portfolio 3 - Number of exceptions
VaR vs Realised P/L Expected shortfall vs Realised P/LNumber ofexceptions
Frequency ofexceptions
Number ofexceptions
Frequency ofexceptions
2011/2012VaR(Black-Scholes) 3 0.017 4 0.022
VaR(Jump) 1 0.006 1 0.006
2012/2013VaR(Black-Scholes) 3 0.017 3 0.017
VaR(Jump) 2 0.011 3 0.017
2013/2014VaR(Black-Scholes) 5 0.031 4 0.022
VaR(Jump) 4 0.025 4 0.022
Table 3.6: Portfolio 3 - Backtesting
Basel approach Kupiec’s test RMSE P/Land expectedshortfall
VPC(VaRperformancecriterion)
2011/2012VaR(Black-Scholes) Green zone 24.993 0.123 0.131
VaR(Jump) Green zone 30.143 0.153 0.157
2012/2013VaR(Black-Scholes) Green zone 20.772 0.058 0.345
VaR(Jump) Green zone 24.929 0.049 0.301
2013/2014VaR(Black-Scholes) Yellow zone 14.288 0.211 0.025
VaR(Jump) Green zone 17.251 0.168 0.010
65
Table 3.7: Parameters of Black-Scholes model
2011/2012 2012/2013 2013/2014
µ 0.07 0.01 0.017
σ 0.29 0.21 0.14
Note: Parameters are estimated from spot prices of wheat delivered to Rouen for three harvests,2011/2012, 2012/2013 and 2013/2014. Parameters are estimated by maximum likelihood
function. All the parameters are significant at at least 10%.
Table 3.8: Parameters of Jump model
2011/2012 2012/2013 2013/2014
˛ 0.09 0.02 0.07
µx 5.29 5.45 5.21
p 0.58 0.4 0.59
η1 27.95 15 40.0
η2 30.93 15.0 29.90
λ 0.06 0.06 0.03
ω 3.19E-06 1.50E-04 9.47E-05
α 0.004 0.15 0.1
β 0.97 0.04 0.11
Note: The specification of volatility is a continuous case of GARCH (Brockwell, Chadraa andLindner 2006). For convenient estimating, we can re-write this as Vt =ω+αε2
t−1+t−1 with εt−1
the error term at time t − 1;ω ≥ 0andα ≥ 0,β ≥ 0; and α + β < 1 (stationary condition). Bytransforming to the continuous form of GARCH, we have kv = 1−α − β and σv = α
√2 (Nelson
1990). Parameters are estimated from spot prices of wheat delivered to Rouen during the thethree harvests.
Table 3.9: Basel Committee’s criterion
Number of exceptions (in 250 days) Multiplication factorGreen zone 4 or less 3
Yellow zone
5 3.46 3.57 3.658 3.759 3.85
Red zone 10 or more 4
Note: The criterion are applied to the VaR with a horizon of 10 trading days: 99% confidenceinterval: observation based on at least one year of historical data.
66
Bibliography
[1] Y.Bao and B.Saltoglu 2006, Evaluating predictive performance of Value-at-Risk models inemerging markets: A reality check. Journal of forecasting Volume.25, pages 101-128.
[2] F. Black and M.Scholes 1973, The pricing of Options and corporate liabilities. The Journal ofPolitical Economy Volume. 81, No. 3, pages 637-654.
[3] M. Boehlje and D.Lins 1998, Risks and risk management in an industrialized agriculture.Agricultural Finance Review Volume 58, pages 1-16.
[4] Peter Brockwell; Erdenebaatar Chadraa and Alexander Lindner.2006, Continuous-timeGARCH processes. The Annals of Applied Probability Volume. 16, No.2 , pages 790-826.
[5] C. Brooks; A. D. Clare; J.W. Dalle Molle and G. Persand 2005, A comparison of extremevalue theory approaches for determining Value-at-Risk. Journal of Empirical Finance Volume12, pages 339-352.
[6] F.Declerck and R.Mauget 2008, Prix moyen de campagne en cooperatives et couverture derisques de prix sur marches a terme. Colloque SFER: "Les entreprises cooperatives agricoles,mutations et perspectives" Paris.
[7] D.Duffie and J.Pan 1999, Transform analysis and asset pricing for affine Jump-diffusions.NBER Working paper series.
[8] R.Fuss;Z.Adams;D.Kaiser 2006, The predictive power of Value-at-Risk models in commodityfutures markets. Working paper.
[9] T.W.Gray and C.A.Kraenzle 2002, Problems and issues facing farmer cooperatives. RuralBusiness/Cooperative Service.
[10] Kou.2002, A Jump-Diffusion Model for Option pricing. Management Science Volume. 48, No.8, pages 1086-1101.
[11] M. R.Manfredo; R.Leuthold 2001, Market risk and the cattle feeding margin: An applicationof Value-at-Risk. Agribusiness Volume 17, Issue 3, pages 333-353.
[12] C.Marshall;M.Siegel 1996, Value-at-Risk: Implementing a Risk measurement standard.Available at SSRN: http://ssrn.com/abstract=1212.
67
[13] Nelson.1990, ARCH models as diffusion approximations. Journal of Econometrics Volume.45, Issues 1-2 , pages 7-38.
[14] C.A. Ramezani;Y. Zeng.1998, Maximum likelihood estimation of asymmetric jump-diffusionprocesses: Application to security prices. Working paper, University of Wisconsin.
[15] M.A.Roussillon-Montfort 2008, Les marchés à terme agricoles en Europe et en France. Noteset études économiques No.30 .
68
Part III
Financialization of commodity markets
69
4Does the impact of index flows on commodities prices
involve stockpiling as a signal? An empirical
investigation
Abstract
A number of prominent authors have recently argued that any abnormal impact of speculatorson commodities prices should involve stockpiling as a signal. Others contend by contrast that,due to the price inelasticity of supply and demand in commodity markets, speculation coulddistort commodity prices without any change in inventories. Motivated by this debate, thispaper examines the impact of commodity index funds flows on commodities’ term structurefor 12 US agricultural commodities, which are included in the two major commodity indexes.By using an autoregressive-lag model, we show that commodity index funds flows do not inmost cases have significant impacts on commodities’ term structure. By contrast, we retrievethe short-term impacts of index flows on certain agricultural commodities’ prices already evi-denced in the literature. Different robustness tests, using more exogenous index flows variables,spot and long-term futures prices, bi-monthly and monthly data, different term structure in-dicators, USDA inventory projections, and non-agricultural commodities, confirm our conclu-sions. Hence, our results suggest that stockpiling is not a necessary ‘signal’ of an abnormalimpact of speculators on commodities prices.
Keywords: Commodity index funds, Inventory,term structureJEL Classification:G1 G11
4.1 Introduction
The financialization of commodity has triggered a heated debate among professionals, researchersand policy markers.In the literature on “the impact of speculation” on commodity futures markets, commodity indexinvesting, a financial innovation, which emerged as a standard investment practice among port-folio managers in the mid-2000s, has attracted particular attention. Index funds track indicescomposed of baskets of commodity futures that are regularly rolled over prior their expiration.While causality tests generally refute the hypothesis that index flows Granger-cause commodi-ties’ prices, Guilleminot et al (2012) have uncovered a contemporaneous impact of weekly indexflows on some agricultural futures prices.Two opinions exist at the theoretical level about the origin of an abnormal speculative impact oncommodities’ prices. Paul Krugman and other authors consider that speculation in commodities’
70
paper markets can only have impact on commodity prices through a change in inventory or inthe commodities’ forward term structure. From this perspective, absent abnormal stockpiling orchange in the term structure, speculation should not be blamed for abnormal price distortionsin commodity markets. By contrast, Pierru & Babusiaux (2010) argue that, due to the short-termprice inelasticity of supply and demand in commodity markets, an abnormal speculative impacton commodities prices can occur without involving stockpiling as a signature. More details ofthese two opinions will be given in the second section.In this paper, we empirically investigate the impact of commodity index funds on both com-modities’ term structure and price returns. For this aim, we assess the impact of weekly indexflows, as reported by the CFTC on 12 US agricultural markets, on the term structure of commod-ity forward curves (defined as the spread between the one-year-out and the first-nearby futurescontracts), which can be considered, by virtue of the theory of storage, as an inventory proxy forstorable commodities. We then carry out robustness tests using bi-monthly and monthly (insteadof weekly) observation times, spot prices (instead of futures prices), alternative futures calendarspreads (instead of one-year-out to first-nearby calendar spreads) and monthly revisions in USDAinventory projections (instead of higher-frequency term structure dynamics).According to our analysis of twelve US-traded agricultural futures contracts, we find that theimpact of index speculation goes directly through prices without any change in inventory nor interm structure. In addition, the impact of index flows on agricultural prices disappears whenwe turn to bi-monthly or monthly observations, suggesting a short-lived market impact followedby a correction. The same conclusion is obtained in the case of several metal and energy futuresmarkets. Overall, our results suggest that speculation can have an impact on commodities priceswithout involving stockpiling as a signature.The rest of paper is organized as follows: section II reviews previous literature; Section III de-scribes the data used in this paper and the methodology; section IV presents and discusses ourresults; section V is devoted to some robustness checks; section VI concludes.
4.2 Theoretical Background
In the discussion of relation between commodity price and speculation, Paul Krugman (2008)mentioned in his New York Times blog that: “The only way speculation can have a persistent ef-fect on oil prices, then, is if it leads to physical hoarding – an increase in private inventories of blackgunk. . . .” In another blog, he argues that the impact is possible only if “someone who actuallyhas oil” sells oil to a long speculator through a forward contract and “holds oil off the marketso he can honor that contract when it comes due”. Let us remark that not only storers but alsoproducers may have the ability to “hold a commodity off the spot market” with a view to honor-ing a short position in the paper market. Because storers and producers must be incentivized tohold the commodity off the spot market, such mechanism should be revealed by a change in theinventory or at least in the forward term structure (if producers are involved rather than storers).Hence, from this perspective, speculation cannot drive prices away from the fundamentals with-
71
out a change in inventories. The argument can be illustrated as follows:
Figure 4.1: Relation model 1
An alternative story along the same line is as follows: if unwarranted speculative activity drivesprices away from the “fundamental” price (which could be defined as the one which balancesproduction with consumption in the medium term), then, if the price and demand functions aresufficiently price elastic, an excess of production should ensue, resulting in turn in inventory ac-cumulation. Hence, also in this case, an abnormal speculative impact should be associated to achange in inventory. However, in this case, inventory accumulation would be the consequence ofthe abnormal price impact, not its cause. (Pirrong 2008), (L.Smith, 2009), as in the case of suffi-ciently large price elasticity of demand and supply, the production reacts to an abnormal changeof prices. The inventory accumulation is due to an excess of production, which is caused by anabnormal price increase. .An opposite opinion proposed by (Pierru & Babusiaux, 2010) suggests that speculation can havean impact on commodity prices without stockpiling as a cause or consequence. The argumentrelies on the low short-term price elasticity of supply and demand in commodity markets, whichleads to a slight impact on demand and minimal effect on inventory in the case of exogenouslydriven price rise. ε (t) is supposed to be an increasing function of time t, with ε′ (t) > 0. Demandcan be assumed to be inelastic in the short run and demand adjusts gradually. Inventory shock4I after an increase of prices due to a speculative shock can be defined as a function of demandelasticity ε (t) and prices changes 4S, with 4I (t) = ε (t)4S. After a speculative shock on the prices,and given short-run price inelasticity, if prices turn to be normal quickly, stockpiling will not beobserved. This economic mechanism might be illustrated as follows:
72
Figure 4.2: Relation model 2
The hypothesis of Babusiaux and Pierru (2010) is relevant for agricultural commodities, sincetheir price elasticity of the demand is particularly in this case.The question of the price impact of uninformed speculative flows on commodity prices has re-ceived an important attention in the recent literature. Some studies (see in particular Irwinand Sanders, 2011; Büyüksahin and Harris, 2011; Capelle-Blancard and Coulibaly, 2011) havetried to identify the direction of the causality between prices and flows by performing a Grangercausality test between investment flows and prices or alternatively by analyzing the relationshipbetween flows and subsequent price returns at the cross-sectional level (see Irwin and Sanders2010-2012). Most studies in this strand of literature reach the conclusion that flows do not“Granger cause” price changes. However, temporal precedence is not equivalent to causation.Singleton (2011) observes that the 13-weeks rolling cumulative index flows predict the subse-quent oil prices weekly returns after controlling for fundamental financial and non-financialvariables (convenience yield, equities prices returns, financing conditions offered by large invest-ment banks. . . ). The documented instability of the precedence relation between flows and pricesis an additional concern. For example, Robles et al. (2009), who carry out Granger causality testson sliding 30-months windows, show that the hypothesis that flows “Granger cause” prices issometimes rejected, sometimes validated depending on the period of the test.Gilbert (2010) finds a significant positive association between contemporaneous weekly indexflows and prices variations for a set of energy, metal and agricultural commodities after control-ling for equities price returns. From here, the author evaluates that the maximum price impact ofindex flows may have been to raise prices by the order of 15% in 2008. Hendersen et al. (2012) useCommodity-Linked-Notes (CLN) issues as a plausibly exogenous index flows variable. Throughan event-study analysis, they observe the existence of an impact of CLN issues on commoditiesprices. Guilleminot et al. (2014) also address the question of endogeneity by using index flowsinto twelve agricultural contracts and into three generalist ETF as two possibly exogenous indexflows variables. They find an economically and statistically strong association of weekly indexflows to futures prices returns in some agricultural markets, the effect being reinforced in condi-tions of overall financial stress.Although the literature tends to conclude to the existence of an effect of index flows on commod-ity price returns, on the one hand; the relationship between commodity index funds and inven-
73
tory changes or spread of futures contracts, on the other hand, has so far received little attention.Mou (2011) shows that a strategy front-running the GSCI investors just a few days before themonthly rolling of the positions yields abnormal returns. The abnormal return disappears if thestrategy is executed on contracts, which are not included in the Goldman Sachs Commodity In-dex. This suggests that GSCI investors cause the spread between first-nearby and second-nearbycontracts to widen (the first-nearby contract, which is sold by the GSCI investor, depreciates withrespect to the second-nearby, which is bought by GSCI investors) at the time of the rolling. Usingavailable oil inventory data, Lombardi Robays (2011), find no significant impact of “destabiliz-ing speculation” on inventories. On a related note, some recent works have recently looked at theimpact of index investors on futures risk premiums (see Brunetti and Reiffen (2011), Acharya,Lochstoer, and Ramadorai (forthcoming), Cheng, Kirilenko, and Xiong (2012), and Hamilton andWu (2013). Only Brunetti and Reiffen (2011) have found some evidence of such impact so far.The relation between inventory and futures term structure is now well established. Working(1933) is the first to identify a strong positive correlation of the futures calendar spreads (differ-ence between long and short dated maturities) to the inventory level in the wheat market. Thisrelation between inventory and futures calendar spreads has been retrieved on different commod-ity markets and more recent periods (see e.g. Fama & French 1987, Gorton et al. 2007, Geman& Ohana 2009). The relation goes both ways. An inventory increase raises the price of storage(i.e. the price to rent a free storage space), which can be readily expressed in terms of the spreadbetween the futures prices of long and short dated maturities. Conversely, an increase in thespread between long and short dated futures prices prompts market participants to execute thecash-and-carry arbitrage, resulting in an increase of inventories.From this fundamental insight, our aim in this paper is to test the two aforementioned hypothe-ses on the impact of uninformed index speculation on commodity prices: (1) this impact goesdirectly through the prices without involving any change in inventories nor the term structure or(2) this impact involves a change in the term structure and inventories as a cause or consequence.Overall, our results give strong support to the second hypothesis. An additional implication ofour observations is that the market impact of index flows vanishes beyond a weekly time horizon.
4.3 Inventory and index funds
Inventory and its determinants
Investigating the impact of index funds on agricultural commodities’ futures calendar spreadsor inventory requires the knowledge of the various control variables that are likely to drive com-modities prices’ term structure and inventory.The possibilities of arbitrage between spot and futures markets for storable commodities stronglyattach the spread between spot and futures prices to the total charge of carrying forward a com-modity from the present date to the futures contract maturity. This charge, in turn, is composedof capital charge (i.e. interest rates), storage costs and the “price of storage” (see Working, 1941),
74
i.e. the cost of renting a free storage space from the present date to the contract maturity. Thisprice of storage naturally increases with the level of inventories, as high inventories imply scarceresidual storing capacities. When inventories are depleted, the price of storage turns negativeas free storage space is abundant and diverse financial benefits (often referred to as the “conve-nience yield”) accrue to the holder of a physical commodity.“Backwardation” corresponds to a situation where the futures term structure curve (the behaviorof futures prices as a function of contract maturity) is downward sloping. According to the theoryof storage, backwardation occurs when the convenience yield is greater than the cost of holdinginventory (interest rate and cost of storage), hence when inventory is low. The opposite situation(futures price higher than the spot price) is referred to as “contango”.Backwardation determines in turn the storage behavior. (Telser, 1958) points out that investorstend to store in the situation of contango and sell inventory when in the context of backwardation.Moreover, since long-dated futures contracts are less expensive, rolling from nearer to longer con-tracts is beneficial for speculators in backwardated markets. Hence, the term structure influencesnot only storers’ but also financial speculators’ behavior in commodity futures markets.In previous literature, the spread between first and second month is often used as an inven-tory proxy. However, short-dated contracts can be highly volatile, (Borovkova, 2003), hence thespread between the two nearest contracts might contain “noise”. Another concern is seasonality:the term structure behavior is affected by the timing of the harvest (e.g. the spread between twocontracts with maturities before the harvest conveys information about pre-harvest inventorywhile the spread between two contracts with maturities before and after the next harvest re-flects post-harvest inventory anticipations) and the spread between different contracts is stronglydiscontinuous at the crossing of rolling dates (i.e. jumps after the nearest contract ceases trad-ing). In order to circumvent these problems, we use the smooth inventory proxy introduced in(Guilleminot, Ohana, J.J, & Ohana, S., 2014), 1 defined from the performance of a strategy short-ing the first maturity after the closest harvest (denoted F1), while buying the first maturity afterthe second closest harvest (F2)2:
I (t) = Πi≤t−∆t
(1 +
F2i+∆t −F2iF2i
− F1i+∆t −F1iF1i
)(4.1)
Due to a lack of data for long-dated maturities for a number of commodities, we will define F1 asthe prompt-month contract and F2 as the contract whose delivery is precisely one year after F1. In the robustness test section, we will use alternative contracts F1 and F2 (spot vs first-nearby,
1 More information about new stock proxy can be referred to Guilleminot et al. (2013). In this article, the newinventory proxy is compared with stock-to-use ratio in USA. Regression analysis between inventory proxy changes andstock-to-use ratio changes has shown a highly significant relation for both end-year data and monthly data.
2For example, in the case of corn (resp. wheat) futures at the CBOT, the strategy shorts the prompt December (resp.July) month and longs the subsequent December (resp. July) month. When reaching the last trading day of the promptDecember contract, the strategy moves to the next two December contracts available. For soybeans, we use the Novembercontract instead of the December contract.We use also alternative contracts F1 and F2(spot vs first-nearby, first-nearby vssecond-nearby spreads). These results can be obtained by requiring authors.
75
first-nearby vs second-nearby spreads).
We are interested specifically in the inventory shock between two dates t and t-1, defined as∆I (t) = I (t)− I (t − 1), the inventory proxy difference of two dates can measure the market condi-tion: ∆I (t) > 0 for an increase, ∆I (t) < 0 for a decrease in inventory projections.As will be shown in the robustness tests section, this smooth inventory shock proxy is highlycorrelated with USDA stock-to-use revisions for corn, soybean and wheat. For non-storable com-modities, such as feeder cattle, lean hogs and live cattle, the defined “inventory proxy” may notbe interpreted any longer as a measure of inventory shocks, but rather as a measure of the one-year-ahead expected spot price variation. In this case, only the impact of commodity index fundson the futures price term structure will be highlighted.
Three variables are included as control variables in the regressions of inventory proxy on indexflows: interest rate, dollar index and hedging pressure.
Interest RateSince interest rates determine the cost of carrying forward a commodity, a positive correlationbetween interest rate changes and inventory shock is expected. Six months government bill rateis chosen in our case.
Dollar indexMost commodities are quoted in US dollars. A dollar depreciation induces an upward pressureon the dollar-libeled price since, everything else equal, commodities become cheaper for investorsoutside the US. In this paper, we use as a proxy for the value of the US dollar, the so-called “DollarIndex”, defined as the value of the US dollar against a trade-weighted basket of other currencies.While a negative relation is expected between dollar index and commodities’ price in dollars, it isharder to build a conjecture on the correlation of dollar index to inventory/term structure shocks.
Hedging PressureHedging pressure is considered as a measure of risk premium, which can be referred to the netdemand for long positions from hedgers in futures markets. (Keynes, 1930) points out that riskaversion of inventory holders and producers induces ‘normal backwardation’, i.e. futures pricesbelow the expected spot prices at maturity. Therefore, hedging pressure from sellers of futurescontracts tends to push the term structure towards backwardation. The traditional measure ofhedging pressure is
Hedging P ressure =Long Hedge P ositions − Short Hedge P ositions
T otal Hedge P ositions(4.2)
In this paper, long (short) hedge positions are the long (short) aggregate commercial positionsin futures markets. Since a long demand from hedgers results in a buying pressure on paper
76
contracts relative to the spot market, we expect hedging pressure to increase the contango of thecurve, hence to have a positive impact on our inventory proxy.
Commodity Index funds
The two most popular commodity indexes are the GSCI (Goldman Sachs Commodity Index) andthe DJ-UBS (Dow Jones- UBS Commodity Index). The GSCI contains 24 commodities nearby fu-tures contracts from several commodity sectors: energy products, industrial metals, agriculturalproducts, livestock products and precious metals. The index rolls the positions every month fromfifth to ninth trading days, at a pace of 20% per day. Agriculture weighs about 10% and livestock3.1% in the index. DJ-UBS includes 20 commodities futures contracts, where agricultural weighsabout 30% and livestock 6%. The 12 agricultural and livestock contracts used in this paper arethe ones included in these two major commodity indexes.
Figure 4.3: Commodity Index Funds Flows
Note: The graph shows the weekly cumulative total CITF and the percentage of position held bylong index traders from 2006 to 2014. Total CITF refers to the net weekly aggregate commodityindex funds positions changes, across the 12 agricultural commodities monitored in the CFTCSupplemental Report. Percentage of position held by long commodity index traders are average
of ratioIndex long positiontOpen interestt
across twelve agricultural and livestock commodity futures by week.
In this research, index funds flows is calculated byIndex net positiontOpen interestt−1
where the numerator
refers to the weekly changes in the index investors’ net position in each futures market, obtained
77
from the weekly CFTC supplemental reports. The aggregate index flows into the 12 agriculturalmarkets presented in the figure 4.3 shows that commodities index funds increased their holdingsof commodities from the beginning of 2006 up to the eve of the 2008 financial crisis. Their longpositions account for about 26% of total open interest. During the second semester of 2008 upto the first quarter of 2009, index investors sharply reduced their commodity exposures but stillhold long position with 28% of total open interest. After that date, a new wave of index invest-ment was observed, up to the spring of 2011, after which index investors’ positions exhibit a slowbut steady downward trend and their long positions occupy about 26% of total open interest.COT report provides only weekly data about index funds positions; as a result, all the variablesconsidered in this paper are computed between the days of publication of the CFTC COT report(every Thursday).
4.4 Empirical Analysis
1) Inventory proxy and index funds flowsTo estimate the impact of index flows on inventories, we perform a simple linear regression ofinventory proxy changes on index funds flows, while including the aforementioned control vari-ables:
∆Inventory P roxyi,t = α+ΣPin=0βm∆Inventory P roxyi,t−m+Σqin=0γnxj,t−n+ΣKκ=0δκIndex Flowsi,t−κ+εi,t(4.3)
Parameters α,β0....βp,γ0....γn,δ0....δk are estimated using standard OLS for every commodity i.The index j refers to the control variables: xj,t−n = (∆interest rate,∆dollar index,∆hedging pressure).
The empirical estimation is conducted as follows: Firstly, lag numbers p and q are chosen accord-ing to the Akaike (AIC) and Bayes (BIC) information criteria 3. Meanwhile, the lag variables withnon-significant coefficients are removed for the sake of robustness. All the other regressions inthis paper use the similar estimation methodology.
a) The results of regression (4.3) are reported in Table 4.1. Weekly observations are used, rang-ing from January 2006 to December 2014. The first regression uses the individual index fundsfrom the COT report as a measure of agricultural commodity index funds flows. Interest ratesand dollar variations do not have a significant impact on the term structure, except for lean hogs.Hedging pressure, which can be interpreted as the pressure from long hedging positions, enterswith a positive sign in most cases. The only cases where a significant negative impact of hedgingpressure is found are the ones of soybeans and feeder cattle.
3Table of choosing lag number is given in appendix.
78
The coefficients of commodity index traders’ flows (denoted CITF) on the inventory proxy arepositive and significant at the 10% level only for cocoa and sugar markets. Hence, we have evi-dence that index funds inflows actually increase the “contango” only for two commodities out ofthe twelve considered. This finding may be explained by the fact that demand and supply elas-ticity is relatively higher in cocoa and sugar markets, so that market participants expect supplyand demand to promptly adjust to a price increase caused by unwarranted speculative activity.This in turn results in higher expected inventories. However for other commodities, this effect isfound to be insignificant. We find a case of negative impact of index flows on calendar spreadsfor a non-storable commodity (live cattle). In this case, however, the calendar spread may notbe interpreted as an “inventory proxy” but rather as an estimate of the futures term premium(expected spot price appreciation plus risk premium).
Overall, the results displayed in Table 4.1 lend little support to the hypothesis that index fundsmodify commodities’ term structure or inventories.
b) In this section, we replace the individual commodity index funds flows with ETF flows andaggregate agricultural index flows. “ETF flows” correspond to index speculators investing inthe three main ETF tracking generalist commodity indices: Goldman Sachs Commodity Index,Bloomberg Commodity Index, (formerly called Dow-Jones UBS Commodity Index), and DeutscheBank Commodity Index. “CITF Total” is defined as the net aggregate index flow over the 12agricultural commodities monitored by the CFTC. The ETF variable is clearly the most exoge-nous since agricultural commodities represent only a small part of the GSCI or the DJ-UBS in-dices, which have much higher correlation to energy markets than to agricultural commodities(Guilleminot, Ohana, J.J, Ohana, S., 2014). However, this variable only captures a small (liquid)component of the total index funds’ activity, hence the choice to include as well the variable CITFTotal.The ETF data only start in July 2006. ETF flows on the three indices4 between t-1 and t are de-fined as:
Flowst,t−1 = Σ3i=3ET F P rice
it
(sharesit − sharesit−1
)(4.4)
where shareit refer to the total number of ETF shares issued for the i-th (i=1,2,3) ETF at date t.The results are displayed in Tables 4.2 and 4.3.ETF flows are found to have a significant lagged impact on inventory changes only for soybeansand cocoa but the direction of the impact is negative in the former case. Finally, when usingaggregate commodity index funds to capture index investment, index flows only impact the soy-beans inventory proxy. We have already detected a positive impact of individual index flows oncocoa and sugar inventory proxies.
4The three global commodity index ETFs considered are: Powershares DB, IPATH Dow-Jones-UBS and ISHARES SPGSCI. They represent the lion’s share of ETF investment into generalist commodity indices.
79
By summarizing the results of the three different regressions (using alternatively individual com-modity index flows, ETF flows and Total commodity index flows to capture index flows), a con-clusive effect of index funds on inventory proxy is found only for sugar, cocoa and soybean. Wefind some weak evidence of a negative impact for live cattle (for which the calendar spread maynot be considered as an inventory proxy) and no clear evidence of an impact for the other nineagricultural commodities. Hence, we conclude that index funds have little influence on agricul-tural commodities’ term structure.
2) Index funds and futures pricesWe turn to the relationship between index funds and futures prices returns. Most previous liter-ature finds that there is significant relationship between commodity index positions and agricul-tural futures prices.We check again this insight by using a regression framework very similar to the one above butusing instead futures prices returns as the dependent variable:
Futures Returni,t = α+Σpim=0βm∆Inventory proxyi,t−m+Σqin=0γnxj,t−n+ΣKκ=0δκIndex f lowsi,t−κ+εi,t(4.5)
Futures prices returns are computed from the “rolled futures prices series”, which have the prop-erty to be continuous just after the first-nearby ceases trading (Guilleminot, Ohana, J.J, Ohana,S., 2014).Rolling futures prices are defined as:
Pt = Πi≤t−∆t
(1 +
F1i+∆t −F1iF1i
), with F1 the prices of the first-nearby contract.The “rolled futures
return” is in turn computed by:Pt − Pt−1
Pt−1.
Rolling futures price is actually the performance of a strategy longing the first-nearby contract(the notional position being kept constant at 1 dollar and rolling over this long position wheneverthe first-nearby ceases trading).The control variables are the same as before, except that we now include the inventory proxy asa supplementary control variable. As before, three index funds variables (individual commodityindex flows, ETF flows and Total commodity index flows) are successively tested (the results be-ing reported in Tables 4.4,4.5 and 4.6).
Hedging pressure has in most cases a significant positive relation to the futures prices returns,which is intuitive, as a demand for long futures positions from hedgers is due to exert a positivepressure on futures prices. Dollar index and interest rate have a non-significant impact, exceptin the CBOT wheat market, where their impact is significantly negative (as higher rates implyhigher dollar and reinforce the attractiveness of bonds at the expense of real assets).
Inventory proxy enters in most cases with a negative sign, as could be expected since an inventory
80
increase (resp. decrease) commands a price decrease (resp. rise) to alleviate the supply glut (resp.the supply shortage). However, for non-storable commodities (animal products), the inventoryproxy has a different interpretation, hence the inconsistent signs observed for the impact of theinventory proxy.
We observe a significant effect of index funds on futures prices for a significant number of com-modities. Coffee, Cotton are impacted in three specifications; cocoa, lean hogs, soybean, soybeanoil are impacted in two specifications out of three; sugar, feeder cattle and CBOT wheat are im-pacted in one specification only. Finally, live cattle, corn and Kansas wheat are not impacted inany of the three specifications. The effects are the largest when using aggregated index flows asa proxy for index investment. In the case of ETF flows, the impact is significant only for coffee,cotton, soybeans and soybean oil.
3) Empirical results and Theoretical debateFrom the previous analysis, we draw two conclusions. On the one hand, index funds impact thefutures prices of a significant number of commodities: coffee, cotton, cocoa, lean hogs, soybeansand soybean oil exhibit the clearest evidence of impact. On the other hand, we find little evi-dence of an impact of index funds on inventory: such impact is found only in the sugar, cocoaand soybean markets. There is therefore little evidence of a joint effect of index funds flows onboth futures prices and inventory.The observations lend support to the hypothesis of (Pierru & Babusiaux, 2010): uninformed spec-ulative activity might influence commodities futures prices but not necessarily through the chan-nel of inventory. Index funds can distort futures prices without stockpiling as a signature. Theseresults are in line with previous empirical literature on the oil market (see Lombardi & Robays,2011 and Hamilton, 2009 in particular).
4.5 Robustness Analysis
In this section, we carry out four different robustness tests of the previous analysis.
1) Longer horizon (Data Frequency)Due to the price inelasticity of supply and demand of agricultural commodities, it is possible thatinventory reacts with a time lag of a few weeks to an abnormal price change in the paper market.For this purpose, we apply the same regression analysis by using bi-monthly and monthly datainstead of weekly data. In the bi-monthly regression, observations are made every other Thurs-day. In the monthly specification, observations are made at the last trading day of each month.
In Table 4.7, 4.8, 4.9 and 4.10, For the sake of clarity, we report only with a blue color the com-modities for which a significant positive impact is found at the 10% level (indicating if the im-pact is negative). Although the price elasticity of supply and demand increases at bi-monthly or
81
monthly frequency, the impact of index funds on inventory is still small. At the bi-monthly fre-quency, the commodities whose inventory proxy is positively impacted by index flows are sugar(using individual index flows), soybeans (using ETF flows) and lean hogs (using aggregate indexflows). We note that, for these three commodities, the impact is found in only one model specifi-cation. At the monthly level, we find again sugar in the list of commodities whose term structureis impacted. Live cattle and corn newly appear in respectively one and two specifications out ofthree, the former with a lag of one month. At both frequencies, there is again little evidence ofa joint impact of index flows on prices and inventories. When using lower frequency data, theimpact on futures prices subsides, which suggests that prices revert to an equilibrium after beingpushed away from their fundamental value by uninformed speculative flows in the short term.
2)Specific period studyWe implemented the same analysis for four sub-samples: 06/2006-30/06/2008 (commodities’boom), 07/2008-31/01/2009 (commodities’ collapse), 02/2009-31/03/2011 (recovery) and 04/2011-2014 (a period marked by a steady fall in commodities prices).
Index funds have more impact on the inventory proxy from 2006 to 2008. Again, it is rare tofind evidence of simultaneous impact of index funds on prices and inventory proxy: this happensonly:- for coffee and corn in the first period- for feeder cattle (an unstorable commodity) and soybean oil in the third periodAnd no simultaneous effect observed in the forth period. Therefore, our observations lend sup-port again to the argument made by (Pierru & Babusiaux, 2010).
3) Non-agricultural commoditiesOne of the possible explanations of our results lies on price-inelasticity of supply and demandfunctions in agricultural markets. Indeed, most agricultural commodities are non-substitutableproducts with very limited supply flexibility. The case may be different for non-agricultural com-modities, whose supply and demand are more price elastic. For this purpose, we choose sevennon-agricultural commodities: Aluminum, Brent, Copper, Gold, Nickel, Silver and Zinc. SinceCFTC index flows data are not reliable for non-agricultural commodity, we use ETF data as anindicator of commodity index funds. We are aware of a potential endogeneity problem, in partic-ular when assessing the impact of ETF flows on energy prices, as ETF investors may be partiallydriven by the oil prices.
Table 4.19,4.20 show the results of CITF-inventory and CITF-futures returns regressions on weekly,bi-monthly and monthly data. None of the energy or metal commodities exhibit an impact of ETFflows on the inventory proxy. However, ETF flows do have an impact on futures prices, which sub-sides again when the data frequency decreases.Overall, our results suggest that index funds flows have a significant short-term impact on com-modities’ futures prices but that this impact does not go through inventory changes. These find-
82
ings are consistent with our previous results on agricultural commodities and previous studieson the oil market.
4) Additional robustness tests on corn, soybean and wheat marketsIn this last section, we report additional tests using inventories estimations from USDA monthlyreports and alternative calendar spreads that could serve as inventory proxies. All these supple-mentary tests apply to three US grain markets only: corn, soybean and wheat. Regarding USDAinventories data, we use the monthly USDA (WASDE) Stock-to-Use projections at the end of thecurrent or following marketing year. For example, in the case of corn and soybeans, from May toAugust (resp. September to April), the projections concern the residual stock- to-use at the end ofthe following (resp. current) marketing year. As regards alternative inventory proxies, we use thefirst to second-nearby contracts calendar spreads and the spread between spot and first-nearbycontract prices. Lastly, we reexamine the question of the price impact of index flows by consider-ing grain (US Gulf) spot prices instead of first-nearby paper contracts at the CBOT or KCBT.
Correlation between inventory proxy and stock-to-useIn Table 4.21 ,we observe that the relation between stock-to-use and one-year-distance calendarspread is significantly positive for the three grain markets. The relation between the stock-to-useand the second to first-nearby calendar spread is significantly positive only for corn and soybeans.Finally, the relation between the stock-to-use and the first-nearby to spot prices calendar spreadis positive only for corn but not significantly. Therefore, the one-year-distance calendar spreadappears as the best inventory proxy in the three considered grain markets.Impact of index flows on USDA inventory data and on alternative calendar spreadsFrom table 4.22 to 4.24, Index flows do not have any significant impact on the second to first-nearby calendar spread. However, in Table 4.22, an impact of index flows on the spot to first-nearby calendar spread is found on the wheat and corn markets, using the two CFTC flows datain the former case and ETF flows in the latter case. These impacts persist at the bi-monthlyfrequency but vanish at the monthly time horizon (Table 4.24). However, direct regressions ofmonthly stock-to-use revisions on index flows do not lend support to the hypothesis of an impactof index flows on inventories (Table 4.25). In addition, corn and wheat one-year-distance calendarspreads (which are better inventory proxies than the first-nearby to spot calendar spread) are notimpacted by index flows (Tables 4.1 to 4.3). Hence, the impact of index flows on the first-nearbyto spot calendar spread may be interpreted as a distortion of the term structure induced by indexinvestors’ activity on the first-nearby contract rather than the reflection of an effect of index flowson inventories.Impact of index flows on the spot priceWe lastly examine the impact of index flows on the spot prices in the three considered grain mar-kets (Table 4.26). Aggregate index flows are found to significantly impact the three grain spotprices. However, ETF flows only impact the soybeans spot market. As regards individual indexflows, they have a contemporaneous positive impact and a lagged negative impact on the grainspot prices, which suggests the presence of an abnormal contemporaneous price impact of index
83
flows followed by a correction. Again, the impact recedes on longer time horizons.
4.6 Conclusion
Commodity index speculation has garnered a particular attention in the debate concerning thefinancialization of commodity markets.Motivated by the debate on the channel of the impact of index funds activities on commodities’prices, this paper investigates the influence of index flows on inventories. We have studied theimpact of index funds on an inventory proxy constructed as a futures calendar spread, after con-trolling for lagged inventory proxy variations and the different factors driving commodities’ termstructure.
Our results can be summarized as follows. Firstly, index funds impact the inventory proxy onlyfor three agricultural commodities (cocoa, sugar and soybean). However, an impact of indexfunds on the futures prices is found for a significant number of commodities. These empirical re-sults support the theoretical hypothesis of (Pierru & Babusiaux, 2010) that index funds flows mayhave a direct impact on the futures prices that does not need to transit through inventories. Ourempirical conclusions are robust to the use of two other index funds indicators, namely flows onthe three main generalist commodity ETFs and aggregate index flows on the twelve agriculturalcontracts monitored by the CFTC.Secondly, according to some additional tests, the impact on the futures prices is observed only onthe short term, since, when running the regression on bi-monthly or monthly observations, theinfluence of index flows on futures prices fades out.
Similarly to agricultural commodities, the futures prices of non-agricultural commodities are im-pacted by index flows but not their inventory proxy. Again, the effect of index funds on theprices recedes when the time distance between observations is more than one week. We concludefrom this additional test that the absence of impact of index flows on inventories is not a specificproperty of agricultural markets (which could be linked to the particular price inelasticity of thesupply and demand of agricultural products).Additional robustness tests, performed on corn, wheat and soybeans markets, using alternativeinventory proxies and US Gulf spot prices, uncover an impact of index flows on the first-nearbyto spot prices calendar spread in the corn and wheat markets. However, there is no clear evidencesupporting the impact of index flows on inventories.
As a conclusion, in commodities markets, index speculation can influence the futures prices with-out stockpiling as a signature. There are two implications of this result. First, it sheds light onthe price formation mechanisms in spot and paper commodity markets. We now have strongevidence that spot and futures prices may be influenced by factors that are purely exogenous tothe physical supply and demand conditions in commodity markets and that this impact does not
84
even need to go through physical forces in the spot market. This calls for the introduction of newmodels where commodities’ spot and futures prices are driven in a direct line by speculative ac-tivity in the paper markets. Second, from a regulatory perspective, the evidence that commodityindex funds impact futures prices at the weekly horizon but not at longer time scales suggests acertain short-term inefficiency of commodities markets, as informed traders seem to take a cer-tain time to correct the artificial price moves induced by uninformed speculators.
85
Table 4.1: Inventory proxy and CITF
∆InventoryP roxy(−1) CITF CITF(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.030 0.09 0.002 -0.01 -0.014 0.10P-value 0.538 0.063* 0.96 0.836 0.774
Coffee Estimate 0.002 -0.046 0.268 0.038 -0.059 0.071P-value 0.974 0.351 1.13e-07*** 0.435 0.222
Cotton Estimate -0.092 0.024 0.23 -0.025 -0.061 0.069P-value 0.058* 0.62 4.16e-06*** 0.599 0.205
Sugar Estimate -0.054 0.095 0.04 0.023 -0.063 0.022P-value 0.28 0.064* 0.432 0.64 0.204
Feeder Cattle Estimate -0.05 0.025 -0.146 -0.038 -0.023 0.023P-value 0.31 0.62 0.004*** 0.44 0.64
Lean Hogs Estimate 0.045 0.032 0.37 -0.121 -0.112 0.163P-value 0.321 0.48 5.19e-15*** 0.008*** 0.014**
Live Cattle Estimate 0.035 -0.085 0.504 -0.05 -0.068 0.252P-value 0.412 0.054* <2e-16*** 0.249 0.116
Corn Estimate -0.037 -0.033 0.18 0.003 0.015 0.03P-value 0.44 0.527 0.0005*** 0.94 0.76
Soybeans Estimate -0.19 0.056 -0.145 0.003 0.01 0.05P-value 0.001*** 0.285 0.006*** 0.95 0.83
Soybean Oil Estimate -0.007 -0.03 0.186 0.037 -0.06 0.037P-value 0.881 0.553 0.0002*** 0.441 0.221
Wheat CBOT Estimate -0.144 -0.005 0.107 0.056 -0.584 0.04P-value 0.003*** 0.925 0.03** 0.258 0.23
Wheat KCBOT Estimate -0.055 0.04 0.275 -0.045 -0.002 0.08P-value 0.247 0.403 1.29e-08*** 0.347 0.634
Note:This table gives the result of the following regression:
∆Inventory P roxyi,t = α +ΣPin=0βm∆Inventory P roxyi,t−m +Σ
qin=0γnxi,t−n +ΣKκ=0δκIndex Flowsi,t−κ + εi,t
CITF refers to the ratio of weekly index flows to lagged Open Interest for commodity i in week t from the CFTCSupplemental Report, and CITF(-1) is its lagged value.
∆Inventory proxyi ,t refers to inventory shock for commodity i in week t. The inventory proxy is obtained from thecumulative return of a strategy shorting the first-nearby futures contracts and longing the futures contract deliveringone year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the above strategy between twoconsecutive dates of the CFTC supplemental report.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate. ∆Dollar
represents the dollar index return. ∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position.
The number of lags for CITF is chosen according to the AIC/BIC criterion. Weekly data from 2006 to 2014 are used. Bothcoefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardized.Number of observations: 417
86
Table 4.2: Inventory proxy and ETF
∆InventoryP roxy(−1) ETF flows ETF flows(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.04 0.211 0.002 -0.016 -0.007 0.47P-value 0.416 2.22e-05*** 0.975 0.749 0.896
Coffee Estimate -0.006 -0.03 0.27 -0.34 -0.06 0.076P-value 0.898 0.497 5.99e-08*** 0.48 0.175
Cotton Estimate -0.134 -0.047 0.2 -0.048 0.06 0.067P-value 0.006*** 0.33 6.72e-05*** 0.32 0.22
Sugar Estimate 0.42 0.78 0.06 -0.047 -0.07 0.02P-value 0.40 0.12 0.241 0.34 0.14
Feeder Cattle Estimate -0.06 -0.001 -0.147 -0.04 -0.026 0.026P-value 0.22 0.9 0.003*** 0.41 0.60
Lean Hogs Estimate 0.044 0.005 0.36 -0.133 -0.116 0.16P-value 0.343 0.92 1.9e-14*** 0.004*** 0.012**
Live Cattle Estimate 0.031 -0.02 0.5 -0.035 -0.07 0.25P-value 0.52 0.66 0.004*** 0.73 0.72
Corn Estimate 0.007 -0.015 0.07 -0.01 0.014 0.06P-value 0.98 0.775 0.158 0.84 0.775
Soybeans Estimate -0.13 0.03 -0.3 -0.012 0.004 -0.015 0.03P-value 0.009*** 0.6 0.013** 0.02** 0.94 0.77
Soybean Oil Estimate -0.006 0.06 0.19 -0.04 -0.06 0.04P-value 0.907 0.27 0.0002*** 0.371 0.19
Wheat CBOT Estimate -0.154 -0.03 0.09 -0.05 -0.06 0.04P-value 0.002*** 0.56 0.06* 0.27 0.23
Wheat KCBOT Estimate -0.063 -0.06 0.286 -0.05 -0.022 0.087P-value 0.194 0.214 5.96e-09*** 0.283 0.651
Note:This table gives the result of the following regression:
∆Inventory P roxyi,t = α +ΣPin=0βm∆Inventory P roxyi,t−m +Σ
qin=0γnxi,t−n +ΣKκ=0δκET F Flowsi,t−κ + εi,t
ET F f lowst denotes the net weekly inflows into the three ETF tracking generalist commodity indices: Goldman SachsCommodity Index, Bloomberg Commodity Index, and Deutsche Bank Commodity Index in week t, and ETF (-1) is thelagged value.
∆Inventory proxyi ,t refers to inventory shock for commodity i in week t. The inventory proxy is obtained from thecumulative return of a strategy shorting the first-nearby futures contracts and longing the futures contract deliveringone year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the above strategy between twoconsecutive dates of the CFTC supplemental report.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate. ∆Dollar
represents the dollar index return.∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position.
The number of lags for ETF flows is chosen according to the AIC/BIC criterion. Weekly data from 2006 to 2014 are used. Bothcoefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardized.Number of observations: 408
87
Table 4.3: Inventory proxy and CITF Total
∆InventoryP roxy(−1) CITF Total CITF Total(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.031 0.007 0.008 -0.028 -0.010 0.002P-value 0.535 0.891 0.878 0.577 0.847
Coffee Estimate 0.008 -0.063 0.266 -0.043 -0.062 0.070P-value 0.869 0.202 9.21e-08 *** 0.373 0.198
Cotton Estimate -0.090 -0.002 0.230 -0.031 -0.064 0.070P-value 0.0638 * 0.965 3.32e-06 *** 0.528 0.184
Sugar Estimate -0.047 0.038 0.057 -0.034 -0.071 0.015P-value 0.344 0.456 0.261 0.504 0.152
Feeder Cattle Estimate -0.050 0.019 -0.141 -0.038 -0.025 0.020P-value 0.312 0.706 0.004 *** 0.451 0.608
Lean Hogs Estimate 0.048 -0.043 0.040 0.373 -0.125 -0.112 0.160P-value 0.295 0.371 0.430 3.11e-15 *** 0.008 *** 0.014 **
Live Cattle Estimate 0.038 -0.051 0.498 -0.044 -0.072 0.250P-value 0.385 0.251 <2e-16 *** 0.315 0.097 *
Corn Estimate -0.018 0.015 0.215 -0.031 0.015 0.070P-value 0.702 0.770 1.56e-05 *** 0.529 0.761
Soybeans Estimate -0.186 0.094 -0.146 0.012 -0.007 0.060P-value 0.0001 *** 0.06 * 0.004 *** 0.813 0.887
Soybean Oil Estimate -0.005 -0.033 0.185 -0.043 -0.062 0.040P-value 0.912 0.517 0.0002 *** 0.382 0.205
Wheat CBOT Estimate -0.143 -0.006 0.108 -0.062 -0.059 0.040P-value 0.003 ** 0.901 0.028 ** 0.212 0.224
Wheat KCBOT Estimate -0.045 -0.069 -0.040 0.286 -0.061 -0.020 0.090P-value 0.341 0.159 0.420 3.9e-09 *** 0.204 0.678
Note:This table gives the result of the following regression:
∆Inventory P roxyi,t = α +ΣPin=0βm∆Inventory P roxyi,t−m +Σ
qin=0γnxi,t−n +ΣKκ=0δκCIT F T otali,t−κ + εi,t
CIT F totalt is the ratio of aggregate index flows to aggregate open interests for the 12 agricultural commoditiesmonitored by the CFTC Supplemental Report.
∆Inventory proxyi ,t refers to inventory shock for commodity i in week t. The inventory proxy is obtained from thecumulative return of a strategy shorting the first-nearby futures contracts and longing the futures contract deliveringone year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the above strategy between twoconsecutive dates of the CFTC supplemental report.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate. ∆Dollar
represents the dollar index return.∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position.
The number of lags for CITF Total is chosen according to the AIC/BIC criterion. Weekly data from 2006 to 2014 are used. Bothcoefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardized.Number of observations: 417
88
Table 4.4: Futures returns and CITF
∆InventoryP roxy CITF CITF(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.273 0.14 -0.007 0.515 -0.008 -0.09 0.38P-value 7.79e-12*** 0.00054 *** 0.846 <2e-16 *** 0.452 0.02**
Coffee Estimate -0.26 0.04 -0.102 0.64 0.038 -0.04 0.4P-value 1.5e-10*** 0.34 0.012** <2e-16 *** 0.32 0.91
Cotton Estimate -0.26 0.09 0.321 0.03 0.02 0.15P-value 5.7e-08*** 0.07* 1.9e-10*** 0.54 0.72
Sugar Estimate -0.38 0.08 0.46 0.04 0.011 0.34P-value <2e-16*** 0.05** <2e-16 *** 0.34 0.78
Feeder Cattle Estimate 0.06 0.007 0.022 -0.08 -0.01 0.01P-value 0.218 0.882 0.665 0.121 0.826
Lean Hogs Estimate -0.69 0.003 0.47 0.004 0.022 0.47P-value <2e-16*** 0.93 <2e-16 *** 0.92 0.529
Live Cattle Estimate 0.02 0.04 -0.07 -0.006 0.002 0.004P-value 0.67 0.457 0.249 0.903 0.97
Corn Estimate 0.004 0.014 -0.001 -0.04 0.01 0.002P-value 0.93 0.79 0.98 0.407 0.83
Soybeans Estimate -0.07 0.05 0.61 0.011 0.09 0.43P-value 0.08* 0.19 <2e-16 *** 0.77 0.014**
Soybean Oil Estimate -0.12 -0.034 -0.63 -0.002 0.055 0.39P-value 0.003*** 0.39 <2e-16 *** 0.95 0.80
Wheat CBOT Estimate 0.09 -0.04 0.05 -0.08 -0.22 0.07P-value 0.054* 0.42 0.36 0.081* 5.9e-06***
Wheat KCBOT Estimate -0.006 0.02 0.022 -0.03 -0.008 0.002P-value 0.89 0.743 0.68 0.49 0.87
Note: The table show the results of regression:
Futures Returni,t = α +Σpim=0βm∆Inventory proxyi,t−m +Σ
qin=0γnxj,t−n +ΣKκ=0δκIndex f lowsi,t−κ + εi,t
∆Futures returnsi ,t represents the return of the rolled futures prices series of commodity i in week t.
CIT Fi refers to the ratio of weekly index flows to lagged Open Interest for commodity i in week t from the CFTCSupplemental Report, and CITF(-1) is its lagged value.The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate.∆Dollarrepresents the dollar index return. ∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position. , ∆Inventory proxyi refers to inventory shock for commodity i in week t. The inventory proxy isobtained from the cumulative return of a strategy shorting the first-nearby futures contracts and longing the futurescontract delivering one year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the abovestrategy between two consecutive dates of the CFTC supplemental report.
The number of lag for CITF is chosen according to the AIC/BIC and the co-correlation test. Weekly data from 2006 to 2014 areused. Both coefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardizedNumber of observation: 417
89
Table 4.5: Futures returns and ETF
∆InventoryP roxy ETF flows ETF flows(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.26 0.06 0.04 0.54 -0.02 -0.11 0.36P-value 2.2e-10*** 0.15 0.28 < 2e-16 *** 0.6 0.004***
Coffee Estimate -0.26 0.07 0.64 0.048 0.013 0.39P-value 1.6e-10*** 0.07* < 2e-16 *** 0.22 0.74
Cotton Estimate -0.29 0.08 0.009 0.38 0.02 0.008 0.18P-value 1.7e-09*** 0.083* 0.83 4.9e-15*** 0.65 0.86
Sugar Estimate -0.39 0.01 0.003 0.47 0.06 0.007 0.35P-value <2e-16*** 0.88 0.95 < 2e-16 *** 0.11 0.86
Feeder Cattle Estimate 0.05 0.02 0.032 -0.07 -0.03 0.01P-value 0.24 0.74 0.52 0.135 0.6
Lean Hogs Estimate -0.71 -0.03 0.09 0.45 0.002 0.03 0.47P-value <2e-16*** 0.39 0.02** < 2e-16 *** 0.96 0.45
Live Cattle Estimate 0.02 0.003 -0.06 -0.013 0.004 0.003P-value 0.71 0.96 0.29 0.79 0.93
Corn Estimate 0.01 0.025 0.002 -0.04 0.02 0.003P-value 0.84 0.54 0.97 0.38 0.81
Soybeans Estimate -0.12 0.07 0.64 0.01 0.09 0.43P-value 0.002*** 0.078* < 2e-16 *** 0.77 0.012**
Soybean Oil Estimate -0.12 0.066 0.62 0.0012 0.05 0.39P-value 0.003*** 0.09* < 2e-16 *** 0.97 0.16
Wheat CBOT Estimate 0.09 0.02 0.04 -0.08 -0.22 0.06P-value 0.052* 0.65 0.44 0.11 9.1e-06***
Wheat KCBOT Estimate -0.006 0.0003 0.03 -0.04 -0.007 0.002P-value 0.89 0.99 0.61 0.47 0.89
Note: The table show the results of regression:
Futures Returni,t = α +Σpim=0βm∆Inventory proxyi,t−m +Σ
qin=0γnxj,t−n +ΣKκ=0δκET Fi,t−κ + εi,t
∆Futures returnsi ,t represents the return of the rolled futures prices series of commodity i in week t.
ET F f lowst denotes the net weekly inflows into the three ETF tracking generalist commodity indices: Goldman SachsCommodity Index, Bloomberg Commodity Index, and Deutsche Bank Commodity Index in week t, and ETF (-1) is thelagged value.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate.∆Dollarrepresents the dollar index return. ∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position. , ∆Inventory proxyi refers to inventory shock for commodity i in week t. The inventory proxy isobtained from the cumulative return of a strategy shorting the first-nearby futures contracts and longing the futurescontract delivering one year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the abovestrategy between two consecutive dates of the CFTC supplemental report.
The number of lag for ETF flows is chosen according to the AIC/BIC and the co-correlation test. Weekly data from 2006 to 2014are used. Both coefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardizedNumber of observation: 408
90
Table 4.6: Futures returns and CITF Total
∆InventoryP roxy CITF Total CITF Total(-1) ∆HP ∆Dollar ∆r R2
Cocoa Estimate -0.275 0.100 -0.056 0.532 -0.010 -0.112 0.390P-value 6.31e-12 *** 0.016 ** 0.167 < 2e-16 *** 0.802 0.004 ***
Coffee Estimate -0.257 0.087 -0.082 0.635 0.054 0.010 0.410P-value 2.01e-10 *** 0.031 ** 0.04 ** < 2e-16 *** 0.167 0.788
Cotton Estimate -0.262 0.130 0.337 0.037 0.012 0.160P-value 3.48e-08 *** 0.005 *** 3.05e-12 *** 0.417 0.799
Sugar Estimate -0.393 0.033 0.442 0.060 0.017 0.330P-value <2e-16 *** 0.422 <2e-16 *** 0.146 0.673
Feeder Cattle Estimate 0.060 0.125 0.015 -0.055 -0.012 0.030P-value 0.224 0.012 ** 0.761 0.273 0.806
Lean Hogs Estimate -0.696 0.041 0.468 0.013 0.024 0.480P-value <2e-16 *** 0.266 <2e-16 *** 0.730 0.505
Live Cattle Estimate 0.019 -0.016 -0.058 -0.016 0.004 0.003P-value 0.734 0.758 0.308 0.744 0.943
Corn Estimate 0.006 0.012 -0.001 -0.040 0.011 0.002P-value 0.908 0.827 0.992 0.430 0.820
Soybeans Estimate -0.073 0.100 0.607 0.027 0.096 0.440P-value 0.05 ** 0.011 ** <2e-16 *** 0.471 0.011 **
Soybean Oil Estimate -0.111 0.174 0.591 0.031 0.055 0.410P-value 0.004 *** 1.14e-05 *** < 2e-16 *** 0.421 0.146
Wheat CBOT Estimate -0.255 0.144 0.365 0.000 -0.021 0.210P-value 2.04e-08 *** 0.002 *** 4.63e-15 *** 0.995 0.640
Wheat KCBOT Estimate -0.004 0.044 0.022 -0.027 -0.007 0.004P-value 0.936 0.385 0.673 0.587 0.886
Note: The table show the results of regression:
Futures Returni,t = α +Σpim=0βm∆Inventory proxyi,t−m +Σ
qin=0γnxj,t−n +ΣKκ=0δκCIT F T otali,t−κ + εi,t
∆Futures returnsi ,t represents the return of the rolled futures prices series of commodity i in week t.
CIT F T otalt is the ratio of aggregate index flows to aggregate open interests for the 12 agricultural commoditiesmonitored by the CFTC Supplemental Report.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate.∆Dollarrepresents the dollar index return. ∆HP refers to the change in hedging pressure, which is calculated from variation ofthe ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregatehedging position. , ∆Inventory proxyi refers to inventory shock for commodity i in week t. The inventory proxy isobtained from the cumulative return of a strategy shorting the first-nearby futures contracts and longing the futurescontract delivering one year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the abovestrategy between two consecutive dates of the CFTC supplemental report.
The number of lag for CITF Total is chosen according to the AIC/BIC and the co-correlation test. Weekly data from 2006 to 2014are used. Both coefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)All variables are standardizedNumber of observation: 417
91
Table 4.7: Bi-monthly regression results comparison (a)
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT -
Wheat KCBOT
Table 4.8: Bi-monthly regression results comparison (b)
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Note: The blue boxes indicate the case where index flows have significant positive impact on the inventory proxy (uppertable) or the futures price return (bottom table). The retained significance threshold is the 90% level.
Both the impact of contemporaneous flows (lag=0) and lagged flows (-1) (lag=1) are reported. Sign – refers to thenegative identified relation. The sign is positive unless otherwise reported.
Three index indicators are used: CITF is the net weekly index flows divided by lagged Open Interest for individualcommodities; ETF refers to the inflows into the three ETF tracking generalist commodity indices; and CITF totalrepresents the ratio of weekly net aggregate index flows to lagged aggregate open interests for the 12 agriculturalcommodities. The inventory proxy is obtained from the cumulative return of a strategy shorting the first-nearby futurescontracts and longing the futures contract delivering one year after the first-nearby contract. Inventory proxyi iscomputed as the return of the above strategy between two consecutive dates of the CFTC supplemental report.
Return represents the return of the rolled futures prices series for each individual commodity.Prices and flows are observed every other Thursday (day of the week where the CFTC Supplemental data is published).
92
Table 4.9: Monthly regression results comparison (a)
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Table 4.10: Monthly regression results comparison (b)
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Note: The blue boxes indicate the case where index flows have significant positive impact on the inventory proxy (uppertable) or the futures price return (bottom table). The retained significance threshold is the 90% level.Both the impact of contemporaneous flows (lag=0) and lagged flows (-1) (lag=1) are reported. Sign – refers to thenegative identified relation. The sign is positive unless otherwise reported.
Three index flows indicators are used: CITF is the net weekly index flows divided by lagged Open Interest for individualcommodities (from the CFTC Supplemental Report); ETF refers to the inflows into the three ETF tracking generalistcommodity indices; and CITF total represents the ratio of weekly net aggregate index flows to lagged aggregate openinterests across the 12 agricultural commodities monitored by the CFTC Supplemental Report. The inventory proxy isobtained from the cumulative return of a strategy shorting the first-nearby futures contracts and longing the futurescontract delivering one year after the first-nearby contract. Inventory proxyi is computed as the return of the abovestrategy between two the last trading dates of each month.
Return represents the return of the rolled futures prices series for each individual commodity.Prices and flows are observed at the last trading day of each month.
93
Table 4.11: Specific period study: 06/2006-30/06/2008 (Commodities’ boom) a
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Table 4.12: Specific period study: 06/2006-30/06/2008 (Commodities’ boom) b
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffee -CottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Note:We repeat the analyses above for four sub-samples: Table 4.11 and 4.12: 06/2006-30/06/2008 (commodities’boom), Table 4.13 and 4.14: 07/2008-31/01/2009 (commodities’ collapse), Table 4.15 and 4.16: 02/2009-31/03/2011(recovery) and Table 4.17 and 4.18: 04/2011-2014 (fall in commodities prices).The blue boxes indicate the case where index flows have significant positive impact on the inventory proxy (upper table)or the futures price return (bottom table). The retained significance threshold is the 90% level.Both the impact of contemporaneous flows (lag=0) and lagged flows (-1) (lag=1) are reported. Sign – refers to thenegative identified relation. The sign is positive unless otherwise reported.Three index flows indicators are used: CITF is the net weekly index flows divided by lagged Open Interest for individualcommodities; ETF refers to the inflows into the three ETF tracking generalist commodity indices; and CITF totalrepresents the ratio of net weekly aggregate index flows to lagged aggregate open interests across the 12 agriculturalcommodities monitored by the CFTC Supplemental Report.The inventory proxy is obtained from the cumulative return of a strategy shorting the first-nearby futures contracts andlonging the futures contract delivering one year after the first-nearby contract. Inventory proxyi is computed as thereturn of the above strategy between two consecutive dates of the CFTC supplemental report.Return represents the return of the rolled futures prices series for each individual commodity.
94
Table 4.13: Specific period study: 07/2008-31/01/2009 (Commodities’ collapse) a
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Table 4.14: Specific period study: 07/2008-31/01/2009 (Commodities’ collapse) b
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
95
Table 4.15: Specific period study: 02/2009-31/03/2011 (Recovery) a
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Table 4.16: Specific period study: 02/2009-31/03/2011 (Recovery) b
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
Cocoa -CoffeeCottonSugar
Feeder Cattle -Lean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
96
Table 4.17: Specific period study: 04/2011-2014 (Fall in commodities prices) a
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
Table 4.18: Specific period study: 04/2011-2014 (Fall in commodities prices) b
CITF on proxy-inventory ETF on proxy-inventory CITF Total on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CocoaCoffeeCottonSugar
Feeder CattleLean HogsLive Cattle
CornSoybeans
Soybean OilWheat CBOT
Wheat KCBOT
97
Table 4.19: No-agricultural commodities (a)
Weekly Bi-monthly MonthlyETF on proxy-inventory ETF on proxy-inventory ETF on proxy-inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
AluminiumBrent
CopperGold
NickelSilverZinc
Table 4.20: No-agricultural commodities (b)
Weekly Bi-monthly Monthly
ETF on returns ETF on returns ETF on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
AluminiumBrent
CopperGold
NickelSilverZinc
Note: The tables 4.19 and 4.20 summarize the results of the above regressions for seven non-agricultural commodities:Aluminum, Brent, Copper, Gold, Nickel, Silver and Zinc. The blue boxes indicate the case where index flows havesignificant positive impact on the inventory proxy (upper table) or on the futures price returns (bottom table). Theretained significance threshold is the 90% level.
Both the impact of contemporaneous flows (lag=0) and lagged flows (-1) (lag=1) are reported. Sign – refers to thenegative identified relation. The sign is positive unless otherwise reported.
As index flows data is not available for non-agricultural commodities, only one specification using ETF flows is used.
ETF refers to the inflows into the three ETF tracking generalist commodity indices.
The inventory proxy is obtained from the cumulative return of a strategy shorting the first-nearby futures contracts and
longing the futures contract delivering one year after the first-nearby contract. Inventory proxyi is computed as the
return of the above strategy between two consecutive releasing dates of the CFTC report (weekly specification), between
or between the last trading dates of each month (monthly specification).
Return represents the return of the rolled futures pries of every commodity.
The analysis is conducted with data ranging from 2006 to 2014 and successively uses alternatively weekly, bi-monthly or
month data.
For the weekly specification, we observe data every Thursday (releasing date of the CFTC report). For the bi-monthly
specification, we observe data every other Thursday, and for the monthly specification, the last Thursday of every month.
98
Table 4.21: Correlation between Stock-to-use and different inventory proxy methods
Proxy using spot prices Proxy using M2/M1 Proxy using one-year-distance contracts
Stock-to-useCorn 0.01 (0.90) 0.30 (0.003) 0.26 (0.01)
Soybean -0.13 (0.13) 0.56 (7.5e-09) 0.34 (0.00008)Wheat -0.08 (0.34) 0.12 (0.28) 0.76 (<2.2e-16)
Note: This table reports the correlation coefficient between monthly calendar spreads changes and monthly stock-to-userevisions (from the USDA WASDE report) for corn, soybean and wheat. From May to August, the data refer to endinginventory and consumption estimates for the following campaign. From September to April, the data refer to the endinginventory and consumption estimations for the current campaign.
Three different spreading strategies are considered: 1) The inventory proxy using the first-nearby and spot prices isobtained from the return of a strategy longing the first-nearby futures contracts and shorting the spot price. 2) Theinventory proxy using first and second nearby futures prices is calculated from the return of a strategy longing thesecond-nearby futures contracts and shorting the first-nearby futures contracts. 3) The inventory proxy with one-yeardistance contracts is obtained from return of a strategy longing the shorting the first-nearby futures contract and longingthe contract delivering exactly one year after the first-nearby contract.
The correlation is calculated based on the monthly variations of the stock-to-use ratio and the monthly returns of thedifferent spreading strategies from 2006 to 2014.The number in parenthesis is the p-value of the test of significance of the correlation coefficient.
99
Table 4.22: Relationship between CITF and inventory proxy using alternative calendar spreads a)
M1/spot∆Inventory(−1) CITF CITF(-1) ∆HP ∆Dollar ∆r R2
CornEstimate 0.400 0.060 0.160 0.020 0.040 0.180P-value <2e-16 *** 0.340 0.001 *** 0.680 0.390
SoybeanEstimate -0.040 -0.034 0.140 0.020 0.120 0.040P-value 0.420 0.510 0.008 *** 0.750 0.013 **
WheatEstimate -0.030 0.090 0.150 0.060 0.060 0.040P-value 0.490 0.06* 0.230 0.230 0.190
∆Inventory(−1) CITF total CITF total(-1) ∆HP ∆Dollar ∆r R2
CornEstimate 0.390 0.060 0.150 0.030 0.040 0.190P-value <2e-16 *** 0.180 0.001*** 0.550 0.360
SoybeanEstimate -0.037 0.080 0.110 7.000 0.120 0.040P-value 0.460 0.110 0.03** 0.890 0.01**
WheatEstimate -0.030 0.120 0.009 0.150 0.070 0.060 0.050P-value 0.510 0.026 ** 0.870 0.003*** 0.140 0.210
∆Inventory(−1) ETF flows ETF flows(-1) ∆HP ∆Dollar ∆r R2
CornEstimate 0.410 0.130 0.150 0.020 0.040 0.200P-value <2e-16 *** 0.006 *** 0.0008 *** 0.640 0.390
SoybeanEstimate -0.040 -0.060 0.130 -0.020 0.120 0.040P-value 0.380 0.220 0.007 *** 0.720 0.013
WheatEstimate -0.030 0.020 -0.050 0.170 0.040 0.060 0.030P-value 0.480 0.740 0.300 0.0007 *** 0.410 0.220
Note: The table show the results of regression:
∆Inventory proxyi,t = α +Σpim=0βm∆Inventory proxyi,t−m +Σ
qin=0γnxj,t−n +ΣKκ=0δκIndex f lowsi,t−κ + εi,t
∆Inventory proxyi ,t refers to inventory shock for commodity i in week t. The inventory proxy is obtained from thecumulative return of a strategy shorting the first-nearby futures contracts and longing the futures contract deliveringone year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the above strategy betweentwo consecutive dates of the CFTC supplemental report. The control variables Xi , t − n = (∆r,∆Dollar,∆HP ) . ∆r is thereturn in the six-month government bill rate. ∆Dollar represents the dollar index return. ∆HP refers to the change inhedging pressure, which is calculated from variation of the ratio of difference between long and short hedging position(commercial position) in futures markets to the aggregate hedging position.
Three specifications of index flows are used: CITF is the net weekly index flows for individual commodities (CFTCSupplemental Report); ETF refers to the index flows into the three main ETF tracking generalist commodity indices;and CITF total represents the ratio of aggregate index flows to aggregate lagged open interests across the 12 agriculturalcommodities.
The number of lag is chosen according to the AIC/BIC. Weekly data from 2006 to 2014 are used. Both coefficient and P-value arepresented in the table. (*** significant at 1%, ** significant at 5Allvariablesarestandardized.Numberof observations : 417
This table (M1-Spot), the dependent variable is the difference in return between first-nearby rolled futures prices series and spotprices.In the following table 4.23 (M2-M1), the dependent variable is the difference in return between second-nearby and first-nearbyrolled futures prices series.
100
Table 4.23: Relationship between CITF and inventory proxy using alternative calendar spreadsb)
M1-M2∆Inventory(−1) CITF CITF(-1) ∆HP ∆Dollar ∆r R2
CornEstimate -0.037 -0.033 0.18 0.003 0.015 0.031P-value 0.44 0.527 0.0005*** 0.94 0.759
SoybeanEstimate -0.19 0.056 -0.145 0.003 0.01 0.054P-value 0.0009*** 0.285 0.0056*** 0.948 0.83
WheatEstimate -0.144 -0.005 0.107 0.056 -0.584 0.04P-value 0.003*** 0.925 0.03** 0.258 0.23
∆Inventory(−1) CITF total CITF total(-1) ∆HP ∆Dollar ∆r R2
CornEstimate -0.036 -0.03 0.173 0.002 0.01 0.03P-value 0.463 0.549 0.0004*** 0.97 0.81
SoybeanEstimate -0.185 0.036 -0.133 0.004 0.004 0.06P-value 0.0001*** 0.228 0.007*** 0.93 0.94
WheatEstimate -0.143 -0.046 -0.044 0.108 0.07 -0.07 0.045P-value 0.003*** 0.368 0.387 0.027** 0.16 0.17
∆Inventory(−1) ETF ETF(-1) ∆HP ∆Dollar ∆r R2
CornEstimate 0.007 -0.015 0.07 -0.01 -0.01 0.06P-value 0.98 0.775 0.158 0.84 0.78
SoybeanEstimate -0.13 0.03 -0.3 -0.012 0.004 -0.02 0.03P-value 0.009*** 0.6 0.013** 0.02** 0.94 0.77
WheatEstimate -0.154 -0.03 0.09 -0.05 -0.06 0.04P-value 0.002*** 0.56 0.06** 0.27 0.23
101
Table 4.24: Results with bi-monthly and monthly data using M1/spot spread as inventory proxy
By-monthly
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
Corn -Soybeans -
Wheat
CITF on inventory ETF on inventory CITF Total on inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CornSoybeans
Wheat
Monthly
CITF on returns ETF on returns CITF Total on returnslag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CornSoybeans
Wheat
CITF on inventory ETF on inventory CITF Total on inventorylag=0 lag=1 lag=0 lag=1 lag=0 lag=1
CornSoybeans
Wheat
Note: The blue boxes indicate the case where, index speculation has significant impact on the inventory (upper table) oron the futures prices (bottom table).
Both the impact of contemporaneous flows (lag=0) and lagged flows (-1) (lag=1) are reported. Sign – refers to the sign ofthe identified relation. The sign is positive unless otherwise reported.Three index indicators are used: CITF is the index flows for relative commodity; ETF refers to the index investing inthe three ETF tracking generalist commodity indices; and CITF total represents the ratio of aggregate index flows toaggregate open interests for the total 12 agricultural commodities.Calendar spread is the return of a strategy longing the first-nearby futures contracts and shorting the spot price.Spot prices return represents the return of the spot pries of each commodity.
For bi-monthly data, we observe data every other Thursday; and for monthly data, the last Thursday of every month.
102
Table 4.25: The impact of CITF using stock-to-use ratio
CITF total Intercept R-squared
Stock-to-useWheat 0.31(0.26) 0.001(0.78) 0.02Corn -0.17(0.4) -0.0005(0.87) 0.01
Soybean 0.07(0.65) -0.001(0.49) 0.004
CITF Intercept R-squared
Stock-to-useWheat 0.13(0.11) 0.02(0.72) 0.10Corn -0.13(0.15) -0.08(0.34) 0.05
Soybean 0.06(0.62) -0.06(0.30) 0.01
ETF Intercept R-squared
Stock-to-useWheat 0.02(0.79) 0.01(0.75) 0.001Corn -0.10(0.27) -0.08(0.38) 0.03
Soybean -0.13(0.21) -0.09(0.41) 0.08
Note:The regression specification is:
∆stock − to −uset = α + βIndexf lowsi,t + εt
Stock-to-use ratio is the ratio of inventory to total use, taken from the monthly USDA WASDE report. From May to
August, the data refer to ending inventory and consumption estimates for the following campaign. From September to
April, the data refer to the ending inventory and consumption estimations for the current campaign.
Three specifications of index flows are used: CITF is the net weekly index flows for individual commodities (CFTC
Supplemental Report); ETF refers to the index flows into the three main ETF tracking generalist commodity indices;
and CITF total represents the ratio of aggregate index flows to aggregate lagged open interests across the 12 agricultural
commodities.
Both coefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%)
All variables are standardized
Number of observation: 99 monthly observations from 2006 to 2014;
103
Table 4.26: Relationship between prices returns and CITF using spot prices
∆Inventory(−1) CITF CITF(-1) ∆HP ∆Dollar ∆r R2
CornEstimate -0.02 -0.08 0.5 0.05 0.03 0.26P-value 0.65 0.06* <2e-16*** 0.31 0.42
SoybeanEstimate -0.14 0.06 -0.04 0.61 0.01 0.06 0.41P-value 0.00004*** 0.15 0.29 <2e-16*** 0.79 0.06*
WheatEstimate 0.16 0.04 -0.09 0.31 -0.05 -0.03 0.15P-value 0.001*** 0.43 0.04** 1.4e-10*** 0.32 0.54
∆Inventory(−1) CITF total CITF total (-1) ∆HP ∆Dollar ∆r R2
CornEstimate -0.03 0.09 -0.06 0.48 0.07 0.04 0.26P-value 0.53 0.06* 0.21 <2e-16*** 0.13 0.4
SoybeanEstimate -0.15 0.11 0.6 0.04 0.08 0.41P-value 0.0001*** 0.008*** <2e-16*** 0.34 0.05**
WheatEstimate 0.14 0.14 0.31 -0.01 -0.04 0.16P-value 0.003*** 0.003*** 1.9e-10*** 0.78 0.44
∆Inventory(−1) ETF flows ETF flows (-1) ∆HP ∆Dollar ∆r R2
CornEstimate -0.02 0.02 0.5 0.06 0.04 0.25P-value 0.68 0.62 <2e-16*** 0.16 0.38
SoybeanEstimate -0.14 0.08 0.62 0.02 0.07 0.41P-value 0.001*** 0.04** <2e-16*** 0.66 0.07*
WheatEstimate 0.15 0.02 0.32 -0.05 -0.04 0.15P-value 0.001*** 0.72 4.1e-11*** 0.32 0.43
Note: The table show the results of regression:
Futures Returni,t = α +Σpim=0βm∆Inventory proxyi,t−m +Σ
qin=0γnxj,t−n +ΣKκ=0δκIndex f lowsi,t−κ + εi,t
∆P ricesreturni,t ,t represents the return of the rolled futures prices series of commodity i in week t.
The control variables Xi ,t − n = (∆r,∆Dollar,∆HP )). ∆r is the return in the six-month government bill rate.∆Dollar
represents the dollar index return. ∆HP refers to the change in hedging pressure, which is calculated from variation of
the ratio of difference between long and short hedging position (commercial position) in futures markets to the aggregate
hedging position. , ∆Inventory proxyi refers to inventory shock for commodity i in week t. The inventory proxy is
obtained from the cumulative return of a strategy shorting the first-nearby futures contracts and longing the futures
contract delivering one year after the first-nearby contract. ∆Inventory proxyi is computed as the return of the above
strategy between two consecutive dates of the CFTC supplemental report.
Three specifications of index flows are used: CITF is the net weekly index flows for individual commodities (CFTC
Supplemental Report); ETF refers to the index flows into the three main ETF tracking generalist commodity indices;
and CITF total represents the ratio of aggregate index flows to aggregate lagged open interests across the 12 agricultural
commodities.
The number of lag for CITF is chosen according to the AIC/BIC and the co-correlation test. Weekly data from 2006 to 2014 are
used. Both coefficient and P-value are presented in the table. (*** significant at 1%, ** significant at 5%, *significant at 10%) All
variables are standardized. Number of observation: 417
104
Appendix
Table 4.27: Table of AIC for choosing lag number in the regression
AICCITF on proxy-inventory ETF on proxy-inventory CITF total on proxy-inventory
CITF CITF +CITF(-1) CITF(-1) ETF ETF + ETF(-1) ETF(-1) CITF Total CITF Total +CITF Total(-1) CITF Total(-1)Cocoa -33.07 -31.7 -32.07 -32.6 -28.69 -32.59 -34.65 -31.63 -33.51Coffee -15.48 -13.51 -15.51 -19.33 -14.23 -18.7 -13.99 -13.89 -11.01Cotton 4.06 2.65 2.06 1.69 3.47 2.07 1.47 4.49 2.5Sugar -40.36 -37.69 -38.41 -26.87 -25.62 -18.4 -51.69 -50.7 -47.1
Feeder Cattle 2.44 2.65 4.16 1.42 2.7 3.7 1.99 3.04 2.29Lean Hogs -2.15 -2.53 -3.91 -3.58 -2.29 -2.79 -1.69 -3.91 -3.08Live Cattle -6.36 -3.24 -2.64 0.8 0.97 0.89 -1.5 0.39 -1.03
Corn -12.04 -10.05 -10.56 -3.79 0.21 -3.06 -11.93 -9.61 -11.26Soybeans -36.17 -35.86 -34.02 -28.1 -30.1 -27.8 -34.05 -30.7 -29.6
Soybean Oil -58.52 -56.8 -55.14 -53.57 -51.36 -50.76 -51.53 - 49.31 -50.9Wheat CBOT -14.31 -12.63 -12.03 -13.71 -11.68 -13.58 -9.83 -14.66 -13.39
Wheat KCBOT -10.33 -9.9 -13.63 -12.17 -10.56 -11.1 -11.25 -10.2 -9.01
AICCITF on returns ETF on returns CITF total on returns
CITF CITF +CITF(-1) CITF(-1) ETF ETF + ETF(-1) ETF(-1) CITF Total CITF Total +CITF Total(-1) CITF Total(-1)Cocoa -25.75 -25.7 -22.84 -13.92 -18.68 -17.1 -2.07 -3.27 -3.14Coffee -38.55 -39.05 -29.02 -29.01 -26.8 -26.29 -31.95 -32.91 -26.04Cotton -14.92 -13.02 -12.01 -4.59 -6.58 -4.58 7.3 -7.1 -6.36Sugar -6.96 -5.54 -2.93 2.48 2.16 4.01 -3.96 -2.54 -3.64
Feeder Cattle -119.49 -207.45 -203.45 -490.91 -492.26 -495.35 -213.98 -206.44 -211.45Lean Hogs -10.51 -9.44 -7.53 -4.81 -10.79 -9.16 -10.94 -9.02 -9.43Live Cattle 3.42 4.45 4.15 4.4 4.7 4.95 4.2 4.51 5.5
Corn 4.01 5.66 4.25 7.5 7.9 7.76 5.65 5.89 5.93Soybeans -46.86 -42.9 -45.9 -23.82 -20.96 -19.5 -30.05 -26.13 -24.98
Soybean Oil -25.31 -23.81 -25.3 -29.33 -28.22 -23.85 -33.9 -33.31 -33.33Wheat CBOT -189.58 -188.79 -178.54 -266.42 -260.92 -194.22 -189.97 -177.82 -189.79
Wheat KCBOT 4.42 4.61 4.77 6.81 7.81 8.66 3.14 6.16 4.34
105
Bibliography
[1] Algieri, B. Feb.2013, A roller Coaster Ride: An empirical investigation of the main diriversof wheat price. ZEF - Discussion papers on Development Policy No. 176 .
[2] Aulerich, N. M., Irwin, S. H., and Garcia, P.2013, Bubbles, Food prices, and Speculation: Ev-idence from the CFTC’s daily large trader data files. The Economics of Food Prices Volatility,University of Chicago Press.
[3] Büyüksahin, B., and Harris, J. H.2011, Do speculators drive crude oil futures prices. EnergyJournal 32.2 , pp. 167-202.
[4] Beckmann, J., and Czudaj, R.2012, Oil price and US dollar exchange rates dynamics. Univer-sity of Duisburg-Essen.
[5] Belke, A., and Potrafke, N. Sep.2012, Does government ideology matter in monetary policy?A data analysis for OECD countries. Journal of International Money and Finance Volume 31(Issue 5), pp. 1126-1139.
[6] Bessembinder, H., and Seguin, P. J. Mar. 1993, Price volatility, Trading Volume, and MarketDepth: Evidence from futures markets. Journal of Financial and Quantitative Analysis Volume28 (Issue 01), pp. 21-39.
[7] Bichetti, D., and Nicolas, M.2012, The synchronized and long-lasting structural change oncommodity markets: Evidence from high frequency data.United Nations Conference on Tradeand Development - UNCTAD.
[8] Borin, A., and Di Nino, V. 2012, The role of financial investments in agricultural commodityderivatives markets. Bank of Italy Temi di discussione (Working Paper) No 849 .
[9] Borovkova, S. Jun.2003, Detecting market transitions: from backwardation to contango andback. EPRM pp. 46-49.
[10] Brunetti, C., and Reiffen, D.2011, Commodity index trading and hedging costs. The FederalReserve Board, Finance and Economics Discussion series 2011-57 .
[11] Burnetti, C., and Buyuksahin, B.2009, Is speculation destabilizing? Available at SSRN1393524 .
106
[12] Cheng, I.H., Kirilenko, A., and Xiong, W.2012, Convective risk flows in commodity futuresmarkets. National Bureau of Economic Research, working paper.
[13] Fama, E. F., and French, K. R.1988, Business cycles and the behavior of metals prices. TheJournal of Finance 43.5 , pp. 1075-1093.
[14] Feldman, B., and Till, H. 2006, Backwardation and commodity futures performance: Ev-idence from evolving agricultural markets. The Journal of Alternative Investments 9.3 , pp.24-39.
[15] Frenk, D., and Turbeville, W. Oct.2011,Commodity index traders and the boom/bust cyclein commodities prices.BETTER MARKETS working paper.
[16] Gibert, C. L., and Pfuderer, S. May.2012, Index funds do impact agricultural prices. Money,Macro and Finance Study Group workshop on commodity markets London, Bank of England ,Vol. 25.
[17] Guilleminot, B., Ohana, J.J, and Ohana, S. Nov.2014, The interaction of speculators andindex investors in agricultural derivatives markets. Agricultural Economics Vol.45,Issue 6,pages 767-792.
[18] Hamilton, J. D., and Wu, J. C. May.2013, Effects of index-fund investing on commodityfutures prices. Working paper.
[19] Irwin, S. H., and Sanders, D. R. 2010, The impact of index and swap funds in commod-ity futures markets. Report for the Organization on Economic Co-Operation and Development(OECD) .
[20] Juvenal, L., and Petrella, I. Oct.2011, Speculation in the oil market. FRB of St. Louis WorkingPaper No. 2011-027E.
[21] Kaufmann, R. K., and Ullman, B. Jul.2009, Oil prices, speculation, and fundamentals: Inter-preting causal relations among spot and futures prices. Energy Economics Volume 31, Issue4, pp. 550-558.
[22] Keynes, J. M. 1930, A treatise on money. The Applied Theory of Money Vol. 2.
[23] Killian, L., and Murphy, D.2010, The role of inventory and speculative trading in the globalmarket for crude oil. Working paper.
[24] Lizenberger, R. H., and Rabinowitz, N.1995, Backwardation in oil futures markets: Theoryand empirical evidence. The Journal of Finance 50.5, pp. 1517-1545.
[25] Lombardi, M. J., and Robays, V. May.2011, Do financial investors destabilize the oil prices?ECB Working Paper N. 1346.
[26] Mou, Y.2010, Limits to arbitrage and commodity index investment: Front- running the Gold-man Roll. Available at SSRN 1716841.
107
[27] Peersman, G., and Robays, I. V. Oct.2009, Oil and the Euro area economy. Economy PolicyVolume 24, Issue 60, pp. 603-651.
[28] Pierru,A. and Babusiaux, D. April 2010, Speculation without oil stockpiling as a signature:A dynamic perspective. Center for Energy and Environmental Policy Research (10-004).
[29] Power, G. J., and Turvey, G. G. 2011, Revealing the impact of index traders on commodityfutures markets. Applied Economics Letters 18.7 , pp. 621-626.
[30] Roache, S. K. Sep.2008, Commodities and the Market Price of risk. IMF Working Paper.
[31] Rothig, A. 2008, The impact of backwardation on hedgers’ demand for currency futures con-tracts: Theory versus empirical evidence. Darmstadt discussion papers in economics No.190.
[32] Russi, L. Dec.2012, A second look at the 2007-08 food price crisis: Considering the impactof endogenous dynamics on food prices. CITYPERC Working Paper Series No. 2012/01 .
[33] Sigl-Grub, C., and Schiereck, D. 2010, Speculation and nonlinear price dynamics in com-modity markets. Investment Management and Financial Innovations Volume 7 Issue 1.
[34] Silvennoinen, A., and Thorp, S. 2012, Financialization, crisis and commodity correlationdynamics. Journal of International Financial Markets, Institutions and Money.
[35] Sornette, D., Woodard, R., and Zhou, W.-X. April 2009, The 2006-2008 oil bubble: Evidenceof speculation, and prediction. Physica A: Statistical Mechanics and its Applications Volume388, Issue 8, pp.1571-1576.
[36] Stoll, H. R., and Whaley, R. E. 2010, Commodity index investing and commodity futuresprices. Journal of Applied Finance 20.1 , pp. 7-46.
[37] Tang, K., and Xiong, W. 2012, Index investment and the Financialization of commodities.Financial Analysts Journal 68,pp. 54-74.
[38] Telser, L. G.1958, Futures trading and the storage of cotton and wheat. The Journal of PoliticalEconomy 66.3 , pp. 233-255.
[39] Till, H. Nov.2009, Has there been excessive speculation in the US oil futures markets?EDHEC-Risk Institute Position Papers.
[40] Tse, Y., and Williams, M. R. 2013, Does index speculation impact commodity prices? Anintraday analysis. Financial Review 48.3, pp. 365-383.
108
5The impact of Central Banks’ announcements on
commodities markets
Abstract
In a context of financialization of commodities markets, this paper assesses the impact ofunexpected monetary announcements on commodities prices. In an era of Zero InterestRate Policy, monetary policy surprises are characterized alternatively through forwardinterest rates, dollar index and inflation swap rate variations during the days withFederal Reserve announcements. Overall, we find that monetary surprises have sizableeffects on commodities markets on the day of the announcement. Negative monetarypolicy surprises have instantaneous as well as delayed effects on commodity prices,which exhibit a negative trend up to ten days after the event. In an additional analysis,we investigate the channels through which monetary policy shocks affect commodityprices. We reject explanations through the inventory or investors’ portfolio rebalancingchannels (Frankel, 2006) and provide evidence in support of the “dollar debasement”channel, whereby commodities prices appreciate in terms of a debased currency in orderto maintain their absolute level in terms of major foreign currencies. Our results haveseveral implications. First, we provide direct evidence that commodity markets maybe impacted by monetary policy announcements. Second, this work may provide anexplanation for the increased correlation between dollar, traditional investment classesand commodity prices over the last decade.
Keywords: Unexpected monetary policies, Commodity prices, Dollar index, Inflation swapsJEL Classification: E44, E58, G23
5.1 Introduction
Commodities prices have sharply increased from March 2009 to the beginning of 2011, onthe back of an exceptionally accommodative monetary policy from the Federal Reserve.Supply and demand fundamentals are probably insufficient to explain this commodities’boom. As a matter of fact, the development of index investing has drastically modified thenature and the motivations of investors in commodities paper markets. Monetary policy,the problematic of inflation hedging and global liquidity conditions in the financial systemare now part of what commodity investors look at when making their decisions.
109
The interaction between commodities prices and monetary policy works both ways. Onthe one hand, Garner (1989) proposes that commodity prices might be a policy targetor information variable. Commodity prices contain useful information for future pricesmovements. As a result, commodity prices determine the Federal Reserve monetary policy.On the other hand, the Federal Reserve can intervene in the commodity markets directlyor indirectly, through the setting of its policy rate, through its unconventional monetarypolicies (Quantitative Easing in particular), and lastly through controlling the activities offinancial institutions who usually take a large stake in commodities markets1. The FederalReserve’s interest rate policy impacts commodities prices through different channels, out-lined in Frankel (2006). Firstly, changes in the short-term rate modify the production andstorage incentives given to commodity players (an increase in the short-term rate promptsplayers to destore and produce, resulting in an increase in the short-term supply and adrop in the spot prices). Secondly, short-term rates determine the relative attractiveness offixed income assets relative to physical assets (an increase in the short-term rate increases,everything else equal, the attractiveness of bonds at the expense of real assets).
In recent years there has been a great amount of work furthering the understanding of thebi-directional interaction between monetary policies and commodity markets. Previouswork has examined in particular the relationships between monetary policy variables, suchas the interest rate (Gospodinov & Jamali, 2013), the exchange rate or the money supply, andcommodity prices. (Anzuini, Lombardi, & Pagano, 2013) find that a monetary policy shockmeasured by a reduction in the federal funds rate leads to an increase in the commodityprice index. (Browne Cronin, 2010) show that commodity prices are proportional to themoney supply in the long run and equilibrium relationship exists between commodityprices and money supply. (Glick & Leduc, 2011) indicate that, in response to the monetaryannouncements, long-term interest rate falls, U.S. dollar depreciates and commoditiesprices decline. However, negative monetary policy surprises result in a joint increase oflong-term yields, dollar index and commodities prices. Previous studies may differ in theirconclusions due to the differences in methodology and data sample used.
To avoid the problem of endogeneity in the relationship between commodity prices andmonetary policy, we use an event-study methodology. Unlike previous event-study ap-proaches to monetary policy announcements, we characterize the “monetary surprise” asthe unexpected event-day change in three key financial variables impacted by the Fed’smonetary policy: dollar index, forward interest rates and swap inflation rates. In a contextof zero interest rate policy (ZIRP), these variables better capture monetary surprise than thesurprise change in the target rate, which is the traditional measure used in the literature,see for example, (Gospodinov & Jamali, 2013). (Gurkaynak, Sack, Swanson, 2005), (Kuttner
1For example, in an advance notice of proposed rulemaking in 2014, the Federal Reserve proposed to restrict theactivities of financial holding companies so as to control any potential risk that might pose a threat to financial stabilityfrom the commodity markets.
110
2001) (Hayo et al. 2011).
After identifying unexpected monetary policy events, we explore the effect of monetarypolicy on commodity prices (global GSCI Index and three sectoral commodity indices).The key findings can be summarized as follows. Firstly, only monetary policy surprisescaptured by the dollar exchange rate and the swap inflation rate impact commodities prices,the role of forward interest rate surprises being not significant. Second, a negative centralbank announcement has an instantaneous as well as delayed effect on commodity prices,which display a decreasing trend after unexpected negative monetary policy events. Third,commodity markets have become highly sensitive to monetary policy surprises, only from2006 onwards; before that date, no significant impact is observed. This fact supports theview that commodity markets have entered a process of financialization and integration intobroad capital markets from 2006 onwards. Finally, we also investigate the way commoditystorers and investors react to unexpected monetary policy shocks (flows into the threemain exchange-traded funds (ETF) tracking generalist commodity indices, speculative andindex flows as reported by the CFTC), with the result that storers and index investors arelargely insensitive and commercial traders sensitive to monetary policy surprises. Instead,we provide evidence in support of the “dollar debasement” channel, whereby commoditiesprices appreciate in terms of a debased currency in order to maintain their absolute level interms of major foreign currencies.The remainder of the paper is organized as follows: section 2 presents the theoreticaland empirical literature on the relationship between commodity market and monetarypolicy; section 3 introduces some preliminary salient facts on commodity markets andthe Federal Reserve’s monetary policy and explains the methodology to capture monetarypolicy surprises; section 4 presents the results of the regressions of commodity prices onmonetary policy surprises; section 5 conducts some robustness tests; section 6 concludes.
5.2 Literature on the interaction between commodity prices andmonetary policy
Theoretical aspects
In a seminal paper, Frankel (1986) points out that “macroeconomic policy may be asimportant a source of fluctuation in agricultural prices as the traditional microeconomicfactors”. He argues that the exchange rate and the interest rate are the main transmissionmechanisms of monetary policy to commodity prices. Based on an overshooting model,Frankel (2006) proposes three channels through which interest rate variations may havean impact on commodities prices: 1. lower interest rates increase the incentive to defer theextraction of depletable commodities, hereby exerting an upward pressure on the prices 2.lower interest rates also increase the incentive to borrow in order to buy the commodity on
111
the spot market and pile up inventories, with the same effect as above on the spot prices3. lower rates finally reduce the attractiveness of fixed income assets at the benefit of realassets and in particular commodity contracts. In the same vein, Pindyck & Rotemberg(1990) attribute the co-movements of commodity prices to macroeconomic variables suchas interest rates, exchange rates etc.
Empirical research
Extensive empirical research exists on the impact of central bank actions on commoditymarkets. One way to resolve the endogeneity problem in the identification of the impactof central bank policies on commodity markets is the use vector autoregressive regression(VAR). Anzuini, Lombardi, & Pagano (2013) investigate the relationship between commod-ity prices and US monetary policy using standard VAR. They find that a laxer monetarypolicy increases commodity index prices, but that this impact is of little significance.Browne & Cronin (2007) support the causal role of money policy in the recent developmentof the commodity market using a VAR model. The opposite result is found by Thomson &Summers(2012), where they estimate VAR and error-correction models on transformed datato avoid possible problems of unit roots or co-integration in the data and find little evidenceof any impact of monetary policy on commodities prices.
Another way to deal with the endogeneity problem is the use of an event-study approach.Basistha & Kurov (2013) examine the impact of money policy surprises on energy pricesusing intraday, daily, and monthly data. They find an immediate response of energy prices tothe unexpected changes in Federal fund rates, especially in the unscheduled Federal OpenMarket Committee (FOMC) meeting; and no significant effects within a monthly horizon.Gospodinov & Jamali (2013) assess the response of commodities prices, convenience yieldsand trading positions to monetary policy surprises. They find that energy and metal futuresprices and traders’ positions respond to monetary policy surprises, concluding that theadjustment of the net long positions of hedgers and speculators appears to be a channelthrough which the monetary policy shocks are propagated to commodity price changes. TheFed’s target rate is considered as an important indicator of monetary policies in general;however, in an era of zero interest rate policy (ZIRP), this indicator cannot be used as aconsistent indicator of the monetary policy stance. Instead, inflexions in Central Banks’forward guidance or asset purchase programs become the key monetary policy eventswithin a ZIRP context. Glick & Leduc (2011) focus on the impact of the surprises relativeto the Bank of England’s and Federal Reserve’s asset purchase programs. They measurethese surprises through intra-day yield variations (across a large range of maturities) on theevent date. They show that positive surprises about these purchases led to lower long-terminterest rates and depreciations of the U.S. dollar and the British pound on announcementdays, while commodity prices generally declined despite this more stimulative financial
112
environment. They attribute the last surprising result to a signalling effect of monetarypolicy surprises on the state of the economy, leading investors to downgrade their U.S.growth forecasts and eventually triggering a decline in commodity prices.Instead of examining the impact on prices, the other strand of empirical research focuseson the influence of monetary policy on prices volatility. Hayo, Kutan, & Neuenkirch (2011)find an increase in commodity prices volatility due to target rate surprises and unorthodoxmonetary policy measures, where they identify a “calming” effect of communicationsduring financial crises. And in Batten, Ciner, & Lucey (2010), gold volatility is influencedby macroeconomic determinants, such as the business cycle and the monetary environment,but the results are not the same for other precious metals.Our paper contributes to the existing literature in three different ways. Firstly, it extendsthe analysis of monetary policy impacts to the post-crisis period (2008 to 2014), a uniqueperiod where commodities markets pursued their process of financialization and the zerointerest rate policy got more entrenched. Second, we introduce new measures of monetarypolicy surprises, using either the dollar index, forward interest rates or swap inflation ratesvariations on the days of important central bank announcements. This is novel as previousworks have focused on unexpected changes of the Fed’s policy rate or of the yields of longermaturities, which may prove a restrictive measure in a context of entrenched ZIRP. Finally,our study provides ancillary empirical evidence on the effect of monetary policy shocks onspeculators and index traders’ position, as well as inventory variations (captured by thechanges in the commodities’ futures prices term structure). This allows us to shed light onthe mechanisms through which monetary policy affects commodity prices.
5.3 Preliminary observation
General observation
In the aftermath of the 2008 financial crisis, the Federal Reserve has applied exceptionallyaccommodative monetary policies. Figure 5.1 and 5.2 display the joint evolution of theS&P GSCI index, the Eurodollar futures prices2, the dollar index3 and inflation swap.The launching dates of three important monetary policy operations are also represented– Quantitative Easing 1 (QE1) on November 25, 2008, QE2 on November 3, 2010, andQE3 on September 12, 2012. However, in most cases, Central Banks give first hints of theirintended actions several months before the launching date of the program. The dotted linesin figures refer to dates where web articles start evoking the possibility of a QE program inthe near future, hence when investors started incorporating the effect of such programs onthe market prices. The visual inspection of the dollar index and the GSCI Index trajectories
2The Eurodollar futures contract refers to a financial futures contract based upon the deposits denominated in USdollars at banks outside the US. 3-months term Eurodollar futures contracts are used in our analysis. Eurodollar futuresprices have a negative relationship with the interest rate.
3The dollar index measures the exchange rate of the dollar against a trade-weighted basket of foreign currencies.
113
reveals that dollar commodities prices generally evolve in opposite directions. QE1 seemsto have had an immediate depreciating effect on the dollar but only a deferred effect oncommodities prices, which bottomed only in the end of the first quarter of the year 2009.The program seems to have had a durable impact on the prices, which maintain theirrespective trends up to the fall of 2009. As for QE2, its debasement effect on the dollar andits stimulus effect on commodities prices seem to have been anticipated from the summer2010 and continue to manifest themselves up to the first quarter of the year 2011. Lastly,the possible impact of QE3 has been largely frontloaded, since the negative dollar trendand the positive GSCI trend, which originated in the summer 2012, stopped when the firsthints of the program were announced in the press These effects can be also observed ininflation swap rates, which evolve in sync with S&P GSCI prices. As a conclusion, it seemsthat the potential impacts of the successive QE programs have been weaker and weaker interms of size, more and more advanced in terms of timing and less and less durable in termsof impact range. However, the problem of endogeneity makes it impossible to draw firmconclusions on the impact of monetary policy at this level of analysis: there are plenty ofvariables other than monetary policy that could account for the evolution of commoditiesprices on the time horizons considered (from a quarter to one year).
Another observation concerns the correlation among the studied variables, including theGSCI return, the Eurodollar futures return, the dollar index return, and the inflationswap return. In Table 5.1, we observe that dollar index has exhibited a significant negativecorrelation with the GSCI index return after 2005. A significant positive relationshipis also found between inflation swap and GSCI returns. The Eurodollar futures returnhas a positive relation with the GSCI return, which means that commodities prices arenegatively correlated with interest rate variations; however, their correlation is not signif-icant according to our t-test. This suggests that commodity prices vary closely with othermacroeconomic indicators, although this correlation does not necessarily imply causality.
Observation with unexpected central bank announcements
A failure to distinguish unexpected monetary policy events from other expected announce-ments may lead us to falsely reject the hypothesis that commodity prices do not react tocentral bank announcements. Indeed, expected policy changes should be already incor-porated into asset prices. Therefore, the first step in this study is to identify unexpectedmonetary policies. The measure of monetary surprise in this study is different from thestandard event-study approaches, which usually measure the unexpected component bythe deviation of a realized benchmark variable to its “expected” value. Kilian & Vega (2008)define the “surprise” as the deviation of the realized US macroeconomic fundamentals tothe expected data provided by the International Money Market Services. SimilarlyCavallo& Wu (2006) measure an oil supply shock as the difference of the realized inventory to the
114
anticipated level found in the oil sector newspapers. Gospodinov & Jamali (2013) definethe monthly policy surprise as the difference between the average federal fund rate and theone-month ahead futures.In our case, we analyse daily data on the dollar index return in conjunction with the listof monetary policy announcements of the Federal Reserve, which we henceforth refer toas “monetary policy events”. The list of monetary policies announcements is found fromthe Federal Reserve website. An “unexpected announcement” is defined as a monetaryannouncement date when the dollar index exhibits an “abnormal” change. A change in thedollar index is considered “abnormal” at time t if:
| Returnt−Mean returns of previous 250 days |> 1.65 Standard deviation of previous 250 days(∗)
The days where dollar indexes satisfy condition (*) but where no Central Bank announce-ment has been made are likely to have been caused by other types of events.In different robustness tests, we test an alternative threshold of 1.28 (instead of 1.65) toselect “abnormal” changes and replace the dollar index definition of policy surprises with aone applied to US inflation swap rates.
The list of Federal Reserve events are gathered from the monetary announcement datesfrom 16/09/2008 to 31/12/2014. With the methodology explained above, we found 39unexpected events, a high proportion of which correspond to the release of Federal ReserveOpen Market Committee (FOMC) reports. Another important part of the announcementscorresponds to the publication of forecasts relative to the economic growth and inflationin the US. Some others are linked to interest rate forward guidance and other less relevantannouncements relate to banking and market regulation4. Table 5.10 in the appendix givesdetails on some of the unexpected monetary policy announcements.
Figure 5.3 presents the relationship between GSCI returns and dollar index returns in thewhole sample, the unexpected event days being highlighted. In a high number of cases,abnormal dollar index event-day returns coincide with very large GSCI returns in theopposite direction. In particular, four monetary events have been associated with a dollarindex decrease and a GSCI increase of about 3 standard deviations or more.
4We carefully check the information content in each identified unexpected communications day. Four events corre-sponding to “fixed-rate offering of term deposits” are less likely than others to represent monetary surprises. However,we retain these four events in order to preserve a significant number of unexpected monetary events and further therobustness of the statistical relation identified between commodities prices returns and monetary surprises.
115
5.4 Response of the commodity market to the Federal Reserve monetarypolicy
The above mentioned selection process of unexpected monetary events allows us to alleviatethe endogeneity problem in the assessment of the impact of monetary policy on commodi-ties prices. We conduct an OLS regression of four different commodities indices returnssuccessively on the dollar index, Eurodollar futures prices and inflation swap returns duringthe selected event dates. The four commodity indices used are the general S&P GSCI, theagricultural GSCI, the industrial metals GSCI, and the energy GSCI. This approach allowsus to test the reaction of the commodity market to the unexpected monetary event, and alsotheir sensitivity to the size of the monetary surprise.
Elementary event study
In the elementary event study, we examine the reaction of the main commodities index tothe dollar index return and the Eurodollar futures return during the unexpected event days.The model is specified as below:
Yi = α + βreturni + εi (5.1)
With Yi , commodity indexes returns (S&P GSCI, GSCI industrial, GSCI energy and GSCIagriculture); “returnsi” refer alternatively to dollar index, Eurodollar futures or inflationswaps returns, as indicated in Table 5.2 for the regression (a), (b), and (c).
When a threshold of 1.65 standard deviations is used to define monetary surprises, wefind no evidence of an impact of Eurodollar futures returns on commodities prices, butcommodities prices do react to the changes in the dollar index. An increase in 1 standarddeviation in the dollar index causes a decrease of about 0.2 standard deviations in the GSCIindex. The impact size is similar across different commodity sectors. If a 1.28 threshold isused instead of 1.65, dollar index returns still have a significant impact on commoditiesprices, the size of the impact being similar to the former case with a threshold of 1.65. Noevidence of impact of Eurodollar futures returns on commodity market is found again
An expansionary monetary policy aimed at stimulating the economy can therefore boostcommodities prices, a channel of this impact being the depreciation of the dollar againstother currencies. On the unexpected event days, Eurodollar futures returns do not havea significant impact on the commodity index. However, this fact does not mean that theinterest rate may not have any impact on commodity prices in any period since the periodunder analysis has been characterized by exceptionally low and stable level of rates. Turningto the impact of inflation swaps, the energy index and broad GSCI index with a large weightof energy commodities, are affected by inflation swap returns, but not the agricultural
116
nor the industrial metals prices. However, this phenomenon is not found when using athreshold of 1.28 for the selection of monetary surprises.
Asymmetric and delayed effect
So far we have looked simply at the impact of monetary surprises, as captured by dollarindex or inflation swap variations during important monetary announcement dates, oncommodities prices. However, the analysis has thus far not taken into account the directionof the monetary surprise (positive or negative). Following an earlier investigation of thecommodity markets’ asymmetric responses to monetary policy, Glick & Leduc (2011) showthat negative monetary surprises lead to a weak increase in interest rate and commoditiesprices, and positive surprises have much stronger impact on commodities prices. Theresearch of Gospodinov & Jamali (2013) also shows that commodity groups responddifferently to both positive and negative monetary shocks.
Therefore, in this section we are interested in discriminating the impact of positive andnegative monetary surprises on commodities prices. Negative events may be defined asthose announcement dates when the dollar index abnormally increases, reflecting a morerestrictive monetary stance than the one expected before the announcement.
To complement this analysis, we observe the mean and median commodities returns5. 10days before and after the unexpected event: the reaction of commodities prices to monetarysurprises may indeed be delayed. We calculate the delayed (resp. advanced) return n daysafter (resp. before) the event day as:
n days delayed returni =Commodity index pricen,i −Commodity index pricev,i
Commodity index pricev,i(5.2)
With n = the number of days before or after the unexpected announcement; We standardizethe obtained returns, expressing them in daily ex ante standard deviations (using again a250 day window before the event). In each case, we successively compute the mean andmedian of the delayed/advanced returns across the positive or negative monetary eventdays to obtain a mean and median price trajectory before and after the event.
The results are presented in Figure 5.4 and 5.5. During positive events, general, metal andenergy index prices decrease before the event day, increase immediately on the event day,then decrease again after the event. Agricultural commodities exhibit no evident trendbefore and after the event but increase as well during the event (as could be expected fromthe results of Table 5.2). In the case of negative events, the price trend after the event is more
5The calculation of delayed returns is based on the standardized returns series for every index as in the regressionanalysis
117
pronounced and shared across commodity sectors. A decreasing trend is observed on the10 days preceding and following the negative unexpected event for all of the commoditiesindexes.
The negative trend before the monetary surprises suggests that participants have been themost surprised by monetary announcements (positively or negatively) during periods ofcommodity meltdowns, where the decrease in inflation anticipations and the fear of a liq-uidity crunch justify a firm response from the Central Bank. In the case of positive monetarysurprises, the anticipation built my market participants before the announcement is de-ceived in the announcement day, but not beyond that date, as the pre-announcement trendresumes afterwards. In the case of negative monetary surprises, the pre-announcementpessimistic mood of market participants is confirmed by the announcement. So in all thecases, the Central Bank’s responses to deflationary episodes have tended to confirm thenegative market participants’ anticipations over the considered horizon. This observationis not necessarily contradictory with the visual impression that the successive QuantitativeEasing programs (and in particular QE1 and QE2) have successfully and durably changedthe trends of commodities prices (Figure 5.1). Indeed, the horizon of the analysis is oneday to two weeks in our framework, as opposed to at least a quarter when the effectsof successive QEs are analyzed. Moreover, our framework of analysis mixes extremelysignificant monetary surprises like the announcements of full-fledged QE with much lesssignificant monetary announcements. Lastly, the changes in the trends of commoditiesprices observed at the occasion of QE1 and QE2 may be the result of a consistent set ofmonetary surprises over the course of several weeks or months rather than the effect of onesingle monetary announcement.
Evidence before 2008
After the fall of Lehman Brothers in September 2008, monetary policy and its impact onthe market experienced significant changes. In this section we conduct the same analysisfor two distinct periods before the 2008 financial crisis: the first period is 2006–2008and the second 2000–2005. The choice of separating the periods before and after 2005is motivated by the turning point in the monetary policy after 2006 (the Fed stoppedhiking the target rate after June 2006) and the marked increase of correlations betweencommodities and the US dollar after 2006. Using the same methodology as before todefine the monetary surprises, we find 21 unexpected event days for the period 2005–2008and 16 unexpected event days for the period 2000–2005. To assess the impact of mon-etary surprises on commodities surprises, we use the same specification as in regression(5.1).
The regression results are reported in Table 5.3 and 5.4. A significant relationship betweencommodity prices and the dollar index already exists for the period 2005–2008, and the
118
influence is even stronger than after 2008. A one standard deviation negative monetarysurprise causes on average a 0.4 standard deviation drop of the GSCI Index. The relationshipbetween the commodity index and the Eurodollar futures return is again inexistent.
The relationship between the broad commodity The relationship between the broadcommodity index and the dollar index disappears in the 2000–2005 period, as only theindustrial index is sensitive to dollar index changes. The GSCI index and the energy indexreacted positively to Eurodollar futures prices returns (hence negatively to surprise forwardinterest rate variations), a feature which is in line with the impacts of conventional interestrate-based monetary policy on inflation expectations. This conclusion confirms that abreaking point occurred around 2006 as regards the form taken by the Fed’s monetarypolicy and its impact on commodities markets.
To sum up, commodities prices have been negatively correlated with monetary surprisescaptured by dollar index returns since 2006. Before that date, only a relation betweencommodities prices and Eurodollar futures returns is found on unexpected monetaryannouncement dates. In addition, contrary to positive monetary surprises, whose impact isfelt the very day of the announcement and is inverted afterwards, negative surprises have alasting effect on commodity indexes returns, up to 10 days after the announcement.
5.5 Additional tests
In this section, we carry out several robustness tests. The first test consists in definingmonetary surprises through inflation swaps instead of dollar returns. The second set oftests aim at determining the channels of transmission by which monetary policy impactscommodity prices.
Inflation swap
Barsky & Kilian (2004) argue that monetary policy impacts commodity prices indirectly byits effect on the expected future inflation rate. In Rigobon & Sack (2008), the authors findthat twelve macroeconomic variables have prompted a significant reaction in the Eurodollarfutures rates. Among these variables, are the dollar exchange rate and the inflation rate.Hence, inflation anticipations are a crucial indicator monitored and regulated by centralbanks. As a matter of fact, in the list of unexpected monetary announcements reportedin Table 5.10, the news concerning prices stability and inflation rate expectations has animportant place. In order to assess the robustness of our previous results, we re-estimateregression (5.1) by using the change in the inflation swap rate6 instead of the dollar index
6Inflation swap rates are not available before 2008.
119
returns on the date of the announcement.
We now find 26 unexpected event days from 09/2008 to 10/2014, which are in most casesidentical to those found with the dollar index. We then regress commodities prices returnson monetary surprises (regression (5.1)) on these 26 unexpected event days. As before, mon-etary surprises are defined alternatively through inflation swap returns, Eurodollar futuresreturns and dollar index return on the day of the unexpected monetary announcement.The results, reported in Table 5.5, indicate that the monetary surprises defined throughthe dollar index keep a significant impact on most commodity indices when unexpectedmonetary announcements are identified through inflation swaps data. And in this case,monetary surprises defined through the Eurodollar futures returns display a positiveimpact on the industrial metals commodity index. When monetary surprises are definedthrough the inflation swap rate, the impact is again found only on the industrial metals andthe broad commodity indices.
Testing the transmission channel in agricultural commodities
- Financial investments and inventory channelsAs noted above, Frankel (1989) proposes three channels through which changes in theinterest rate may have an influence on commodity markets. Gospodinov & Jamali, (2013)find an increase in speculating pressure for the metal and energy commodities as the resultsof expansionary monetary policy but no evidence as regards the storing behaviour. Theseresults are obtained in other markets; see for example, Anzuini, Lombardi, & Pagano (2013)in the oil market (transmission channel of inventory is significant but definitely small) andGlick & Leduc (2011) in the metal market. In this section, we test whether the transmissionchannels envisioned in Frankel (1989) may explain the impact of monetary surprises oncommodities prices.
The framework of the analysis is the same as above, replacing the commodity prices returnswith inventory and investors’ investment positions. The measure of investors’ behavioruses a weekly report from the US Commodity Futures Trading Commission (CFTC), whichgives the weekly positions of commercial, non-commercial, and index investors for 12agricultural commodities. The behavior of traders is defined by the net long position foreach type of trader and of commodities as a percentage of total open interest:
long position in f utures and optionsi − short positions in f utures and optionsiOpen interesti
.
Exchange-traded fund (ETF) flows7 are also examined. We consider daily investment
7ETF flow is calculated as: Flowst,t−1 = Σ3i=1ET FP rice
it
(sharesit − shares
it−1
).
120
inflows into the three major ETF tracking generalist commodity indices: DBC–US, UJ–UBS,and GSCUS.
As for the inventory data, we compute an investment proxy, as defined in Guilleminotet al. (2014), for 12 agricultural commodities. This “inventory proxy” is defined as theperformance of a strategy selling the first maturity after the closest harvest and buying thefirst maturity after the second-closest harvest:
I (t) = Πi≤t−∆t
(1 +
F2i+∆t −F2F2i
− F1i+∆t −F1iF1i
)(5.3)
With F2 the price of the second-closest harvest futures contract. F1 is the price of the closestharvest contract. Inventory in the following study is the averaged inventory across the 12agricultural commodities monitored in the CFTC Supplemental Report. The “inventoryshock” is defined as ∆I(t) = I(t)− I(t − 1) where the time t refers to days.
Our regression analysis is based on weekly data on CFTC traders’ positions and daily ETFflows, inventory changes, dollar index, and Eurodollar futures returns. In the case of CFTCpositions only, monetary surprises are defined on a weekly basis: a “monetary surpriseweek” is defined as the period stretching between during two consecutive releasing dates ofthe weekly CFTC Supplemental Report and containing at least one unexpected monetaryannouncement.
A regression of the inventory proxy on the monetary surprise suggests that monetary policydoes not impact commodities’ prices through the channel of inventories piling. Turning tothe trading position channel, the results show that the dollar index does not have any impacton trading position – including index flows, commercial position, speculative position.Only monetary surprises captured by forward interest rate changes have a relatively smallnegative impact8 on commercial trading positions. As a result, our results does not lendsupport to the view that monetary policy affects commodities prices through the channel ofinventory piling or investors’ portfolio rebalancing decisions.
- Test of the inventory channel for five specific commoditiesWe also choose test the inventory channel for three agricultural commodities (wheat, corn,soybeans) and two non-agricultural commodities (Brent and copper). Inventory proxy ofBrent and copper is calculated also with the equation (5.3), except that F2 the price ofthe second-closest futures contract, as non-agricultural commodities do not have harvestseasonality.Our results are presented in Table 5.7. The relation of inventory proxies to dollar index
8Eurodollar futures returns have a positive relation to net commercial positions but are negatively related to forwardinterest rate changes.
121
returns is negative in all cases bar wheat. However, none of these relations is significantat the 10% level. The soybeans inventory proxy is negatively associated (with weak sig-nificance), and the copper inventory proxy positively associated, to forward interest rates(only a negative association being consistent with the inventory channel). Hence, we findagain little support for the inventory channel driving the impact of monetary surprises oncommodities prices.
- Role of dollar in the variation of commodity pricesOur last test aims at investigating whether the impact of monetary policy on commodityprices is the result of a monetary effect (commodities prices remain constant in foreigncurrencies but rise in terms of a debased dollar currency). For this purpose, we reapplythe regression (5.2), but using as dependent variable the commodities prices expressed inforeign currencies:Commodity index (in f oreign currencies) = commodity index × dollar indexResults are showed in table 5.8. We find no impact of the monetary surprise (be it expressedthrough dollar index, Eurodollar futures or inflation swaps variations on the announcementdates) on “absolute” commodities prices (i.e. the prices expressed in foreign currencies).As a result, the impact of unexpected monetary announcement on commodities prices isdriven by a the dollar debasement effect rather than the change in the “absolute” level ofcommodities prices in terms of major foreign currencies.
5.6 Conclusion
Defining monetary surprises as the dollar index, Eurodollar futures or inflation swapreturns on the dates of major monetary announcements, this paper studies the impact ofmonetary surprises on commodities prices. Our results show that commodity markets havebeen sensitive to dollar index and inflation swap variations on monetary announcementdates from 2006 onwards; before that, only the interest rate variations during monetaryannouncement dates had a notable impact. Second, the effect of negative monetary surprisesis not only instantaneous but also delayed, as a decreasing trend is observed up to 10 daysafter the announcement. Positive surprises have an instantaneous effect but this effectis reversed after the day of the announcement. Thirdly, we perform one robustness testdefining the monetary surprise through the change in the inflation swap rate, which leads tovery similar conclusions as above. Lastly, our results do not lend support to the hypothesisof Frankel (2008) that monetary policy may affect commodities prices through inventorypiling or investors’ portfolio rebalancing decisions. Instead, we find evidence that theimpact of unexpected monetary announcements on commodities prices is mainly driven bya dollar “debasement” (or revaluation) effect, whereby commodities prices are unaffectedby the Fed’s surprise moves once expressed in terms of major foreign currencies.Our results have several implications. First, we provide direct evidence that commodity
122
markets may be impacted by factors, which are exogenous to physical supply and demandfundamentals (in this case, monetary policy announcements). This effect works through thecanal of dollar depreciation against other major currencies. Second, this work may providean explanation for the increased correlation between dollar, traditional investment classesand commodity prices over the last decade, which could be traced back to the investors’anticipation of monetary surprises.
123
Figure 5.1: Commodities prices,Interest rate and Dollar index (a) and (b)
(a)
(b)
124
Figure 5.2: Commodities prices,Interest rate and Dollar index (c)
Note: Commodities prices are represented by SP GSCI index prices; Interest rate is noted byEurodollar futures prices, which have a negative relationship with the interest rate; dollar indexmeasures the value of dollar against a trade-weighted weighted basket of foreign currencies.Three important Fed operations are noted in the graph, with Quantitative Easing 1(QE1) beginson 25/11/2008: Fed initiated purchase of 500 billion dollars in mortgage-backed securities; QE2 on 3/11/2010: began the purchase of 600 billion dollars of longer-term Treasury securitiesand QE3 on 12/09/2012: planed to buy another 40 billion dollars in mortgage-backed invest-ment each month. The dotted line gives the possible reaction of hints, such as on 27/08/2010Bernanke’s speech on Jackson Hole hinted the QE2 and on 22/08/2012, Fed released FOMC asthe first indication of considering QE3. Quantitative Easing has weakened US dollar value andincreased commodity index prices as it leases more liquidity. Negative relation between SPGSCIprices and dollar index is more obvious than the relation between SPGSCI and Eurodollar.
125
Table 5.1: Correlation of major variables
GSCI IR Dollar Inflation
2001/2005
GSCI 1 0.039(1.24) -0.13(-4.01**)IR 1 -0.19(-6.06***)
Dollar 1Inflation
2005/2008
GSCI 1 0.024(1.13) -0.36(-18.56***)IR 1 -0.11(-5.02**)
Dollar 1Inflation
2008/2014
GSCI 1 0.038(1.5) -0.38(-16.76***) 0.29(11.86***)IR 1 0.11(-4.35**) 0.06(2.5)
Dollar 1 -0.09(-3.59**)Inflation 1
Note: Correlation matrix of variables GSCI return, Eurodollar futures return, Dollar indexreturn and inflation swap return. For the first two samples, 2001/2005 and 2005/2008, result ofinflation is missing due to lack of inflation swap data in these two periods. Inside the parentheses,it gives the t-test of correlation. (*** significant at 1%, ** significant at 5%, * significant at 10%.).
126
Figure 5.3: GSCI and dollar index of all trading days and of unexpected monetary policy days
Note: Two graphs below show GSCI and dollar index data and the unexpected event days from16/09/2008 to 31/12/2014. GSCI is SP GSCI returns; Dollar is the dollar index, which measuresthe value of dollar against a trade-weighted weighted basket of foreign currencies. GSCI returnand dollar index returns are represented by number of standard deviation.Unexpected event days are defined as the Fed monetary policy announcement days when dollarindex returns excessed 1.65 standard deviation.Blue points are the values of all sample days. Red ones represent both GSCI and dollar value theunexpected event days.
127
Tabl
e5.
2:E
lem
enta
ryev
ent
stu
dy
-Th
resh
old
1.65
Com
mod
ity
Ind
exre
turn
sC
omm
odit
yin
dex
and
dol
lar
ind
exC
omm
odit
yin
dex
and
Euro
dol
lar
futu
res
Com
mod
ity
ind
exan
din
flat
ion
swap
Dol
lar
ind
exIn
terc
ept
R-sq
uar
edEu
rod
olla
rfu
ture
sIn
terc
ept
R-sq
uar
edIn
flat
ion
swap
Inte
rcep
tR-
squ
ared
GSC
IAR
-0.2
560.
172
0.3
-0.1
90.
340.
030.
020.
060.
080.
001*
**0.
390.
260.
150.
760.
76G
SCIE
N-0
.241
-0.0
20.
161
-0.2
70.
30.
040.
060.
360.
250.
009
***
0.91
0.18
0.3
0.00
9***
0.14
*G
SCII
N-0
.222
10.
015
0.15
-0.2
10.
190.
030.
020.
170.
020.
012
**0.
940.
260.
490.
440.
54G
SCI
-0.2
670.
003
0.19
-0.2
60.
330.
040.
050.
320.
230.
004
***
0.98
0.19
0.25
0.01
**0.
17
-
Th
resh
old
of1.
28
Com
mod
ity
Ind
exre
turn
sC
omm
odit
yin
dex
and
dol
lar
ind
exC
omm
odit
yin
dex
and
Euro
dol
lar
futu
res
Com
mod
ity
ind
exan
din
flat
ion
swap
Dol
lar
ind
exIn
terc
ept
R-sq
uar
edEu
rod
olla
rfu
ture
sIn
terc
ept
R-sq
uar
edIn
flat
ion
swap
Inte
rcep
tR-
squ
ared
GSC
IAR
-0.2
30.
060.
17-0
.04
0.04
0.00
1-0
.01
0.07
0.01
0.00
02**
*0.
630.
720.
790.
840.
66G
SCIE
N-0
.25
-0.0
20.
15-0
.06
-0.0
50.
002
0.05
0.09
0.03
0.00
1***
0.88
0.68
0.78
0.22
0.65
GSC
IIN
-0.2
50.
140.
150.
060.
140.
002
0.01
0.22
0.00
10.
001*
**0.
360.
70.
420.
780.
24G
SCI
-0.2
8-0
.000
20.
17-0
.06
-0.0
30.
002
0.04
0.11
0.02
0.00
02**
*0.
980.
70.
880.
30.
56
Not
e:T
heta
ble
give
sth
ere
sult
sof
3re
gres
sion
s,re
gres
sion
(a):Yi
=α
+β
[Dollarindexreturns ]i
+ε i
;re
gres
sion
(b):Yi
=γ
+δ
[Eurodollarfuturesreturns ]i
+θi;
regr
essi
on(c
):Yi
=α
+β
[Infaltionswapreturns ]i
+ε i
.Yi:
chan
ges
ofco
mm
odit
yin
dex
spot
retu
rns
onth
eev
ent
day
s.A
llva
riab
les
are
stan
dar
diz
edan
din
clu
de
the
sam
ple
from
16/0
9/20
08to
31/1
2/20
14w
ith
39u
nexp
ecte
dev
ents
(1.6
5T
hres
hold
)and
wit
h74
une
xpec
ted
even
ts(1
.28
Thr
esho
ld).
The
ind
exof
dep
end
ent
vari
able
sar
eG
SCIA
R:A
gric
ult
ura
lind
ex;G
SCIE
N:E
nerg
yin
dex
;GSC
IIN
:Ind
ust
rial
ind
ex;G
SCI:
Bro
adco
mm
odit
yin
dex
.t-s
tati
stic
sar
egi
ven
ital
iciz
edaf
ter
ever
yp
aram
eter
.(*
**si
gnifi
cant
at1%
,**
sign
ifica
ntat
5%,*
sign
ifica
ntat
10%
.).
128
129
Figure 5.4: Delayed effect - Mean
(a)
(b)
Note: Twenty values are the mean return/standard deviation (upper graph) and median re-turn/standard deviation (bottom graph) of each 10 days after and before unexpected event. “1”is the event day when unexpected event happens; “0” is the day before unexpected event.Unexpected event days are defined as the Fed monetary policy announcement days when dollarindex returns excessed 1.65 standard deviation.Positive event is defined as the event day whenwe have negative dollar return and positive GSCI returns.The returns of commodities indexesare: GSCIAR: Agricultural index; GSCIEN: Energy index; GSCIIN: Industrial index; GSCI: Broadcommodity index. 130
Figure 5.5: Delayed effect - Median
(a)
(b)
131
Table 5.3: Observation before 2008: 2005/2008 (21 unexpected events)
Commodity index returnsCommodity index and dollar index Commodity index and Eurodollar futuresDollar index Intercept R-squared Eurodollar futures Intercept R-squared
GSCIAR -0.21 -0.40 0.30 0.13 -0.37 0.090.01*** 0.04** 0.18 0.09*
GSCIEN -0.34 -0.06 0.40 -0.07 0.07 0.020.002*** 0.80 0.6 0.81
GSCIIN -0.37 -0.56 0.38 0.05 -0.45 0.0060.003*** 0.046** 0.75 0.19
GSCI -0.42 -0.11 0.46 -0.07 0.05 0.0120.0008*** 0.67 0.64 0.89
Table 5.4: Observation before 2008: 2000/2005 (16 unexpected events)
Commodity index returnsCommodity index and dollar index Commodity index and Eurodollar futuresDollar index Intercept R-squared Eurodollar futures Intercept R-squared
GSCIAR 0.13 -0.255 0.116 -0.26 -0.32 0.070.197 0.28 0.1838 0.048**
GSCIEN 0.07 0.014 0.03 0.57 0.07 0.160.53 0.96 0.04** 0.76
GSCIIN -0.33 -0.71 0.25 0.26 -0.296 0.0320.05** 0.06* 0.39 0.21
GSCI 0.029 -0.019 0.005 0.60 0.023 0.180.794 0.94 0.03** 0.91
Note: The table gives the results of 3 regressions, regression (a):Yi = α + βDollar;i +εi ; regression(b):Yi = α + βEurodollar; ;i +εi ;regression(c):Yi = α + βInf lation; ;i +εi Yi : changes of commodityindex spot returns on the event days. All variables are standardized and include two samplesfrom 2005 to 2008 (upper table) and from 2000 to 2005 (bottom table). Unexpected event daysare defined as the Fed monetary policy announcement days when dollar index returns excessed1.65 standard deviation.The index of dependent variables are the returns of: GSCIAR: Agricultural index; GSCIEN: En-ergy index; GSCIIN: Industrial index; GSCI: Broad commodity index.The independent variablesare: Eurodollar futures returns; the dollar index, which measures the value of dollar againsta trade-weighted weighted basket of foreign currencies; inflation swap returns. t-statistics aregiven italicized after every parameter.(*** significant at 1%, ** significant at 5%, * significant at 10%.)
132
Tabl
e5.
5:R
esu
lts
base
don
une
xpec
ted
even
tfo
und
from
infl
atio
nsw
ap
Com
mod
ity
ind
exan
din
flat
ion
swap
Com
mod
ity
ind
exan
dEu
rod
olla
rfu
ture
sC
omm
odit
yin
dex
and
Dol
lar
ind
exC
omm
odit
yin
dex
retu
rns
Infl
atio
nsw
apIn
terc
ept
R-sq
uar
edEu
rod
olla
rfu
ture
sIn
terc
ept
R-sq
uar
edD
olla
rin
dex
Inte
rcep
tR-
squ
ared
GSC
IAR
0.02
0.09
0.08
3-0
.11
0.18
20.
05-0
.29
-0.0
170.
190.
161
0.60
50.
360.
440.
029*
0.92
GSC
IEN
0.06
0.45
0.28
-0.0
51.4
10.
004
0.35
0.24
0.14
90.
007*
**0.
062.
0.78
0.26
0.06
*0.
345
GSC
IIN
0.02
0.29
0.03
20.
330.
078
0.21
-0.3
90.
160.
190.
390.
280.
04**
0.79
0.03
**0.
50
GSC
I0.
055
0.42
0.27
-0.0
20.
380.
001
-0.3
90.
204
0.20
40.
008*
**0.
07*
0.90
0.27
0.02
**0.
38
Not
e:T
heta
ble
give
sth
ere
sult
sof
3re
gres
sion
s,re
gres
sion
(a):Yi
=α
+βDollar;i+ε i
;reg
ress
ion
(b):Yi
=α
+βEurodollar;
; i+ε i
;reg
ress
ion(
c):Yi
=α
+βInflation
;;i+ε i
Yi:
chan
ges
ofco
mm
odit
yin
dex
spot
retu
rns
onth
eev
ent
day
s.A
llva
riab
les
are
stan
dar
diz
edan
din
clu
de
the
sam
ple
from
16/0
9/20
08to
31/1
0/20
14w
ith
26u
nexp
ecte
dev
ents
.U
nexp
ecte
dev
ent
day
sar
ed
efine
das
the
Fed
mon
etar
yp
olic
yan
nou
ncem
ent
day
sw
hen
infl
atio
nsw
apre
turn
sex
cess
ed1.
65st
and
ard
dev
iati
on.
The
ind
exof
dep
end
ent
vari
able
sar
eth
ere
turn
sof
:GSC
IAR
:Agr
icu
ltu
rali
ndex
;GSC
IEN
:Ene
rgy
ind
ex;G
SCII
N:I
ndu
stri
alin
dex
;GSC
I:B
road
com
mod
ity
ind
ex.T
hein
dep
end
ent
vari
able
sar
e:Eu
rod
olla
rfu
ture
sre
turn
s;th
ed
olla
rin
dex
,w
hich
mea
sure
sth
eva
lue
ofd
olla
rag
ains
ta
trad
e-w
eigh
ted
wei
ghte
dba
sket
offo
reig
ncu
rren
cies
;infl
atio
nsw
apre
turn
s.t-
stat
isti
csar
egi
ven
ital
iciz
edaf
ter
ever
yp
aram
eter
.(*
**si
gnifi
cant
at1%
,**
sign
ifica
ntat
5%,*
sign
ifica
ntat
10%
.).N
um
ber
ofob
serv
atio
ns:2
6.
133
Table 5.6: Transmission channels: Inventory and trading position
Dollar index Intercept R-squared Eurodollar futures Intercept R-squaredCITF 0.032 -0.02 0.002 -0.02 -0.003 0.03
0.794 0.022** 0.305 0.035**Non-commercial 0.089 1.427 0.029 0.053 1.422 0.004
0.301 2.05e-13*** 0.693 5.13e-13***Commercial -0.01 -0.04 0.065 0.015 -0.043 0.078
0.117 0.00004*** 0.086* 1.62e-05***ETF -0.05 0.112 0.016 -0.161 0.09 0.04
0.27 0.89 0.36 0.97Inventory 0.21 -8.12e-05 0.046 0.18 2.15e-05 0.32
0.27 0.89 0.36 0.97
Note: Regression (a): Yi = α + βDollar;i +εi ; regression (b): Yi = α + βEurodollar; ;i +εi ;Yi : the transmission channels,which are: Weekly data of CITF: aggregate 12 agricultural Commodity index flows from COT report from CFTC;commercial/ speculative positions of 12 agricultural commodity futures contracts from CFTC report and Daily inventoryare averaged inventory proxy for 12 agricultural commodities. This “inventory proxy” is defined as the performanceof a strategy selling the first maturity after the closest harvest and buying the first maturity after the second-closest
harvest:I (t) = Πi>t−∆t
(1 +
F2i+∆t −F2F2i
− F1i+∆t − 1iF1i
), with F2 the price of the second-closest harvest futures contract.
F1 is the price of the closest harvest contract.Daily ETF is three major ETF tracking generalist commodity indices: DBC–US, UJ–UBS, and GSCUS. All variables arestandardized and include the sample from 16/09/2008 to 31/10/2014. Unexpected event days are defined as the Fedmonetary policy announcement days when inflation swap returns excessed 1.65 standard deviation.The independent variables are: Eurodollar futures returns; the dollar index, which measures the value of dollar against atrade-weighted weighted basket of foreign currencies; inflation swap returns.(*** significant at 1%, ** significant at 5%, * significant at 10%.)
134
Table 5.7: Transmission channels of inventory for five individual commodities
With dollar index (a) With Interest rateInventory proxy Dollar index Intercept R-squared Eurodollar futures Intercept R-squared
Wheat 0.02 6 0.01 0.004 0.14 -0.02 0.040.646 0.95 0.28 0.91
Corn -0.08 -0.03 0.04 0.22 -0.03 0.090.13 0.85 0.101 0.86
Soybean -0.08 -0.028 0.05 0.22 -0.03 0.030.15 0.85 0.09* 0.86
Brent -0.06 -0.003 0.02 0.18 -0.03 0.06
0.39 0.98 0.17 0.86Copper -0.07 -4.00E-03 0.04 -0.27 0.05 0.15
0.29 0.9 0.03** 0.78
Note: Regression (a): Yi = α + βDollar;i +εi ; regression (b): Yi = α + βEurodollar; ;i +εi ; With Yi : the inventory shock ofcommodity i. Daily inventory proxy is used which is defined as the performance of a strategy selling the first maturityafter the closest harvest and buying the first maturity after the second-closest harvest (agricultural commodities) orselling the first maturity after the closest harvest and buying the second maturity (non-agricultural commodities).Five commodities are chosen: three grains commodities: wheat, corn and soybean; two non-agricultural commodities:Brent and copper. All variables are standardized and include the sample from 16/09/2008 to 31/10/2014. Unexpectedevent days are defined as the Fed monetary policy announcement days when inflation swap returns excessed 1.65standard deviation.The independent variables are: Eurodollar futures returns; the dollar index, which measures the value of dollar against atrade-weighted weighted basket of foreign currencies; inflation swap returns (*** significant at 1%, ** significant at 5%, *significant at 10%.)
135
Tabl
e5.
8:R
ole
ofd
olla
rin
the
vari
atio
nof
com
mod
ity
pri
ces
Com
mod
ity
Ind
exre
turn
s(in
fore
ign
curr
enci
es)
Com
mod
ity
ind
exan
dd
olla
rin
dex
Com
mod
ity
ind
exan
dEu
rod
olla
rfu
ture
sC
omm
odit
yin
dex
and
infl
atio
nsw
apD
olla
rin
dex
Inte
rcep
tR-
squ
ared
Euro
dol
lar
futu
res
Inte
rcep
tR-
squ
ared
Infl
atio
nsw
apIn
terc
ept
R-sq
uar
edG
SCIA
R-0
.02
0.13
0.01
-0.3
40.
180.
020.
010.
150.
004
0.8
0.57
0.17
0.38
0.71
0.5
GSC
IEN
-0.0
6-0
.02
0.01
-0.2
60.
002
0.02
0.07
0.09
0.07
0.6
0.94
0.39
0.99
0.14
0.74
GSC
IIN
-0.0
5-0
.08
0.01
-0.4
20.
090.
030.
02-0
.04
0.01
0.6
0.75
0.14
0.72
0.69
0.88
GSC
I-0
.04
0.02
0.00
4-0
.32
0.06
0.03
0.06
0.13
0.06
0.7
0.93
0.31
0.82
0.18
0.66
Not
e:T
heta
ble
give
sth
ere
sult
sof
3re
gres
sion
s,re
gres
sion
(a):Yi
=α
+βDollar;i+ε i
;reg
ress
ion
(b):Yi
=α
+βEurodollar;
; i+ε i
;reg
ress
ion(
c):Yi
=α
+βInflation
;;i+ε iYi:
chan
ges
ofco
mm
odit
yin
dex
spot
retu
rns
onth
eev
entd
ays.
All
vari
able
sar
est
and
ard
ized
and
incl
ud
eth
esa
mp
lefr
om16
/09/
2008
to31
/10/
2014
wit
h39
une
xpec
ted
even
ts.U
nexp
ecte
dev
ent
day
sar
ed
efine
das
the
Fed
mon
etar
yp
olic
yan
nou
ncem
ent
day
sw
hen
infl
atio
nsw
apre
turn
sex
cess
ed1.
65st
and
ard
dev
iati
on.T
hein
dex
ofd
epen
den
tva
riab
les
are
the
retu
rns
info
reig
ncu
rren
cies
of:
GSC
IAR
:Agr
icu
ltu
ral
ind
ex;G
SCIE
N:E
nerg
yin
dex
;GSC
IIN
:Ind
ust
rial
ind
ex;G
SCI:
Bro
adco
mm
odit
yin
dex
.Com
mod
ity
ind
exp
rice
sar
etr
ansf
orm
edin
fore
ign
curr
enci
esby
mu
ltip
lyin
gd
olla
rin
dex
.The
ind
epen
den
tvar
iabl
esar
e:Eu
rod
olla
rfu
ture
sre
turn
s;th
ed
olla
rin
dex
,whi
chm
easu
res
the
valu
eof
dol
lar
agai
nst
atr
ade-
wei
ghte
dw
eigh
ted
bask
etof
fore
ign
curr
enci
es;i
nflat
ion
swap
retu
rns.
t-st
atis
tics
are
give
nit
alic
ized
afte
rev
ery
par
amet
er.
Sign
ifica
ntco
de:
“*”
sign
ifica
ntat
10%
,“**
”si
gnifi
cant
at5%
,“**
*”si
gnifi
cant
at1%
.N
um
ber
ofob
serv
atio
ns:3
9
136
Tabl
e5.
9:Sy
nthe
sis
tabl
e
un
exp
ecte
dev
ent
sele
ctio
ncr
iter
iaD
olla
rEu
rod
olla
rfu
ture
sIn
flat
ion
swap
c
Regressionvariables
2001
-200
5(1
6ev
ents
)20
05-2
008(
21ev
ents
)20
08-2
014(
39ev
ents
)20
01-2
005(
5ev
ents
)20
05-2
008(
28ev
ents
)20
08-2
013(
21ev
ents
)20
01-2
005
2005
-200
820
08-2
014
Dol
lar
Sign
ifica
ntne
ga-
tive
Sign
ifica
ntne
ga-
tive
**;e
xcep
tE
TF
flow
s
Sign
ifica
ntne
g-at
ive;
GSC
Ian
dE
nerg
y
Sign
ifica
ntne
ga-
tive
;Ind
ust
rySi
gnifi
cant
nega
-ti
ve*
Euro
dol
lar
futu
res
Sign
ifica
ntp
os-
itiv
e;G
SCI
and
Ene
rgy(
Onl
y5
un-
exp
ecte
dev
ents
)In
flat
ion
swap
sSi
gnifi
cant
pos
i-ti
ve*;
ET
F,E
nerg
yan
dG
SCI
Sign
ifica
ntp
os-
itiv
e*;E
nerg
yIn
dust
ryan
dG
SCI
Sign
ifica
ntp
osi-
tive
**;E
nerg
yan
dG
SCI
Not
e:U
nexp
ecte
dev
ent
day
sar
ed
efine
das
the
Fed
mon
etar
yp
olic
yan
nou
ncem
ent
day
sw
hen
one
ofth
eth
ree
ind
icat
ors
(Dol
lar
ind
exre
turn
s,Eu
rod
olla
rfu
ture
sre
turn
sor
infl
atio
nsw
apre
turn
s)ex
cess
ed1.
65st
and
ard
dev
iati
on.T
heta
ble
give
sth
ere
sult
sof
3re
gres
sion
s,re
gres
sion
(a):Yi
=α
+βDollar;i
+ε i
;reg
ress
ion
(b):Yi
=α
+βEurodollar;
; i+ε i
;reg
ress
ion(
c):Yi
=α
+βInflation
;;i
+ε iYi:
chan
ges
ofco
mm
odit
yin
dex
spot
retu
rns
onth
eev
ent
day
s.E
ach
grid
give
sth
esi
gnifi
canc
eeff
ect
onth
ere
turn
sof
:G
SCIA
R:A
gric
ult
ura
lin
dex
;GSC
IEN
:Ene
rgy
ind
ex;G
SCII
N:I
ndu
stri
alin
dex
;GSC
I:B
road
com
mod
ity
ind
ex.B
lue
grid
sre
fer
toth
eca
sew
ith
sign
ifica
nce
effec
ton
com
mod
ity
ind
ex(o
nal
lfi
veco
mm
odit
ies
ind
exes
ifw
itho
ut
spec
ifica
tion
).R
edgr
ids
are
the
case
sof
non
sign
ifica
nce
effec
tob
serv
ed.
The
ind
epen
den
tva
riab
les
are:
Euro
dol
lar
futu
res
retu
rns;
the
dol
lar
ind
ex,
whi
chm
easu
res
the
valu
eof
dol
lar
agai
nst
atr
ade-
wei
ghte
dw
eigh
ted
bask
etof
fore
ign
curr
enci
es;i
nflat
ion
swap
retu
rns.
All
vari
able
sar
est
and
ard
ized
and
incl
ud
eth
ree
sam
ple
sfr
om20
01to
2005
;fro
m20
05to
2008
and
from
2008
to20
14.R
esu
ltof
infl
atio
nis
mis
sing
due
tola
ckof
infl
atio
nsw
apd
ata
infi
rst
two
per
iod
s.Si
gnifi
cant
imp
act
ism
arke
dby
blu
egr
ids.
Ora
nge
grid
sd
enot
eth
ein
sign
ifica
ntre
sult
s.“*
”si
gnifi
cant
at10
%,“
**”
sign
ifica
ntat
5%,“
***”
sign
ifica
ntat
1%.
137
Table 5.10: List of unexpected monetary defined from dollar index events from 2008 to 2014
Date Activities Type Dollar GSCI2008-9-22 Federal Reserve will offer 75 billion dollars in
28-day credit through its Term Auction Facilitytoday
Guidance -3.929 2.841
2008-10-6 Federal Reserve will offer 150 billion dollars in85-day credit through its Term Auction Facilitytoday
Guidance 3.082 -3.337
2008-10-13 Federal Open Market Committee has autho-rized increases in the sizes of its temporaryswap facilities with the BoE, the ECB, and theSNB
Balance sheet 3.082 -3.337
2008-10-21 Fed conducted an auction of 150 billion dollarsin 28-day credit through its Term Auction Facil-ity AND the creation of the Money Market In-vestor Funding Facility (MMIFF)
Balance sheet/ Guidance 1.776 -1.029
2008-10-22 Fed will alter the formula used to determine theinterest rate paid to depository institutions onexcess balances
Balance sheet/IR 2.734 -2.865
2008-10-29 FOMC statement indicates that It lowers its tar-get federal funda rate; Economic activity ap-pears to be slow and Energy and other com-modities prices declined
IR/Guidance -3.865 3.310
2008-11-4 Fed conducted an auction of 150 billion dollarsin 84-day credit through its Term Auction Facil-ity
Balance sheet -3.013 3.556
2008-11-24 Quantitative Easing 1 (QE1): It initiate a pro-gram to purchase the direct obligations ofhousing-related government-sponsored enter-prises (GSEs)–Fannie Mae, Freddie Mac, andthe Federal Home and announce the creation ofthe Term Asset-Backed Securities Loan Facility(TALF)
Balance sheet -3.815 3.125
138
2008-11-25 Fed conducted an auction of 150 billion dol-lars in 13-day credit through its Term Auc-tion Facility AND Minutes of Board dis-count rate meeting/ discount rate and pri-mary credit rate. Il released also the cre-ation of the Term Asset-Backed Securities LoanFacility (TALF) and Fed will initiate a pro-gram to purchase the direct obligations ofhousing-related government-sponsored enter-prises (GSEs)–Fannie Mae, Freddie Mac, andthe Federal Home Loan Banks–and mortgage-backed securities (MBS) backed by Fannie Mae,Freddie Mac, and Ginnie Mae
Balance sheet/IR -1.978 -1.827
2008-12-15 Federal Reserve will offer 150 billion dollars in28-day credit through its Term Auction Facilitytoday
Guidance -2.443 -0.421
2008-12-16 FOMC statement: Economic outlook has weak-ened/Inlfation pressure declined and is ex-pected to moderate further/ Committee’s pol-icy going to purchase large quantities of agencydebt and mortgage-backed securities to providesupport to the mortgage and housing markets,and it stands ready to expand its purchases ofagency debt and mortgage-backed securities asconditions warrant
Balance sheet/Guidance -2.817 0.006
2008-12-19 Fed released revised terms and conditions;questions and answers detailing operational as-pects of the Term Asset-Backed Securities LoanFacility (TALF) and It will conduct three auc-tions of 28-day credit and three auctions of 84-day credit through its Term Auction Facility(TAF)
Guidance 3.174 0.075
2009-1-13 Fed conducted an auction of 150 billion dollarsin 28-day credit through its Term Auction Facil-ity;It release the Minutes of Board discount ratemeeting
Balance sheet/IR 1.980 0.536
2009-3-18 FOMC statement: Economic continues to con-tract and inflation could persist for a time be-low rates that best foster economic growth andprice stability in the longer term
Balance sheet/Guidance -3.608 -0.501
139
2009-3-19 The set of eligible collateral for loans extendedby the Term Asset-Backed Securities Loan Fa-cility (TALF) is being expanded to include fouradditional categories of asset-backed securities(ABS)
Balance sheet/Guidance -2.289 2.220
2009-12-4 It announced the adoption of a final rule thatwould establish a process by which the FederalReserve Bank of New York may determine theeligibility of credit rating agencies for the TermAsset-Backed Securities Loan Facility (TALF)
Guidance 2.491 -0.476
2010-9-7 Minutes of Board discount rate meetings IR 1.759 0.0542010-9-21 FOMC statement: Pace of recovery continue to
be slow and longer-term inflation expectationshave remained stable
Balance sheet/Guidance -2.153 -0.738
2010-10-19 Minutes of Board discount rate meetings IR 3.066 -2.1802010-11-23 FOMC minutes Balance sheet 2.287 -0.2672011-9-6 Minutes of Board discount rate meetings IR 2.149 -0.4612013-1-3 FOMC minutes Balance sheet 1.770 -0.3542013-2-20 FOMC minutes Balance sheet 2.008 -1.0332013-4-16 Minutes of Board’s discount rate meetings IR -2.292 0.2102013-6-19 FOMC statement: Economic activity expanded
at a modes pace/Inflation is below its 2% ob-jective and is anticipated to move back over themedium term
Balance sheet/Guidance 2.659 0.411
2013-9-18 Fed released economic projections from theSeptember 17-18 FOMC meeting
Guidance -2.817 1.826
2014-3-19 FOMC statement: Economic activity slowedduring the winter months; Inflation persistentlybelow 2% objective could pose risks to eco-nomic performance
Balance sheet/ Guidance 1.885 -0.109
2014-5-9 Federal Reserve plans to conduct a series ofeight consecutive operations offering seven-dayterm deposits through its Term Deposit Facility(TDF)
Guidance 1.935 -0.459
2014-9-4 Federal Reserve plans to conduct a series ofeight consecutive seven-day term deposit oper-ations through its Term Deposit Facility (TDF)beginning in October and to Reform U.S. DollarLIBOR
Guidance 4.452 -1.006
2014-10-15 Fed conducted a fixed-rate offering of term de-posits through its Term Deposit Facility
Balance sheet -2.899 -1.685
140
2014-10-29 FOMC statement: Economic activity is expand-ing at a moderate pace and Inflation will likelybe held down by lower energy prices and otherfactors
Balance sheet/ Guidance 2.192 1.808
2014-12-2 Fed conducted a fixed-rate offering of term de-posits through its Term Deposit Facility
Balance sheet 2.514 -2.635
2014-12-17 FOMC statement: Economic activity is expand-ing at a moderate pace and inflation is expectedto rise gradually toward 2%
Balance sheet/ Guidance 3.541 1.163
141
Table 5.11: List of unexpected monetary defined from inflation swaps from 2008 to 2014
Date Activities Type Inflation swap GSCI2008-11-19 Minutes of Federal Open Market Committee,
September 29 October 7 and October 28 in 2008Balance sheet -0.17 -0.24
2008-11-24 the Federal Reserve will offer 150 billion dollarsin 13-day credit through its Term Auction Facil-ity
Balance sheet -0.94 3.12
2008-11-25 Fed conducted an auction of 150 dollar billionin 13-day credit through its Term Auction Facil-ity AND Minutes of Board discount rate meet-ing/ discount rate and primary credit rate. Itreleased also the creation of the Term Asset-Backed Securities Loan Facility (TALF) and Fedwill initiate a program to purchase the di-rect obligations of housing-related government-sponsored enterprises (GSEs) and mortgage-backed securities (MBS) backed by Fannie Mae,Freddie Mac, and Ginnie Mae
Balance sheet/IR -4.33 -1.83
2008-12-1 the Federal Reserve will offer 150 billion dollarsin 84-day credit through its Term Auction Facil-ity
Balance sheet -17 -2.64
2008-12-2 In light of continuing strains in financial mar-kets, the Federal Reserve on Tuesday an-nounced the extension through April 30, 2009,of three liquidity facilities: the Primary DealerCredit Facility (PDCF), the Asset-Backed Com-mercial Paper Money Market Fund LiquidityFacility (AMLF), and the Term Securities Lend-ing Facility (TSLF)
Balance sheet/Guidance 16.94 -1.20
2008-12-29 the Federal Reserve will offer 150 billion dollarsin 83-day credit through its Term Auction Facil-ity.
Balance sheet 4.33 1.25
2009-1-6 minutes of the Committee meeting held on De-cember 15-16, 2008
Balance sheet 7 0.68
2010-4-6 Minutes of the Federal Open Market Commit-tee, March 16, 2010
Balance sheet 0.03 0.01
2010-5-10 The Federal Reserve Board has authorized up tofive small-value offerings of term deposits un-der the Term Deposit Facility (TDF) to be con-ducted in coming months.
Balance sheet 0.02 1.12
142
2010-5-19 Minutes of the Federal Open Market Commit-tee, April 27-28, 2010
Balance sheet -0.02 -0.65
2010-7-20 Minutes of Board discount rate meetings, May24 through June 21, 2010; Federal Reserve an-nounces agreement with the Treasury Depart-ment regarding a reduction of credit protectionprovided for the Term Asset-Backed SecuritiesLoan Facility (TALF)
Balance sheet/IR -0.04 0.31
2010-9-7 Minutes of Board discount rate meetings IR -0.05 0.052011-6-29 Federal Reserve and other central banks an-
nounce an extension of the existing tempo-rary U.S. dollar liquidity swap arrangementsthrough August 1, 2012
Balance sheet 0.04 1.50
2011-9-6 Minutes of Board discount rate meetings IR -0.04 -0.462011-11-30 Coordinated central bank action to address
pressures in global money marketsGuidance 0.05 0.47
2012-1-4 Federal Reserve offers 3 billion dollars in 28-dayterm deposits through its Term Deposit Facility
Balance sheet 0.04 0.36
2012-10-4 Minutes of the Federal Open Market Commit-tee, September 12-13, 2012
Balance sheet 0.05 2.13
2013-3-13 Federal Reserve announces time changes forFOMC statements and news conferences
0.03 -0.40
2013-5-1 Federal Reserve issues FOMC statement Balance sheet 0.02 -2.052013-5-28 Minutes of the Board’s discount rates meetings
from April 8 and April 29, 2013IR -0.02 0.95
2013-7-31 Federal Reserve issues FOMC statement Balance sheet 0.03 1.232013-9-18 Fed released economic projections from the
Septermber 17-18 FOMC meetingGuidance 0.02 1.83
2014-3-20 Federal Reserve offers seven-day term depositsfrom its Term Deposit Facility
Balance sheet -0.02 0.31
2014-6-17 Federal Reserve announces results of offering ofseven-day term deposits on June 16
Balance sheet 0.04 0.09
2014-6-18 Federal Reserve Board and Federal Open Mar-ket Committee release economic projectionsfrom the June 17-18 FOMC meeting
Balance sheet/Guidance 0.02 1.03
2014-9-17 Federal Reserve issues FOMC statement on pol-icy normalization principles and plans;FederalReserve Board and Federal Open Market Com-mittee release economic projections from theSeptember 16-17 FOMC meeting
Balance sheet -0.04 -0.31
143
Bibliography
[1] Anzuini Alessio, Lombardi Macro J., and Pagano Patrizio.2013, The impact of monetarypolicy shocks on commodity prices. International Journal of Central Banking Vol. 9, No.3.
[2] Anzuini Alessio, Pagano Patrizio, and PisaniMassimiliano. 2007, Oil supply news in aVAR: Information from financial markets. Banca d’Italia.
[3] Basistha Arabinda, and Kurov Alexander.2015, The impact of Monetary policy sur-prises on energy prices.The Journal of Futures Markets Vol.35, No.1, 87-103
[4] Batten Jonathan A., Ciner Cetin, and Lucey Brian M.2010, The macroeconomic deter-minants of volatility in precious metals markets. Ressources Policy Vol. 35, pp 65-71.
[5] Bernanke S.Ben, and Kuttner N.Kenneth.2005, What explains the stock market’s reac-tion to Federal Reserve Policy? The Journal of Finance Vol.60, No.3, 1221-1257.
[6] Browne Frank, and Cronin David.2007, Commodity prices, Money and Inflation. Euro-pean Central Bank, Working paper series No. 738.
[7] Cavallo Michele, and Wu Tao.2006, Measuring oil-price shocks using market-basedinformation. Federal Reserve Bank of San Francisco.
[8] Frankel A.Jeffrey.1985, Commodity prices and money: Lessons from international fi-nance. American Journal of Agricutural Economics Vol. 66, No. 5.
[9] Frankel A.Jeffrey.1986, Expectations and Commodity Price Dynamics: The overshoot-ing model. American Agricultural Economics Assocation Vo.68,No. 2.
[10] Frankel A.Jeffrey.2006, The effect of monetary policy on real commodity prices. NBERWorking papers series 12713.
[11] Frankel A.Jeffrey, and Hardouvelis Gikas A.1985, Commodity prices, Money surprisesand Fed credibility. Journal of Money, Credit and Banking Vol. 17, No. 4.
[12] Garner C.Alan.1989, Commodity prices: Policy target or information variable? Journalof Money, Credit and Banking Vol. 21, No. 4, pp. 508-514.
[13] Gillman Max, Nakov Anton, and de Espana Banco.2008, Monetary effects on nominaloil prices. Banco de Espana Documentos de Trabajo No.0928.
144
[14] Glick Reuven, and Leduc Sylvain. 2011, Central bank announcement of asset purchasesand the impact on global financial and commodity markets. Federal reserve bank of SanFrancisco, Working Paper series.
[15] Gospodinov Nikolay, and Jamali Ibrahim.2013, Monetary policy surprises, Positions oftraders, and changs in commodity futures prices. Federal reserve bank of Atlanta, Work-ing paper series.
[16] Gregorio De Jose.2012, Commodity prices, Monetary policy and Inflation. Conferenceon Policy Responses to Commodity Price Movements.
[17] Gurkaynak S Refet, Sack Brian, and Swanson TEric.2005, Do acitions speak louderthan words? The response of asset prices to monetary policy actions and state-ments.International Journal of Central Banking Vol.1, No.1, pp 55-93.
[18] Hayo Bernd, Kutan Ali M., and Neuenkirch Matthias.2011, Communication Matters:U.S. Monetary policy and commodity price volatility. Working paper,http://www.uni-marburg.de/fb02/makro/forschung/magkspapers/05-2011hayo.pdf .
[19] Kilian Lutz.2008, The Economic effects of Energy price shocks. Journal of Economic Lit-erature 46:4, 871-909.
[20] Kilian Lutz, and VegaClara.2008, Do energy prices respond to U.S. macroeconomicnews? A test of the hypothesis of predetermined energy prices. Bord of Governors ofthe Federal Reserve system, International Fiance Discussion Papers No. 957.
[21] Kormilitsina Anna.2011, Oil price shocks and the optimality of monetary policy.Reviewof Economics Dynamics(Special issue: Source of Business Cycles) Vol.14, Issue 1, pp 199-223.
[22] Marteau Didier.2014, Étonnantes corrélations entre politique monétaire et cours desmatières premières. Revue Banque n.773.
[23] Pindyck Robert S., and Rotemberg Julio J. 1990, The excess co-movement of commodityprices. The Economic Journal Vol. 100, No. 403, pp. 1173-1189.
[24] Rigobon Roberto, and Sack Brian.2008, Noisy macroeconomic annoucements, mone-tary policy, and asset prices. Asset prices and monetary policy, University of Chicago Press,pp. 335-370.
[25] Thomson Amanda S., and Summers Peter M.2012, The effect of Monetary policy onreal commodity prices: A Re-examination. Journal of Economics Vol. 38, Issue 1, p 1-21.
145
Part IV
General conclusion
146
6General conclusion
This thesis provides an overview on the evolution of commodity markets and focuseson the price behaviour in both the commodity physical and the financial derivativesmarkets. Instead of directly analysing regulation in the commodity markets, this thesishighlights the market changes and specific market behaviour, which gives potentialimplications for market regulation.
Main results of thesisThis research on regulation of commodity markets has been conducted within twodifferent markets: commodity physical markets and commodity derivatives markets. i)Results in physical markets confirm the evidence of jump in agricultural commodityprices. Price volatility varies with time and is not constant. Commodity markets’ –at least agricultural markets’- prices oscillated during the period 2007/2008, whichcoincided with financial crisis. Relatively high frequent and small jumps in commodityprices are probably due to financial market factors instead of market fundamentals.The implication of these results in risk management of agricultural cooperatives isthat, although taking into account jump in risk measure, such as VaR does not out-perform traditional VaR with normal distribution. Considering this kind of extremeprice variation can be complementary for risk managers when facing a highly volatilecommodity market. ii) Findings for financial derivatives markets lead to two con-clusions. Firstly, commodity index funds are confirmed again as having a short-termimpact on futures prices in most products. Based on the theoretical conclusion aboutthe intermediary role of inventory on the impact of speculation on the commoditymarket, we have found that commodity index can influence commodity futures priceswithout necessarily passing through inventory changes, which is probably due toprice-inelasticity. Secondly, in relation to the impact of central bank announcementson commodity prices, the study shows that the central bank can be an alternativeregulator to influence the commodity market. Additionally, it shows that commodityprices include the information from macroeconomic factors, such as currency rate andinflation rate.
Potential Regulatory implicationBased on the results, suggestions regarding several regulatory implications are made.
147
In general, two aspects should be considered.
• On the one hand, regulation in the commodity market needs more transparent in-formation (lack of transparent information about strategies, positions etc.), whichneeds to be improved in both physical markets and commodity derivatives ex-changes and should include more information on market participants and positionin the financial market, and supply, demand, inventory, etc., in the spot market.
• On the other hand, more generations of commodity investment products exist inthe commodity market. Increasing interconnectedness between the commoditymarket and other financial markets, and the commodity market and the macroeco-nomic environment is becoming evident. Regulatory proposals for the commoditymarket should consider the stability of the global financial system and not belimited to the commodity market itself.
Other potential regulatory implication could be proposed based on our results.
• Enforce risk management inside cooperative companies and other producers. This studyapproves Basel III on the modelling standards for specific risks to banking organ-isations. Our results suggest that similar requirements might be imposed in thecommodity industry. As extreme market event is regularly observed in the com-modity market. Cooperative companies could apply alternative risk measures (ex:Improved Value-at-Risk) as complementary to recent standard methodologies.
• Surveillance of innovative investment products from commodity derivatives markets.In the financial market, commodity index funds are justified to have a short-termeffect on the market prices. An initial step that should be carried out in order tomonitor the index fund activities would be to report transparent information abouttheir market position and strategies. Regulatory authorities could be mandated tointervene directly in exchange trading of both traditional derivatives contracts andinnovative investment products to address the occurrence of market distortion.
• Improving policy coordination of other regulatory authorities in relation to commoditymarkets. Commodity markets are found to react to the central bank’s activities.This evidence suggests that the central bank uses a macroeconomic policy thatmight provide another regulatory authority to control commodity markets. Onthe other hand, monetary decision of central bank on global economy should notignore the potential influence on commodity markets.
148