Вы находитесь на странице: 1из 120

Graduate School ETD Form 9

(Revised 12/07)

PURDUE UNIVERSITY
GRADUATE SCHOOL
Thesis/Dissertation Acceptance

This is to certify that the thesis/dissertation prepared

By Chen Liu

Entitled US Dollar Exchange Rate and Crude Oil Price: A Common Driver Explanation

For the degree of Master of Science

Is approved by the final examining committee:


Wallace Tyner
Chair
Philip Abbott

Matthew Holt

To the best of my knowledge and as understood by the student in the Research Integrity and
Copyright Disclaimer (Graduate School Form 20), this thesis/dissertation adheres to the provisions of
Purdue University’s “Policy on Integrity in Research” and the use of copyrighted material.

Wallace Tyner
Approved by Major Professor(s): ____________________________________
____________________________________

Approved by: Kenneth Foster Oct. 6, 2010


Head of the Graduate Program Date
Graduate School Form 20
(Revised 1/10)

PURDUE UNIVERSITY
GRADUATE SCHOOL

Research Integrity and Copyright Disclaimer

Title of Thesis/Dissertation:
US Dollar Exchange Rate and Crude Oil Price: A Common Driver Explanation

Master of Science
For the degree of ________________________________________________________________

I certify that in the preparation of this thesis, I have observed the provisions of Purdue University
Teaching, Research, and Outreach Policy on Research Misconduct (VIII.3.1), October 1, 2008.*

Further, I certify that this work is free of plagiarism and all materials appearing in this
thesis/dissertation have been properly quoted and attributed.

I certify that all copyrighted material incorporated into this thesis/dissertation is in compliance with
the United States’ copyright law and that I have received written permission from the copyright
owners for my use of their work, which is beyond the scope of the law. I agree to indemnify and save
harmless Purdue University from any and all claims that may be asserted or that may arise from any
copyright violation.

Chen Liu
______________________________________
Printed Name and Signature of Candidate

10/06/2010
______________________________________
Date (month/day/year)

*Located at http://www.purdue.edu/policies/pages/teach_res_outreach/viii_3_1.html
US DOLLAR EXCHANGE RATE AND CRUDE OIL PRICE - A COMMON DRIVER EXPLANATION

A Thesis

Submitted to the Faculty

of

Purdue University

by

Chen Liu

In Partial Fulfillment of the

Requirements for the Degree

of

Master of Science

December 2010

Purdue University

West Lafayette, Indiana


UMI Number: 1490676

All rights reserved

INFORMATION TO ALL USERS


The quality of this reproduction is dependent upon the quality of the copy submitted.

In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.

UMI 1490676
Copyright 2011 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.

ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106-1346
ii

ACKNOWLEDGMENTS

The two-year graduate study at Purdue University is an important chapter in my life. I would have
stumbled all the way without the help from many people. Firstly, I want to express my sincere appreciation to
my major professor, Dr. Wallace Tyner. What I received from him is not only guidance in research, but also
understanding, encouragement and support. I would also like to thank Dr. Philip Abbott and Dr. Matthew
Holt, who gave me strong support throughout my research. I am a sensitive person not very good at
communication. But all my committee members are very patient with me. I learned much from them.
I am also thankful for the precious graduate study opportunity that the Department of Agricultural
Economics offered me. It changed my life. My sincere thanks also go to Lou Ann Baugh, David Ortega and
many other people in the Department of Agricultural Economics for their kind help in the past two years.
I would like to thank my wife, my parents and my parents-in-law. Their love and encouragement always
give me strength and courage.
At last, I would like to thank God for leading me and offering me the wisdom to complete this thesis.
iii

TABLE OF CONTENTS

Page

LIST OF TABLES...........................................................................................................................................v

LIST OF FIGURES.........................................................................................................................................vi

ABSTRACT...................................................................................................................................................vii

CHAPTER I – INTRODUCTION...................................................................................................................1
1.1 Background..................................................................................................................................1
1.1.1 Brief Market Movement Review.................................................................................1
1.1.2 Brief Introduction of Previous Research Outcomes....................................................1
1.1.3 Motivation................................................................................................................... 2
1.2 Problem Statement.......................................................................................................................2
1.2.1 Major Flaws of Previous Research..............................................................................2
1.2.2 A Common Driver Perspective: Possibility and Necessity..........................................3
1.2.3 Further Consideration on Candidate Common Driver.................................................3
1.3 Objectives....................................................................................................................................5
1.4 Hypotheses..................................................................................................................................5
1.5 Organization of the Thesis............................................................................................................5

CHAPTER II – LITERATURE REVIEW.........................................................................................................6


2.1 Theoretical Studies of the Nexus..................................................................................................6
2.1.1 The Impact of the Dollar Exchange Rates on Crude Oil Prices…..…………………..6
2.1.1.1 The Supply Side Purchasing Power Channel.............................................6
2.1.1.2 The Demand Side Purchasing Power Channel...........................................7
2.1.1.3 The Asset Channel.....................................................................................7
2.1.1.4 The Monetary Policy Channel....................................................................7
2.1.1.5 The Market Adjustment Channel...............................................................8
2.1.2 The Impact of the Crude Oil Prices on the Dollar Exchange Rates.............................8
2.1.2.1 Terms of Trade.......................................................................................... 8
2.1.2.2 Balance of Payments................................................................................. 9
2.2 Empirical Studies of the Nexus.....................................................................................................9
2.3 Assessment.................................................................................................................................13
2.4 Studies on the Determinants of Each Variable............................................................................14
2.4.1 Main Determinants of Dollar Exchange Rate............................................................14
2.4.2 Main Determinants of Crude Oil Price......................................................................21
2.4.2.1 Macroeconomic Fundamentals................................................................ 21
2.4.2.2 Supply Related Factors............................................................................ 26
2.4.2.3 Other Factors............................................................................................30
2.5 Studies on Gold Price..................................................................................................................31
2.6 Summary.....................................................................................................................................33

CHAPTER III – METHODOLOGY...............................................................................................................34


iv

Page

3.1 Theoretical Hypotheses..............................................................................................................34


3.1.1 Hypothesis One......................................................................................................... 34
3.1.2 Hypothesis Two.........................................................................................................35
3.1.3 Hypothesis Three.......................................................................................................35
3.2 Measurement of Correlation and Volatility - Exponential Smoothing........................................36
3.3 Description of Models and Test Procedures...............................................................................38
3.3.1 Data Stationarity........................................................................................................38
3.3.2 Cointegration Test.................................................................................................... 39
3.3.3 Stability Test of the Estimated Cointegration Relationships.....................................40
3.3.4 Timing of Structural Breaks..................................................................................... 41
3.3.5 Granger Non-Causality Test......................................................................................41

CHAPTER IV – DATA..................................................................................................................................44
4.1 Considerations on Dollar Exchange Rate Data...........................................................................44
4.1.1 Justification for Dollar/Euro Exchange Rate.............................................................44
4.1.2 Nominal Exchange Rate vs. Real Exchange Rate.....................................................45
4.1.3 Bilateral Exchange Rate, NEER, REER and Commodity Currency.........................45
4.1.4 Spot Exchange Rate vs. Forward Exchange Rate......................................................46
4.2 Considerations on Crude Oil Price Data.....................................................................................46
4.2.1 Brent, West Texas Intermediate and Dubai Crude Oil Price.....................................46
4.2.2 Nominal Price vs. Real Price.....................................................................................46
4.2.3 Spot Price vs. Futures Price.......................................................................................46
4.3 Considerations on Gold Price Data.............................................................................................47
4.3.1 London Gold Fix....................................................................................................... 47
4.3.2 London AM Fix vs. London PM Fix.........................................................................47
4.4 Data Frequency..........................................................................................................................47
4.5 Description of Data Sources.....................................................................................................47
4.6 Data Preparation........................................................................................................................48

CHAPTER V – EMPIRICAL RESULTS........................................................................................................49


5.1 Correlation Coefficients..............................................................................................................49
5.1.1 Dollar Exchange Rate and Gold Price.......................................................................49
5.1.2 Crude Oil Price and Gold Price..................................................................................50
5.1.3 Dollar Exchange Rate and Crude Oil Price...............................................................52
5.1.4 Summary of Correlation Coefficient Analysis..........................................................53
5.2 Volatilities..................................................................................................................................54
5.2.1 Dollar Exchange Rate................................................................................................54
5.2.2 Crude Oil Price..........................................................................................................55
5.2.3 Gold Price..................................................................................................................56
5.2.4 Summary of Volatility Analysis................................................................................57
5.3 Econometric Results...................................................................................................................57
5.3.1 Data Stationarity........................................................................................................58
5.3.2 Co-integration Test....................................................................................................59
5.3.3 Parameter Stability Test............................................................................................ 59
5.3.4 Timing of Structure Breaks.......................................................................................71
5.3.5 Granger Non-Causality (GNC) Test..........................................................................80
5.3.6 Discussion................................................................................................................. 93

CHAPTER VI –SUMMARY & CONCLUSION...........................................................................................98


6.1 Conclusion.................................................................................................................................98
6.2 Limitations and Future Research................................................................................................99

LIST OF REFERENCES.............................................................................................................................101
v

LIST OF TABLES

Table Page

2.1 A Partial List of Empirical Studies on Exchange Rate and Crude Oil Price......................................11
2.1 A Partial List of Empirical Studies on Exchange Rate and Crude Oil Price......................................11
2.2 A Brief Summary of Table 2.1............................................................................................................13
2.3 Macroeconomic Fundamentals Used in Selective Literature on Exchange Rate................................18
2.4 Macroeconomic Fundamentals Used in Selective Literature on Crude Oil Price..............................23
2.5 Supply Related Factors Used in Selective Literature on Crude Oil Price...........................................29
2.6 Other Factors Used in Selective Literature on Crude Oil Price..........................................................32
3.1 Summary of Possible Outcomes.........................................................................................................36
5.1 Average Correlation Coefficient of Dollar Exchange Rate and Gold Price........................................51
5.2 Average Correlation Coefficient of Crude Oil Price and Gold Price..................................................52
5.3 Average Correlation Coefficient of Dollar Exchange Rate and Crude Oil Price...............................54
5.4 Correlation Coefficient Matrices........................................................................................................55
5.5 Average Volatility of Dollar Exchange Rate......................................................................................56
5.6 Average Volatility of Crude Oil Price................................................................................................56
5.7 Average Volatility of Gold Price........................................................................................................57
5.8 The Results of ADF Test and PP Test.................................................................................................59
5.9 Results of Co-integration Test............................................................................................................60
5.10 Results of Parameter Stability Tests...................................................................................................61
5.11 Results of Co-integration Test on Subsample of Crude Oil Price and Gold Price.............................72
5.12 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Gold Price....................72
5.13 Results of Co-integration Test on Subsample of Crude Oil Price and Dollar Exchange Rate...........74
5.14 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Dollar Exchange Rate. 74
5.15 Results of Co-integration Test on Subsample of Crude Oil Price and Dollar Exchange Rate...........76
5.16 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Dollar Exchange Rate. 76
5.17 Temporary Reciprocal Currency Arrangements Timeline................................................................ 80
5.18 U.S. Government Assets Other Than Official Reserve Assets during 2008-2009.............................81
5.19 AIC Values and Lag Length k Selection............................................................................................82
5.20 Results of GNC Test on Dollar Exchange Rate and Gold Price.........................................................83
5.21 Results of GNC Test on Crude Oil Price and Gold Price (Subsample One)......................................84
5.22 Results of GNC Test on Crude Oil Price and Gold Price (Subsample Two)......................................84
5.23 Results of GNC Test on Crude Oil Price and Gold Price (Full Sample)............................................85
5.24 Results of GNC Test on Subsample One of Dollar Exchange Rate and Crude Oil Price...................86
5.25 Results of GNC Test on Subsample Two of Dollar Exchange Rate and Crude Oil Price..................87
5.26 Results of GNC Test on Subsample Three of Dollar Exchange Rate and Crude Oil Price................88
5.27 Results of GNC Test on Subsample Four of Dollar Exchange Rate and Crude Oil Price..................89
5.28 Results of GNC Test on Entire Sample of Dollar Exchange Rate and Crude Oil Price.....................90
5.29 GNC Test Result Summary............................................................................................................... 93
vi

LIST OF FIGURES

Figure Page

4.1 OPEC Major Trade Partners and Shares 1980-2007.........................................................................46


5.1 Correlation Coefficients between Dollar Exchange Rate and London PM Fix Gold Price...............50
5.2 Correlation Coefficients between Crude Oil Price and London PM Fix Gold Price.........................52
5.3 Correlation Coefficients between Dollar Exchange Rate and Crude Oil Price.................................53
5.4 Exponentially Smoothed Annualized Volatilities of Three Prices....................................................55
5.5 F Statistic Sequence from Regression of Dollar Exchange Rate on Gold Price................................62
5.6 Gold Price, Actual Dollar Exchange Rate and Fitted Dollar Exchange Rate......................................62
5.7 U.S. Historical 12-Month Percent Change of CPI in 1991-2010........................................................63
5.8 F Statistic Sequence from Regression of Crude Oil Price on Gold Price............................................64
5.9 Gold Price, Actual Crude Oil Price and Fitted Crude Oil Price...........................................................64
5.10 Historical Daily Prices of Crude Oil and Gold 1986-2010..................................................................65
5.11 Co-movement Days of Dollar Exchange Rate and Gold Price (2006.5-2007.10)...............................67
5.12 Industrial Production: Gross Value of Products 1999-2010................................................................68
5.13 U.S. Crude Oil Consumption 1999-2010 (Thousand Barrels/Day)....................................................68
5.14 F Statistic Sequence from Regression of Crude Oil Price on Dollar Exchange Rate..........................70
5.15 Dollar Exchange Rate and Crude Oil Price.........................................................................................70
5.16 Target Federal Funds Rate (%)...........................................................................................................71
5.17 F Statistic Sequence from Subsample Regression of Crude Oil Price on Gold Price..........................73
5.18 F Statistic Sequence from Subsample Regression of Crude Oil Price on Dollar Exchange Rate........75
5.19 F Statistic Sequence from Subsample Regression of Crude Oil Price on Dollar Exchange Rate........77
5.20 U.S. Total Banking Claims on Foreigners 2003-2009........................................................................78
5.21 Historical Movements of Dollar Exchange Rate and Crude Oil Price during 2008 – 2009.................78
vii

ABSTRACT

Chen Liu. M.S., Purdue University, December 2010. US Dollar Exchange Rate and Crude Oil Price - A
Common Driver Explanation. Major Professor: Dr. Wallace Tyner.

This study analyzes the relationship between dollar exchange rate and crude oil price. Unlike
previous literature which established a bilateral causal relationship between the two prices, this paper
proposes that gold price is likely to be a common driver of the two prices and therefore, the bilateral
relationship is a part of a triangular system of gold price, dollar exchange rate and crude oil price. The results
show that while in the short run gold price movement does not Granger cause the movement of the dollar
exchange rate and crude oil price, it is the Granger causality between the two prices in the long run. Causality
is also identified from dollar exchange rate to crude oil price in both short run and long run. But crude oil
price is much less influential in its bilateral relationship with dollar exchange rate. As a result, a triangular
system of gold price, dollar exchange rate and crude oil price is established. Dollar exchange rate not only has
direct impacts on crude oil price, but also functions as an intermediary vehicle through which gold price
movement indirectly affects crude oil price. However, the relationship between dollar exchange rate and
crude oil price is unstable with multiple structural breaks over the sample period. This partially explains why
previous studies have mixed results.
1

CHAPTER I. INTRODUCTION

Dollar exchange rate and crude oil price are of great importance to today’s world economy. As the start
of the study, this chapter provides a brief background introduction of the relationship between the two prices,
a statement of problems, objectives, major hypotheses and thesis organization.

1.1. Background

1.1.1. Brief Market Movement Review


In July 2008, the crude oil price in the international market stopped its more than one year rapid
escalation and started to decline. After the daily closing spot WTI price hit the bottom of 30.28 dollars/barrel
on December 23, 2008, it gradually rebounded and reached around 78 dollars/barrel in November 2009.
During this period, the U.S. dollar exchange rate (against the Euro) first appreciated from 1.59 in July 2008 to
1.25 in November 2008. Then starting from February 2009, the dollar exchange rate slowly depreciated to
1.50 in November 2009. Not surprisingly, there seems to be a negatively correlated relationship between
crude oil price and U.S. dollar exchange rate. That is, the weaker the U.S. dollar exchange rate, the stronger
the crude oil price. But this is not the only co-movement pattern shown by these two prices. During
December 2009 and April 2010, the monthly crude oil price increased from $74 per barrel to $84 per barrel.
The U.S. dollar appreciated from around 1.50 to 1.33. The two prices exhibited a positively correlated
relationship. As indicated by historical data starting from 1999, such a positive correlation relationship also
appeared in 1999, 2000 and 2005. In fact, U.S. dollar exchange rate and crude oil price show both positive
and negative correlation relationships, which switched back and forth in the past decade.

1.1.2. Brief Introduction of Previous Research Outcomes


The interesting relationship between U.S. dollar exchange rate and crude oil price has attracted great
research attention. Previous studies work hard to find causality in the relationship and to provide reasons to
justify it. However, the findings of theoretical analyses and empirical studies are mixed. While some studies
argue that exchange rate leads oil price, other studies assert the opposite (see, for example, Yousefi and
2

Wirjanto (2004) and Amano and van Norden (1998) for contrary results). As for the reasons behind the
causality, those studies in favor of the former present five channels through which a fall in the value of U.S.
dollar can raise crude oil prices. The counterpart studies also address two ways in which oil price fluctuation
moves exchange rate. While all these studies improve people’s knowledge about the relationship, the
discussion on this issue is still far from conclusive.
Besides the work mentioned above, there are related research papers focusing on the formation and
dynamics of exchange rate and crude oil price respectively that indirectly contribute to the understanding of
their relationship. The research outcomes of these studies will be discussed in detail in the next chapter.

1.1.3. Motivation
Perhaps U.S. dollar exchange rate and crude oil price are two of the most influential asset prices in
today’s world economy. The U.S. dollar is the major foreign reserve currency of almost all countries and the
payment vehicle that facilitates a predominant proportion of international transactions. The changes of dollar
exchange rate affect not only the values of global products, assets and wealth, but also the investment
decisions of individuals, businesses and sovereign wealth funds. Crude oil, on the other hand, plays a role of
key source of energy, indispensable supplier of various chemical raw materials, and increasingly important
investment commodity. As a result, the crude oil price drives the price level of a variety of inputs and outputs
in the economy, as well as inter-market and international capital flows.
Because of the functions of dollar exchange rate and crude oil price in the global economy and their ever
changing co-movement patterns, it is necessary to explore further their relationship and reveal the reasons
behind the numbers and charts. A better understanding of the linkage between dollar exchange rate and crude
oil price will be valuable for investors in commodity and foreign exchange markets to shape reasonable
anticipations of market movements and for researchers to comprehend the dynamics of the two prices
respectively.

1.2. Problem Statement

1.2.1. Major Flaws of Previous Research


While straightforward and appealing, previous studies are flawed in several aspects.
By considering the linkage between U.S. dollar exchange rate and crude oil price a bilateral
relationship, previous research underestimates the effect of all other factors that may drive these two prices.
The observed correlation relationships between the two prices are probably just a part of a dynamic triangular
relationship among U.S. dollar exchange rate, crude oil price and a common driver or common drivers.
3

Therefore, explorations within the scope of two variables may not well explain the actual relationships
between them.
Moreover, by proposing possible explanations and performing empirical studies, researchers try to
establish a causal relationship between U.S. dollar exchange rate and crude oil price (see, for example,
Bénassy-Quéré et al.(2005) and Amano and van Norden(1998)). But the theoretical reasoning and statistical
results of these studies are mixed and do not lend enough support to what they propose. If a triangular system
is the case, it is possible that the two prices are causally related with a third factor respectively but not so
themselves. Their correlation relationships, then, are merely correlation relationships.
Besides, although US dollar exchange rate and crude oil price show both positive and negative
correlation relationships, most of the theoretical analyses focus on explanation of the negative ones. The
periods when the two prices move together have not received enough academic attention. The stories in such
periods have not been fully revealed.

1.2.2. A Common Driver Perspective: Possibility and Necessity


U.S. dollar exchange rate and crude oil price movements are multi-factor determined outcomes.
According to the exchange rate determination theories, exchange rate is driven by a variety of variables, such
as the relative economic performance, interest rate differentials and productivity differentials. The crude oil
price is also influenced by issues like micro-market structure and world economic growth. Moreover, both
prices are closely related to the gold price, the safe haven of capital. Therefore, it is highly likely that there are
factors qualified as common drivers of U.S. dollar exchange rate and crude oil price.
However, almost no previous research tries to view U.S. dollar exchange rate and crude oil price in a
triangular system or to explore candidate common drivers, necessitating the discussions in this study.
Another advantage of a triangular system is that the relationships of the common driver with the two prices
respectively may provide an explanation to both the negative and positive correlation relationships observed
between the two prices. So this study tries to fill the “blank” of the literature and present a new perspective on
the relationship between U.S. dollar exchange rate and crude oil price.

1.2.3. Further Consideration on Candidate Common Driver


To narrow down the scope of possible common drivers, it is important to set several criteria. The
candidates are expected to meet at least two requirements: 1) The candidates have to be related to the demand
side or supply side or both sides of the two markets; 2) Ideally, the candidates are expected to be exogenous to
the formation process of the two prices.
4

The distinction between exogenous and endogenous variables in a model has led to extensive literature
1
and discussion. This study is not intended to go far on this topic but to apply the concept of “exogeneity” in
a practical way while maintaining its economic sense. In economics, a variable in the model is called
exogenous if it could reasonably be expected to vary “autonomously,” independently of the other variables in
the model (Greene, 2002). Unfortunately, this neat classification is of fairly limited use in macroeconomics,
where almost no variable can be said to be truly exogenous in the fashion that most observers would
understand the term. As a result, researchers draw the distinction largely on statistical grounds (Greene,
2002). Engle, Hendry, and Richard (1983) define a set of variables Xt in a parameterized model to be weakly
exogenous if the full model can be written in terms of a marginal probability distribution for Xt and a
conditional distribution for Yt | Xt such that estimation of the parameters of the conditional distribution is no
less efficient than estimation of the full set of parameters of the joint distribution. In this study, the variables
will be analyzed using the statistically defined exogeneity but be explained using the economically defined
exogeneity.
Two kinds of variables immediately meet the first requirement: macroeconomic fundamentals and gold
price. Clearly, macroeconomic fundamentals such as interest rate, output and money supply have
considerable impact on dollar exchange rate and crude oil price. But a major problem with them is their
inherent endogeneity, which may result in causality interpretation difficulty and other problems (see, for
example, Chen, Rogoff and Rossi, 2008). Gold price also has the potential to be a qualified candidate. Firstly,
gold price is closely related to dollar exchange rate. Gold is mainly priced and traded in dollars in the
international market. The demand for dollars may be affected by gold demand changes. In recent decades,
gold is more and more frequently used to hedge inflation, political unrest and currency risk. So once there are
exogenous shocks to the world economy, the gold price tends to respond quickly. Gold price has deviated
from the inner value of gold especially after the year 2001 and is more likely to reflect investors’ anxiety
about economic prospects and capital safety. Dollar exchange rate to some extent mirrors the health of U.S.
economic development and of world economic development. Gold price is likely to go up if dollar exchange
rate is expected to depreciate due to factors that may cause economic slowdown and vice versa. Second, gold
price is closely related to crude oil price. Both gold and crude oil are increasingly important investment
options often added to investor asset portfolios. Their prices tend to move together driven by investment
purposes. As a valuable and standard commodity, gold is also used to hedge inflationary risk. The upsurge of
gold price may be interpreted as the rise of inflation expectation. As a result, capital flows will rush to the
gold market, crude oil market as well as other commodity markets and push up the prices of these
commodities. Since gold price is heavily driven by demand for investment, inflation hedging and capital
safety, it may be regarded as exogenous and be used directly in structural models and econometric models.

1. See, for example, Granger (1969) and Engle, Hendry, and Richard (1983).
5

1.3. Objectives
The objective of this study is to understand the relationship between U.S. dollar exchange rate and crude
oil price in a triangular framework by evaluating the role of gold price as part of the oil price-dollar exchange
rate linkage.
This study will focus on the linkage between U.S. dollar exchange rate and gold price and the linkage
between crude oil price and gold price. In particular, this study will try to find answers to the following
questions: is gold price the reason for the movements of dollar exchange rate and crude oil price? Do the
relationships of gold price with the two prices respectively help explain the ever switching correlation
relationship of the two prices?

1.4. Hypotheses
To fulfill the objective, this study will mainly test three hypotheses:
Hypothesis One: The U.S. dollar exchange rate and gold price have a statistical and economically
plausible causality relationship running from gold price to U.S. dollar exchange rate.
Hypothesis Two: The crude oil price and gold price have a statistical and economically plausible
causality relationship running from gold price to crude oil price.
Hypothesis Three: The dollar exchange rate and crude oil price have a statistical causality relationship. 2

1.5. Organization of the Thesis


The paper is organized as follows. Chapter 2 introduces a review of the theoretical and empirical
literature on the relation between exchange rates and crude oil price. Related studies about the formation and
dynamics of U.S. dollar exchange rate and crude oil price will be discussed as well. Chapter 3 analyzes the
role of gold price as a common driver and explains the empirical methods and approaches used. Chapter 4
describes data sources and considerations. Chapter 5 presents and interprets econometric results. Chapter 6
provides a summary of results together with limitations of this research and proposed directions for future
research.

2. Section 3.1.3 will discuss the Hypothesis Three in detail.


6

CHAPTER II. LITERATURE REVIEW

The nexus between crude oil prices and dollar exchange rate has been noted by a number of researchers
and discussed widely in previous literature. However, because of the differences in theoretical standpoint
adopted, methodology employed, period of analysis, and data collected, the results of these studies are
contradictory and far from conclusive. This section will briefly review and summarize previous theoretical
and empirical research on the nexus of the two prices. Then the studies on the determination of each price will
be presented. The studies on gold price will also be introduced.

2.1. Theoretical Studies of the Nexus


Because of the divergence on the question of which one causes the other, theoretical interpretations of
the linkage between crude oil prices and dollar exchange rates have followed two avenues, with the first one
emphasizing the impact of exchange rate movements on crude oil price movements and the second one
stating the reverse.

2.1.1. The Impact of the Dollar Exchange Rates on Crude Oil Prices
There are at least five possible channels through which a fall in the value of the dollar can raise crude oil
prices. They are supply side purchasing power channel, demand side purchasing power channel, asset
channel, monetary policy channel, and market adjustment channel.

2.1.1.1. The Supply Side Purchasing Power Channel


Because almost all of standard commodities, including crude oil, are priced and settled in dollars, a
change in dollar exchange rates alters the terms of trade of all the countries. The extent of this change
depends on the proportion of "dollar goods" relative to "non-dollar goods" in their trade structure
(Schulmeister, 2000). For oil-exporting countries, their export revenues are predominantly dollar
denominated assets but their imports, largely from European countries, Japan and other non-U.S. areas, are
7

mainly Euro-denominated. 3 So once the dollar depreciates, to maintain the purchasing power of their dollar
export revenues, countries with pricing power tend to increase oil prices. Moreover, Cheng (2008) mentions
that producers outside the dollar area also have price pressures facing declining profits in local currency
caused by dollar depreciation.

2.1.1.2. The Demand Side Purchasing Power Channel


Fluctuations in the exchange rate of the U.S. dollar create disequilibria in the market for crude oil. Since
crude oil is priced and settled in dollars, dollar depreciation makes it less expensive for consumers in
non-dollar regions, thereby increasing their demand, which eventually causes the oil price to go up (Austvik,
1987; Cheng, 2008).

2.1.1.3. The Asset Channel


A depreciation of U.S. dollar means decrease of the returns on dollar-denominated financial assets in
foreign currencies, which can make commodities a more attractive class of alternative assets to foreign
investors. Moreover, dollar depreciation raises risks of inflationary pressure in the United States, prompting
investors to move toward real assets—such as commodities—to hedge against inflation. For example,
commodity markets rallied in the 1970s amid high inflation. Cheng (2008) uses six commodities in a study of
the relationship between dollar exchange rate and commodity prices and finds that among crude oil, gold,
aluminum, copper, corn, and wheat, the former two are more suitable than others as “store of value.” In
general, nonrenewable commodities such as crude oil are a better store of value than perishable or renewable
commodities. Zhang et al. (2008) also argues that when the U.S. dollar is expected to decrease, a large
amount of money will flow to the oil market. Both new investment and speculation opportunities can be
derived in the process for market participants.

2.1.1.4. The Monetary Policy Channel


This channel argues that a fall of the value of dollar could lead to loose monetary policy in other
economies, especially in countries with currencies pegged to the dollar. This could result in lower interest
rates and increased liquidity, thereby stimulating demand for commodities, including crude oil, and other
assets (Cheng, 2008).

3. For example, in the 1970s, OPEC members like Algeria, Iran, Iraq and Kuwait purchased more than 70%
of their import needs in the West European and Japanese markets. Meanwhile, those like Saudi Arabia,
Kuwait, and some other Persian Gulf oil exporters hold the bulk of their accumulated reserves in U.S.
securities and dollar accounts (Amuzegar, 1978). Although OPEC’s import share from Europe has decreased
over time, Europe still has the largest share of OPEC’s total imports.
8

2.1.1.5. The Market Adjustment Channel


If the foreign exchange market is more efficient than the crude oil market, the former market absorbs
new information and adjusts more quickly than the latter one. Then exchange rate may be useful for
forecasting crude oil price movements, and exchange rate changes will likely affect crude oil price. This idea
receives support from several studies. In his famous paper on exchange rate, Dornbusch (1976) proposes that
exchange rates and asset markets adjust fast relative to goods markets. Chen et al., (2008), find that
"commodity currency" exchange rates have strong power in predicting global commodity prices but the
reverse relationship is less robust. They explain that the exchange rate is fundamentally a forward-looking
variable that likely embodies information about future commodity price movements. In contrast, commodity
prices tend to be quite sensitive to current conditions because both demand and supply are typically quite
inelastic. In addition, financial markets for commodities tend to be far less developed than for the exchange
rate. Therefore, as interpreted by Breitenfellner et al. (2008), foreign exchange markets are possibly more
efficient than oil markets and can anticipate developments in the real economy that affect the demand for and
supply of oil.
Clearly, most of the above reasoning implies that a causality relationship will go from the dollar to oil
prices and that dollar exchange rate and oil price are negatively correlated. 4 That is, crude oil price increases
when dollar depreciates. These possible channels may be at work at the same time, but it is difficult to tell
which one dominates.

2.1.2. The Impact of the Crude Oil Prices on the Dollar Exchange Rates
There are mainly two lines of research addressing the impact of the crude oil prices on the dollar
exchange rate. The first strand focuses on the terms of trade, and the second on the balance of payments.

2.1.2.1. Terms of Trade


Zhou (1995) proposes a model with two countries, both of which have tradable sector and non-tradable
sector. If the home country is more dependent on imported oil, a real oil price rise may increase the prices of
tradable goods in the home country by a greater proportion than in the foreign country, and thereby cause a
real depreciation of the home currency. 5 McGuirk (1983), Zhou (1995) and Chen and Chen (2007) argue that
a rise in the real price of oil will worsen the terms of trade of a net oil-importing country. In order to improve
competitiveness, the home country would have to raise the nominal exchange rate, which would lead to a
further real depreciation. Amano and van Norden (1998) reason that higher oil prices lead to an appreciation

4. The market efficiency channel explains how exchange rate is likely to affect crude oil price, but may not
clearly indicate a negatively correlated relationship between the two.
5. See, for example, Chen and Chen (2007) for a similar variant.
9

of the U.S. dollar in the long run, because of an often-mentioned hypothesis of relative terms of trade effect.
This effect posits that while the United States is a significant energy importer, it is less dependent on imports
than most of its major trading partners (apart from Canada and Mexico). Therefore, while the dollar should
depreciate in absolute terms, it should be expected to depreciate less than the currencies of its major trading
partners. In other words, while higher oil prices lower the U.S. absolute terms of trade, it raise the U.S. terms
of trade relative to its industrialized trading partners. Thus higher oil prices will lead to an appreciation of
dollar. Cashin et al. (2004) develops a model that embodies both the Balassa–Samuelson effect and terms of
trade. In their model, an increase in the international price of the primary commodity will increase wages in
the commodity sector. Since wages are assumed to be equal across sectors, the increase in wages will raise
the relative price of the non-traded good and, therefore, appreciate the real exchange rate.

2.1.2.2. Balance of Payments


Golub (1983) views a rise in oil prices as a wealth transfer from oil importing countries to oil-exporting
ones. The impact on exchange rates then depends on the distribution of oil imports across oil-importing
countries and on portfolio preferences of both oil-importing countries (whose wealth declines) and
oil-exporting ones (whose wealth increases). In a three-country model, Krugman (1980) argues that in the
short run, the real dollar exchange rate depends on OPEC’s portfolio choices. But since the spending of
OPEC’s accumulated wealth on imports of goods from industrial countries rises over time, in the long run the
real exchange rate will depend on the geographic distribution of OPEC imports. Assuming that oil-exporting
countries have a strong preference for dollar-denominated assets but not for U.S. goods, an oil price increase
will lead first to dollar appreciation and later to even greater dollar depreciation. Be´nassy-Que´re´ et al (2007)
extend Krugman’s model to the case of four countries (the United States, the Eurozone, OPEC and China)
and only one exchange rate (the dollar against the euro) and achieve similar outcomes. The emergence of
China in both oil and foreign exchange markets could strengthen the positive causality found from the oil
price to the dollar in the short run but reverse its sign in the long run.
The above two theories seem to give contradictive predictions of the impact of crude oil price changes
on exchange rate. The first one indicates a positive causality relationship from oil price to dollar exchange
rate, while the third one shows that both positive and negative causality relationships may occur.

2.2. Empirical Studies of the Nexus


The outcomes of empirical studies are even more confusing and contradictory than the predictions of
theoretical studies. Table 2.1 lists 13 highly relevant pieces of the empirical literature on the exchange rate-oil
price linkage. While all these empirical studies suggest unidirectional causality relationship, because of the
differences in theoretical standpoint, methodology, target period and data collection, these studies display a
10

great variety of estimation results. By summarizing all the outcomes in Table 2.2, however, we can loosely
infer that most studies on the causality relationship from oil price to exchange rate obtain a positive
correlation between the two prices and that most studies on the causality relationship from exchange rate to
oil price document a negative correlation. This is, to some extent, consistent with the main predictions of
theoretical studies.
11
11

Table 2.1 A Partial List of Empirical Studies on Exchange Rate and Crude Oil Price
Parameter
Econometric Exchange Rate
Study Theory Stability Period Oil Data Frequency Causality Sign
Method Data
Treatment
1973:Q1- Oil Price
Zhou (1995) Relative price VAR, VECM, CO No Real bilateral Real Dubai Quarterly +
1993:Q2 Exchange rate
Amano and van 1973:01- Oil Price
Terms of trade CO,GC Yes REER Real WTI Monthly +
Norden, (1998) 1993:06 Exchange rate
Chaudhuri and
1973:01- Oil Price
Daniel Relative price CO, ECM No Real bilateral Real Dubai Monthly +
1996:02 Exchange rate
(1998)
Sadorsky VAR, CO, VECM, 1987:01-
- Yes EER USD Future Monthly USDOil price -
(2000) GC 1997:09
Commodity Commodity
Chen, Rogoff 1973Q1-
Terms of trade DOLS,GMM Yes Real dollar Price Quarterly PriceExchange
(2003) 2000Q1*
Indices Rate
Commodity Commodity
Cashin et 1980:01 -
Terms of trade CO,GC, ECM Yes REER Price Monthly PriceExchange
al.(2004) 2002:03
Indices Rate
Spot WTI,
Yousefi and 1989-
Purchasing Power GMM No REER USD Brent, Monthly USDOil price -
Wirjanto (2004) 1999
OPEC
Bergvall 1975:Q1- Oil Price
Terms of trade VAR, VECM, CO No REER Real OPEC Quarterly +
(2004) 2001:Q1 Exchange rate
Real WTI,
Chen and Chen CO,FMOLS, 1972:01- Monthly Oil Price
Terms of trade Yes Real Brent, +
(2007) DOLS, PMG 2005:10 Panel Exchange rate
Dubai
Benassy-Quere Balance of 1974:01-
CO, VECM, GC No REER USD Real Monthly Oil PriceUSD +
et al (2007) payment 2004:11

11
11
11
12

Table 2.1 continued


Parameter
Econometric Exchange Rate
Study Theory Stability Period Oil Data Frequency Causality Sign
Method Data
Treatment
Exchange Rate
Chen, Rogoff Commodity Commodity
1972Q1-  Not
and Rossi Market Adjustment GC Yes Currency Nominal Price Quarterly
2008Q1* Commodity indicated
(2008) Spot dollar Indices
Price
Purchasing power,
NEER USD REER USD
Cheng (2008) asset, monetary CO, DOLS, ECM No 1980-2007 Real Spot Monthly -
USD Oil price
policy
Zhang et VAR, VaR 2000.1.4 – Nominal Spot USD
Asset No Spot WTI Daily -
al(2008) GC,TGARCH, 2005.5.31 EUR/USD Oil price
Note: VAR refers Vector Auto Regression Model; VECM refers to Vector Error Correction Model; CO refers to Cointegration; GC refers to Granger Causality
Test; GMM refers to Generalized Method of Moments; DOLS refers to Dynamic OLS; FMOLS refers to fully modified OLS; PMG refers to Pooled mean
group estimation; VaR refers to Value at Risk; TGARCH refers to Threshold GARCH; REER refers to real effective exchange rate; NEER refers to nominal
effective exchange rate; “+” means positive correlation, “-” means negative correlation.
12
13

Table 2.2 A Brief Summary of Table 1


Causality
Correlation Oil PriceExchange Rate Exchange Rate  Oil Price

Positive Correlation 6 0
Negative Correlation 0 4

2.3. Assessment
The previous theoretical and empirical research efforts have greatly improved our understanding of the
nexus between exchange rates and crude oil prices. However, there are several aspects to which current
literature has not paid enough attention:
1) Historical data starting in the 1970s shows that exchange rate and crude oil price have both positive
and negative correlation relationships and that the sign of such relationships switch back and forth over time.
Although many of current studies focus on the period starting from the 1970s, none of them, either theoretical
or empirical, provides an acceptable explanation to this phenomenon.
2) By proposing possible theoretical channels, researchers tend to establish bilateral causality
relationships between exchange rate and crude oil price. This arbitrarily rules out the possibility that a
common factor may drive both variables. Also, the wide application of Granger causality test as empirical
support is often associated with at least two problems. Firstly, Granger causality does not imply true causality.
Secondly, if both X and Y are driven by a common third process, but with a different lag, there would be
Granger causality. Therefore, the Granger causality test may provide misleading results when a common
third factor is the real cause. As far as the author knows, from all of the literature, only Amano and van
Norden (1998) have discussed this problem. The reason that they believe a bilateral causality relationship
between exchange rate and crude oil price is appropriate in their study is that the behavior of oil prices over
their sample period (1973:01 -1993:06) is dominated by major persistent supply shocks that have been
exogenous in the sense of most macroeconomic models. Demand-side factors, however, are small relative to
the supply-side shocks experienced over those 20 years. Accordingly, few macroeconomic insights are likely
to be gained from a search for a co-determinant of exchange rates and oil prices. While this may be close to
true for the study period, it may no longer hold in the latest decade. The recent oil price escalation starting
from the end of the 1990s is likely to have resulted in part from the rapid increase of oil demand from
emerging economies like China, India and Brazil. So the oil price movements are no longer determined by
just the supply side, but also the demand side. Such market structure change makes consideration of a
common third factor necessary.
3) Target period. Since early 1970s, the world economic structure, monetary system, and international
trade development have experienced profound changes. An example is the appearance of European Union
14

and the introduction of Euro. So the studies that include earlier data may not provide an accurate depiction of
recent years’ economic characteristics and obtain inaccurate statistical inferences.
4) Most of current studies use monthly or quarterly data in the empirical work. The amount of research
based on daily data is really small. While monthly and quarterly data reflects long-run trends clearly, it may
also entail the risk of information lost, because the movements of dollar exchange rate and crude oil price
have become more volatile since the collapse of Bretton Woods System and the first Oil Price Shock in the
early 1970s.
5) The consideration of Structural breaks. More and more scholars have concluded that due to political,
monetary, and other exogenous shocks, exchange rate determinants may have varied over time. 6 Also, since
1970s, the crude oil price has become highly volatile. Thus the test of structural breaks should be considered
when analyzing time series data of exchange rate and crude oil price. However, not all studies have drawn
enough attention to this fact.
6) Most studies use real instead of nominal data. Breitenfellner et al (2008) regard nominal data as more
appropriate since oil prices contribute directly and indirectly via other input costs, such as energy or other
commodities, to inflation. Thus, inflation adjustment removes some important information of this relative
price. Real exchange rates, which are sometimes used as well, mask the fact that exchange rate and inflation
are interdependent.

2.4. Studies on the Determinants of Each Variable


Dollar exchange rate and crude oil price are both multifactor-determined outcomes. While the literature
on their co-determinants is rare, there are relatively rich studies discussing the determination of each price.
The following section presents the main determinants of each price that are studied in previous research. This
review is not intended to show exhaustively all relevant factors but to depict the general scope of past
research.

2.4.1. Main Determinants of Dollar Exchange Rate


Theoretical attempts to explain exchange rate movements have had a long history. Most of classic and
traditional exchange rate determination theories and models developed are formulations of the relationship
between macro variables and the exchange rate. For example, The Purchasing Power Parity (PPP) theory is
about relative price level and exchange rate; Uncovered Interest rate Parity (UIP) describes the relationship
between interest rate differentials and exchange rates; the Balassa–Samuelson effect rationalizes the effect of
differences in the relative productivity of the tradable and non-tradable sectors on exchange rate movements;
the simple monetary model with price flexibility (Mussa, 1976; Frenkle, 1976), the Mundell-Fleming model

6. See, for example, Mark (2001), Cashin et al (2004) and Rossi (2006).
15

(Fleming, 1962; Mundell, 1964, 1968) and the Dornbusch Overshooting model (Dornbusch, 1976; a simple
monetary model with price stickiness) emphasize the role of monetary policy in determining exchange rate
changes. Whereas these theories, models and the subsequent studies developed based on them are impressive
and partially effective, they did not gain full support in terms of data verification.
When it comes to the effect of macroeconomic fundamentals in explaining and predicting exchange rate
movements, one important paper (Meese and Rogoff, 1983) reached the conclusion that a naïve random walk
predicts exchange rates better than structural macroeconomic models. This result, having been proved robust
over the decades and having become a benchmark for evaluating the performance of new exchange rate
models, is one of the biggest challenges posed for researchers. In the words of Frankel and Rose (1995, p.
1704), this negative result has had a “pessimistic effect on the field of empirical exchange rate modeling in
particular and international finance in general. The pessimistic effect has been with us 20 years.” However,
some recent studies have reported positive effects of fundamentals in predicting exchange rate changes.
MacDonald and Taylor (1994), Chinn and Meese (1995), Mark (1995) and Mark and Choi (1997) find
success in forecasting exchange rates at longer horizons when imposing long-run restrictions from monetary
models. Groen (2000) and Mark and Sul (2001) find greater success using panel methods. Engel et al. (2007)
assert that standard models imply near random walk behavior in exchange rates, so that their power to “beat
the random walk” in out-of-sample forecasts is low. In sample fit, rather than out of sample fit, should be used
to evaluate exchange rate models. The application of two kinds of data, order flow data 7 and real time data 8,
also sheds light on the role of macro fundamentals in affecting exchange rate changes. Evans and Lyons
(2002) have shown that exchange rates at short horizons are to a significant extent driven by order flow.
Evans and Lyons (2008) find that much of this order flow is closely linked to news about macro fundamentals.
Andersen et al. (2003) find strong evidence of exchange rate reaction to macro news in intraday data.
Therefore, macroeconomic indicators and models still have great merit in studying exchange rate movements.
Macroeconomic fundamentals studied in exchange rate economics are introduced roughly from four sources:
 structural models of exchange rate determination theories. Meese and Rogoff (1983), in comparison
with time series models, use a structural model that has a general form of flexible and sticky price
monetary models with variables of relative money supply, relative real income, short term interest rate
differential, expected long-run inflation differential and foreign trade balances. Obstfeld and Rogoff
(1995) develop a two-country model of exchange rate determination with an explicit role for current
account imbalances. Mark and Choi (1997) examine seven competing macro model specifications for
real exchange rate prediction that include a variety of fundamentals. Mark (1995), Mark and Choi

7. Evans and Lyons (2002) define order flow as the net of buyer-initiated and seller-initiated orders. It is a
measure of net buying pressure. They adopt a microstructure finance perspective on the determination of
exchange rate using order flow data.
8. According to Croushore and Stark (2001), real time data refers to “original, unrevised data available to
economic agents who were around at the time”. Such data is in contrast with “(final, revised) data available
from government statistical agencies today”.
16

(1997), Mark and Sul (2001) find that a linear combination of differentials in log money and log real
income has predictive power. Also using narrow money supply and real GDP, Taylor and Peel (2000)
contend that a simple monetary fundamentals model displays strong nonlinearity. Cavallo and Ghironi
(2002) provide a role for net foreign assets in the determination of exchange rates in the sticky-price
optimizing framework of Obstfeld and Rogoff (1995). Hau and Rey (2004, 2006) connect exchange rate
movements with equity returns, equity prices and capital flows. Gourinchas and Rey (2007) reveal that
today’s external imbalances forecast exchange rate movements at short, medium, and long horizons
both in and out of sample. In a rational expectations present-value framework, Engel and West (2005)
and Engel et al. (2007) discuss two important models used in macro: the money income model and
Taylor rule model and find that while the causality from current economic fundamentals to exchange
rate is weak, the expectations of future fundamentals have great weight in determining the exchange rate
in standard models. Cheung et al. (2005) evaluate three models proposed in the 1990s along with the
PPP and sticky price monetary models against the random walk model and reach a conclusion that
models that work well in one period do not necessarily work well in another period. More recently,
Bacchetta and Wincoop (2009) use money supply, industrial production, unemployment rate, interest
rate and oil price to study the exchange rate and find that the relationship between a forward looking
variable like the exchange rate and macro fundamentals is determined not by the structural parameters
themselves, but rather by the expectations of these structural parameters.
 survey results obtained from foreign exchange market participants. Cheung and Chinn (2001) report
findings from a survey of United States foreign exchange traders, who think unemployment, the interest
rate, the trade deficit, inflation, GDP, and the money supply are the six most important macro
fundamentals.
 macro news effect studies. Andersen et al. (2003) characterize the response of exchange rate market to
the release of 28 U.S. and 13 German macroeconomic announcements using 5 minute frequency data.
Using a similar data set and approach, Andersen et al. (2007) extend their previous study to stock and
bond market. Ehrmann and Fratzscher (2005) analyze the effect of macroeconomic news on the
exchange rate changes using real time data of 13 types of U.S. announcements and 12 of German
announcements. Faust et al. (2007) document the response of exchange rates and interest rates to 10
kinds of U.S. macro news using intraday frequency data.
 other empirical studies. For example, Wright (2008) pays special attention to financial variables such as
relative stock price, relative annual stock price growth rate and relative term spread in addition to the
prevailing macro fundamental variables used in other studies.
Table 2.3 provides a list of fundamentals used in selective literature mentioned above, where 33
variables are included in 11 categories, namely, price, monetary, output, international trade, capital market,
labor market, consumption, investment, government purchases, forward looking and commodity market.
Consistent with classical exchange rate determination theories, structural models focus mainly on such
17

categories as price, monetary, output, international trade, capital market, labor market and government
purchases. Macro news effect studies, however, reveal some other fundamentals that may be important, such
as nonfarm payroll 9, retail sales, consumer confidence index and NAPM index. All these variables are proved
to be related to the movements of exchange rate.

9 . Actually, as noted by Andersen et al. (2007), the nonfarm payroll is one of the most significant
announcements for all markets, and it is often referred to as the “king” of announcements by market
participants; see also Andersen and Bollerslev (1998).
18
11

Table 2.3 Macroeconomic Fundamentals Used in Selective Literature on Exchange Rate


Study Structural models of exchange rate determination theories Survey Macro news effect studies Other
Obstfeld Engel Bacchetta Cheung Ehrmann
Meese and Mark Hau and Cheung Gourinchas Anderson Faust
and and and and and Wright
Variables Rogoff and Choi Rey et al. and Rey et al. et al.
Rogoff West Wincoop Chinn Fratzscher (2008)
(1983)M (1997)M (2004)M (2005)Q (2007)Q (2003)I (2007)I
(1995) (2005)Q (2009)M (2001) (2005)D
CPI/Inflation √ √ √ √ √ √* √* √ √
Price
PPI √* √ √
Money M1,M2, M1,M2,
M1 M1,M4 √ M1 √ √
Supply M3 M3*
IR S SL S SL √ √ √* SL
Monetary
Target
federal funds √* √*
rate
GDP √ R √ √* √* √* √
Industrial
√ √ √ √* √
Production
Capacity
Output √
Utilization
Productivity √ √ √
Personal
√ √
Income
Trade
√ √ √ √* √ √* √
International Balance
Trade Terms of

Trade
Net Foreign
√ √
Asset
Capital Equity
√ √ √
Market Return
Net Capital

Flow
18
1911

Table 2.3 continued


Study Structural models of exchange rate determination theories Survey Macro news effect studies Other
Meese Obstfeld Mark Engel Bacchetta Cheung Ehrmann
Hau and Cheung Gourinchas Anderson Faust
and and and and and and and Wright
Variables Rey et al. and Rey et al. et al.
Rogoff Rogoff Choi West Wincoop Chinn Fratzscher (2008)
(2004)M (2005)Q (2007)Q (2003)I (2007)I
(1983)M (1995) (1997)M (2005)Q (2009)M (2001) (2005)D
Retail Sales √* √ √*
Consumer

Credit
Consumption Personal
Consumption √
Expenditures
New Home

Sales
Unemployment
√ √ √* √*
Rate
Nonfarm
√* √*
Payroll
Nonfarm
Labor Payroll √*
Market Employment
Initial
unemployment √* √*
claims
Average
√*
Workweek
Durable Goods
√*
Orders
Construction
√*
Investment Spending
Factory Orders √*
Business

Inventories
19
20
11

Table 2.3 continued


Study Structural models of exchange rate determination theories Survey Macro news effect studies Other
Meese Obstfeld Engel Bacchetta Cheung Ehrmann
Mark and Hau and Cheung Gourinchas Anderson Faust
and and and and and and Wright
Variables Choi Rey et al. and Rey et al. et al.
Rogoff Rogoff West Wincoop Chinn Fratzscher (2008)
(1997)M (2004)M (2005)Q (2007)Q (2003)I (2007)I
(1983)M (1995) (2005)Q (2009)M (2001) (2005)D
Government
Government Purchase √ √ √
/Debt/Deficit
Consumer
Confidence √* √*
Forward Index
Looking NAPM Index √* √*
Hoursing Starts √ √ √*
Index of
Leading √
Indicators
Commodity
Oil Price s
Market
Note: IR-S= Short term interest rate; IR-L=Long run interest rate; GDP-R=Real GDP; Oil Prices=Spot crude oil price; I, D, M and Q refers to Intraday, Daily,
Monthly and Quarterly frequency data set; Asterisks in the cells of “macro news effect studies” columns mean that the coefficients of variables are statistically
significant at the 5-percent level.
20
21

2.4.2. Main Determinants of Crude Oil Price


Factors that affect the crude oil price can be roughly divided into three categories, macroeconomic
fundamentals, supply related factors and others.

2.4.2.1. Macroeconomic Fundamentals


The literature on the relationship between macroeconomic fundamentals and crude oil price is quite
large. The bulk of these works consider crude oil price movements exogenous shocks to the economy and
focus exclusively on the effects of energy price shocks on the economy and on the possible transmission
mechanisms (See, among others, Hamilton (1996), Blanchard and Galí (2007), Kilian (2008), Kilian et al.
(2009) and Park and Ratti (2008); Jones et al. (2004) present a review of theoretical and empirical
developments since 1996). However, to this author’s knowledge, there is almost no comprehensive study of
the effects of various macroeconomic fundamentals on the crude oil price movements.
As partial contributions, many studies show that global GDP is a major determinants of crude oil price
(e.g. Baldwin and Prosser, 1988; Dahl and Yucel, 1990; Bacon, 1991; EMF11, 1992; Krichene, 2002; Barsky
and Kilian, 2002, 2004; Dees et al. 2007; Hamilton 2008). Krichene (2007) asserts that changes in the rate of
world economic growth would cause significant changes in demand for oil and gas and subsequently could
cause large upturns or downturns in oil and gas prices. Kilian (2008) considers the endogeneity of energy
prices and points out the fact that not only do energy prices affect the U.S. economy, but that there is reverse
causality from U.S. and more generally global macroeconomic aggregates to the price of energy. Interest rate
is considered another factor that drives crude oil price (e.g. Dahl and Yucel, 1990; Barsky and Kilian, 2002,
2004; Krichene, 2005, 2007). Frankel (2006) analyzes commodity prices in the overshooting framework and
finds that monetary policies have significant influence on crude oil price. A decrease of interest rate could
lead to a surge in the oil price. Besides the relationship studies, exchange rate is also included in other studies
on crude oil price. Baldwin and Prosser (1988), Dees et al. (2007) and Krichene (2007) build simultaneous
equations of crude oil demand and supply to determine the price and use exchange rate as an explanatory
variable. Futures prices are important references for spot prices. Brook et al. (2004) carefully discuss the
implication of strong crude oil market backwardation. 10 Dees et al. (2008) study the impact of conditions on
the futures markets on crude oil price changes. They use the difference between four-month and near month
contract for WTI on the New York Mercantile Exchange (NYME) as an explanatory variable and find that
contango has a positive effect on real crude oil prices. In addition to futures prices, researchers also try to
extract valuable information from trade volume. The volume of oil futures and options contracts traded on the
NYME is an often-quoted gauge of speculative pressure. NYME roughly separated traders into commercial

10. If the spot price or near month contract price exceeds the futures price, the market is described as being in
strong backwardation. If the spot price is less than the futures price, but exceeds the discounted futures price,
the market is described as weak backwardation. If the spot price is less than the futures price, the market is
described as being in contango.
22

and non-commercial categories according to the registration of traders. Brook et al. (2004) argue that net
positions held by non-commercial traders can be very significant and any sudden changes in these net
positions could have an important influence on prices from time to time. Möbert (2007) uses a relevant
variable, the difference between long and short positions of crude oil futures contracts on the NYME, to study
crude oil price determination. He finds that if the difference of long minus short positions increase in the
futures market then the oil price is positively affected. Guo and Kliesen (2005) use past realized oil variance,
the oil price change, realized stock market variance, stock market return, the default premium, the term
premium 11, the growth rate of real GDP, the growth of fixed nonresidential business investment and changes
in the federal funds rate target to test whether standard macro variables forecast one-quarter-ahead realized
oil futures variance. They report that real GDP growth and the growth of fixed nonresidential business
investment forecast oil price volatility but lose the predictive power after lagged dependent variables are
included. The other macro variables either do not relate to oil price volatility or have negligible forecasting
power. Darrat et al. (1996) establish a quarterly multivariate VAR model to investigate the existence and
direction of causality between oil prices, oil consumption, industrial production, the monetary base, the
federal budget deficit and short-term interest rates. They find that oil consumption, the federal budget deficit
and short-term interest rates are useful in predicting crude oil prices.
Hamilton (1983) and Kilian and Vega (2008) are two of the few exceptions. In identifying a set of
influences that were responsible for both the oil price increases and the subsequent recessions over the period
1948-1972, Hamilton conducts a thorough investigation of the relationship between crude oil price and
various macro fundamentals. His variable set includes real and nominal GNP, index of leading indicators,
inventories/sales ratio, Index of Capacity Utilization, strikes, unemployment, hourly wage, M1, interest rate,
BAA Bond Yields, Dow-Jones Industrial Average, implicit price deflator for nonfarm business income,
import prices, wholesale prices and prices of nonferrous metals, iron and steel, farm products, lumber
products and coal. However, he did not find much evidence that these variables can help predict crude oil
price movements. Kilian and Vega use a similar data set as used by Andersen et al. (2003, 2007) to explore
the response of crude oil price to macroeconomic news. They conclude that oil prices, unlike financial asset
prices, do not respond instantaneously to domestic macroeconomic news.
Some other recent studies emphasize the increasing impact of emerging Asia in general and China and
India in particular on global oil demand and thus oil price (see e.g. Brook et al., 2004; Benassy-Quere et al.,
2007; Möbert, 2007). Elekdag et al. (2007) attribute the oil price increase since 2003 primarily to increased
productivity in the oil-importing regions, particularly emerging Asia, with its increased intensive use of oil in
the production of tradable goods. Table 2.4 below briefly summarizes the previous studies mentioned here.

11. Guo et al. (2005) define the default premium as the difference between the yield on Baa- and Aaa-rated
corporate bonds and the term premium as the difference between the yield on 10-year Treasury notes and
3-month Treasury bills. Many studies suggest that yield spreads like these contain valuable information about
current and prospective business conditions (e.g. Dueker (1997) and the references therein).
23
11

Table 2.4 Macroeconomic Fundamentals Used in Selective Literature on Crude Oil Price
Baldwin and Darrat Barsky and Brook Guo and Elekdag Dees Kilian and Dees
Hamilton Krichene Möbert
Variables Prosser et al. Killian et al. Kliesen et al. et al. Vega et al.
(1983) (2007) (2007)
(1988) (1996) (2002) (2004) (2005) (2007) (2007) (2008) (2008)
CPI/Inflation √ √
Price Level PPI √
Exchange Rate √ √ √
Money Supply M1 M0
Monetary IR √ √* √ √
Policy
Target Federal
√ √
Funds Rate
GDP/GDP
√ √ √ √ √ √ √
Growth Rate
Industrial
√ √
Production
Output Capacity
√ √
Utilization
Productivity √
Personal

Income
Trade Balance √
International
Trade Oil Emerging OECD,
√ √*
Consumption China China, India
Stock Market
√ √
Return
Bond Yield √ √
Capital Futures and
Market Spot Price √ √
Spread
Net Long
√ √
Position
23
11
24

Table 2.4 continued


Baldwin Darrat Barsky and Brook Guo and Elekdag Dees Kilian Dees
Hamilton Krichene Möbert
Variables and Prosser et al. Killian et al. Kliesen et al. et al. and Vega et al.
(1983) (2007) (2007)
(1988) (1996) (2002) (2004) (2005) (2007) (2007) (2008) (2008)
Retail Sales √
Consumer

Credit
Consumption Personal
Consumption √
Expenditures
New Home

Sales
Unemployment
√ √
Rate
Nonfarm

Labor Payroll
Market Initial
unemployment √
claims
Hourly Wage √
Durable Goods

Orders
Construction

Spending
Factory Orders √
Business
Investment √
Inventories
Fixed
nonresidential

business
investment
Inventory/Sales

Ratio
24
11
25

Table 2.4 continued


Baldwin Darrat Barsky and Brook Guo and Elekdag Dees Kilian Dees
Hamilton Krichene Möbert
Variables and Prosser et al. Killian et al. Kliesen et al. et al. and Vega et al.
(1983) (2007) (2007)
(1988) (1996) (2002) (2004) (2005) (2007) (2007) (2008) (2008)
Government Government
√* √
Purchase Debt/Deficit
Consumer
Confidence √
Index
Forward NAPM Index √
Looking Hoursing Starts √
Index of
Leading √ √
Indicators
nonferrous

metals
Iron and Steel √
Commodity
Farm Products √
Market
Lumber

Products
Coal √
Note: Asterisks in the cells of “macro news effect studies” columns mean that the coefficients of variables are statistically significant at 5% level or lower.
25
26

2.4.2.2. Supply Related Factors


Since crude oil is a standard commodity, any factor that affects the supply of crude oil affects its price.
Along this line, a considerable number of papers focus on the supply side of the crude oil market.
There are a variety of factors that catch the attention of researchers. As a direct constitution of crude oil
price, production costs and other related costs are pointed out in many studies (e.g. EMF11, 1992; Lynch,
2002). Specifically, researchers analyze the oil exploration costs (Dahl and Yucel, 1990; Bacon, 1991), oil
extraction / production costs (Dahl and Yucel, 1990; Bacon, 1991; Dees et al. 2007) and oil transportation
costs (very large crude carrier rates; see Brook et al., 2004; Möbert, 2007). To evaluate the impact of field
production on crude oil price, many studies use the number of active oil rigs (Lynch, 2002; Möbert, 2007;
Hamilton, 2008) and the number of wells drilled (Dahl and Yucel, 1990) as relevant variables. Brook et al.
(2004) also use the number of active oil rigs as a proxy of the active exploration and development activities.
Geographical factors are also influential. EMF11(1992) and Lynch (2002) stress that crude oil
production depends not only on discovery, but also on the amount of capacity lost due to depletion effects.
Hamilton (2008) argues that the declining production from the mature Chinese fields has been a factor
influencing the recent course of world oil prices. When modeling the world crude oil market, Baldwin and
Prosser (1988) and Dahl and Yucel (1990) include both proven crude oil reserve and additions to the reserve
in their equations. Krichene (2007) develops a simultaneous equations model to study crude oil price. He
finds that an increase of proven reserve tends to stimulate oil output and depress oil prices. Besides the
proven and additional reserves, Brook et al. (2004) also qualitatively discuss the impact of strategic
petroleum reserve (SPR) on crude oil markets. They argue that SPR has the potential to impact oil markets, at
least in the short term, even in the case of small adjustments, via its effect on market psychology. It is,
however, often very difficult to quantitatively study the role of SPRs, because governments are loath to report
them (Chevillon and Rifflart, 2009). According to US Geological Survey (2000), there is considerable scope
for substantial additions to reserves. Probabilistic estimates suggest that undiscovered reserves (combined
with growth of existing reserves) could double current proved reserves. However, newly-discovered
resources have tended to be smaller and more expensive to develop, being increasingly offshore, and the
costs of exploration, development and production are higher than in the reserve-rich Middle East. Therefore,
investments in the global oil sector are needed to expand supply capacity, to promote technological progress
and to replace existing and future supply facilities. When analyzing the crude oil price development in the
recent decade, both Brook et al. (2004) and Elekdag et al. (2007) argue that the insufficient and lagging
investment in the crude oil industry results in low increase of additional capacity and low excess capacity.
The crude oil supply in the recent decade has become noticeably rigid and thus excessively vulnerable to
even minor disruptions. As a result, the crude oil price is higher and more sensitive than before.
In addition to the above factors, the microstructure of the crude oil market is an issue that attracts much
research attention. Researchers discuss heavily the OPEC and the Non-OPEC production and pricing
behavior. For example, Dees et al. (2007) model the world crude oil supply and demand and develop a price
27

rule equation, which include OPEC production quota, the difference between OPEC production and OPEC
quotas (defined as “Cheat”) and OPEC production capacity utilization. Baldwin and Prosser (1988), Dahl and
Yucel(1990) and Dees et al. (2007) address the role of Non-OPEC in crude oil market competition. Many
studies also analyze production capacity as a set of influencing factors. Four often seen types of variables
under the production capacity category are OPEC or Non-OPEC total production capacity (EMF 11, 1992;
Dees et al., 2007), OPEC spare capacity (e.g. Elekdag et al. 2007), OPEC or Non-OPEC capacity addition
(Bacon, 1991; Lynch, 2002) and OPEC or Non-OPEC capacity utilization (Baldwin and Prosser, 1988;
Lynch, 2002; Dees et al., 2007; Möbert, 2007). Dees et al. (2008) assert that oil prices would be more
sensitive to supply as production approaches capacity. Therefore, they study the non-linear effect of OPEC
capacity utilization and find that non-linear relationships between OPEC spare capacity and oil prices
account for changes in real oil prices and allow their model to perform well. In their study, Dees et al. (2008)
also argue that a lack of spare refining capacity is likely to be one cause for the rise in the price of motor
gasoline and crude oil. So they use refinery utilization rates as explanatory variables in their model. Results
indicate that the refining sector plays an important role in the crude oil price increase. The relationship is
negative such that higher refinery utilization rates reduce crude oil prices. They explain that this effect is
associated with shifts in the production of heavy and light grades of crude oil and price spreads between them.
Möbert (2007) empirically studies the effect of refinery capacity utilization rate on crude oil price when it
reaches 97%, 98% and 99%. He finds that crude oil price rises if the utilization rate is above 97% but the
magnitude decreases as free refinery capacity further reduces.
Inventory is another important class of price drivers. Pindyck (1994, 2001) conducts theoretical analysis
on the role of inventory on crude oil price determination. Ye et al. (2002, 2005) present a short-term
forecasting model of monthly West Texas Intermediate crude oil spot price using OECD petroleum inventory
levels. Brook et al. (2004) use the number of days of forward cover provided by OECD industry stocks
(OECD inventory/consumption), OECD inventory and the difference between actual and desired inventories
to analyze the impact of inventory on crude oil price. Chevillon and Rifflart (2009) specify inventory using
the number of days of forward cover provided by OECD industry stocks, OECD inventory and Non-OECD
inventory.
Since the crude oil production is sensitive to global political, geological, economic and natural
conditions, many studies develop dummy variables to allow for such exogenous shocks to crude oil price. For
example, Dees et al. (2007) use dummy variables to represent the Persian Gulf War, Mexico Peso crisis and
other geological and institutional determinants of crude oil production. Möbert (2007) develops two dummy
variables, one for positive events on the supply side such as U.S. SPR release and the other for negative
events on the supply side such as hurricane Katrina and Rita. All the influencing factors mentioned above and
selective papers which contain the factors are summarized in Table 2.5 below.
2811

Table 2.5 Supply Related Factors Used in Selective Literature on Crude Oil Price
Baldwin Dahl Ye Chevillon
Brook Dees Dees
and and Bacon EMF11 Lynch et al. Elekdag Krichene Möbert Hamilton and
Variables et al. et al. et al.
Prosser Yucel (1991) (1992) (2002) (2002, (2007) (2007) (2007) (2008) Rifflart
(2004) (2007) (2008)
(1988) (1990) 2005) (2009)
cost √ √ √ √ √ √ √
active oil rigs √ √ √ √
number of wells drilled √
Oil depletion √ √ √
Proven √ √ √ √
Reserve addition √ √ √ √
strategic √
investment √ √ √ √
Inventory √ √ √ √ √ √
Refining capacity √ √
production
√ √ √
quota
OPEC production √ √ √ √
Cheat √ √
Non-OPEC production √ √ √ √
OPEC total or
Non- OPEC √ √
total
Production
OPEC spare √
capacity
addition √ √
utilization √ √ √ √ √
28
29
11

Table 2.5 continued


Baldwin Dahl Ye Chevillon
Brook Dees Dees
and and Bacon EMF11 Lynch et al. Elekdag Krichene Möbert Hamilton and
Variables et al. et al. et al.
Prosser Yucel (1991) (1992) (2002) (2002, (2007) (2007) (2007) (2008) Rifflart
(2004) (2007) (2008)
(1988) (1990) 2005) (2009)
political,
dummy geological,
√ √ √ √
variable economic,
natural
29
30

2.4.2.3. Other Factors


In addition to the above two categories, there is another class of variables that do not fit conveniently
into the above categories. Most of them are likely to affect both the demand and supply of crude oil.
Crude oil Price volatility would be one of these variables. Pindyck (2001, 2004) and Brook et al. (2004),
among others, analyze the impact of volatility on crude oil price. As argued by Brook et al., volatility affects
the level of oil prices in two main ways. First, high price volatility will drive refiners and consumers to have
a high desired level of inventories, which, ceteris paribus, raises prices in the short run. Second, high
volatility raises the value of the call option held by oil producers. This increases the opportunity cost of
current production and can result in decreased oil supply. Higher demand for inventories and reduced supply
will thus push prices up. Although the impact of the first channel will be temporary, a high price that results
from the second channel will persist as long as the higher level of volatility persists. Besides, price volatility
compounded by geopolitical instabilities raises uncertainty about underlying price trends and has tended to
depress oil exploration. As a result, the increase of production capacity is slower than the increase of global
energy demand, causing low excess capacity and upsurge of oil price.
Technological progress is another important factor to both crude oil supply and demand (EMF 11, 1992).
On the demand side, technology advances contribute to the improvement of oil use efficiency, utilization of
alternative energy sources and gradual shift of the demand for oil (Bacon, 1991; Brook et al., 2004) On the
supply side, technological progress increases the success rate of exploration, improves drilling and extraction
productivities and pushes up the global recovery rate (Lynch, 2002; Brook et al., 2004). In general,
technology advances tend to reduce the dependence on crude oil and the energy intensities and increase crude
oil supply, providing downward pressure on crude oil prices.
Another source of pressure on crude oil price is inter-fuel substitution. The price and production of
alternative energy sources are likely to affect both the supply of and the demand for crude oil. Many studies
evaluate the impact of natural gas, either its price or production, on crude oil price (e.g. EMF 11, 1992;
Krichene, 2005; Dees et al., 2007). In addition to natural gas, Bacon (1991) also discusses coal and nuclear
power as substitutes of crude oil.
Two other factors appeared in crude oil price related literature are seasonality and population growth
rate. Since a fraction of crude oil is used to produce heating oil, the demand for oil is partially subject to the
weather changes and thus exhibits seasonality. Many studies include seasonality specifications in crude oil
price models (e.g. Ye et al., 2002, 2005; Dees et al. 2007). Möbert (2007) develop variables that represent
spring months and summer months. He finds that in the months March, April, and May the demand for oil
was significantly smaller than in the rest of the year and that in the summer months price increases are more
common. This is due to the fact that consumbers drive more in summer vacation months. Population is an
important demand variable. Krichene (2007) mentions it but does not include it in his model. Population is
not often seen directly in models but is shown sometimes indirectly in the GDP per capita variable. Table 2.6
below is a summary of the variables and selective literature mentioned above.
31

Table 2.6 Other Factors Used in Selective Literature on Crude Oil Price
Pindyck Ye et al. Brook Dees
Bacon EMF11 Lynch Krichene Krichene Möbert
Variables (2001, (2002, et al. et al.
(1991) (1992) (2002) (2005) (2007) (2007)
2004) 2005) (2004) (2007)
Price Volatility √ √
Technology Progress √ √ √ √ √
Natural
√ √ √ √
Gas
Inter-fuel
Substitution Coal √
Nuclear √
Seasonality √ √ √ √
Population/Growth

Rate

2.5. Studies on Gold Price


To the author’s knowledge, there has been no previous study that carefully analyzes the role of gold
price in the formation of both dollar exchange rate and crude oil price. Neither does past literature interpret
the relationship between the two prices from a triangular system perspective or examine specifically the
interaction of gold price with the two prices. Instead, many studies on gold price focus on two issues: the role
of gold as a hedge against risk and the role of gold as a safe haven.
In practice, gold price is often considered a hedge against inflation, currency risk, political unrest and
military tension. In economic studies, the first two, inflation risk and currency risk have drawn the most
research attention. The literature on the relationship between gold price and inflation is quite rich and the
results are generally consistent. Numerous studies show that gold is a useful hedge against inflation (e.g.
Sherman, 1982; Worthingtona and Pahlavani, 2007; Bruno and Chincarini, 2010). Particularly, Chua and
Woodward (1982) examine the effectiveness of gold as an inflation hedge for investors in six major industrial
countries over the period 1975 to 1980. They find that gold has been an effective inflationary hedge only in
the US. Similarly, Levin et al. (2006) argue that gold is a long-term hedge against U.S. inflation, but there is
no significant relationship between changes in the price of gold and changes in world inflation, world
inflation volatility and world income. There are also studies on the relationship between gold and inflation
expectation. For example, Adrangi, Chatrath, and Raffiee (2003) find that real gold return has a positive
relationship with expected inflation but is not significantly related to unexpected inflation. They explain that
this finding verifies that gold investment is a hedge against inflationary threats. Compared with the rich
empirical studies, theoretical studies are quite rare. As one of them, Ghosh et al. (2004) develops a theoretical
model to show that the price of gold rises over time at the general rate of inflation and hence is an effective
hedge against inflation. Although there is little explanation about the reasons why gold is a useful hedge
against inflation risk, previous literature does provide some possible explanations. Bruno and Chincarini
32

(2010) provide an interesting explanation that if everyone believes that gold is a good hedge, then it will tend
to rise as demand for gold rises with expected oncoming inflation. Worthingtona and Pahlavani (2007)
introduce the “conventional wisdom” that because commodities are physical assets, they are the best way to
hedge against rising prices. Unlike most commodities, gold is durable, relatively transportable, universally
acceptable and easily authenticated. These characteristics make gold a good hedge against inflation. There
are fewer studies on the role of gold as a hedge against currency risk. Capie et al. (2005) analyze the role of
gold as a hedge against the dollar and find evidence of the exchange-rate hedging potential of gold. They
explain that gold serves as a hedge because it is a homogeneous asset and therefore is easily traded in a
continuously open market. Moreover, gold cannot be produced by the authorities that produce currencies.
This means that those who can increase the supply of money and therefore, from time to time, debase its
value cannot by similar means debase the value of gold.
Although gold is frequently described as a “safe haven” in financial press and public media, it is much
less carefully studied as a safe haven in economic research. Even the term “safe haven” is not clearly defined
until recently. In studying the role of gold, Baur and Lucey (2010) define and distinguish the concepts of safe
haven, hedge and diversifier. According to their definition, a safe haven is an asset that is uncorrelated or
negatively correlated with another asset or portfolio in times of market stress or turmoil. A hedge is an asset
that is uncorrelated or negatively correlated with another asset or portfolio on average. 12 A diversifier is an
asset that is positively (but not perfectly correlated) with another asset or portfolio on average. Based on these
concepts, they find gold is a safe haven for stocks in the United States, in the United Kingdom and in
Germany. Gold is also a hedge for stocks in the United States and the United Kingdom. However, gold is
nowhere a safe haven for bonds; nor is it a bond hedge in the United States or United Kingdom. Furthermore,
the safe haven property of gold is short-lived. Gold is not a safe haven for stocks at all times but only after
extreme negative stock market shocks. It is not a safe haven in periods of rising stock markets. They argue
that gold is among the first forms of money and is traditionally used as an inflation hedge. Baur and
McDermott (2010) extend the definition of safe haven and hedge in Baur and Lucey (2010) by differentiating
between a strong hedge and a weak hedge and a strong safe haven and a weak safe haven. A strong (weak)
hedge is defined as an asset that is negatively correlated (uncorrelated) with another asset or portfolio on
average. A strong (weak) safe haven is defined as an asset that is negatively correlated (uncorrelated) with
another asset or portfolio in certain periods only, e.g. in times of falling stock markets. Based on further
defined concepts, they find that gold is a weak safe haven for some emerging markets but a strong safe haven
for most developed markets during the peak of the recent financial crisis. Investors react to short-lived and
extreme shocks by seeking out the safe haven of gold. But gradual trends in stock markets, like weekly or
monthly losses, do not appear to elicit the same impulsive response from investors. In addition, they find that
gold is a safe haven for increased levels of global uncertainty not confined to specific crisis periods.

12. Ranaldo and Söderlind (2010) use a similar definition in their study of safe haven currencies.
33

2.6. Summary
This chapter reviews the previous studies on dollar exchange rate, crude oil price and gold price. Clearly,
as the overlapped factors, macroeconomic fundamentals have the potential to be common drivers of dollar
exchange rate and crude oil price. However, as mentioned in Chapter I, macroeconomic variables are usually
inherently endogenous and thus fail to meet the requirements of common drivers. Therefore, this study has to
exclude macroeconomic fundamentals from the set of candidate common drivers. The studies of gold and
gold price introduced above document the role of gold as a hedge against risk and as safe haven, providing
support to the argument that gold price is driven by need for risk hedging and capital safety in addition to
consumption. So the gold price is considered an exogenous variable in this study and a candidate common
driver of dollar exchange rate and crude oil price. This paper tries to establish a triangular system among
dollar exchange rate, crude oil price, and gold price, hoping to fill the gap in past literature and present a new
way to understand the connection between dollar exchange rate and crude oil price.
34

CHAPTER III. METHODOLOGY

After literature review of the linkage between dollar exchange rate and crude oil price and of their major
determinants in Chapter II, this chapter turns to the empirical side of the relationship study and discusses how
to assess the relationship quantitatively with the help of econometric tools and historical data. Firstly, three
hypotheses about dollar exchange rate, crude oil price and gold price are presented to clarify objectives.
Along with the hypotheses is a brief look at the possible test results and corresponding implications. To
provide a vivid illustration of the connections among the three prices and supportive evidence for the
procedure design of empirical tests, analyses of correlation coefficients and volatilities over time are
conducted. Then a detailed description of the models and procedures developed to test the hypotheses will be
introduced.

3.1. Theoretical Hypotheses


This section outlines three hypotheses with respect to the relationship among dollar exchange rate,
crude oil price and gold price and discusses briefly the possible test results and implications.

3.1.1. Hypothesis One


Hypothesis One: The U.S. dollar exchange rate and gold price have a statistical causality relationship
running from gold price to dollar exchange rate.
This is one of the three pillars of the triangular system proposition. The term “statistical causality
relationship” here refers to the causality in the sense of Granger (1969), which is identified typically by
Granger Non-causality Test. There are up to four outcomes that may come from the test. In the context of
exchange rate (e) and gold price (g), the result can be unidirectional causality relationship from exchange rate
to gold price or the reverse (eg or ge), bidirectional causality relationship between exchange rate and
gold price (eg) or no causality relationship. A confirmation of the hypothesis one would mean that gold
price is a driver of dollar exchange rate and that the null hypothesis of Granger non-causality test that no
causality relationship exists will not hold.
35

3.1.2. Hypothesis Two


Hypothesis Two: The crude oil price and gold price have a statistical causality relationship running from
gold price to crude oil price (o).
This hypothesis serves as another pillar of the triangular system proposition. A confirmation of it would
mean that gold price is a driver of crude oil price (go). Again, if the test results support this hypothesis, the
null hypothesis of Granger non-causality test will be rejected.

3.1.3. Hypothesis Three


Hypothesis Three: The dollar exchange rate and crude oil price have a statistical causality relationship.
This hypothesis serves as the third pillar of the triangular system proposition. The direction of causality
is not specified in this hypothesis because, according to Chapter II, both dollar exchange rate and crude oil
price can affect each other through several possible channels. It is difficult at this point to prejudge which
direction is dominant or both directions are significant. This hypothesis is therefore like a “bottom line” for
the relationship between the two prices, simply indicating that they are causally connected.
Clearly, there are eight possible outcome combinations of the three hypotheses above. The Table 3.1
below lists these possibilities using different scenarios.

Table 3.1 Summary of Possible Outcomes

Hypothesis One Hypothesis Two Hypothesis Three Triangular System


Scenario 1 Accepted Accepted Accepted Hold
Scenario 2 Accepted Accepted Rejected Hold
Scenario 3 Accepted Rejected Accepted Fail
Scenario 4 Accepted Rejected Rejected Fail
Scenario 5 Rejected Accepted Accepted Fail
Scenario 6 Rejected Accepted Rejected Fail
Scenario 7 Rejected Rejected Accepted Fail
Scenario 8 Rejected Rejected Rejected Fail

Among all eight possibilities, scenario 1 and 2 are of special importance because the triangular system
proposition holds under these two circumstances. But these two scenarios have quite different implications. If
the test results show that the scenario 1 is the case, gold price is a direct common driver of dollar exchange
rate and crude oil price. Either dollar exchange rate or crude oil price is an intermediary vehicle through
36

which gold price movement indirectly affects crude oil price or dollar exchange rate. 13 If, however, the test
results show that the scenario 2 is the case, gold price is still a direct common driver of dollar exchange rate
and crude oil price. But the two prices per se are not causally connected as argued in previous literature but
merely seemingly correlated as a result of their common driver.

3.2. Measurement of Correlation and Volatility—Exponential Smoothing


Although correlation is not a sufficient condition of causality, it is a necessary one. Since gold price is
considered a potential common driver of dollar exchange rate and crude oil price, the first step is to check the
connection of gold price with the two prices. A effective way to see correlation between two variables is to
calculate their correlation coefficient. A relatively high correlation coefficient provides the foundation for
further empirical studies. Volatility is another tool to reveal data characteristics. It measures how dispersed
the price changes are. Abrupt volatility changes may indicate important changes of market situation. The
movement pattern of correlation coefficient and volatility over time may shed light on the procedure design
of further tests. Therefore, this section calculates correlation coefficients between dollar exchange rate and
gold price and between crude oil price and gold price, as well as the volatility of each price.
The data used in calculation is exponentially smoothed daily data. A favorable characteristic of this way
of data processing is that it not only shows the movement of correlation coefficient over time, but also reveals
major trends within the movement by muting noise in data. The following part describes the steps of
calculating exponentially smoothed correlation coefficients of the first order differences of dollar and gold,
dollar and oil and gold and oil.
Let lndollart, lngoldt and lnoilt denote the natural logarithm of daily price of dollar, gold and oil. Then
their first order differences are Δlndollart, Δlngoldt and Δlnoilt, which are daily percentage changes or
������������� , �����������
logarithmic returns. Let Δlndollar𝑡𝑡 Δlngold𝑡𝑡 and ���������
Δlnoil𝑡𝑡 be the arithmetical average of {Δlndollart},
{Δlngoldt} and {Δlnoilt} series and {edollart}, {egoldt} and {eoilt} be the deviation series of the first order
difference series and the arithmetical average series. That is,
���������������������),
������������� (=d(dollar) / dollar – d(dollar)/dollar
edollart = Δlndollart - Δlndollar𝑡𝑡
Δlngold𝑡𝑡 (=d(gold) / gold – �����������������
egoldt = Δlngoldt - ����������� d(gold)/gold), and
���������
eoilt = Δlnoilt - Δlnoil𝑡𝑡 ������������).
(=d(oil) / oil – d(oil)/oil
The general idea of exponential smoothing is explained below. The raw data sequence is often
represented by {xt}, and the output of the exponential smoothing algorithm is commonly written as {st}
which may be regarded as the best estimate of what the next value of x will be. When the sequence of
observations begins at time t = 0, the simplest form of exponential smoothing is given by the formulae

13. It is also possible that dollar exchange rate and crude oil price mutually cause each other. In this case, both
prices are intermediary vehicles through which gold price movement indirectly affects each other.
37

s0 = x0,
st = αxt + (1-α)st-1,
where α is the smoothing factor, and 0 < α < 1 (Roberts, 1959). Values of α close to one have less of a
smoothing effect and give greater weight to recent changes in the data, while values of α closer to zero have a
greater smoothing effect and are less responsive to recent changes.
The exponential smoothing technique is applied to two kinds of calculations here, the variances and the
covariances of series {Δlndollart}, {Δlngoldt}and {Δlnoilt}.
2 2 2
Let 𝜎𝜎dollart , 𝜎𝜎goldt and 𝜎𝜎oilt denote the variance of Δlndollart, Δlngoldt and Δlnoilt. Let cov denote
covariance. Then the formulas for the calculation of variances and covariances are:
2 2
𝜎𝜎dollar 0 = edollar0 ,

σ2gold 0 = egold02,
σ2goldt = α *egoldt2 + (1-α)*σ2gold (t−1) ,
σ2oil 0 = eoil02,
σ2oilt = α *eoilt2 + (1-α)*σ2oil (t−1) ,
cov(Δlndollar0, Δlngold0) = edollar0*egold0,
cov(Δlndollart, Δlngoldt) = α*edollart*egoldt + (1-α)*cov(Δlndollart-1, Δlngoldt-1),
cov(Δlndollar0, Δlnoil0) = edollar0*eoil0,
cov(Δlndollart, Δlnoilt) = α*edollart*eoilt + (1-α)*cov(Δlndollart-1, Δlnoilt-1),
cov(Δlngold0, Δlnoil0) = egold0*eoil0,
cov(Δlngoldt, Δlnoilt) = α*egoldt*eoilt + (1-α)*cov(Δlngoldt-1, Δlnoilt-1).
Let {ρdollart, goldt}, {ρdollart, oilt} and {ρgoldt, oilt} denote the correlation coefficient series of series
{Δlndollart} and {Δlngoldt}, {Δlndollart} and {Δlnoilt} and {Δlngoldt} and {Δlnoilt} respectively. Then the
formulas for the calculation of variances and covariances are:
ρdollart, goldt = cov(Δlndollart, Δlngoldt)/σdollart *σgoldt,
ρdollart, oilt = cov(Δlndollart, Δlnoilt)/σdollart *σoilt, and
ρgoldt, oilt = cov(Δlngoldt, Δlnoilt)/σgoldt *σoilt.
These three correlation coefficients show the co-movement of the daily percentage changes of two
prices and thus depict the co-movement of the two prices.
Let𝜎𝜎𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 , 𝜎𝜎𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 and 𝜎𝜎𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 denote the standard deviations of Δlndollart, Δlngoldt and Δlnoilt, 𝑉𝑉𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 ,
𝑉𝑉𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 and 𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 denote the annualized volatilities of Δlndollart, Δlngoldt and Δlnoilt. The formulas for the
three volatilities are:
𝑉𝑉𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 = 𝜎𝜎𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑 / �1/252 ,

𝑉𝑉𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 = 𝜎𝜎𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔𝑔 / �1/252 , and

𝑉𝑉𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 = 𝜎𝜎𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 / �1/252.


38

To reveal long run movement patterns and to identify clear signals, a small α, 0.005, is chosen. 14 The
cost of such choices is that first several hundred numbers in the result sequences are not appropriate to use,
because the long run smoothing effect is not fully realized at the very beginning.

3.3. Description of Models and Test Procedures


After clarifying the objectives of empirical tests in Section 3.1 and exploring the co-movement patterns
of the three variables in 3.2, this section turns to the description of models and test procedure. Given the time
series nature of variables, the empirical study will begin with an analysis of data stationarity.

3.3.1. Data Stationarity


The stationarity of time series data is a key issue to regression analysis. Without considering the data
stationarity, spurious regression problem will rise, which refers to the case that even when two variables have
no meaningful relationship with each other, the regression can still report an extremely high R-square,
misleading the interpretation of regression results. Also, because non-stationary series have changing
properties over time, Ordinary Least Square (OLS) estimations will fail. Therefore, it is necessary to examine
the stationarity of variables.
A popular way of testing data stationarity is unit root test. This paper uses two frequently adopted tests,
Augmented Dickey-Fuller (ADF) Test and Phillips-Perron (PP) Test (Dickey and Fuller, 1979, 1981; Phillips
and Perron, 1988), to examine data stationarity.
The ADF test and PP test are based on the regression
Yt = μ + βt + δYt-1 + ∑𝑘𝑘𝑖𝑖=1 𝛾𝛾𝑖𝑖 ∆𝑌𝑌𝑡𝑡−𝑖𝑖 + εt , (4.1)
where Δ is the first-order difference operator, Y is the variable under consideration, e, g or o, μ, β, δ and γs are
parameters to be estimated, and ε is an error term. When μ = β =0, the above equation shows the random walk
form. When β =0, the equation shows the random walk with drift. The trend stationary model leaves both
parameters free. The ADF test is carried out by testing the null hypothesis δ =1 against δ <1. While the PP
approach is based on the same equation, it makes corrections to the statistics used in Dickey-Fuller test to
capture the effect of autocorrelation. For both tests, the lag term k is determined by the Akaike Information
Criterion. A t-statistic larger in absolute value than the critical values results in a rejection of the null
hypothesis of unit root in favor of the stationarity alternative. If the variable is not stationary in level but
stationary after kth order difference, the variable is said to be integrated and denoted as an I(k) process.

14. The author calculated the correlation coefficients and volatilities at both α = 0.005 level and α = 0.01 level.
The figures under two cases are similar except that the curves are smoother when α is set to be 0.005. To save
space, only the results calculated at α = 0.005 level will be shown in this study.
39

3.3.2. Cointegration Test


As will be shown in Chapter V, all three variables are not stationary in level but stationary after first
order difference. All three variables are I (1) processes. Although regressions of non-stationary variables may
lead to a spurious regression, it might be the case that the non-stationary time series still have a meaningful
long run stable relationship. The implication would be that the series are drifting together at roughly the same
rate. Two series that satisfy this requirement are said to be cointegrated (Greene, 2002). The investigation of
cointegration relationships between variables is important for several reasons. Besides revealing long run
equilibrium relationships between variables as mentioned above, the existence of cointegration implies that
there must be Granger causality between some of the variables (Engle and Granger, 1987). Moreover, if the
non-stationary variables have cointegration relationships, it is more appropriate to develop a Vector Error
Correction (VEC) Model than to develop a Vector Autoregression (VAR) Model in first order differences.
While both of them can be constructed to be stationary, the latter one would be counterproductive, since it
would result in information loss and obscure the long-run relationship between variables.
There are two broad approaches to test cointegration relationships, the Engle-Granger (EG) two-step
method and the Johansen and Juselius (1990) (JJ) approach. The EG method assesses the stationarity of
equilibrium errors estimated by single-equation regression. The JJ method is based on the Vector
Autoregression (VAR) approach. While both methods are effective, the EG method is the simplest
cointegration test for bivariate models. Therefore, this paper uses EG method to test cointegration
relationships between dollar exchange rate and gold price, between crude oil price and gold price and
between dollar exchange rate and crude oil price.
The EG method is based on assessing whether single equation estimates of the equilibrium errors appear
to be stationary. The null hypothesis of no cointegration is tested against the alternative that the variables are
cointegrated. This test requires that all the individual variables are in the same order of integration. It starts
from using Ordinary Least Squares to estimate the regression equation
Yt = β0 + β1Xt + εt. (4.2)
In this study, X and Y are g and e, g and o or e and o. Equation 4.2 can also be written as
εt =Yt - β0 - β1Xt, (4.3)
where the residual ε can be considered a linear combination of variables and its estimated values can be
calculated from the regression of equation 4.2. The second step is to perform a unit root test on the residual
series, using an ADF test, to see whether it is an I (0) process. If the residual series is stationary, the two
non-stationary time series Y and X are said to have a meaningful long-run relationship and vice versa.
For consistency of the parameter estimates, the right hand side variables in equation 4.2 should be
weakly exogenous with respect to the cointegrating parameters. Here the X variable is assumed to be weakly
exogenous and will be tested later.
40

3.3.3. Stability Test of the Estimated Cointegration Relationships


The main idea of cointegration is that a linear combination (the cointegrating vector) of non-stationary
variables may be stationary. The standard test for cointegration above presumes that the cointegrating vector
is time-invariant under the alternative hypothesis. However, given the movement patterns of dollar exchange
rate, crude oil price and gold price in the sample period and the results of section 3.2 that will be shown in
Chapter V, it is not guaranteed that the linear combinations established are stable without shifts at unknown
points. Therefore, it is necessary to entertain the possibility of cointegration with structural changes.
Gregory and Hansen (1996) propose three tests for cointegration which allow for the possibility of
regime shifts. They argue that both the standard cointegration test and their tests examine the null of no
cointegration, so rejection by either test implies that there is a long-run relationship in the data. If the standard
cointegration test does not reject, but their tests do, this implies that structural change in the cointegrating
vector may be important. If both the standard cointegration test and their tests reject, no inference that
structural change has occurred is warranted from this piece of information alone, since their tests are
powerful against conventional cointegration. In this case, further investigation is necessary to differentiate
between cointegration with stable parameters and cointegration with a structural change, as the null
hypothesis of no cointegration is rejected in comparison with either alternative hypothesis. Gregory and
Hansen suggest using the parameter instability tests of Hansen (1992) to determine whether the cointegrating
relationship identified is subject to a regime shift.
According to the results of EG cointegration test that will be shown in Chapter V, each price is
cointegrated with the other two prices separately, which implies the existence of cointegration relationships.
In this context, even the Gregory and Hansen tests reject the null hypothesis, it is still necessary to examine
parameter instability. Therefore, this study performs directly the parameter instability tests of Hansen (1992).
Hansen (1992) proposes three tests for parameter instability examination, SupF test, MeanF test and Lc
test, which are all based on the residuals of a Fully Modified (FM) least-squares regression previously used in
Phillips and Hansen (1990). All three tests have the same null hypothesis of parameter stability, but differ in
the alternative hypothesis. So the choice of an appropriate test statistic depends on the purpose of the test.
Specifically, the SupF test is used to detect whether there is a sharp shift in regime. The MeanF test is used to
test whether the specified model captures well a stable relationship, because the notion of martingale
parameters captures gradual changes in the regression coefficients. If the likelihood of parameter variation is
relatively constant throughout the sample, then the Lc statistic is the appropriate choice. Critical values for
the three tests are provided in Hansen (1992). Since there are noticeable quick ups and downs in the three data
series, SupF and MeanF tests are used in the quantitative analysis.
The SupF test dates back to Quandt (1960). It assumes that there is one structural break at unknown time.
Consider a regression equation as follows:
Yt = AtXt + εt, (4.4)
where the coefficient matrix A is not constant but time dependent, so that
41

Ai = Ap, i ≤ t and Ai = Aq, i > t,


where 1< t < n, n is the sample size. The null hypothesis is H0: Ap = Aq, and the alternative hypothesis is H1:
Ap ≠ Aq, t/n ∈ τ, where τ is set to be [0.15, 0.85] as recommended by Andrews (1993). An F statistic can be
constructed based on equation 4.4 which is equivalent to the statistic of the classical Chow test. Specifically,
the F statistic is computationally equivalent to estimating the coefficient matrices Ap and Aq using two
sub-samples divided at time t and testing their equivalence using a Wald test. A SupF test of H0 against H1 is
performed by computing a series of F statistics under different time t and comparing the largest F statistic to
the corresponding critical values. The variance-covariance matrix used to construct the test statistic is the
quadratic spectral (QS) kernel based VAR prewhitened heteroskedasticity and autocorrelation consistent
(HAC) covariance matrix estimator developed by Andrews and Monahan (1992). 15 If the SupF value
exceeds critical values, the null hypothesis is rejected and a structure break exists.
The MeanF test statistic is derived from a different hypothesis structure but is calculated as simply the
average of F statistics obtained in the SupF test.

3.3.4. Timing of Structural Breaks


A subsequent question to the identification of structure breaks is the timing of breaks. If the SupF value
exceeds corresponding critical values, the date that exhibits this largest F statistic is considered a break date
estimator. However, it is possible that there is more than one F statistic that exceeds critical values and thus
multiple break dates may exist. In this case, the multiple break dates will be estimated sequentially. The idea
is that if there are multiple break dates, the series of F statistics may have both local maximum and global
maximum. The global maximum (SupF) is viewed as a break date estimator and the local maximums are
viewed as candidate break date estimators. The sample is divided at the date of global maximum and same
analysis repeats in the sub-samples until no significant F statistic is found.

3.3.5. Granger Non-Causality Test


The notion of Granger Non-Causality is firstly developed by Granger (1969). Suppose there are two
variables x and y, causality in the sense of Granger is inferred to run from x to y if the lagged values of x have
explanatory power in a regression of y on the lagged values of y and x. The Granger Non-Causality Test is

15. Actually, the author calculated four different variance-covariance matrices to compare the estimation
performance: standard covariance matrix used in least squares regressions, heteroskedasticity consistent
covariance matrix, HAC covariance matrix developed by Newey and West (1987) and HAC covariance
matrix developed by Andrews and Monahan (1992). As will be shown in Chapter V, the residuals from the
estimation of cointegration equations in section 3.3.2 show heteroskedasticity and strong autocorrelation.
Compared with the Newey and West estimator, Andrews and Monahan’s estimator makes use of all
autocovariances, and thus captures more information and has better performance.
42

widely used in economic studies to analyze the relationships between economic variables and thus is an
important statistical inference tool. According to Engle and Granger (1987), a bivariate cointegrated system
must have a causal ordering in at least one direction. They also demonstrate that the presence of cointegration
in a system implies that a valid error correction representation exists. For a two-variable system as will be
developed in this study, an error correction model relates the change in one variable to both past equilibrium
errors and past changes in two variables. In other words, an error correction model may reveal both short run
dynamics and long run equilibrium of variables. Granger non-causality test can be performed in an error
correction framework as presented below.
∆et = μ1 + ∑ki=1 α11𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + ∑ki=1 α12𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ1 ( et-1 - β0 - β1gt-1 ) + ε1t, (4.5)
∆gt = μ2 + ∑ki=1 α21𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + ∑ki=1 α22𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ2 ( et-1 - β0 - β1gt-1 ) + ε2t, (4.6)
∆ot = μ3 + ∑ki=1 α31𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑ki=1 α32𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ3 ( ot-1 - δ0 - δ1gt-1 ) + ε3t, (4.7)
∆gt = μ4 + ∑ki=1 α41𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑ki=1 α42𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ4 ( ot-1 - δ0 - δ1gt-1 ) + ε4t, (4.8)
∆ot = μ5 + ∑ki=1 α51𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑ki=1 α52𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ5 ( ot-1 - η0 - η1et-1 ) + ε5t, (4.9)
∆et = μ6 + ∑ki=1 α61𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑ki=1 α52𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ6 ( ot-1 - η0 - η1et-1 ) + ε6t, (4.10)
where ∆ is the first order difference operator, μi is a constant and εit is an error term, i=1,2,3,4,5,6. The three
lagged terms, (et-1 - β0 - β1gt-1), (ot-1 - δ0 - δ1gt-1) and (ot-1 - η0 - η1et-1) are called error correction terms, which
are lagged residuals from the cointegration regressions between e and g, between o and g and between o and
e. k is the optimal lag length and is decided using Akaike Information Criterion (AIC).
In equation (4.5)-equation (4.10), the change of left-hand side variable is affected by right-hand side
variables either through the lagged terms or through corresponding lagged error correction term. In many
empirical works, researchers study two kinds of Granger causality, long-run Granger causality and short-run
Granger causality. The long-run Granger causality can be tested by performing weak exogeneity test to
determine whether the long run relationship, captured by the error correction term, has a non-zero coefficient
that is statistically significant. Engle, Hendry, and Richard (1983) define a set of variables Xt in a
parameterized model to be weakly exogenous if the full model can be written in terms of a marginal
probability distribution for Xt and a conditional distribution for Yt | Xt such that estimation of the parameters
of the conditional distribution is no less efficient than estimation of the full set of parameters of the joint
distribution (Greene, 2002). Take gold price as an example, the weak exogeneity of gold price can be
examined by testing the null hypothesis of γ2 = 0 and γ4 =0 using the Likelihood Ratio (LR) test. The short-run
Granger causality can be examined by testing whether the lagged terms other than the lagged error correction
terms have explanatory powers in the regression. For instance, the short-run causality from gold price to
dollar exchange rate and crude oil price can be examined by testing whether all the α12i and α32i, i=1,…,k, are
not equal to zero. The short run Granger causality can also be examined using the LR test.
If no structure breaks are identified, the Granger Non-Causality Test will be performed on the entire
sample. If structure breaks do exist, the Granger Non-Causality Test will be performed on both sub-samples
and the entire sample to make comparisons.
43

After the methodology description here, the next chapter will move on to describe the data set used in
this study.
44

CHAPTER IV. DATA

This chapter introduces the data set used in the analysis. Considerations on choosing appropriate dollar
exchange rate, crude oil price and gold price are discussed. Data sources and preliminary data processing are
also described.

4.1. Considerations on Dollar Exchange Rate Data


There are four major concerns about the choice of dollar exchange rate. Discussions on each of the
concerns are presented below.

4.1.1. Justification for Dollar/Euro exchange rate


The reason this paper studies the Dollar/Euro exchange rate is threefold. Firstly, crude oil price is
denominated by dollars in the world crude oil market. Most of the crude oil transactions are settled in dollars.
Second, by providing crude oil, oil-exporting countries, especially the OPEC, hold a huge amount of
petrodollars. But these countries spend much more in Euro area than U.S. for their imports. As shown in
Figure 4.1 below, as of 2007, Euro area constitutes 27.95% of the OPEC’s total import, compared with 9.73%
of U.S. The OPEC’s average import shares from these two partners during 1980 to 2007 are 36.22% and
13.33%. Each year huge amount of dollars is converted to Euros for the settlement of OPEC’s imports. So the
trade between Euro area and OPEC makes Euro an important currency when studying oil price. Finally, the
U.S. is the largest oil consumption country in the world. In 2008, the U.S. oil consumption was 22.5% of
world total oil consumption. European Union (E.U.) also consumed 17.9% 16. Since both the U.S. and the E.U.
are large oil consumers and both of their currencies are closely related to oil transactions in the world market,
the Dollar/Euro exchange rate is a proper variable to study the relationship between dollar exchange rate and
crude oil price.

16. U.S. and E.U. oil consumption data is collected from BP statistical review of world energy 2009.
45

100.00%
90.00%
80.00%
70.00%
60.00%
50.00%
40.00%
30.00%
20.00%
10.00%
0.00%
1980

1982

1984

1986

1988

1990

1992

1994

1996

1998

2000

2002

2004

2006
U.S. E.U. Japan Rest of the World

Note: From bottom to top: U.S., E.U., Japan and Rest of the World. The data used in this figure is
calculated by author based on data from Intermational Monetary Fund Direction of Trade Statistics.
Figure 4.1 OPEC Major Trade Partners and Shares 1980-2007

4.1.2. Nominal Exchange Rate vs. Real Exchange Rate


As shown in Table 1 of Chapter II, both nominal exchange rate and real exchange rate are studied by
researchers. However, Breitenfellner et al (2008) argue that nominal data is more appropriate since oil prices
contribute directly and indirectly via input costs, such as energy or other commodities, to inflation. Inflation
adjustment removes some important information of this relative price. Real exchange rates, which are often
used as well, mask the fact that exchange rate and inflation are interdependent. So in this study nominal
exchange rate will be used.

4.1.3. Bilateral Exchange Rate, NEER, REER and Commodity Currency


Recall Table 1 of Chapter II, bilateral exchange rate, NEER and REER are all often used in literature to
analyze the connection between exchange rate and crude oil price. Also, researchers’ attention has been
increasingly attracted by the Commodity Currency. While it is incorrect to simply contend that one of these
currencies is better than the others, the focus of this study is the relationship between exchange rate and crude
oil price. NEER, REER and Commodity currency are weighted averages of a basket of exchange rates against
the currencies of major trade partners. So in this study, bilateral exchange rate will be adopted.
46

4.1.4. Spot Exchange Rate vs. Forward Exchange Rate


The foreign exchange markets provide both spot exchange rates and forward exchange rates with
different expiration dates. To make the dollar exchange rate easy to compare with other two prices, spot
exchange rate is used in this study. Another reason for this choice is that if forward exchange rates are used,
prices of commodity futures are needed for consistency purpose. Unlike commodity futures prices, forward
exchange rates are priced without taking into account the convenience yield.

4.2. Considerations on Crude Oil Price Data


There are three major concerns about the choice of crude oil price. Discussions on each of the concerns
are presented below.

4.2.1. Brent, West Texas Intermediate and Dubai Crude Oil Price
The most important three types of crude oil prices are Brent price, West Texas Intermediate (WTI) price
and Dubai price. These three prices serve as benchmarks for other types of crude oil and downstream
products. As noted by Yousefi & Wirjanto (2004), WTI crude oil price is supposed to be one of the primary
representatives of an international crude oil benchmark price. Spot trading of WTI crude oil has been
flourishing, and regarded as the principal facet in international oil markets. Since this study focuses on the
Dollar/Euro exchange rate, the WTI price is used.

4.2.2. Nominal Price vs. Real Price


Both nominal and real oil price are widely used in economic studies. However, Blanchard and Gali
(2007) suggest using the dollar price of oil rather than the real price of oil, to avoid dividing by an
endogenous variable, the GDP deflator. Similar argument is also seen in the discussion of choice of nominal
exchange rate against real exchange rate. So this study uses nominal crude oil price.

4.2.3. Spot Price vs. Futures Price


Crude oil has both spot price and several futures prices with different expiration dates. All of these
prices are commonly used by researchers. Futures price and spot price are frequently seen to exhibit
co-movement and sometimes interchangeable in quantitative analysis. But given the discussion in section
4.1.4 and for comparability and consistency purpose, spot crude oil price will be used in analysis.
47

4.3. Considerations on Gold Price Data


There are two major concerns about the choice of gold price. Discussions on each of the concerns are
presented below.

4.3.1. London Gold Fix


The London gold fix is the procedure by which the price of gold is determined twice each business day
on the London market by the five members of The London Gold Market Fixing Ltd. It is designed to fix a
price for settling contracts between members of the London bullion market, but the gold fix also provides a
recognized rate that is used as a benchmark for pricing the majority of gold products and derivatives
throughout the world's markets.

4.3.2. London AM Fix vs. London PM Fix


The gold fix is conducted in United States dollars (US$), Pound sterling (GBP), and the euro (€) daily at
10:30AM and 3:00PM London time (5:30AM and 10:00AM U.S. Eastern Time), via a telephone conference
of the five members of The London Gold Market Fixing Ltd. So there are two gold fix prices that can be
called London AM Fix and London PM Fix. This study uses the London PM Fix price for analysis. The
reason is explained in section 4.5.

4.4. Data Frequency


The range of data used in this study is from year 1999, when euro is officially introduced to the world, to
year 2010. The data of all three prices is available at daily frequency, monthly frequency, quarterly frequency
and yearly frequency. Usually, the higher the data frequency, the more information that data contains. But
noise also rises as the data frequency gets higher and higher. So this study uses monthly data in the
econometric analysis. For the correlation and volatility analysis, exponentially smoothed daily data is used to
achieve a balance between avoiding information loss and reducing noise.

4.5. Description of Data Sources


According to section 4.4, nominal spot Dollar/Euro exchange rate is chosen for the econometric
analysis. The data is collected from the Federal Reserve Statistical Release H.10, Foreign Exchange Rate
(daily), which is noon buying exchange rate in New York for cable transfers payable in euro. The exchange
rate represents dollars per euro. So if the exchange rate goes up, dollar depreciates. Daily data is from January
4, 1999 to April 30, 2010 and is averaged according to the number of trading days in corresponding month to
48

obtain monthly dollar/euro exchange rate. Daily data is used for the calculation in section 3.2. Monthly data
is used in econometric analysis.
The crude oil price, according to section 4.4, is nominal spot WTI price, explained as dollars per barrel
crude oil. Daily data is daily closing Free on Board spot price and monthly data is the mathematical average
of daily data. Both sequences are collected from Energy Information Agency. Daily data is used for the
calculation in section 3.2. Monthly data is used in econometric analysis.
The London PM Fix gold price sequences are used in this study. The price is nominal gold price
explained as dollars per troy ounce gold. Daily and monthly data is collected from the London Bullion
Market Association, which provides data by permission of the London Gold Market Fixing Ltd. Daily data is
used for the calculation in section 3.2. Monthly data is used in econometric analysis.
Compared with the London AM Fix gold price, the London PM Fix gold price is the latest innovation to
dollar exchange rate and crude oil price. The London AM and PM Fix gold prices are set at 5:30AM and
10:00AM U.S. Eastern time respectively. The dollar exchange rate is noon (U.S. Eastern time) buying
exchange rate in New York for cable transfers payable in euro and the crude oil price is daily closing spot
price. The foreign exchange market and the crude oil market absorb information quickly. So the London PM
Fix gold price, which contains relatively new information, may correlate closer with noon dollar exchange
rate and closing crude oil price. Along this line of reasoning, the author conducted the correlation and
volatility analysis using both London AM Fix gold price and London PM Fix gold price. The results indicate
that the London PM Fix gold price has noticeably stronger correlation with dollar exchange rate and crude oil
price than the London AM Fix gold price does. So in this study the London PM Fix gold price is used.

4.6. Data Preparation


For the calculation in section 3.2, although daily data of all three prices is available, it is not always
available together on the same trading day. Possible reasons are not limited to difference of legal holidays
between U.S. and E.U. countries and data unavailability. If dollar exchange rate or crude oil price of certain
date is unavailable, the data of that date is deleted. If both the London AM Fix and PM Fix are unavailable,
the data of that date is deleted. But if only the London PM Fix gold price is unavailable, the London AM Fix
gold price is used instead. Such case happens on Dec.24 and Dec.31 of each year when the London Gold
Market closes in the afternoon. The purpose of substitution is to avoid information loss. The data adjustment
described above provides a daily data set that consists of 2769 observations in each price sequence. The
monthly data set does not need any data adjustment. The sample period of January 1999 - April 2010 contains
136 observations in total.
Both daily and monthly data are transformed to natural log data before used in analysis.
49

CHAPTER V. EMPIRICAL RESULTS

This chapter displays and interprets the results yielded from statistical and econometric analysis. The
outcomes of correlation calculation are presented first and are followed by the volatility analysis. The
econometric results will be discussed in detail subsequently.

5.1. Correlation Coefficients


Correlation is a necessary condition of causality. The correlation coefficient measures the direction and
closeness of the co-movement of two variables. Below are the statistical results of correlation coefficient
calculation.

5.1.1. Dollar Exchange Rate and Gold Price


This section presents figures of exponentially smoothed correlation coefficients between daily dollar
exchange rate and London PM Fix gold price. The smoothing factor α is set to be 0.005. Analysis will focus
on data starting from January 2, 2002, the 733th number in the correlation coefficient sequence, to allow the
effect of exponential smoothing to be fully realized.

1.00
0.80
0.60
0.40
0.20
0.00
-0.20
1999/1/5
1999/8/5
2000/3/5

2001/5/5

2002/7/5
2003/2/5
2003/9/5
2004/4/5

2005/6/5
2006/1/5
2006/8/5
2007/3/5

2008/5/5

2009/7/5
2010/2/5
2000/10/5

2001/12/5

2004/11/5

2007/10/5

2008/12/5

-0.40
-0.60
-0.80
-1.00

Figure 5.1 Correlation Coefficients between Dollar Exchange Rate and London PM Fix Gold Price
50

Figure 5.1 provides following information:


(1) Within the period of January 2, 2002 - April 30, 2010, dollar exchange rate and gold price were
negatively correlated. The reason is the dollar exchange rate is expressed as dollars per euro. The positive
correlation coefficients shown in this figure mean that dollar appreciation (depreciation) and gold price
decrease (increase) are coincident.
(2) The correlation coefficient increased from 0.227 to 0.581 during January 2002 - May 2005. Then it
started to decrease. In 2006 and the first nine months of 2007, it maintained below 0.400 most of the time. In
October, the correlation coefficient entered into a quick increase. The highest correlation coefficient was
0.601, realized on August 18, 2008. As of April 30, 2010, the correlation coefficient was 0.450. Table 5.1
below provides the average correlation coefficients in different time periods.

Table 5.1 Average Correlation Coefficient of Dollar Exchange Rate and Gold Price

2002.1.2-2007.9.28 2007.10.1-2010.4.30 2002.1.2-2010.4.30


Dollar and London PM Fix 0.395 0.461 0.415

The information summarized above can be interpreted as follows:


(1) Since dollar appreciation (depreciation) and gold price decrease (increase) are coincident, it is
reasonable to expect that a causal relationship may exist between the two.
(2) Although the correlation coefficient was constantly positive during the period of January 2, 2002 -
April 30, 2010, it had ups and downs. The increase of correlation coefficient in a certain period means that the
frequency or magnitude (or both) of opposite direction movement of dollar exchange rate and gold price is
higher or greater (or both) than that of same direction movement. Similarly, the reason for the decrease of
correlation coefficient in a certain period is that dollar exchange rate and gold price move more frequently or
with greater magnitude (or both) in the same direction---they increase or decrease in value at the same time.
However, the decreases of correlation coefficient were just episodes in the historical movement of dollar
exchange rate and gold price. The negative correlation relationship between the two prices is still the main
theme.
(3) In Table 5.1, the correlation coefficients in the second sample period is not considerably higher than
that in the first sample period. It is possible that the relationship between dollar exchange rate and gold price
did not experience any structural changes in the sample period. But this conjecture needs to be tested.

5.1.2. Crude Oil Price and Gold Price


This section presents figures of exponentially smoothed correlation coefficients between crude oil price
and London PM Fix gold price. The smoothing factor α is set to be 0.005. Analysis will focus on data starting
from January 2, 2002 to allow the effect of exponential smoothing to be fully realized.
51

1.00
0.80
0.60
0.40
0.20
0.00
-0.20
1999/1/5
1999/8/5
2000/3/5

2001/5/5

2002/7/5
2003/2/5
2003/9/5
2004/4/5

2005/6/5
2006/1/5
2006/8/5
2007/3/5

2008/5/5

2009/7/5
2010/2/5
2000/10/5

2001/12/5

2004/11/5

2007/10/5

2008/12/5
-0.40
-0.60
-0.80
-1.00

Figure 5.2 Correlation Coefficients between Crude Oil Price and London PM Fix Gold Price

Figure 5.2 provides following information:


(1) Within the period of January 2, 2002 - April 30, 2010, crude oil price and gold price were weakly
positively correlated most of the time.
(2) The correlation coefficient was below 0.200 most of the time before the second half of 2007. Since
then, it generally stayed above 0.200, especially after the quick increase in the end of 2007. The highest
correlation coefficient was 0.321, realized on September 11, 2008. As of April 30, 2010, the correlation
coefficient was 0.305. Table 5.2 below provides the average correlation coefficients in different time periods.

Table 5.2 Average Correlation Coefficient of Crude Oil Price and Gold Price

2002.1.2-2007.9.28 2007.10.1-2010.4.30 2002.1.2-2010.4.30


Dollar and London PM Fix 0.119 0.241 0.157

The information summarized above can be interpreted as follows:


(1) Crude oil price increase (decrease) and gold price increase (decrease) are coincident. It is reasonable
to expect that a causal relationship may exist between the two.
(2) Although the correlation coefficient was positive most of the time during January 2, 2002 - April 30,
2010, it had ups and downs. Its increase in a certain period means that the frequency or magnitude (or both) of
same direction movement of crude oil price and gold price is higher or greater (or both) than that of opposite
direction movement. Similarly, the reason for its decrease in a certain period is that crude oil price and gold
price move more frequently or with greater magnitude (or both) in the opposite direction. However, the
decreases of correlation coefficient were relatively short-lived in the historical movement of crude oil price
and gold price. The positive correlation relationship between the two prices is still the main trend.
52

(3) According to Table 5.2, starting from October 2007, the co-movement of crude oil price and gold
price was much stronger than before. It is possible that their relationship experienced a structural change
somewhere after the middle of 2007.

5.1.3. Dollar Exchange Rate and Crude Oil Price


This section presents the figure of exponentially smoothed correlation coefficients between dollar
exchange rate and crude oil price. The smoothing factor α is set to be 0.005. Analysis will focus on data
starting from January 2, 2002 to allow the effect of exponential smoothing to be fully realized.

1.00
0.80
0.60
0.40
0.20
0.00
-0.20
1999/1/5
1999/8/5
2000/3/5

2001/5/5

2002/7/5
2003/2/5
2003/9/5
2004/4/5

2005/6/5
2006/1/5
2006/8/5
2007/3/5

2008/5/5

2009/7/5
2010/2/5
2000/10/5

2001/12/5

2004/11/5

2007/10/5

2008/12/5
-0.40
-0.60
-0.80
-1.00

Figure 5.3 Correlation Coefficients between Dollar Exchange Rate and Crude Oil Price

Figure 5.3 provides the following information:


(1) Before year 2003, dollar exchange rate and crude oil price are generally not correlated. Within the
period of January 2, 2003-April 30, 2010, dollar exchange rate and crude oil price are negatively correlated
most of the time. Again, since the dollar exchange rate is expressed as dollars per euro, the positive
correlation coefficients shown in this figure mean that dollar appreciation (depreciation) and crude oil price
decrease (increase) are coincident.
(2) The correlation coefficient was below 0.200 before the second half of 2007. Since then, the
correlation coefficient exhibited a quick escalation. Although there was a sharp decrease during December
2008-early February 2009, the correlation coefficient kept standing above 0.200. The highest correlation
coefficient was roughly 0.516, realized on November 24, 2008. As of April 30, 2010, the correlation
coefficient was roughly 0.374. Table 5.3 shows the average correlation coefficients in different time periods.
53

Table 5.3 Average Correlation Coefficient of Dollar Exchange Rate and Crude Oil Price

2002.1.2-2002.12.31 2003.1.2-2007.9.28 2007.10.1-2010.4.30 2003.1.2-2010.4.30


Dollar and Crude Oil -0.023 0.102 0.308 0.175

The information summarized above can be interpreted as follows:


(1) Since dollar depreciation (appreciation) and crude oil price increase (decrease) are coincident, it is
reasonable to expect that a causal relationship may exist between the two.
(2) The increase of correlation coefficient in a certain period means that the frequency or magnitude (or
both) of opposite direction movement of dollar exchange rate and crude oil price is higher or greater (or both)
than that of same direction movement. Similarly, the reason for the decrease of correlation coefficient in a
certain period is that dollar exchange rate and crude oil price move more frequently or with greater magnitude
(or both) in the same direction---they increase or decrease in value at the same time. There was one noticeable
sharp decrease of correlation coefficient during December 2008-early February 2009. It corresponds to the
movement patterns of dollar exchange rate and crude oil price in that period.
(3) According to Table 5.3, the correlation coefficient of dollar exchange rate and crude oil price
displays three major stages, prior 2003, 2003-2007.9 and 2007.10-2010.4. The co-movement of dollar
exchange rate and crude oil price become stronger and stronger along these three stages. It is possible that
their relationship experienced a structural change or even multiple structural changes in the sample period.

5.1.4. Summary of Correlation Coefficient Analysis


The correlation coefficient analysis in section 5.1.1-5.1.3 indicates that
(1) The movement of gold price correlates with the movement of both dollar exchange rate and crude oil
price and the latter two prices correlate with each other too. These results lay the foundations for the causality
analysis in following sections.
(2) The correlation between dollar exchange rate and gold price is stronger than that between dollar
exchange rate and crude oil price. The correlation between gold price and dollar exchange rate is stronger
than that between gold price and crude oil price. Crude oil price correlates stronger with dollar exchange rate
than with gold price during 2007.10.1-2010.4.30. But for the first subsample period (2002.1.2-2010.4.30)
and the entire sample period, the correlation between crude oil price and gold price is slightly stronger than
that between crude oil price and dollar exchange rate. Table 5.4 shows the correlation coefficients in three
different sample periods.
(3) After September 2007, all three correlation relationships became stronger either slightly or
considerably than before. It is therefore reasonable to entertain the possibility of structural changes in the
three bilateral relationships.
54

Table 5.4 Correlation Coefficient Matrices

2002.1.2 – 2007.9.28 2007.10.1 - 2010.4.30 2002.1.2 – 2010.4.30


Dollar Oil Gold Dollar Oil Gold Dollar Oil Gold
Dollar --- 0.081 0.395 --- 0.308 0.461 --- 0.152 0.415
Oil 0.081 --- 0.119 0.308 --- 0.241 0.152 --- 0.157
Gold 0.395 0.119 --- 0.461 0.241 --- 0.415 0.157 ---

5.2. Volatilities
Volatility measures the dispersion of asset returns. It is another tool to reveal data characteristics. Abrupt
volatility shifts may indicate important changes of market situation. This section provides the statistical
results and analysis of annualized volatilities of dollar exchange rate, crude oil price and gold price. Figure
5.4 shows the exponentially smoothed volatilities of the three prices over time. The smoothing factor α is set
to be 0.005. Analysis will focus on data starting from January 2, 2002 to allow the effect of exponential
smoothing to be fully realized.

80%
70%
60%
50%
40%
30%
20%
10%
0%
1999/1/5
1999/8/5
2000/3/5

2001/5/5

2002/7/5
2003/2/5
2003/9/5
2004/4/5

2005/6/5
2006/1/5
2006/8/5
2007/3/5

2008/5/5

2009/7/5
2010/2/5
2000/10/5

2001/12/5

2004/11/5

2007/10/5

2008/12/5

Dollar PM Gold Price Oil Price

Figure 5.4 Exponentially Smoothed Annualized Volatilities of Three Prices

5.2.1. Dollar Exchange Rate


According to Figure 5.4, within the period of January 2, 2002-April 30, 2010, the annualized volatility
of dollar exchange rate firstly fluctuated mildly around 10.000%, and then decreased slowly to around 6.591%
in September 2007. Starting from October 2007, the volatility showed a quick escalation, increasing from
around 6.552% to around 15.445% in April 2009. The volatility kept decreasing since then and as of April 30,
2010, the annualized volatility of dollar exchange rate was 12.021%. The lowest and highest volatility were
55

6.510% and 15.567%, realized on October 8, 2007 and April 2, 2009. Table 5.5 below provides the average
volatilities in different time periods.

Table 5.5 Average Volatility of Dollar Exchange Rate

2002.1.2-2007.9.28 2007.10.1-2010.4.30 2002.1.2-2010.4.30


Dollar Exchange Rate 9.439% 11.278% 10.013%

The interpretation of information conveyed in Figure 5.4 and Table 5.5 is that before October 2007, the
volatility of dollar exchange rate was relatively stable, fluctuating moderately in a narrow range. Starting
from that month, the volatility increased quickly and more than doubled (15.567% / 6.510% ≈ 2.391) in less
than one and a half years. Such increase of volatility indicates that the dollar exchange rate varies in a wider
and wider range, more sensitive and responsive to foreign exchange market innovations than before. Possible
reasons behind include but are not limited to larger and more frequent foreign exchange market capital
inflows and outflows, greater divergence in the prospect for dollar exchange rate movement, wilder
speculation and investors overreaction, which may all affect significantly the exchange rate formation.
Although the volatility decreased quickly after March 2009, it stayed above 12% all the time with a volatility
level higher than that most of the time prior to year 2009 as well as the historical average, a signal of close
investor attention to the dollar exchange rate.

5.2.2. Crude Oil Price


According to Figure 5.4, within the period of January 2, 2002-April 30, 2010, the annualized volatility
of crude oil price decreased slowly from around 50.000% in January 2002 to around 30.000% in May 2008.
In Jun 2008, the volatility started to increase and soared after August, increasing from 34.450% on September
2, 2008 to 71.833% on March 12, 2009, which is also the highest volatility in the entire sample period. The
volatility declined since then and as of April 30, 2010, the annualized volatility of crude oil price was
46.090%. The lowest volatility is 28.770%, realized on October 23, 2007. Table 5.6 below provides the
average volatilities in different time periods.

Table 5.6 Average Volatility of Crude Oil Price

2002.1.2-2008.5.30 2008.6.2-2010.4.30 2002.1.2-2010.4.30


Crude Oil Price 37.491% 54.797% 41.503%

Figure 5.4 and Table 5.6 together show that before June 2008, the volatility of crude oil price decreased
slowly and steadily. Starting from that month, the volatility became higher and higher and more than doubled
56

(71.833% / 34.450% ≈ 2.085) in less than seven month (2008.9.2 -2009.3.12). Such increase of volatility
indicates that the crude oil price varies in a wider and wider range, more sensitive and responsive to world
economy and energy market changes than before. A variety of factors may result in the escalation of crude oil
price volatility. Possible reasons include but are not limited to larger and more frequent crude oil market
capital inflows and outflows, greater divergence in the prospect for crude oil price movement, wilder
speculation and investor overreaction. In addition, different from dollar exchange rate, crude oil price is also
significantly influenced by the physical stock level, or more general, the current and future availability. A low
stock level, implying possible current crude oil scarcity, may cause nonlinear price reaction in the crude oil
market. Although the volatility declined quickly after mid-March 2009 and reached 46.090% on April 30,
2010, this volatility level is higher than that most of the time prior to year 2009 as well as the historical
average. In other words, the crude oil price nowadays is much more volatile than it used to be.

5.2.3. Gold Price


According to Figure 5.4, within the period of January 2, 2002-April 30, 2010, the annualized volatility
of London PM Fix gold price was relatively stable in year 2002-2004, fluctuating mildly around 15.00%. In
year 2005, the volatility slightly declined to around 12.300% in the first eight months. One month later, in
October, the volatility started to increase. At the beginning of August 2006, the volatility reached around
22.358%. Three months later, at the beginning of November, the gold price volatility entered a new round of
decrease, from around 22.492% to around 16.819% at the end of October 2007. Starting from November
2007, the volatility of gold price stepped into a quick escalation, which ended with 31.000% in March 2009.
The volatility declined after the peak and as of April 30, 2010, the annualized volatility of gold price was
21.708%. The lowest volatility is 12.218%, realized on August 26, 2005. Table 5.7 below provides the
average volatilities in different time periods.

Table 5.7 Average Volatility of Gold Price

2002.1.2-2005.9.30 2005.10.3-2007.10.31 2007.11.1-2010.4.30 2002.1.2-2010.4.30


London PM Fix 14.071% 18.615% 24.357% 18.304%

Figure 5.4 and Table 5.7 together show that within January 2002-April 2010, the development of gold
price volatility experiences three major phases: the first phase is from January 2002 to September 2005. In
this period, the volatility of gold price was relatively stable, moving in a narrow range around 15%. The
trough was 12.218%, realized in August 2005. The average volatility was 14.071%. The second phase is from
October 2005 to October 2007. Within these 25 months, the volatility firstly went up to around 22.492% and
then went down to around 16.819%. The average volatility in this period was 18.615%. The third phase starts
from November 2007 and ends in April 2010. In this period, the movement pattern of volatility was similar to
57

that of the second phase but the magnitude was greater. The peak, trough and average are 31.350%, 21.708%
and 24.357%. Clearly, as the volatility go through the three phases, the peaks, troughs and averages all rise.
Take the troughs as an example, they were around 12.200%, 16.800% and 21.700% in three phases,
increasing by 37.705% from phase one to phase two and 29.167% from phase two to phase three. The
increase of volatility indicates that the gold price varies in a wider and wider range, more sensitive and
responsive to world economy innovations than before. A variety of factors may result in the escalation of gold
price volatility. Possible reasons include but are not limited to larger and more frequent gold market capital
inflows and outflows, greater investor divergence in the prospect for gold price movement, wilder
speculation and investor overreaction. As of April 30, 2010, the volatility of gold price, also the trough in
phase three, was 21.708%, close to the peak in phase two. An interesting inference based on this fact and the
example above is that after each unprecedented volatile period, the endurance and cognition of investors
about gold price variability adjust and grow. As a result, investors and the market as a whole gradually get
used to a “wilder and wilder gold price”.

5.2.4. Summary of Volatility Analysis


The volatility analysis in section 5.2.1-5.2.3 indicate that
(1) The development of dollar exchange rate volatility and crude oil price volatility can both be roughly
divided into two stages, the pre-2008 period and the post-2008 period. The volatility of gold price, however,
experiences three major phases. The volatilities of dollar exchange rate and crude oil price steadily decreased
in their first stage. The volatility of gold price was stable in phase one and became volatile in phase two,
roughly year 2006-year 2007.
(2) All three prices were much more volatile after the second half of 2007. Their volatilities all showed
dramatic increase after 2007 and dramatic decrease starting from the first half of 2009. As of April 30, 2010,
their volatilities were higher than their respective historical average, as well as the respective volatilities in
most of the sample period. This finding lends further support to the motive for detecting structural changes.
(3) Crude oil price has the highest volatility and dollar exchange rate has the lowest volatility.
(4) A variety of factors may contribute to high price volatility. Besides the factors that are pertinent to all
three prices, crude oil price is also subject to the stock level.
(5) The development path of gold price volatility also seems to be the growth path of investor endurance
and cognition about gold price variability.

5.3. Econometric Results


According to the results and analysis in section 5.1 and section 5.2, it is reasonable to explore causality
between gold price and dollar exchange rate and between gold price and crude oil price. Moreover, it is also
58

justifiable to detect the existence of structural changes in the relationships of gold price with dollar exchange
rate and crude oil price. Therefore, this section presents and interprets the results of econometric tests
described in Chapter III.

5.3.1. Data Stationarity


As introduced in Chapter III, ADF test and PP test are used to examine stationarity of the three price
series. The results are summarized in Table 5.8 below.
Although test statistics are different, the results of both the ADF test and the PP test are similar. In the
level estimations of the variables, the null hypotheses that e, o and g series have unit roots are not rejected,
which means the three variables are not stationary in levels. Further tests are performed on the first order
differences of the three variables. The results show that the null hypotheses are rejected this time, indicating
that the three variables are stationary in the first order difference form. So the three variables, e, o and g, are
not I(0) process but I(1) process.

Table 5.8 The Results of ADF Test and PP Test

Variables Test Statistics Probability Critical Value (10%;5%;1%) Test Form Conclusion
ADF Test
e -2.776 0.209 -3.147;-3.444;-4.028 C,T,1 Do not reject H0
o -2.226 0.198 -2.578;-2.883;-3.480 C,0,1 Do not reject H0
g -2.746 0.220 -3.147;-3.444;-4.028 C,T,0 Do not reject H0
∆e -8.272 0.000 -1.615;-1.943;-2.582 0,0,0 Reject H0
∆o -8.589 0.000 -1.615;-1.943;-2.582 0,0,0 Reject H0
∆g -10.595 0.000 -1.615;-1.943;-2.582 0,0,0 Reject H0
PP Test
e -2.834 0.188 -3.147;-3.444;-4.028 C,T,3 Do not reject H0
o -3.099 0.111 -3.147;-3.444;-4.028 C,T,5 Do not reject H0
g -2.725 0.228 -3.147;-3.444;-4.028 C,T,4 Do not reject H0
∆e -8.131 0.000 -1.615;-1.943;-2.582 0,0,5 Reject H0
∆o -8.647 0.000 -1.615;-1.943;-2.582 0,0,2 Reject H0
∆g -10.646 0.000 -1.615;-1.943;-2.582 0,0,6 Reject H0
Note: ∆ is the first-order difference operator. Test form is expressed by C, T and K, where C denotes
constant term, T denotes trend term and K denotes number of lags. For C and T, 0 means the term is not
included. For K, 0 means no autoregressive terms.
59

5.3.2. Co-integration Test


The purpose of co-integration test is to see whether two or more non-stationary time series still have a
meaningful long run stable relationship. According to section 3.3.2, the Engle-Granger (EG) two-step
method is used to test co-integration relationships, which is based on assessing whether single equation
estimates of the equilibrium errors appear to be stationary. The estimation results are shown in Table 5.9.
The null hypothesis of the EG two-step method is that no co-integration relationship exists, which
means that the regression residual series from step one is not stationary. Rejection of the null hypothesis of
ADF Test indicates that the residual variable is stationary, which in turn, indicates that a co-integration
relationship exists. According to Table 5.8, for the residual variables from both co-integration equations, the
null hypothesis of the ADF Test is rejected. Therefore, two co-integration relationships are identified. Dollar
exchange rate and gold price are co-integrated. Crude oil price and gold price are also co-integrated.

Table 5.9 Results of Co-integration Test

Results from Step One


Co-integration Equation 𝑒𝑒�𝑡𝑡 = -1.831 + 0.323 gt 𝑜𝑜�𝑡𝑡 = -2.221 + 0.974 gt 𝑜𝑜�𝑡𝑡 = 3.370 + 2.526 et
Standard Error (0.089) (0.014) (0.295) (0.048) (0.035) (0.152)
t-Statistic (-20.629) (22.418) (-7.519) (20.300) (96.191) (16.571)
Probability (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Results from Step Two
Variables Test Statistics Probability Critical Value (10%;5%;1%) Test Form Conclusion
εeg -2.078 0.037 -1.615;-1.943;-2.582 0,0,1 Reject H0
εog -3.716 0.000 -1.615;-1.943;-2.582 0,0,2 Reject H0
εoe -4.322 0.000 -1.615;-1.943;-2.582 0,0,1 Reject H0
Note: εeg and εog denote the residual variables from the two co-integration equations. Test form is
expressed by C, T and K, where C denotes constant term, T denotes trend term and K denotes number of lags.
For C and T, 0 means the term is not included. For K, 0 means no autoregressive terms.

5.3.3. Parameter Stability Test


The Engle-Granger two-step procedure for co-integration detection presumes that the co-integrating
vector is time-invariant under the alternative hypothesis. However, given the results and analysis in section
5.2, it is necessary to examine whether the co-integration relationships established are stable without shifts at
unknown points. To cope with this issue, this section presents in Table 5.10 the results of parameter stability
tests described in section 3.3.3.
60

Table 5.10 Results of Parameter Stability Tests

Sample Period Observations τ Bounds Possible Breakdates No. of Breakdates


1999.1-2010.4 136 [0.15, 0.85] [21, 115] 2000.9-2008.7 95
Co-integration
𝑒𝑒�𝑡𝑡 = -1.831 + 0.323 gt 𝑜𝑜�𝑡𝑡 = -2.221 + 0.974 gt 𝑜𝑜�𝑡𝑡 = 3.370 + 2.526 et
Equation
SupF 5.346 (10.6;12.4;16.2) 21.333 (10.6;12.4;16.2) 29.664 (10.6;12.4;16.2)
MeanF 1.353 (3.73;4.57;6.78) 3.355 (3.73;4.57;6.78) 8.992 (3.73;4.57;6.78)
Note: The numbers in parentheses are critical values at 10% level, 5% level and 1% level.

Spanning over 11 years, the full sample has 136 observations. After sample truncation according to τ, 95
possible breakdates are selected, which are from the 21st observation to the 115th observation in the sample,
corresponding to the period from September 2000 to July 2008.
As shown in Table 5.10, for the regression of dollar exchange rate on gold price, both SupF value and
MeanF value are lower than respective critical value at 10% level. So the null hypothesis of parameter
stability is not rejected. Since no structural change is identified, the co-integration relationship between dollar
exchange rate and gold price is considered stable over the sample period. For the regression of crude oil price
on gold price, the story is a little different. The SupF value is noticeably higher than the critical value at 1%
level. So the null hypothesis of parameter stability is rejected and at least one structural change occurred in
the sample period. The MeanF test, however, provides a seemingly contradictive result. The MeanF value is
lower than the critical value at 10% level. So for the MeanF test, the null hypothesis of parameter stability is
not rejected, suggesting that no structural change exists in the sample period. To solve this contradiction,
recall the difference between SupF test and MeanF test mentioned in section 3.3.3. The SupF test is used to
detect whether there is a sharp shift in regime. The MeanF test is used to capture gradual changes in the
regression coefficients. Therefore, the test results suggest an abrupt change in the co-integration relationship
but are not in favor of gradual changes in the regression coefficients. For the regression of crude oil price on
dollar exchange rate, both SupF value and MeanF value are higher than respective critical value at 1% level.
So the null hypothesis of parameter stability is rejected. The co-integration relationship between crude oil
price and dollar exchange rate is considered unstable over the sample period.
Besides the SupF and MeanF test statistics, the sequences of F statistic also provide useful information.
Figure 5.5 below exhibits the F statistic sequence obtained from regression of dollar exchange rate on gold
price at different breakdates.
61

20

15

10

Oct-2002
Mar-2003

Mar-2008
Jan-2004
Jun-2004

Oct-2007
May-2002

Apr-2005

May-2007
Aug-2003
Sep-2000
Feb-2001
Jul-2001

Nov-2004

Sep-2005
Feb-2006
Jul-2006
Dec-2001

Dec-2006
F statistic 10% 5% 1%

Figure 5.5 F Statistic Sequence from Regression of Dollar Exchange Rate on Gold Price

In Figure 5.5, the highest F value, SupF, is 5.346, realized in October 2005. Although this date does not
provide a SupF value high enough to suggest parameter instability, it is still of importance in the historical
price development of dollar exchange rate and gold.
Figure 5.6 below shows the gold price, actual dollar exchange rate, and fitted dollar exchange rate.

0.5 7.5
0.4
7
0.3
0.2 6.5

0.1 6
0
5.5
May-2000

May-2002

May-2004

May-2006

May-2008
Sep-1999

Sep-2003

Sep-2005

Sep-2007

Sep-2009
Sep-2001
Jan-1999

Jan-2001

Jan-2003

Jan-2005

Jan-2007

Jan-2009

-0.1
-0.2 5

Fitted Dollar Actual Dollar Gold Price (Right Axis)

Note: The vertical dotted line corresponds to October 2005, the date with the highest F value. Axes are
labeled with natural logarithm values.
Figure 5.6 Gold Price, Actual Dollar Exchange Rate and Fitted Dollar Exchange Rate

In Figure 5.6, October 2005 is a milestone for all three series. Soon after that month, the dollar stopped
appreciation and started its 30 months of depreciation, decreasing by 33.63% from 1.179 in November 2005
to 1.575 in April 2008. For gold price, it fluctuated slightly around $430 in the first eight months of 2005 but
62

started to increase thereafter. In October 2005, the gold price was $469.898. After 29 months, the gold price
increased by 106.09% to $968.434 in March 2008. Moreover, gold price is more volatile after October 2005.
What happened around October 2005? Among various events, Hurricanes Katrina (on August 29), Rita (on
September 24) and Wilma (on October 24) are of special relevance. These hurricanes severely hit Louisiana,
Texas, Florida and other states and caused heavy casualty and huge economic loss. The U.S. crude oil
production and distribution were also seriously affected. As a result, energy prices soared and Consumer
Price Index (CPI) rose sharply in September and remained noticeable increase in October. Figure 5.7 shows
the 12-month percent change of CPI over 1991-2010. In September and October 2005, the CPI increased 4.74%
and 4.35% respectively, the first and second highest increase in August 1991 – October 2005. The 12-month
CPI percent change did not fall below 3.00%, a relatively high level of inflation for U.S. over past 20 years,
until September 2006. Worries about the impacts of these natural disasters on U.S. economy, especially on
price level and consumer spending, were likely to push down dollar value and drive up gold price.
Figure 5.8 exhibits the F statistic sequences obtained from regression of crude oil price on gold price at
different breakdates. Clearly, the highest F value is 21.333, realized in June 2008. Since it is well above the
critical value at 1% level, the parameters in the co-integration relationship between crude oil price and gold
price are considered instable. Besides the SupF value, there are also several F values that are statistically
significant at different levels or that are not statistically significant but are local maxima. The following part,
coupled with Figure 5.9 and Figure 5.10, will focus on three particular time points, which are June 2004, a
local maximum without statistical significance, March 2007, a local maximum exceeding the critical value at
10% level and June 2008, the global maximum of the F sequence.

7.00%
6.00%
5.00%
4.00%
3.00%
2.00%
1.00%
0.00%
-1.00%
Jan/91
Jan/92
Jan/93
Jan/94
Jan/95
Jan/96
Jan/97
Jan/98
Jan/99
Jan/00
Jan/01
Jan/02
Jan/03
Jan/04
Jan/05
Jan/06
Jan/07
Jan/08
Jan/09
Jan/10

-2.00%
-3.00%

Data source: The Bureau of Labor Statistics. Consumer Price Index – All Urban Consumers, U.S. City
Average, All items; Seasonally Adjusted.
Figure 5.7 U.S. Historical 12-Month Percent Change of CPI in 1991-2010
63

25

20

15

10

Oct-2002
Mar-2003

Mar-2008
Jan-2004
Jun-2004

Oct-2007
May-2002

Apr-2005

May-2007
Aug-2003
Sep-2000
Feb-2001
Jul-2001

Nov-2004

Sep-2005
Feb-2006
Jul-2006
Dec-2001

Dec-2006
F statistic 10% 5% 1%

Figure 5.8 F Statistic Sequence from Regression of Crude Oil Price on Gold Price

In June 2004, the crude oil price was $38.03 and the gold price is $392.37. Soon after this month, the y
stood above $40 and $400 respectively. The monthly crude oil price was higher than $40 in most of the
sample period except in February 2009. The gold price, however, never went back to below $400 again.
For crude oil price, $40 is special because it was a possible “psychological threshold” to investors. As
shown in Figure 5.9, the crude oil price had started to increase since January 2002, when the crude oil price
was around $20. By April 2004, the crude oil price had been constantly below $40. Although it once reached
$35.83 in February 2003, a local maximum, it soon decreased to below $30. In May 2004, the monthly crude
oil price reached $40.28, slightly higher than the threshold price. But one month later, in June, the crude oil
price declined to below $40 and reached as low as $35.60 (daily closing spot price on June 29, 2004). In July,
investors finally accepted a price level of crude oil price that is above $40 and the crude oil market stepped
over the threshold price. As further evidence of the “psychological threshold”, Figure 5.10 below provides
historical prices of crude oil from 1986-2010. 17 In fact, before May 2004, there are only two days when the
nominal spot crude oil price exceeded $40, which are October 9 and 11, 1990. The prices were 40.73 and
41.07. A major reason for the high crude oil price at that time is the Persian Gulf War. So June 2004 is a
milestone for the crude oil price development. After that month, $40 is no longer a ceiling for crude oil price,
which instead entered a four-year long escalation ending in July 2008 with a price level higher than $145.

17. The spot FOB WTI crude oil price data series kept by the Energy Information Agency starts from year
January 2, 1986.
64

5 7.5
4.5 7
4
6.5
3.5
6
3
2.5 5.5

2 5

Mar-2000

Mar-2007
Jan-1999

Oct-2000

Jun-2005
Jan-2006

Oct-2007
May-2001
Aug-1999

Apr-2004

May-2008
Aug-2006
Nov-2004
Jul-2002
Feb-2003
Sep-2003

Feb-2010
Jul-2009
Dec-2001

Dec-2008
Fitted Oil Price Actual Oil Price Gold Price

Note: From left to right, the three dashed lines correspond to June 2004, March 2007 and June 2008.
Axes are labeled with natural logarithm values.
Figure 5.9 Gold Price, Actual Crude Oil Price and Fitted Crude Oil Price

160 1400
140 1200
120 1000
100
800
80
600
60
40 400
20 200
0 0
1/2/1986

1/2/1988

1/2/1990

1/2/1992

1/2/1994

1/2/1996

1/2/1998

1/2/2000

1/2/2002

1/2/2004

1/2/2006

1/2/2008

1/2/2010

Crude Oil Price (Left Axis) Gold Price (Right Axis)

Figure 5.10 Historical Daily Prices of Crude Oil and Gold 1986-2010

For gold price, $400 to gold price is not as extreme as $40 to crude oil price. But it is still an important
price level. In Figure 5.10, the gold price fluctuated around $400 most of the time during 1986 - 1995.
Starting from March 1996, the gold price stayed below $400 and did not go back to $400 until December
2003. After moving slightly above $400 for five months, in May 2004, the gold price decreased again and
stayed below $400 for three months and reached as low as $384.95 on June 10. The investors seemed to need
some time to digest the fact that gold price went back to $400 again. 18 After the short “setback” from May to
July, the gold price stood above $400 again and entered a new round of escalation. This interpretation is

18. It is certainly possible that someone would like to interpret such price movement in the way that $400 is a
psychological ceiling price at which investors sell gold to lock the profits earned in earlier months.
65

partially supported by the findings of Aggarwal and Lucey (2007). In their study of gold market, they find
that “psychological barriers” exist at the 100’s digits (price levels such as $400) of daily gold prices.
Different from June 2004, March 2007 corresponds to an F value of 12.022, a local maximum with
statistical significance at 10% level. In Figure 5.9, the crude oil price in March 2007 just went out of a local
minimum realized in January 2007. During July 2004-July 2006, the crude oil price continued the upward
movement trend that started from year 2002 and increased to $74.41. Then the price declined to $54.51 in
January 2007. March 2007 was a start of the subsequent 13 - month rapid price escalation of crude oil price.
The gold price declined to its local minimum and rallied earlier than the crude oil price. In March 2007, it was
on a mild upward trend which started in October 2006. Before then, the gold price peaked at $725 in May and
later fell down to around $570. The gold price kept increasing in April 2007 and then fluctuated narrowly
between $650 and $700 for three months. In mid-August 2007, the gold price began to soar.
The gold price movement during May 2006-August 2007 can be associated with a series of events
occurred then. At least as early as in year 2005, warning about housing market bubble, Collateralized Debt
Obligations (CDOs) and credit default swaps arose. In May 2006, as early signals of the looming subprime
crisis, Ameriquest, a subprime mortgage market player, announced large scale job cuts and close of branches
(Wikipedia, 2010). Another market participant, the Merit Financial Inc., filed for bankruptcy (Cox and Glapa,
2009). In the fall, J.P. Morgan started to reduce its exposure to subprime mortgages. So did Goldman-Sachs in
December and many other financial institutions in the spring and summer 2007 (Wikipedia, 2010). While it
was possible that the early signals were not captured or well interpreted by the gold market, the upward trend
of gold price starting from October 2006 was likely to be the reaction to the potential danger of subprime
mortgages. However, perhaps because investors did not fully realize how serious the problem was, the gold
price only showed a moderate increase. In early February 2007, HSBC warned that its mortgage service
operation was much worse than current market estimates (Cox and Glapa, 2009). Soon after that, the
subprime mortgage industry began to collapse. Many subprime lenders declared bankruptcy or reported
significant losses (Wikipedia, 2010). Such news fueled the gold price increase in February, March and April.
Since there was no further shocking news in May, the nerves of investors eased and the gold price reduced.
But in June, the redemption halt and huge loss of two funds of Bear Stearns & Co hit the global financial
market and panic began to germinate (Cox and Glapa, 2009). In August, liquidity began freezing and panic
widely spread across markets. As the subprime crisis aggravated, capital safety became increasingly
important. As a result, a huge amount of money rushed to gold and other commodity markets. Not
surprisingly, starting from August, the gold price soared. Figure 5.11 shows the numbers and percentages of
days when dollar exchange rate and gold price increased or decreased together in each month during May
2006 – October 2007. According to section 5.1.1, these two prices are negatively correlated over time. But in
Figure 5.11, the more shock to the market, the more likely dollar exchange rate and gold price move in the
same direction. In March and August 2007, the two prices move in the same direction in nearly half of all
trading days in a month, a phenomenon that is likely to be driven by the capital safety concern of investors.
66

12 60.00%
10 50.00%
8 40.00%
6 30.00%
4 20.00%
2 10.00%
0 0.00%
May/06

May/07
Aug/06

Apr/07

Aug/07
Jul/06

Sep/06

Nov/06

Feb/07

Jul/07

Sep/07
Dec/06
Jun/06

Oct/06

Jan/07

Jun/07
Mar/07

Oct/07
Number of Days Percentage of Corresponding Month

Data source: Author’s own calculation.


Figure 5.11 Co-movement Days of Dollar Exchange Rate and Gold Price (2006.5-2007.10)

In Figure 5.8, the F value of June 2008 is the global maximum and exceeds the critical value at 1% level.
So the co-integration relationship between crude oil price and gold price is considered instable in parameters.
This result is supported by Figure 5.10, in which June 2008 is a clear turning point in the price histories of
crude oil and gold. Soon after that month, both prices decreased sharply. The crude oil price declined from
monthly average price $133.37 in July 2008 to $39.09 in February 2009. The gold price reduced from
monthly average price $939.77 in July 2008 to $760.86 in November 2008. The sharp decreases of the two
prices after June 2008 were the largest declines in both magnitude and pace in the price histories of crude oil
and gold over the sample period.
Such price movements reflect the market situation in that period. Before July, the crude oil price was
driven up by many factors such as concern of decline in petroleum reserves, Middle East tension, speculation
and the subprime crisis starting from 2007, in which investors switched to commodity markets to seek capital
security. But fundamentally, since the main role of crude oil is a major energy source, its price directly
depends on the crude oil consumption and supply. On July 3, 2008, the daily closing crude oil price set a new
record of $145.31. On July 11, the intraday crude oil price further climbed to $147.27 in response to the
Iranian Missiles Tests. Such high price levels seriously jeopardized the potential of global economic growth
which relies heavily on fossil fuel and also greatly depressed crude oil consumption. Figure 5.12 presents the
U.S. industrial production during 1999-2010. The industrial production plummeted after July 2008 and did
not stop decreasing until June 2009. 19 Figure 5.13 shows the U.S. crude oil consumption during 1999-2010.
The crude oil consumption started to decline after June 2008 and dived in September 2008. Although the
consumption partially rebounded in October, it fell again soon and stayed at low levels in both 2009 and 2010.
The consumption in almost every month of both 2009 and 2010 is lower than the consumption in

19. It is possible for industrial production to have a short lag when responding to the crude oil price.
67

corresponding month of previous nine years. Without strong demand, the crude oil price fell down from the
peak and reached as low as $30.28 on December 23, 2008, around only one fifth of the peak price.

3100
3000
2900
2800
2700
2600
2500
2400
May/01

May/08
Aug/99

Apr/04
Nov/04

Aug/06
Jul/02
Feb/03
Sep/03

Jul/09
Feb/10
Dec/01

Dec/08
Jan/99

Jun/05
Mar/00
Oct/00

Jan/06

Mar/07
Oct/07
Data source: Federal Reserve Statistical Release G.17, Seasonally Adjusted, Billions of 2000 dollars.
Figure 5.12 Industrial Production: Gross Value of Products 1999-2010

17500
17000
16500
16000
15500
15000
14500
14000
13500
13000
May/01

May/08
Aug/99

Apr/04
Nov/04

Aug/06
Jul/02
Feb/03
Sep/03

Jul/09
Feb/10
Dec/01

Dec/08
Jan/99

Oct/00

Jun/05
Mar/00

Jan/06

Mar/07
Oct/07

Data source: Author’s own calculation based on raw data from Energy Information Agency. 20
Figure 5.13 U.S. Crude Oil Consumption 1999-2010 (Thousand Barrels/Day)

20. Monthly crude oil consumption data is not available. The author uses the following formula to calculate
approximate crude oil consumption: U.S. total crude oil production + U.S. net crude oil import – U.S. crude
oil stock change + Petroleum Adjustments = U.S. crude oil consumption. The data of the four variables on the
left hand side of the equation are available from Energy Information Agency.
68

The movement of gold price in 2008 was another story. Due to continuous huge capital loss of financial
institutions, widespread liquidity deficiency problems and rating downgrade of large bond insurers, the gold
price went up from $850 at the beginning of the year to $1000 in mid-March. After the second 75 basis points
cut of the target federal funds rate and the bailout of Bear Stearns, the gold price declined. But shocked by
events like further asset write-downs of large financial institutions, rating downgrade of the Municipal Bond
Insurance Association and the American Municipal Bond Assurance Corporation, the world’s largest and
second largest bond insurers, and capital concerns on Fannie Mae and Freddie Mac, the gold price climbed up
again in May 2008 and reached $986 on July 15, 2008. The panic among investors was then somewhat
relieved. In the second half of July, several large financial institutions reported better than expected
performance in the second quarter of 2008 and the U.S. House of Representatives passed a housing bill which
contained a rescue plan for Freddie Mac and Fannie Mae (Cox and Glapa, 2009). In August, Citigroup,
Merrill Lynch and UBS agreed to buy back $36.7 billion worth of auction-rate securities from their clients
(Cox and Glapa, 2009). 21 From roughly mid-July to mid-September, the gold price declined quickly and
reached $740.75 on September 11, 2008. But it climbed up to $900 in two weeks because of the market panic
caused by the bankruptcy of Lehman Brothers, the acquisition of Merrill Lynch by Bank of America, the
rating downgrade of American International Group, the freeze of commercial paper market and the collapse
of Washington Mutual. The gold price then went down to $828 in the first week of October partially because
the approval of revised bailout plan and a series of nationalization of financial institutions in European
counties (Cox and Glapa, 2009). But it went back to $900.5 in the following week because of worries of
worsening credit crisis and global recession. The Dow Jones Industrial Average Index and the Standard &
Poor's 500 index had their worst week since 1933, losing 22.1% and 18.2% respectively (Wikipedia, 2010).
As the governments of major economies unveiled their rescue plans and started international cooperation to
deal with the global credit crisis, the gold price fell down quickly to $712.5 on October 24 and fluctuated
between $710 and $750 for roughly one month. In December 2008, Standard and Poor’s rating agency
downgraded the credit rating of Russia and of eleven of the world’s largest banks (Cox and Glapa, 2009). At
least five major economies further cut interest rates, among which, the Federal Reserve lowered interest rates
to a range of 0 - 0.25% (Cox and Glapa, 2009). Three largest automobile manufacturers in U.S., General
Motors, Chrysler, and Ford, were in the danger of collapse. The credit crisis that originally broke out within
the financial system had gradually infected other industries and transformed into a global economic crisis.
The economic growth slow-down was inevitable. In this month, the gold price rose again and increased to
$865 as of December 31, 2008. Throughout 2008, the gold price was very sensitive and vulnerable, heavily
subject to changes of the global financial system and the world economy. The gold price was much less an
accurate measurement of the value of gold as goods but more a degree indicator of investor anxiety. Gold was
considered a harbor against financial surges and was invested to meet the need for capital security.

21. The total amount of buybacks breaks down into $7.3 billion for Citigroup, $10 billion for Merrill Lynch
and $19.4 billion for UBS.
69

Figure 5.14 below exhibits the F statistic sequences obtained from regression of crude oil price on dollar
exchange rate at different breakdates. Clearly, the highest F value is 29.664, realized in May 2001. Since it is
well above the critical value at 1% level, the parameters in the co-integration relationship between crude oil
price and dollar exchange rate are considered instable. Besides the SupF value, there are also many dates after
May 2001 with F values that are statistically significant at different levels. The following part, coupled with
Figure 5.15 and Figure 5.16, will focus on the date with the highest F value. The rest dates will be discussed
in section 5.3.4 if they are confirmed to be structural break date estimators.

35
30
25
20
15
10
5
0
May-2002

Aug-2003

Apr-2005

May-2007
Nov-2004
Sep-2000
Feb-2001
Jul-2001

Sep-2005
Feb-2006
Jul-2006
Dec-2001

Dec-2006

Mar-2008
Oct-2002
Mar-2003

Jan-2004
Jun-2004

Oct-2007
F Statistic 10% 5% 1%

Figure 5.14 F Statistic Sequence from Regression of Crude Oil Price on Dollar Exchange Rate

2 160
140
1.5 120
100
1 80
60
0.5 40
20
0 0
Jan-1999

Jan-2001

Jan-2003

Jan-2005

Jan-2007

Jan-2009
May-2000

May-2002

May-2004

May-2006

May-2008
Sep-1999

Sep-2001

Sep-2003

Sep-2005

Sep-2007

Sep-2009

Dollar(Left Axis) Crude Oil Price(Right Axis)

Figure 5.15 Dollar Exchange Rate and Crude Oil Price

In Figure 5.15, the dollar exchange rate and crude oil price were generally positively correlated before
May 2001. Dollar appreciation (depreciation) concurred with crude oil price increase (decrease). However,
70

during June 2001-April 2010, the two variables were generally negatively correlated. Dollar depreciation
(appreciation) coincided with crude oil price increase (decrease). Several reasons may account for the change.
During 1997-1998, the Asian financial crisis broke out. As its subsequent impact, global economic growth
slowed down. The crude oil price once reached as low as $10.82 per barrel on December 10, 1998. As the
global economy gradually recovered, crude oil price started to increase. The U.S. was not as heavily affected
by the financial crisis as Asian countries. Also, during 1999 – 2000, the Federal Reserve Bank increased 6
times the target federal funds rate from 4.75% to 6.5%. This rate was maintained until the beginning of 2001.
The high interest rate might motivate interest arbitrage and thus pushed dollar exchange rate up. Besides,
Euro was officially introduced in 1999. It is possible that investors and foreign exchange markets were
discovering the proper price level of Euro at the beginning 22. All these reasons may explain that within the
period of 1999.1-2000.11 both crude oil and dollar increased in value. Between December 1999 and May
2001, crude oil price went down and dollar exchange rate stopped appreciation. A relevant reason is that
during March 2001 - November 2001, U.S. went into a short recession 23. The “great moderation” ceased and
energy consumption was affected. In late 2001, dollar depreciation began, and crude oil price started to
increase. Such price movement partially resulted from the monetary policies at that time. Figure 5.16 shows
the target federal funds rate during 1999-2010. From 2001 to mid-2004, the Federal Reserve Bank reduced
target federal funds rate from 6.5% to 1%. Particularly, the Federal Reserve Bank cut the rate eleven times in
2001, trying to stimulate the U.S. economy. As a result, both M2 and M3 increased quickly and the dollar
became the low interest rate currency in the interest arbitrage, pushing dollar to depreciate. The crude oil
price, however, driven by gradually recovered energy demand and loose monetary policy, started to increase.

7
6
5
4
3
2
1
0
1999/1/1
1999/8/1
2000/3/1

2001/5/1

2002/7/1
2003/2/1
2003/9/1
2004/4/1

2005/6/1
2006/1/1
2006/8/1
2007/3/1

2008/5/1

2009/7/1
2010/2/1
2000/10/1

2001/12/1

2004/11/1

2007/10/1

2008/12/1

Data source: Federal Reserve Bank of St. Louis. Starting from December 15, 2009, the target rate is not
a fixed number but a narrow range of 0-0.25%.
Figure 5.16 Target Federal Funds Rate (%)

22. The author thanks Dr. Abbott for the Euro introduction explanation.
23. Detailed recession data and time is available at The National Bureau of Economic Research.
71

5.3.4. Timing of Structure Breaks


According to section 5.3.3, the relationship between dollar exchange rate and gold price is tested to be
stable over the sample period. However, for the relationship between crude oil price and gold price and the
relationship between crude oil price and dollar exchange rate, there are several dates with F values higher
than the critical value at 10% level. So multiple break dates may exist in these two bilateral relationships.
Recall Figure 5.8, the F sequence obtained from the regressions between crude oil price and gold price
has a global maximum at the date of June 2008 and a local maximum at the date of March 2007, both of
which are statistically significant. Therefore, June 2008 is inferred as a break date estimator and March 2007
a candidate break date estimator. The sample is split at June 2008 and the parameter stability tests are
performed on the first subsample, which starts from January 1999 and ends in June 2008. Table 5.11 and
Table 5.12 below present results of co-integration test and parameter stability tests on the first subsample.

Table 5.11 Results of Co-integration Test on Subsample of Crude Oil Price and Gold Price

Results from Step One


Co-integration Equation 𝑜𝑜�𝑡𝑡 = -3.873 + 1.257gt
Standard Error (0.302) (0.050)
t-Statistic (-12.820) (25.002)
Probability (0.000) (0.000)
Results from Step Two
Variables Test Statistics Probability Critical Value (10%;5%;1%) Test Form Conclusion
εog -3.758 0.000 -1.615;-1.944;-2.586 0,0,0 Reject H0
Note: εog denotes the residual variable from the co-integration equation. Test form is expressed by C, T
and K, where C denotes constant term, T denotes trend term and K denotes number of lags. For C and T, 0
means the term is not included. For K, 0 means no autoregressive terms.

Table 5.12 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Gold Price

Sample Period Observations τ Bounds Possible Breakdates No. of Breakdates


1999.1-2008.6 114 [0.15, 0.85] [18, 96] 2000.6-2006.12 79
1999.1-2008.6 114 [0.1, 0.9] [12, 102] 1999.12-2007.6 91
Co-integration Equation 𝑜𝑜�𝑡𝑡 = -3.873 + 1.257 gt
SupF 4.178(10.6;12.4;16.2)
MeanF 1.027 (3.73;4.57;6.78)
Note: The numbers in parentheses are critical values at 10% level, 5% level and 1% level.
72

The Table 5.11 above shows that in the subsample period, crude oil price and gold price are still
co-integrated, but the co-integration parameters are different. The constant term is smaller and the slope term
is larger in the co-integration relationship established based on the subsample.
Table 5.12 presents two scenarios of possible breakdate range in the upper part of the table. The
difference between the two scenarios is the choice of τ. In the first case, τ is [0.15, 0.85], consistent with the
choice in the previous section. But the resulting possible breakdate range is June 2000-December 2006,
excluding March 2007, the candidate break date estimator. So in the second case τ is [0.1, 0.9], which yields
a wide enough possible breakdate range to include March 2007. The results of parameter stability tests in the
lower part of Table 5.12 are calculated using the τ in the second scenario. The results indicate that both the
SupF statistic and the MeanF statistic are well below the critical values at 10% level. Figure 5.17 below
provides the sequence of F values obtained from the subsample regressions. It shows that all F values are
much lower than the critical value at 10% level. The highest F value corresponds to the date of March 2004.
The F value of March 2007, the candidate breakdate estimator, is close to zero. The inference based on Table
5.12 and Figure 5.17 is that the co-integration relationship between crude oil price and gold price in the first
subsample period is stable without structural breaks. Therefore, the relationship between the two prices has
only one structural break in the entire sample period, which occurred in June 2008.

20

15

10

0
Oct-2000
Mar-2001

Jan-2002
Jun-2002

Oct-2005
Mar-2006

Jan-2007
Jun-2007
May-2000

May-2005
Aug-2001

Apr-2003
Nov-2002

Aug-2006
Sep-2003
Feb-2004
Jul-2004
Dec-1999

Dec-2004

F Statistic 10% 5% 1%

Figure 5.17 F Statistic Sequence from Subsample Regression of Crude Oil Price on Gold Price

In Figure 5.14, the F sequence obtained from regressions between crude oil price and dollar exchange
rate has a global maximum at the date of May 2001 and a local maximum at the date of November 2001, both
of which are statistically significant at 1% level. Therefore, May 2001 is inferred as a break date estimator
and November 2001 a candidate break date estimator. The sample is split at May 2001 and the parameter
stability tests are performed on the second subsample, which is from June 2001 to April 2010. Table 5.13 and
Table 5.14 present results of co-integration test and parameter stability tests on the second subsample.
73

Table 5.13 Results of Co-integration Test on Subsample of Crude Oil Price and Dollar Exchange Rate

Results from Step One


Co-integration Equation 𝑜𝑜�𝑡𝑡 = 3.368 + 2.688et
Standard Error (0.035) (0.137)
t-Statistic (96.688) (19.675)
Probability (0.000) (0.000)
Results from Step Two
Variables Test Statistics Probability Critical Value (10%;5%;1%) Test Form Conclusion
εog -3.062 0.000 -1.615;-1.944;-2.586 0,0,1 Reject H0
Note: εog denotes the residual variable from the co-integration equation. Test form is expressed by C, T
and K, where C denotes constant term, T denotes trend term and K denotes number of lags. For C and T, 0
means the term is not included. For K, 0 means no autoregressive terms.

The Table 5.13 shows that in the subsample period, crude oil price and dollar exchange rate are still
co-integrated, but the co-integration parameters are different. The constant term is slightly smaller and the
slope term is larger in the co-integration relationship established based on the subsample.

Table 5.14 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Dollar Exchange Rate

Sample Period Observations τ Bounds Possible Breakdates No. of Breakdates


2001.6-2010.4 107 [0.15, 0.85] [17, 90] 2002.10-2008.11 74
2001.6-2010.4 107 [0.1, 0.9] [11, 96] 2002.4-2009.5 86
2001.6-2010.4 107 [0.05, 0.95] [6, 101] 2001.11-2009.10 96
Co-integration Equation 𝑜𝑜�𝑡𝑡 = 3.368 + 2.688 et
SupF 73.602(10.6;12.4;16.2)
MeanF 4.948 (3.73;4.57;6.78)
Note: The numbers in parentheses are critical values at 10% level, 5% level and 1% level.

Table 5.14 presents three scenarios of possible breakdate range in the upper part of the table. When τ is
set to be [0.15, 0.85] or [0.1, 0.9], the possible breakdate range does not include November 2001, the
candidate break date estimator. In the third scenario, τ is [0.05, 0.95]. The resulting possible breakdate range
is November 2001 – October 2009, exactly including November 2001. The results of parameter stability tests
in the lower part of Table 5.14 are calculated using the τ in the third scenario. The results indicate that the
SupF statistic is well above the critical values at 1% level and the MeanF statistic is above the critical values
at 5% level. So the co-integration relationship between crude oil price and dollar exchange rate are
considered instable during June 2001- April 2010. Figure 5.18 below provides the sequence of F values
obtained from the subsample regressions. The highest F value corresponds to the date of September 2009.
The F value of November 2001, the candidate breakdate estimator, is 10.570, slightly lower than 10.6, the
74

critical value at 10% level. The F value of February 2004 is 10.140, a local maximum clearly smaller than
10.6. The inference based on Table 5.14 and Figure 5.18 is that the co-integration relationship between crude
oil price and dollar exchange rate in June 2001- April 2010 is instable with a clear structural break at
September 2009. Why would this date be another structural break date estimator? Recall Figure 5.15, soon
after September 2009, dollar exchange rate began to appreciate. During the same period, crude oil price also
showed a generally upward movement trend. A relevant reason for the movement crude oil price in this
period is that (at least part of) investors optimistically believed that the worst time resulted from the financial
crisis had passed. The global economy, including U.S. economy, had started to recover. So crude oil price
slowly increased. An important reason for the appreciation of dollar exchange rate was the European
sovereign debt crisis, which germinated at least as early as November 2009 24. Worries about the financial
status of Greek as well as other European countries and speculations together forcefully drove down euro
against dollar. Gold price also increased in this same period.

80
70
60
50
40
30
20
10
0
Apr-2002

May-2004

Apr-2007

May-2009
Nov-2001

Aug-2005

Nov-2006
Sep-2002
Feb-2003
Jul-2003

Sep-2007
Feb-2008
Jul-2008
Dec-2003

Dec-2008
Oct-2004
Mar-2005

Jan-2006
Jun-2006

Oct-2009
F Statistic 10% 5% 1%

Figure 5.18 F Statistic Sequence from Subsample Regression of Crude Oil Price on Dollar Exchange Rate

Since 10.570 becomes 10.6 when only one digit is maintained after the decimal point, to be prudent,
November 2001 is considered a local maximum with the F value reaches the critical value at 10% level. So a
new round of parameter stability test has to be performed on the relationship between crude oil price and
dollar exchange rate. The subsample this time is from November 2001 to September 2009. Table 5.15 and
Table 5.16 below present results of co-integration test and parameter stability tests on the above subsample.

24. For detailed information, see, for example, the explanation of term “2010 European sovereign debt crisis
timeline” from Wikipedia.
75

Table 5.15 Results of Co-integration Test on Subsample of Crude Oil Price and Dollar Exchange Rate

Results from Step One


Co-integration Equation 𝑜𝑜�𝑡𝑡 = 3.365 + 2.678et
Standard Error (0.036) (0.144)
t-Statistic (94.656) (18.597)
Probability (0.000) (0.000)
Results from Step Two
Variables Test Statistics Probability Critical Value (10%;5%;1%) Test Form Conclusion
εog -3.016 0.003 -1.615;-1.944;-2.586 0,0,1 Reject H0
Note: εog denotes the residual variable from the co-integration equation. Test form is expressed by C, T
and K, where C denotes constant term, T denotes trend term and K denotes number of lags. For C and T, 0
means the term is not included. For K, 0 means no autoregressive terms.

Table 5.16 Results of Parameter Stability Tests on Subsample of Crude Oil Price and Dollar Exchange Rate

Sample Period Observations τ Bounds Possible Breakdates No. of Breakdates


2001.6-2009.9 100 [0.05, 0.95] [6, 95] 2001.11-2009.4 90
Co-integration Equation 𝑜𝑜�𝑡𝑡 = 3.365 + 2.678 et
SupF 26.030(10.6;12.4;16.2)
MeanF 5.165 (3.73;4.57;6.78)
Note: The numbers in parentheses are critical values at 10% level, 5% level and 1% level.

The Table 5.15 above shows that in the subsample period of June 2001 – September 2009, crude oil
price and dollar exchange rate are still co-integrated. The co-integration coefficients are similar to those
obtained from subsample June 2001-April 2010.
Table 5.16 presents parameter stability test results. The results are calculated using the τ of [0.05, 0.95].
The reason is still to include the date of November 2001. The results indicate that the SupF statistic is well
above the critical values at 1% level and the MeanF statistic is above the critical values at 5% level. So the
co-integration relationship between crude oil price and dollar exchange rate are considered instable during
June 2001- September 2009. Figure 5.19 below provides the sequence of F values obtained from the
subsample regressions. The highest F value corresponds to the date of June 2008. The F value of November
2001, the candidate breakdate estimator, is 9.122, clearly smaller than 10.6, the critical value at 10% level.
The F value of January 2004 is 9.664, a local maximum clearly smaller than 10.6. Another local maximum is
at the date of January 2009. The F value is 22.349, also higher than the critical value at 1% level. The
inference based on Table 5.16 and Figure 5.19 is that the co-integration relationship between crude oil price
and dollar exchange rate in June 2001- September 2009 is instable with a clear structural break at June 2008.
November 2001 is confirmed not to be a structural break date estimator but instead January 2009 is a new
76

candidate breakdate estimator. The reason why June 2008 is another structural break date estimator is
relatively clear. As discussed earlier in the analysis of the relationship between crude oil price and gold price,
crude oil price experience a sharp and abrupt decline right after June 2008. The world economy cannot
sustain at a price level of $145 or higher. Economic slowdown and crude oil demand shrink resulted in price
dive. For dollar exchange rate, it appreciated quickly after June 2008. Several reasons may account for this.
Firstly, to some extent, dollar is a relatively safe asset. Although the age of Gold Standard has become history,
dollar exchange rate is backed up by U.S. economy, the largest and most influential economy in the world.
Investors believe and are willing to believe that U.S. economy is more vigorous with great potential than
other economies even in the worst time. Therefore, investors would like to possess or convert into dollar
assets when elsewhere in the financial markets is insecure. 25 Moreover, many U.S. companies are
multinational companies who have oversea subsidiaries or financial companies who have huge financial
investment in other countries. As the financial crisis deteriorated, more and more capitals and investment
were repatriated by oversea subsidiaries or withdrawn from foreign countries to save or support U.S. parent
companies. Such capital inflows were likely to push up dollar exchange rate. Besides, the appreciation of
dollar caused by capital repatriation may further enhance the judgment of global investors that dollar is
relatively safe and thus result in larger and larger dollar demand. Figure 5.20 provides U.S. total claims on
foreigners during 2003-2009. In 2008, the total claims kept declining after the first quarter, which means that
U.S. institutions or individuals reduced their positions of foreign assets and increased their positions of dollar
assets.

30
25
20
15
10
5
0
Mar-2005
Oct-2004

Jan-2006
Jun-2006
Apr-2002

May-2004

Apr-2007
Nov-2001

Aug-2005
Sep-2002

Nov-2006
Feb-2003
Jul-2003

Sep-2007
Feb-2008
Jul-2008
Dec-2003

Dec-2008

F Statistic 10% 5% 1%

Figure 5.19 F Statistic Sequence from Subsample Regression of Crude Oil Price on Dollar Exchange Rate

25. For example, Kaul and Sapp (2006) study the bid-ask spread in the Euro - U.S. dollar spot and forward
markets around year 2000 and conclude that dollar is a safe haven currency during periods of uncertainty.
77

4,500,000
4,000,000
3,500,000
3,000,000
2,500,000
2,000,000
1,500,000
1,000,000
500,000
0
2003-03

2003-09

2004-03

2004-09

2005-03

2005-09

2006-03

2006-09

2007-03

2007-09

2008-03

2008-09

2009-03
Data source: Department of The Treasury, Treasury International Capital Reporting System.
Figure 5.20 U.S. Total Banking Claims on Foreigners 2003-2009

According to Figure 5.19, since the date of January 2009 corresponds to an F value that is higher than
the critical value at 1% level, it is a candidate break date estimator. It is reasonable to continue parameter
stability tests on the subsample of July 2008 – September 2009. But this subsample has only 15 observations,
too limited to further perform parameter stability test. So the parameter stability test on the relationship
between dollar exchange rate and crude oil price stops here. However, the movement of dollar exchange rate
and crude oil price around January 2009 is still worth discussion. Figure 5.21 below shows the historical
movements of the two prices during 2008-2009. Clearly, from mid-December 2008 to the beginning of
March 2009, the dollar exchange rate appreciated from around 1.44 to 1.25. The crude oil price increased
from around $30 to well above $40. The two prices again moved in the same direction.

1.7 160
140
1.6
120
1.5 100
80
1.4 60
40
1.3
20
1.2 0
2008/1/2

2008/3/2

2008/5/2

2008/7/2

2008/9/2

2009/1/2

2009/3/2

2009/5/2

2009/7/2

2009/9/2
2008/11/2

2009/11/2

dollar exchange rate (Left Axis) crude oil price (Right Axis)

Figure 5.21 Historical Movements of Dollar Exchange Rate and Crude Oil Price during 2008-2009
78

There are several reasons that may account for the dollar appreciation in this period. The first one is the
frozen interbank borrowing and lending market. A majority of international transactions are completed using
the dollar as the settlement currency. As the financial crisis aggravated, more and more financial institutions
reported or were revealed to have severe asset loss and liquidity deficiency. Commercial banks and other
financial institutions were reluctant to provide short term financing support for each other. As a result, both
domestic and international interbank borrowing/lending markets became frozen. Commercial banks and
other financial institutions in many countries gradually ran out of dollars to finish transactions. Such upsurge
of dollar demand for transactions caused the dollar to appreciate.
The second possible reason is the temporary reciprocal currency arrangements (swap lines) between the
Federal Reserve Bank (FED) and other major central banks. In response to the dollar funding deficiency of
domestic financial institutions, the central banks in major developed countries established temporary
reciprocal currency arrangements (swap lines) with the FED. As stated by FED (2008), “These facilities, like
those already established with other central banks, are designed to help improve liquidity conditions in global
financial markets and to mitigate the spread of difficulties in obtaining U.S. dollar funding in fundamentally
sound and well managed economies.” Table 5.17 below provides the timeline of temporary reciprocal
currency arrangements. There were several arrangements with large amounts of dollars due on Jan.30, 2009,
which means the corresponding central banks had to repurchase dollars in the foreign exchange market to pay
back the FED. Table 5.18 provides the U.S. government assets other than official reserve assets during
2008-2009. As explained by the Bureau of Economic Analysis (2009), U.S. government assets other than
official reserve assets decreased $244.1 billion in the first quarter, following an increase of $265.3 billion in
the fourth. The shift resulted from a shift from net drawings to net repayments on temporary reciprocal
currency arrangements between the U.S. Federal Reserve System and foreign central banks. Therefore, the
dollar exchange rate was pushed up. In fact, the dollar appreciation in September 2008 and April 2009 and the
depreciation in September 2009 can all be associated with the repayments of temporary reciprocal currency
arrangements. In addition, from December 2008 to the end of February 2009, gold price increased from
around $750 per troy ounce to around $990 per troy ounce. The quick and strong increase of gold price may
also contribute to the demand for the dollar and thus push up dollar exchange rate.
For crude oil price, as more and more governments declared economy rescue plans, it touched the
bottom in December 2008 and began to rebound. But since the economic prospect was still gloomy in that
period, the crude oil price increased very slowly.
In sum, when more data is available, it is possible that January 2009 or other dates around it may
provide an F value higher enough to indicate parameter instability.
79

Table 5.17 Temporary Reciprocal Currency Arrangements Timeline

From To Amount (billion) Start End


European Central Bank $20 Dec.12, 2007 Six Months
FED
Swiss National Bank $4 Dec.12, 2007 Six Months
European Central Bank $30(+10) Mar.11, 2008 Sep.30,2008
FED
Swiss National Bank $6(+2) Mar.11, 2008 Sep.30,2008
European Central Bank $50(+20) May.2, 2008 Jan.30, 2009
FED
Swiss National Bank $12(+6) May.2, 2008 Jan.30, 2009
FED European Central Bank $55(+5) Jul.30, 2008 Jan.30, 2009
European Central Bank $110(+55) Sep.18, 2008 Jan.30, 2009
Swiss National Bank $27(+15) Sep.18, 2008 Jan.30, 2009
FED Bank of Japan $60 Sep.18, 2008 Jan.30, 2009
Bank of England $40 Sep.18, 2008 Jan.30, 2009
Bank of Canada $10 Sep.18, 2008 Jan.30, 2009
European Central Bank $120(+10) Sep.26, 2008 Jan.30, 2009
FED
Swiss National Bank $30(+3) Sep.26, 2008 Jan.30, 2009
Reserve Bank of Australia $10 Sep.24, 2008 Jan.30, 2009
Sveriges Riksbank $10 Sep.24, 2008 Jan.30, 2009
FED
Danmarks National Bank $5 Sep.24, 2008 Jan.30, 2009
Norges Bank $5 Sep.24, 2008 Jan.30, 2009
European Central Bank Unlimited Oct.13, 2008 Apr.30, 2009
FED Swiss National Bank Unlimited Oct.13, 2008 Apr.30, 2009
Bank of England Unlimited Oct.13, 2008 Apr.30, 2009
FED Bank of Japan Unlimited Oct.14, 2008 Apr.30, 2009
FED Reserve Bank of New Zealand $15 Oct.28, 2008 Apr.30, 2009
Banco Central do Brasil $30 Oct.29, 2008 Apr.30, 2009
Banco de Mexico $30 Oct.29, 2008 Apr.30, 2009
FED
Bank of Korea $30 Oct.29, 2008 Apr.30, 2009
Monetary Authority of Singapore $30 Oct.29, 2008 Apr.30, 2009
Bank of England £30 Apr.6, 2009 Oct.30, 2009
European Central Bank €80 Apr.6, 2009 Oct.30, 2009
FED
Bank of Japan ¥10000 Apr.6, 2009 Oct.30, 2009
Swiss National Bank CHF 40 Apr.6, 2009 Oct.30, 2009
Note: The numbers in amount column are the total amount of temporary reciprocal currency
arrangements at the start dates. The numbers in the parentheses are the new additions at the start dates, which
are also the press release dates. Data Source: Author’s own compilation based on various monetary policy
press releases of the Federal Reserve Bank in 2007-2009.
80

Table 5.18 U.S. Government Assets Other Than Official Reserve Assets during 2008-2009

2008:I 2008:II 2008:III 2008:IV 2009:I 2009:II 2009:III 2009:IV


U.S. government assets,
other than official reserve 3268 -41592 -225997 -265293 244102 193750 57736 45754
assets
Note: “+” means credits or more explicitly capital inflow; “-” means debits or capital outflow. Numbers
are in millions of dollars, seasonally adjusted. Data Source: U.S. International Transactions, Bureau of
Economic Analysis.

To sum up, three structural breaks are identified in the relationship between dollar exchange rate and
crude oil price over the sample period of January 1999 – April 2010. The estimated break dates are May 2001,
June 2008 and September 2009. These three break dates divide the whole sample into four subsamples. Note
that the three estimated break dates are of two kinds. May 2001 and September 2009 are in the same category
because the direction of correlation prior and post these two break date estimators are different. For example,
before May 2001, the dollar exchange rate and crude oil price are generally positively correlated. Dollar
exchange rate appreciation (depreciation) concurred with crude oil price increase (decrease). But after May
2001, the two prices became negatively correlated. Dollar exchange rate appreciation (depreciation)
coincided with crude oil price decrease (increase). June 2008 belongs to a second category because the
direction of correlation did not change before and after June 2008 but both prices experienced a big jump
after June 2008. The two categories above are an example of different kinds of structural breaks.
Something worth noting is the estimation accuracy. Section 5.3.3 and section 5.3.4 perform parameter
stability test and detect structural breaks in the relationship between crude oil price and gold price and
between crude oil price and dollar exchange rate. While the observations of the full sample are relatively
sufficient, the observations in subsamples are limited. Clearly, as a sample becomes shorter and shorter, the
estimation accuracy is likely to be lower and lower. The entire section of 5.3.4, particularly the analysis of the
relationship between crude oil price and dollar exchange rate, may be subject to this problem.

5.3.5. Granger Non-Causality (GNC) Test


This section provides and interprets the results of Granger Non-Causality Test. Since the co-integration
relationship between dollar exchange rate and crude oil price is stable, the GNC Test is performed on the
entire sample. The co-integration relationship between crude oil price and gold price has a structural break in
June 2008, so the GNC Test is performed on the two subsamples as well as the entire sample to make
comparisons. The co-integration relationship between crude oil price and dollar exchange rate has three
structural breaks, so the GNC Test is performed on the four subsamples as well as the entire sample to make
comparisons.
81

As introduced in Chapter III, in this study the GNC analysis is conducted in the error correction
framework. An important issue to the estimation of error correction models is the choice of proper lag length.
This study uses the AIC criterion to decide the suitable lag length. Table 5.19 below provides the AIC values
obtained from regressions with different lag lengths.

Table 5.19 AIC Values and Lag Length k Selection

e and g o and g
Sample
1999.1-2010.4 1999.1-2008.6 2008.7-2010.4 1999.1-2010.4
k
1 -8.348 -5.922 -4.246 -5.537
2 -8.326 -5.904 -4.605 -5.575
3 -8.284 -5.832 -4.301 -5.547
4 -8.239 -5.775 -4.399 -5.522
5 -8.291 -5.783 -4.301 -5.491
6 -8.249 -5.782 -4.212 -5.446
7 -8.235 -5.756 -4.537 -5.409
8 -8.183 -5.698 - -5.377
9 -8.241 -5.841 - -5.455
10 -8.206 -5.841 - -5.438
optimal k 1 1 2 2
o and e
Sample
1999.1-2001.5 2001.6-2008.6 2008.7-2009.9 2008.7-2010.4 1999.1-2010.4
k
1 -6.591 -7.315 -4.583 -5.154 -6.684
2 -6.492 -7.313 -5.104 -5.492 -6.652
3 -6.241 -7.241 -4.698 -5.225 -6.589
4 -6.025 -7.386 -4.748 -5.040 -6.546
5 -5.969 -7.366 - -4.961 -6.508
6 -6.528 -7.278 - -5.089 -6.483
7 - -7.290 - - -6.428
8 - -7.316 - - -6.483
9 - -7.245 - - -6.482
10 - -7.294 - - -6.456
optimal k 1 4 2 2 1
82

According to Table 5.19, the lag length for the analysis of dollar exchange rate and gold price is one.
The lag length choices for the analysis of crude oil price and gold price are one for the first subsample and
two for the second subsample and the entire sample. The lag length choice for the analysis of crude oil price
and dollar exchange rate ranges from one to four. Note that the fourth subsample is supposed to be October
2009 – April 2010. With only seven observations, this subsample is too small to conduct error correction
analysis. So the author combines the third and the fourth sample as the new fourth sample which is from July
2008 to April 2010 with 22 observations in total.
With the choices of k, the error correction models presented in section 3.3.5 can be further specified.
The tables below exhibit the regression equations, restrictions (null hypotheses) and results of the GNC Test.

Table 5.20 Results of GNC Test on Dollar Exchange Rate and Gold Price

Unrestricted Model ∆et = μ1 + α11 ∆𝑒𝑒𝑡𝑡−1 + α12 ∆𝑔𝑔𝑡𝑡−1 + γ1 ( et-1 - β0 - β1gt-1 ) + ε1t
Estimation Result ∆𝑒𝑒�𝑡𝑡 = 0.001 + 0.365∆𝑒𝑒𝑡𝑡−1 - 0.033∆𝑔𝑔𝑡𝑡−1 - 0.053 (et-1 + 1.831 - 0.323gt-1 )
t-ratio 0.595 3.963 -0.585 -1.975 -20.63 22.42
p-value 0.553 0.000 0.559 0.050 0.000 0.000
Restrictions (H0s) Short Run Causality: α12 =0 Long Run Causality: γ1 = 0
LR Statistic p-value LR Statistic p-value
Likelihood Ratio Test 0.353 0.553 3.960 0.047
Do not reject H0 Reject H0
Unrestricted Model ∆gt = μ2 + α21 ∆𝑒𝑒𝑡𝑡−1 + α22 ∆𝑔𝑔𝑡𝑡−1 + γ2 ( et-1 - β0 - β1gt-1 ) + ε2t
Estimation Result �𝑡𝑡 = 0.010 + 0.030∆𝑒𝑒𝑡𝑡−1 + 0.016∆𝑔𝑔𝑡𝑡−1 - 0.017(et-1 + 1.831 - 0.323gt-1 )
∆𝑔𝑔
t-ratio 2.725 0.183 0.159 -0.350 -20.63 22.42
p-value 0.007 0.855 0.874 0.727 0.000 0.000
Restrictions (H0s) Short Run Causality: α21 =0 Long Run Causality: γ2 = 0
LR Statistic p-value LR Statistic p-value
Likelihood Ratio Test 0.034 0.853 0.126 0.723
Do not reject H0 Do not reject H0

Table 5.20 indicates that


(1) In the short run, the movement of gold price does not cause the movement of dollar exchange rate in
the Granger sense. In the long run, the movement of gold price is the Granger cause of the movement of
dollar exchange rate.
83

(2) In both short run and long run, the movement of dollar exchange rate is not the Granger cause of the
movement of gold price
(3) The gold price is weakly exogenous to the bivariate error correction system of dollar exchange rate
and gold price.
(4) In the long run, gold price increase will cause dollar depreciation. A 10% rise of gold price will cause
a 0.1624% (= -0.0527 × (-0.3234) × (ln1.1gt-1 - lngt-1)) depreciation of dollar exchange rate.

Table 5.21 Results of GNC Test on Crude Oil Price and Gold Price (Subsample One)

Unrestricted Model ∆ot = μ3 + α31 ∆𝑜𝑜𝑡𝑡−1 + α32 ∆𝑔𝑔𝑡𝑡−1 + γ3 ( ot-1 - δ0 - δ1gt-1 ) + ε3t

Estimation Result ∆𝑜𝑜�𝑡𝑡 = 0.020 + 0.102∆𝑜𝑜𝑡𝑡−1 - 0.032∆𝑔𝑔𝑡𝑡−1 - 0.147 (ot-1 + 3.873 - 1.257gt-1 )

t-ratio 2.802 1.126 -0.183 -4.010 -12.82 25.00

p-value 0.006 0.263 0.856 0.000 0.000 0.000

Restrictions (H0s) Short Run Causality: α32 =0 Long Run Causality: γ3 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 0.035 0.853 15.547 0.000

Do not reject H0 Reject H0

Unrestricted Model ∆gt = μ4 + α41 ∆𝑜𝑜𝑡𝑡−1 + α42 ∆𝑔𝑔𝑡𝑡−1 + γ4 (ot-1 - δ0 - δ1gt-1 ) + ε4t

Estimation Result �𝑡𝑡 = 0.009+ 0.025∆𝑜𝑜𝑡𝑡−1 + 0.015∆𝑔𝑔𝑡𝑡−1 + 0.021 (ot-1 + 3.873 - 1.257gt-1 )
∆𝑔𝑔

t-ratio 2.323 0.496 0.157 1.045 -12.82 25.00

p-value 0.022 0.621 0.876 0.298 0.000 0.000

Restrictions (H0s) Short Run Causality: α41 =0 Long Run Causality: γ4 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 0.255 0.613 1.128 0.288

Do not reject H0 Do not reject H0

Table 5.21 shows the results obtained from subsample period January 1999 - June 2008. The results
indicate that
(1) In the short run, the movement of gold price does not cause the movement of crude oil price in the
Granger sense. In the long run, the movement of gold price is the Granger cause of the movement of crude oil
price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of gold price
84

(3) The gold price is weakly exogenous to the bivariate error correction system of crude oil price and
gold price.
(4) In the long run, gold price increase will cause crude oil price increase. A 10% rise of gold price will
cause a 1.7651% (= -0.1473 × (-1.2573) × (ln1.1gt-1 - lngt-1)) increase of crude oil price.

Table 5.22 Results of GNC Test on Crude Oil Price and Gold Price (Subsample Two)

Unrestricted Model ∆ot = μ5 + ∑2i=1 α51𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α52𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ5 ( ot-1 - θ0 -θ1gt-1 ) + ε5t

∆𝑜𝑜�𝑡𝑡 = - 0.024 + 0.224∆𝑜𝑜𝑡𝑡−1 + 0.565∆𝑜𝑜𝑡𝑡−2 + 0.639∆𝑔𝑔𝑡𝑡−1 - 0.660∆𝑔𝑔𝑡𝑡−2


Estimation Result
- 0.415 (ot-1 + 0.125 - 0.635gt-1 )

t-ratio -1.254 1.530 4.135 1.351 -1.485 -4.585 -0.033 1.142

p-value 0.232 0.150 0.001 0.200 0.161 0.001 0.974 0.267

Restrictions (H0s) Short Run Causality: α521 = α522 = 0 Long Run Causality: γ5 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 5.022 0.025 18.280 0.000

Reject H0 Reject H0

Unrestricted Model ∆gt = μ6 + ∑2i=1 α61𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α62𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ6 (ot-1 -θ0 -θ1gt-1 ) + ε6t

�𝑡𝑡 = 0.012 + 0.055∆𝑜𝑜𝑡𝑡−1 - 0.087∆𝑜𝑜𝑡𝑡−2 + 0.110∆𝑔𝑔𝑡𝑡−1 - 0.174∆𝑔𝑔𝑡𝑡−2


∆𝑔𝑔
Estimation Result
- 0.048 (ot-1 + 0.125 - 0.635gt-1 )

t-ratio 1.072 0.652 -1.111 0.406 -0.685 -0.927 -0.033 1.142

p-value 0.303 0.526 0.287 0.691 0.505 0.370 0.974 0.267

Restrictions (H0s) Short Run Causality: α611 = α612 = 0 Long Run Causality: γ6 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 1.765 0.184 1.218 0.270

Do not reject H0 Do not reject H0

Table 5.22 shows the results obtained from subsample period July 2008- April 2010. The results
indicate that
(1) In both short run and long run, the movement of gold price is the Granger cause of the movement of
crude oil price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of gold price
85

(3) The gold price is weakly exogenous to the bivariate error correction system of crude oil price and
gold price.
(4) In the long run, gold price increase will cause crude oil price increase. A 10% rise of gold price will
cause a 2.5141% (= -0.4154 × (-0.6350) × (ln1.1gt-1 - lngt-1)) increase of crude oil price.

Table 5.23 Results of GNC Test on Crude Oil Price and Gold Price (Full Sample)

Unrestricted Model ∆ot = μ7 + ∑2i=1 α71𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α72𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ7 ( ot-1 - η0 -η1gt-1 ) + ε7t

∆𝑜𝑜�𝑡𝑡 = 0.0085 + 0.2483∆𝑜𝑜𝑡𝑡−1 + 0.1600∆𝑜𝑜𝑡𝑡−2 + 0.1469∆𝑔𝑔𝑡𝑡−1 - 0.1160∆𝑔𝑔𝑡𝑡−2


Estimation Result
- 0.1086 (ot-1 + 2.2206 - 0.9743gt-1 )

t-ratio 1.090 2.978 1.908 0.812 -0.640 -3.549 -7.519 20.300

p-value 0.278 0.004 0.059 0.419 0.523 0.001 0.000 0.000

Restrictions (H0s) Short Run Causality: α721 = α722 = 0 Long Run Causality: γ7 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 1.117 0.291 12.577 0.000

Do not reject H0 Reject H0

Unrestricted Model ∆gt = μ8 + ∑2i=1 α81𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α82𝑖𝑖 ∆𝑔𝑔𝑡𝑡−𝑖𝑖 + γ8 (ot-1 -η0 -η1gt-1 ) + ε8t

�𝑡𝑡 = 0.0119 + 0.0457∆𝑜𝑜𝑡𝑡−1 - 0.0888∆𝑜𝑜𝑡𝑡−2 + 0.0215∆𝑔𝑔𝑡𝑡−1 - 0.1151∆𝑔𝑔𝑡𝑡−2


∆𝑔𝑔
Estimation Result
+ 0.0028 (ot-1 + 2.2206 - 0.9743gt-1 )

t-ratio 3.134 1.130 -2.183 0.245 -1.310 0.186 -7.519 20.300

p-value 0.002 0.261 0.031 0.807 0.193 0.852 0.000 0.000

Restrictions (H0s) Short Run Causality: α811 = α812 = 0 Long Run Causality: γ8 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 5.285 0.022 0.036 0.849

Reject H0 Do not reject H0

Table 5.23 shows the results obtained from the full sample for the purpose of comparison with the
results obtained from the two subsamples. The results in Table 5.16 indicate that
(1) In the short run, the movement of gold price does not cause the movement of crude oil price in the
Granger sense. In the long run, the movement of gold price is the Granger cause of the movement of crude oil
price.
(2) In the short run, the movement of crude oil price is the Granger sense of the movement of gold price.
In the long run, the movement of crude oil price is not the Granger cause of the movement of gold price.
86

(3) The gold price is weakly exogenous to the bivariate error correction system of crude oil price and
gold price.
(4) In the long run, gold price increase will cause crude oil price increase. A 10% rise of gold price will
cause a 1.0085% (= -0.1086 × (-0.9743) × (ln1.1gt-1 - lngt-1)) increase of crude oil price.

Table 5.24 Results of GNC Test on Subsample One of Dollar Exchange Rate and Crude Oil Price

Unrestricted Model ∆ot = μ9 + α91 ∆𝑜𝑜𝑡𝑡−1 + α92 ∆𝑒𝑒𝑡𝑡−1 + γ9 ( ot-1 - λ0 - λ1et-1 ) + ε9t

Estimation Result ∆𝑜𝑜�𝑡𝑡 = 0.032 - 0.050∆𝑜𝑜𝑡𝑡−1 - 0.477∆𝑒𝑒𝑡𝑡−1 - 0.204 (ot-1 - 3.130 + 2.893et-1 )

t-ratio 1.741 -0.246 -0.680 -1.618 110.758 -9.074

p-value 0.095 0.808 0.504 0.119 0.000 0.000

Restrictions (H0s) Short Run Causality: α92 =0 Long Run Causality: γ9 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 0.537 0.464 2.912 0.088

Do not reject H0 Reject H0

Unrestricted Model ∆et = μ10 + α101 ∆𝑜𝑜𝑡𝑡−1 + α102 ∆𝑒𝑒𝑡𝑡−1 + γ10 ( ot-1 - λ0 - λ1et-1 ) + ε10t

Estimation Result ∆𝑒𝑒�𝑡𝑡 = -0.008 + 0.074∆𝑜𝑜𝑡𝑡−1 + 0.307∆𝑒𝑒𝑡𝑡−1 - 0.049(ot-1 - 3.130 + 2.893et-1 )

t-ratio -1.428 1.225 1.466 -1.294 110.758 -9.074

p-value 0.167 0.233 0.156 0.208 0.000 0.000

Restrictions (H0s) Short Run Causality: α101 =0 Long Run Causality: γ10 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 1.708 0.191 1.899 0.168

Do not reject H0 Do not reject H0

Table 5.24 indicates that for the first subsample (1999.1-2001.5)


(1) In the short run, the movement of dollar exchange rate does not cause the movement of crude oil
price in the Granger sense. In the long run, the movement of dollar exchange rate is the Granger cause of the
movement of crude oil price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of dollar exchange rate.
(3) In the long run, dollar depreciation will cause crude oil price decrease. A 10% decrease of dollar
exchange rate will cause a 0.590% (= -0.204 × (2.893) × (ln1.1et-1 - lnet-1)) decrease of crude oil price.
87

Table 5.25 Results of GNC Test on Subsample Two of Dollar Exchange Rate and Crude Oil Price

Unrestricted Model ∆ot = μ11 + ∑4i=1 α111 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑4i=1 α112 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ11 ( ot-1 - π0 - π1et-1 ) + ε11t

∆𝑜𝑜�𝑡𝑡 = 0.028 + 0.165∆𝑜𝑜𝑡𝑡−1 - 0.066∆𝑜𝑜𝑡𝑡−2 - 0.170∆𝑜𝑜𝑡𝑡−3 + 0.058∆𝑜𝑜𝑡𝑡−4 -


Estimation Result 0.116∆𝑒𝑒𝑡𝑡−1 - 0.514∆𝑒𝑒𝑡𝑡−2 + 0.461∆𝑒𝑒𝑡𝑡−3 - 0.859∆𝑒𝑒𝑡𝑡−4
- 0.050 (ot-1 - 3.375 – 2.681et-1 )

2.798 1.420 -0.556 -1.439 0.497 -0.264 -1.110 1.017


t-ratio
-1.937 -1.208 96.75 17.65

0.007 0.160 0.580 0.155 0.621 0.793 0.271 0.313


p-value
0.057 0.231 0.000 0.000

Restrictions (H0s) Short Run Causality: α1121 = α1122 = 0 Long Run Causality: γ11 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 4.963 0.026 1.650 0.199

Reject H0 Do not reject H0

Unrestricted Model ∆et = μ12 + ∑4i=1 α121 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑4i=1 α122 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ12 (ot-1 - π0 - π1et-1 ) + ε12t

∆𝑒𝑒�𝑡𝑡 = 0.004 - 0.032∆𝑜𝑜𝑡𝑡−1 + 0.052∆𝑜𝑜𝑡𝑡−2 - 0.032∆𝑜𝑜𝑡𝑡−3 + 0.084∆𝑜𝑜𝑡𝑡−4 +


Estimation Result 0.389∆𝑒𝑒𝑡𝑡−1 - 0.247∆𝑒𝑒𝑡𝑡−2 - 0.070∆𝑒𝑒𝑡𝑡−3 + 0.161∆𝑒𝑒𝑡𝑡−4
+ 0.008 (ot-1 + 0.1254 - 0.6350et-1 )

1.669 -1.069 1.696 -1.050 2.781 3.432 -2.059 -0.597


t-ratio
1.404 0.729 96.75 17.65

0.100 0.289 0.094 0.297 0.007 0.001 0.043 0.552


p-value
0.165 0.468 0.000 0.000

Restrictions (H0s) Short Run Causality: α1211 = α1212 = 0 Long Run Causality: γ12 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 11.961 0.000 0.606 0.436

Reject H0 Do not reject H0

Table 5.25 shows the results obtained from subsample period June 2001- June 2008. The results indicate
that
(1) In the short run, the movement of dollar exchange rate is the Granger cause of the movement of
crude oil price. In the long run, however, the movement of dollar exchange rate is not the Granger cause of
the movement of crude oil price.
(2) In the short run, the movement of crude oil price is the Granger cause of the movement of dollar
exchange rate. In the long run, however, the movement of crude oil price is not the Granger cause of the
movement of dollar exchange rate.
88

Table 5.26 Results of GNC Test on Subsample Three of Dollar Exchange Rate and Crude Oil Price

Unrestricted Model ∆ot = μ13 + ∑2i=1 α131 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α132 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ13 ( ot-1 - φ0 - φ1et-1 ) + ε13t

∆𝑜𝑜�𝑡𝑡 = - 0.008 + 0.282∆𝑜𝑜𝑡𝑡−1 + 0.633∆𝑜𝑜𝑡𝑡−2 – 1.766∆𝑒𝑒𝑡𝑡−1 – 1.939∆𝑒𝑒𝑡𝑡−2


Estimation Result
- 0.998 (ot-1 - 2.486 – 5.224et-1 )

t-ratio -0.260 1.260 3.178 -1.309 -1.585 -3.612 8.120 5.594

p-value 0.803 0.254 0.019 0.238 0.164 0.011 0.000 0.000

Restrictions (H0s) Short Run Causality: α1321 = α1322 = 0 Long Run Causality: γ13 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 4.740 0.029 13.862 0.000

Reject H0 Reject H0

Unrestricted Model ∆et = μ14 + ∑2i=1 α141 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α142 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ14 (ot-1 - φ0 - φ1et-1 ) + ε14t

∆𝑒𝑒�𝑡𝑡 = 0.004 + 0.033∆𝑜𝑜𝑡𝑡−1 + 0.062∆𝑜𝑜𝑡𝑡−2 + 0.167∆𝑒𝑒𝑡𝑡−1 - 0.290∆𝑒𝑒𝑡𝑡−2


Estimation Result
- 0.023 (ot-1 - 2.486 – 5.224et-1 )

t-ratio 0.315 0.306 0.646 0.257 -0.491 -0.169 8.120 5.594

p-value 0.763 0.770 0.542 0.806 0.641 0.871 0.000 0.000

Restrictions (H0s) Short Run Causality: α1411 = α1412 = 0 Long Run Causality: γ14 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 1.123 0.289 0.057 0.811

Do not reject H0 Do not reject H0

Table 5.26 shows the results obtained from subsample period July 2008- September 2009. The results
indicate that
(1) In both short run and long run, the movement of dollar exchange rate is the Granger cause of the
movement of crude oil price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of dollar exchange rate.
(3) In the long run, dollar depreciation will cause crude oil price increase. A 10% depreciation of dollar
will cause a 5.214% (= -0.998 × (-5.224) × (ln1.1et-1 - lnet-1)) increase of crude oil price.
89

Table 5.27 Results of GNC Test on Subsample Four of Dollar Exchange Rate and Crude Oil Price

Unrestricted Model ∆ot = μ15 + ∑2i=1 α151 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α152 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ15 ( ot-1 - ω0 - ω1et-1 ) + ε15t

∆𝑜𝑜�𝑡𝑡 = - 0.028 + 0.335∆𝑜𝑜𝑡𝑡−1 + 0.559∆𝑜𝑜𝑡𝑡−2 – 0.687∆𝑒𝑒𝑡𝑡−1 – 1.679∆𝑒𝑒𝑡𝑡−2


Estimation Result
- 0.638 (ot-1 - 2.812 – 4.294et-1 )

t-ratio -1.025 1.572 2.967 -0.622 -1.486 -2.954 9.896 5.068

p-value 0.324 0.140 0.011 0.545 0.161 0.011 0.000 0.000

Restrictions (H0s) Short Run Causality: α1521 = α1522 = 0 Long Run Causality: γ15 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 2.997 0.083 9.756 0.002

Reject H0 Reject H0

Unrestricted Model ∆et = μ16 + ∑2i=1 α161 𝑖𝑖 ∆𝑜𝑜𝑡𝑡−𝑖𝑖 + ∑2i=1 α162 𝑖𝑖 ∆𝑒𝑒𝑡𝑡−𝑖𝑖 + γ16 (ot-1 - ω0 - ω1et-1 ) + ε16t

∆𝑒𝑒�𝑡𝑡 = -0.004 + 0.013∆𝑜𝑜𝑡𝑡−1 + 0.057∆𝑜𝑜𝑡𝑡−2 + 0.221∆𝑒𝑒𝑡𝑡−1 - 0.269∆𝑒𝑒𝑡𝑡−2


Estimation Result
- 0.035 (ot-1 - 2.812 – 4.294et-1 )

t-ratio -0.393 0.183 0.881 0.579 -0.688 -0.467 9.896 5.068

p-value 0.700 0.858 0.394 0.572 0.504 0.648 0.000 0.000

Restrictions (H0s) Short Run Causality: α1611 = α1612 = 0 Long Run Causality: γ16 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 1.261 0.261 0.316 0.574

Do not reject H0 Do not reject H0

Table 5.27 shows the results obtained from subsample period July 2008 - April 2010. The results
indicate that
(1) In both short run and long run, the movement of dollar exchange rate is the Granger cause of the
movement of crude oil price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of dollar exchange rate.
(3) In the long run, dollar depreciation will cause crude oil price increase. A 10% depreciation of dollar
exchange rate will cause a 2.740% (= -0.638 × (-4.294) × (ln1.1et-1 - lnet-1)) increase of crude oil price.
90

Table 5.28 Results of GNC Test on Entire Sample of Dollar Exchange Rate and Crude Oil Price

Unrestricted Model ∆ot = μ17 + α171 ∆𝑜𝑜𝑡𝑡−1 + α172 ∆𝑒𝑒𝑡𝑡−1 + γ17 ( ot-1 - ν0 - ν1et-1 ) + ε17t

Estimation Result ∆𝑜𝑜�𝑡𝑡 = 0.012 + 0.243∆𝑜𝑜𝑡𝑡−1 + 0.230∆𝑒𝑒𝑡𝑡−1 - 0.108 (ot-1 - 3.370 - 2.526et-1 )

t-ratio 1.567 3.029 0.776 -4.178 96.19 16.57

p-value 0.120 0.003 0.439 0.000 0.000 0.000

Restrictions (H0s) Short Run Causality: α172 =0 Long Run Causality: γ17 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 0.620 0.431 16.885 0.000

Do not reject H0 Reject H0

Unrestricted Model ∆et = μ18 + α181 ∆𝑜𝑜𝑡𝑡−1 + α182 ∆𝑒𝑒𝑡𝑡−1 + γ18 ( ot-1 - ν0 - ν1et-1 ) + ε18t

Estimation Result ∆𝑒𝑒�𝑡𝑡 = 0.001 + 0.008∆𝑜𝑜𝑡𝑡−1 + 0.324∆𝑒𝑒𝑡𝑡−1 + 0.007(ot-1 - 3.370 - 2.526et-1 )

t-ratio 0.391 0.333 3.879 0.918 96.19 16.57

p-value 0.696 0.740 0.000 0.360 0.000 0.000

Restrictions (H0s) Short Run Causality: α181 =0 Long Run Causality: γ18 = 0

LR Statistic p-value LR Statistic p-value

Likelihood Ratio Test 0.114 0.735 0.866 0.352

Do not reject H0 Do not reject H0

Table 5.28 indicates that for the entire sample period,


(1) In the short run, the movement of dollar exchange rate does not cause the movement of crude oil
price in the Granger sense. In the long run, the movement of dollar exchange rate is the Granger cause of the
movement of crude oil price.
(2) In both short run and long run, the movement of crude oil price is not the Granger cause of the
movement of dollar exchange rate.
(3) In the long run, dollar depreciation will cause crude oil price increase. A 10% decrease of dollar
exchange rate will cause a 0.273% (= -0.108 × (-2.526) × (ln1.1et-1 - lnet-1)) increase of crude oil price.
Table 5.29 below summarizes the results in Table 5.20 - Table 5.28. For dollar exchange rate and gold
price, no Granger causality between dollar exchange rate and gold price is identified in the short run. But
there is a unidirectional Granger causality from gold price to dollar exchange rate in the long run. For crude
oil price and gold price, test results indicate a unidirectional Granger causality from gold price to crude oil
price in the long run in all three samples. But the magnitude of long run impact of gold price on crude oil
price differs in the three samples. The impact in the second subsample is the largest and the impact in the full
91

sample is the smallest. The short run causality results are not consistent. In the first subsample, there is no
Granger causality between crude oil price and gold price in the short run. In the second subsample, there is a
unidirectional Granger causality from gold price to crude oil price in the short run. In the full sample, a
unidirectional Granger causality is also identified in the short run. But it goes from crude oil price to gold
price. For crude oil price and dollar exchange rate, short run Granger causality from dollar exchange rate to
crude oil price is identified in subsample one to four. Short run Granger causality from crude oil price to
dollar exchange rate is identified in subsample two. In the entire sample, no short run causality is found. Long
run causality from dollar exchange rate to crude oil price is identified in all samples except subsample two.
There is no long run causality from crude oil price to dollar exchange rate identified. In the first subsample,
dollar exchange rate depreciation causes crude oil price decrease. In the third and fourth subsample and in the
entire sample, dollar exchange rate depreciation causes crude oil price increase. There is no long run
adjustment between dollar exchange rate and crude oil price in subsample two period. The magnitude of long
run impact of dollar exchange rate on crude oil price differs in the four samples. The impact in the third
subsample is the largest and the impact in the full sample is the smallest. The impact in the fourth subsample
is the larger in absolute value than that in the first subsample.

Table 5.29 GNC Test Result Summary

Variables Sample Short Run Causality Long Run Causality Direction Long Run Elasticity
e and g 1999.1-2010.4 e↛g, g↛e g→e, e↛g - 0.162%
o and g 1999.1-2008.6 o↛g, g↛o g→o, o↛g + 1.765%
o and g 2008.7-2010.4 g→o, o↛g g→o, o↛g + 2.514%
o and g 1999.1-2010.4 o→g, g↛o g→o, o↛g + 1.009%
o and e 1999.1-2001.5 e↛o, o↛e e→o, o↛e + 0.590%
o and e 2001.6-2008.6 e→o, o→e e↛o, o↛e
o and e 2008.7-2009.9 e→o, o↛e e→o, o↛e - 5.214%
o and e 2008.7-2010.4 e→o, o↛e e→o, o↛e - 2.740%
o and e 1999.1-2010.4 e↛o, o↛e e→o, o↛e - 0.273%
Note: “-” means in the long run gold price increase causes dollar exchange rate depreciation or dollar
exchange rate depreciation causes crude oil price increase. “+” means in the long run, gold price increase
causes crude oil price increase or dollar exchange rate depreciation causes crude oil price decrease. For the
“Long run elasticity” column, take the second row, crude oil price and gold price, as an example, the number
1.765% means a 10% increase of gold price will cause a 1.765% increase of crude oil price in the long run.

The implications from the test results are:


(1) All three hypotheses described in Chapter III are supported and the three pillars of the triangular
system proposition are established. Gold price is the driver of both dollar exchange rate and crude oil price.
Gold price contains important information influential in the price formation of dollar exchange rate and crude
92

oil price. Dollar exchange rate and crude oil price also have interactions. The former one seems to be more
influential than the latter one in their bilateral relationship.
(2) It is highly possible that at least during January 1999 – April 2010 the bilateral causal relationship
between dollar exchange rate and crude oil price built in previous literature is a part of a triangular system of
dollar exchange rate, crude oil price and gold price. A gold price increase is found to cause dollar exchange
rate depreciation and crude oil price increase. So even if dollar exchange rate and crude oil price are not
connected, they may still appear to have a negative correlation relationship. The new findings not only
conform with the observed fact that dollar exchange rate and crude oil price are generally negatively
correlated, but also further expand the meaning of this negative correlation relationship.
(3) The relationship between dollar exchange rate and crude oil price is unstable over the sample period.
They were positively correlated prior to May 2001 and negatively correlated thereafter. The impact
magnitude of dollar exchange rate on crude oil price is also time-varying, which became stronger after June
2008. Dollar exchange rate movement is found to Granger cause crude oil price movement either in the short
run or in the long run or both depending on different sample period. But fundamentally, the results confirm
that dollar exchange rate movement is a Granger cause of crude oil price movement. This finding
corresponds to the Scenario One in Chapter III. The implication is that dollar exchange rate not only have
direct impact on crude oil price, but also functions as an intermediary vehicle through which gold price
movement indirectly affects crude oil price. Gold price therefore affects crude oil price both directly and
indirectly. The crude oil price is found to have some short run interaction with dollar exchange rate in the
subsample two period. But it does not have any impacts on dollar exchange rate after June 2008. Also, crude
oil price does not have long run impacts on dollar exchange rate. In all five samples, long run causality from
crude oil price to dollar exchange rate is rejected. Dollar exchange rate is more influential in its bilateral
relationship with crude oil price.
(4) The test results obtained from the full sample of crude oil price and gold price is self-contradictory.
The long run causality test result indicates the exogeneity of gold price to the bivariate system of crude oil
price and gold price. But the result of short run causality test suggests that gold price movement is affected by
crude oil price movement. These problematic results lend support to the structural break suspicion and further
justify the necessity of sample split.
(5) Gold price has stronger impacts on crude oil price than on dollar exchange rate in the long run. A 10%
increase of gold price causes a 0.1624% depreciation of dollar exchange rate but a 1.7651% increase of crude
oil price in the first subsample period and a 2.5141% increase in the second subsample period.
(6) Gold price has stronger impacts on crude oil price in the second subsample period. In the first
subsample period, the movement of gold price and the movement of crude oil price do not affect each other in
the short run. But in the second subsample period, the movement of gold price does affect the movement of
crude oil price in the short run. Moreover, while in both subsample periods the movement of gold price is the
Granger cause of the movement of crude oil price, the impact of gold price on crude oil price is stronger in the
93

second subsample period. A 10% increase of gold price causes a 2.514% increase of crude oil price in the
second subsample period, compared to a 1.765% increase in the first subsample period.
(7) Gold price is weakly exogenous to both the system of dollar exchange rate and gold price and the
system of crude oil price and gold price. Gold price is not affected by the other two prices.
Similar to section 5.3.4, section 5.3.5 is also subject to the estimation accuracy problem. While the
observations of the full sample are relatively sufficient, the observations in subsamples are limited. As a
result, the estimation results may be inaccurate. The choice of a linear model for subsamples may also cause
estimation inaccuracy. For example, for the analysis of crude oil price and gold price in the second subsample
period, both prices experienced sharp decreases and gradually recovered. A nonlinear model may fit the data
movement pattern in this period better than a linear model. While this idea is reasonable, this study does not
adopt this approach. The reason is that the total observations in the second subsample are too limited. A
nonlinear model may also subject to this same problem. The nonlinear estimation results may be more
inaccurate and seriously biased, not better than the estimation performance of a linear model. The analysis of
crude oil price and dollar exchange rate may also be subject to this reservation.
Another problem with the analysis in this section is the use of error correction model in subsample
periods. The error correction model is used to reveal both short run dynamics and long run trend of the
interaction between variables. But some of the subsamples in this study are not very long. It may not be
suitable to conclude that there is a long run trend in such sample period or make long run inferences solely
depending on a short sample period. The estimation results are therefore vulnerable to robustness and
accuracy questions.
Finally, when more data is available, all the above problems can be effectively solved.

5.3.6 Discussion
A question closely related to the implications above is why they would be the case. Specifically, why is
the relationship between dollar exchange rate and crude oil price so unstable? Why did the relationship
between dollar exchange rate and crude oil price experience multiple structural breaks in post-2007 period?
Why are the causality relationships in the four subsamples different from each other? Why does gold price
affect dollar exchange rate and crude oil price but affect them in different ways? Why does the gold price
have stronger impact on crude oil price in the second subsample period? Why is gold price exogenous to the
price formation of dollar exchange rate and crude oil price? This section will discuss in turn these questions.
According to the test results, there are three structural breaks in the relationship between dollar
exchange rate and crude oil price. There is also a candidate structural break date estimator not able to be
confirmed because of limited observations. In addition, the three identified structural break date estimators
can be further divided into two categories, structural break caused by correlation direction shifts and by
abrupt jumps. Why does dollar exchange rate and crude oil price not have a stable relationship? Several
94

reasons may account for this. First, as argued in previous literature, exchange rate per se has a time-varying
nature. Not only the dollar exchange rate may change over time, but also many macroeconomic fundamentals
that are influential on dollar exchange rate may experience shifts or structural changes. Thus it is difficult for
dollar exchange rate to maintain stable relationships with other variables. Secondly, as shown and discussed
in Chapter II, both dollar exchange rate and crude oil price are affected by a variety of factors. Some are
common factors, and others are not. Driven by factors that may have impacts different in direction,
mechanism and magnitude, it is difficult for the two prices to be stably connected with each other. Moreover,
both prices are very sensitive, frequently subject to exogenous shocks. Therefore, the relationship between
dollar exchange rate and crude oil price are not likely to be stable over time.
In the post-2007 period, the relationship between dollar exchange rate and crude oil price has two
structural breaks, June 2008 and September 2009. There is also a candidate structural break date estimator
January 2009 not able to be confirmed because of limited observations. Why would the relationship between
the two prices shift (at least) twice in such a short time period? To answer this question, recall the market
situation at that time. The subprime crisis started as least as early as the second half of 2007. In 2008, the
crisis within the subprime mortgage market gradually transformed into a crisis of the financial system. In late
2008 and 2009, the crisis further exacerbated and transformed into a worldwide economic recession. Dollar
exchange rate and crude oil price are heavily influenced by the global economy. Expanding for years, the
global economy finally entered into an adjustment period. In the two years, the global economy was
consecutively shocked by both positive and negative news and became more fragile, sensitive and vulnerable
than before. A direct reason is that panic spread worldwide and global investors became more sensitive than
before. So it is not simply the relationship between dollar exchange rate and crude oil price that was unstable,
but the global economy was unstable as well. The adjustment will not stop until the global economy returns to
a level on which its development can be supported by substantial economic activities. Before then, the
relationship between crude oil price and dollar exchange rate might have more structural breaks when more
data is added for analysis.
In their four subsample periods, dollar exchange rate and crude oil price exhibit three kinds of causality
relationships. During January 1999 – May 2001, there is no short run causality but a long run positive
causality from dollar exchange rate to crude oil price. That is, dollar exchange rate appreciation causes crude
oil price increase. In the period of June 2001 – June 2008, the two prices are mutually causally connected in
the short run but disconnected in the long run. In the last two subsample periods, July 2008 – September 2009
and July 2008 – April 2010, dollar exchange rate movement negatively causes crude oil price movement in
both short run and long run. The possible reason for the causal relationship of the two prices in the first
subsample period is that in the short run, both prices were driven by a variety of factors. Their mutual impacts
were not significant. In the long run, crude oil price touched the bottom in 1998 because of the Asian
financial crisis. As the global economy gradually recovered in 1999-2001, it began to increase too. The U.S.
is the world largest economy and the world largest crude oil consumer. Dollar exchange rate movement
95

reflects U.S. economic development and energy demand. So dollar exchange rate movement may cause crude
oil price movement in the long run. In the second subsample period, global economy expanded and world
energy demand increased quickly. Particularly, emerging economies showed strong energy demand. As a
result, crude oil price kept increasing. In this period, investors do not need to rely as heavily as before on U.S.
energy demand to estimate world energy demand. So there is no long run causality between the two prices.
However, changes of U.S. energy demand and economic development still can move crude oil price in the
short run. Meanwhile, as crude oil price became higher and higher, its impact on U.S. economy gradually
increased through, for example, terms of trade or balance of payment channels. So the two prices showed
noticeable short run interactions. In the third and fourth subsample periods, the world economy went into
recession and global energy demand shrank. The U.S. economy growth and U.S. energy demand again
became the most important reference in forecasting world economic recovery and world energy demand. The
U.S. economic recovery is a source of investor confidence in global economic recovery. So dollar exchange
rate movement drove crude oil price movement in both short run and long run.
There are several reasons behind the negative connection between gold price and dollar exchange rate.
The increase of gold price, especially in the long run, is likely to be a sign of global inflation. Specifically, it
may be a sign of U.S inflation, because gold price is denominated in dollars. As more and more investors
become aware of and begin to worry about inflation, the dollar, a major currency in the world, will lose its
value. If investors are confident of future inflation or expect the inflation to continue, they may buy gold and
sell dollars now and perform reverse transactions in the future, both of which further intensify the negative
connection between gold price and dollar exchange rate. Unless in extreme market situations, gold price
increase depresses gold demand and encourages gold supply, resulting in less demand for and more supply of
dollars and generating downward pressure on the dollar exchange rate. In addition, gold market is always
regarded as a safety haven. When there is an economic slowdown or turmoil, investors switch from other
markets to the gold market. This procedure may take months when there are gradual changes in the economy
but may take only several days if there is an abrupt change in the economy. The dollar exchange rate reflects
the health condition of U.S. economy (over other economies), which is a vane of the global economic
prospect. Economic slowdown or turmoil in U.S. is often interpreted as the prelude of global economic
slowdown or turmoil. Thus, gold price may reflect the degree of anxiety of investors about the U.S. and world
economy. The higher the gold price, the less confidence of investors about the world economy, the worse the
expectation about economic prospect and thus the lower the dollar exchange rate. However, some of the
reasoning above may not be valid in the very short run. For example, in the short run gold price increase may
not be considered a signal of rise of inflationary expectation. Moreover, in the short run, strong demand for
gold may increase the need of dollar and drive up the prices of both. In fact, the finding that gold price does
not Granger cause dollar exchange rate in the short run is consistent with findings in previous literature that
dollar exchange rate is difficult to forecast, especially in the short run. As mentioned in Chapter II, dollar
exchange rate is affected by a variety of factors in addition to gold price. It absorbs innovations quickly
96

because of the continuous transaction hours and the abundant liquidity of the foreign exchange market.
Therefore, gold price is not likely to significantly affect dollar exchange rate in the short run.
The reasons for the positive Granger causal relationship from gold price to crude oil price are somewhat
different. Because of the non-renewable property and the key role in modern society, crude oil is an attractive
alternative investment option. Besides gold, more and more investors include crude oil into their investment
portfolios. As an imperfect substitute for each other in the sense of investment, gold and crude oil tend to
have same price movement direction. As mentioned above, gold price increase, especially in the long run, is
likely to be a signal of inflation. As the inflationary expectation of investors rises, they may invest in
commodities, including crude oil, to hedge inflation risk. In extreme market situations, gold is invested
heavily to meet the capital security need. As an imperfect substitute, crude oil may also be used to avoid risks.
The relationship between gold price and dollar exchange rate is also a channel through which gold price may
move crude oil price. According to the finding of this study, gold price increase results in depreciation of
dollar. Since crude oil is denominated and traded in dollars, dollar depreciation will then cause crude oil price
to go up. Besides the case of extreme market situation, most of the channels mentioned above are likely to be
effective in the long run. In the short run, the crude oil price is hard to predict. Similar to dollar exchange rate,
crude oil price is also subject to the impacts of a variety of factors. Particularly, because crude oil is the major
energy source, its price is heavily affected by short run demand, supply and stock changes. Gold price is less
likely to move crude oil price in the short run as the test result of first subsample suggests.
A subsequent question is that if gold price is less likely to move crude oil price in the short run, why
does the test result of second subsample indicate short run Granger causality from gold price to crude oil
price? In response to this question, recall the market situation in the second subsample period. In July 2008,
both crude oil price and gold price plummeted sharply. The decreases of crude oil price and gold price then
were the largest declines in both magnitude and pace in the price histories of crude oil and gold over the full
sample period. During July 2008 – April 2010, the crisis within the financial system kept aggravating and
gradually transformed into a worldwide economic recession. The market situation in this period was not
normal but extreme. Crude oil was more an imperfect substitute for gold that was invested to hopefully make
wealth withstand financial surges. Since gold price is an indicator of investor anxiety, the short run movement
of crude oil price followed the movement of gold price. Because gold price and crude oil price related closer
in this period than in the first subsample period, gold price had stronger impacts on crude oil price.
The above discussion provides clues for the finding that gold price is exogenous to both the system of
dollar exchange rate and gold price and the system of crude oil price and gold price. Gold is a special
expensive metallic commodity. It is used to meet the need of consumption, investment, risk hedging and, in
extreme market situation, capital safety. While it is possible to predict the gold supply and demand for
consumption purpose, it is difficult to measure the investment demand and the risk hedging demand and even
more difficult to assess the capital safety demand. As discussed in Chapter II, the need for capital safety rises
in times of uncertainty or in economic turmoil. In both cases, investors may have overreactions and even
97

irrational behavior because of panic. Although often obscured by profitability, capital safety is fundamental
for investors. Therefore, gold price is likely to be exogenous to the price formation of dollar and crude oil.
98

CHAPTER VI. SUMMARY & CONCLUSION

The relationship between dollar exchange rate and crude oil price is both interesting and complicated.
This paper conducts theoretical, statistical and econometric analysis trying to better understand their
relationship. This chapter summarizes the results obtained from previous chapters and draws conclusions of
this study. Section 6.1 introduces major conclusions. Section 6.2 outlines and briefly discusses the limitations
of this study and directions of future research.

6.1. Conclusion
Based on the analysis and discussion in previous chapters, the following conclusions are drawn:
(1) Both the movement of dollar exchange rate and the movement of crude oil price are affected by the
movement of gold price. Gold price contains important information influential in the price formation of
dollar exchange rate and crude oil price. It is highly possible that at least during January 1999 – April 2010
the bilateral causal relationship between dollar exchange rate and crude oil price built in previous literature is
a part of a triangular system of dollar exchange rate, crude oil price and gold price.
(2) Dollar exchange rate and crude oil price are also causally connected. Granger causality from dollar
exchange rate to crude oil price is identified in both short run and long run. Thus, dollar exchange rate not
only has direct impacts on crude oil price, but also functions as an intermediary vehicle through which gold
price movement indirectly affects crude oil price. However, crude oil price is much less influential in the
movement of dollar exchange rate and not influential in the movement of gold price.
(3) The relationship between dollar exchange rate and crude oil price is unstable with multiple structural
breaks over the sample period.
(4) For dollar exchange rate and gold price, in the period of January 1999 – April 2010, no Granger
causality between dollar exchange rate and gold price is identified in the short run. But there is a
unidirectional Granger causality from gold price to dollar exchange rate in the long run. The relationship
between dollar exchange rate and gold price has no structural breaks in the sample period.
(5) For crude oil price and gold price, in the period of January 1999 – April 2010, a structural break is
identified at the date of June 2008. In the first subsample period, there is a unidirectional Granger causality
from gold price to crude oil price in the long run. But there is no Granger causality between crude oil price
and gold price in the short run. In the second subsample period, gold price movement is the Granger cause of
crude oil price movement in both short run and long run.
99

(6) The magnitude of long run impact of gold price differs. Gold price has stronger impacts on crude oil
price than on dollar exchange rate in the long run. Gold price has stronger impacts on crude oil price in the
second subsample period.
(7) Gold price is weakly exogenous to both the system of dollar exchange rate and gold price and the
system of crude oil price and gold price. Gold price is not affected by these two prices.

6.2. Limitations and Future Research


There are several aspects about this study that may need future improvement.
(1) Data limitations. This study performs parameter stability tests and detects structural breaks in the
relationship between crude oil price and gold price and between crude oil price and dollar exchange rate.
While the observations of the full sample are relatively sufficient, the observations in subsamples are limited.
Clearly, limited observations are likely to result in inaccurate estimation results. A simple and effective
solution to this problem is to add more data and redo the analysis. For the analysis of crude oil price and gold
price, something that is worth special attention is that it may not be appropriate to expand the sample period
to include pre-1999 data. The reason is that as long as the τ is set to be or be a subset of [0.15, 0.85], the
expansion of the first subsample period may result in exclusion of year 2008, which contains the breakdate
estimator June 2008. For the analysis of dollar exchange rate and crude oil price, both pre-1999 data and post
April 2010 data may be helpful.
(2) Soundness and robustness of selecting a linear model for the subsamples. It is possible that one could
question the reasonableness of applying a linear model to crude oil price and gold price in the second
subsample period, in which both prices experienced sharp decreases and gradually recovered. It seems that a
nonlinear model may fit the data movement pattern in this period better than a linear model. While this idea
has its merit, this study does not adopt this approach. The reason is that the number of observations in the
second subsample is too limited. A nonlinear model may also subject to this same problem. The nonlinear
estimation results may be more inaccurate and seriously biased, not better than the estimation performance of
a linear model. When more data is available to expand the second subsample period, it would be interesting to
use a nonlinear model to analyze again the relationship between crude oil price and gold price to see whether
a nonlinear model may produce significantly different results. The analysis of crude oil price and dollar
exchange rate may also be subject to this kind of questions.
(3) The use of error correction model in subsample periods. The error correction model is used to reveal
both short run dynamics and long run trend of the interaction between variables. But some of the subsamples
in this study are not very long. It may not be suitable to conclude that there is a long run trend in the sample
period or make long run inferences solely depending on a short sample period. The estimation results are
therefore vulnerable to robustness and accuracy questions. This argument can be a criticism to this study. But
again, when more data is available, the question raised here can be effectively solved.
100

(4) Choice of candidate common driver. So far, this study has collected and presented much evidence in
support of the choice of gold price as a common driver of dollar exchange rate and crude oil price. But
certainly some other variables may also be qualified as the potential candidate common driver. This paper is
not intended to argue that gold price is the sole common driver of dollar exchange rate and crude oil price, but
to assert that gold price may drive both prices and thus provides a new story of the relationship between dollar
exchange rate and crude oil price, different from the traditional idea in previous literature. In fact, interest rate
and expectation on global economic prospect can also affect both dollar exchange rate and crude oil price.
But the problem is that it is difficult to find suitable proxies for these variables. Interest rate and other
macroeconomic indicators are well accepted to be endogenous. It is certainly possible that some instrument
variables may function well. But such instrument variables are hard to find and their qualifications need to be
carefully examined to avoid biases. This will undoubtedly complicate the analysis and is likely to obscure the
original relationship between dollar exchange rate and crude oil price.
(5) Methodology limitation. This study uses the Hansen (1992) method to detect structural breaks and
uses error correction framework to test Granger causality, both of which are commonly used and seen in
literature. A clear drawback from the above discussion is that the structural break test results may not be
satisfactorily robust in the case of small sample. Also there are also other ways to detect Granger causality.
The econometric research is advancing every day. It would be necessary and interesting to use more
advanced methods later to check the estimation results in this study.
This study analyzes that relationship between dollar exchange rate and crude oil price over the sample
period of January 1999 - April 2010. The main valuable insight obtained from this study is that dollar
exchange rate movement is the Granger cause of crude oil price movement both in the short run and in the
long run. Gold price is the common driver of dollar exchange rate and crude oil price. Therefore, the observed
bilateral relationship between the latter two prices is likely to be a part of a triangular system of gold price,
dollar exchange rate and crude oil price. However, the relationship between dollar exchange rate and crude
oil price is quite unstable at least over the sample period of this study. The relationships found in this study
could soon change.
LIST OF REFERENCES
101

LIST OF REFERENCES

Adrangi, B., Chatrath, A., & Raffiee., 2003, Economic activity, inflation, and hedging: The case of gold
and silver investments, The Journal of Wealth Management, 6, 60–77.

Aggarwal, Raj, Brian M. Lucey, 2007, Psychological barriers in gold prices? Review of Financial
Economics 16, 217–230

Amano, R. A. and S. van Norden, 1998, Exchange Rates and Oil Prices, Review of International
Economics, 6(4), 683–694

Amuzegar, J., 1978, OPEC and the Dollar Dilemma. Foreign Affairs 56. 740–750.

Andersen, T., Bollerslev, T., 1998, Deutsche mark–dollar volatility: intraday activity patterns,
macroeconomic announcements, and longer run dependencies. Journal of Finance 53, 219–265.

Andersen, T., Bollerslev, T., Diebold, F., Vega, C., 2003, Micro effects of macro announcements:
real-time price discovery in foreign exchange, American Economic Review 93: 38–62.

Andersen, T., Bollerslev, T., Diebold, F., Vega, C., 2007, Real-time price discovery in global stock, bond
and foreign exchange markets, Journal of International Economics 73, 251–277

Andrews, D. W. K., and J. C. Monahan, 1992, An Improved Heteroskedasticity and Autocorrelation


Consistent Covariance Matrix Estimator, Econometrica, 60, 953-966.

Andrews, Donald W. K., 1993, Tests for Parameter Instability and Structural Change With Unknown
Change Point, Econometrica, Vol. 61, No. 4, 821-856

Austvik, O. G. 1987. Oil Prices and the Dollar Dilemma. OPEC Review 4.

Bacchetta, Philippe, and Eric van Wincoop, 2009, On The Unstable Relationship Between Exchange
Rates And Macroeconomic Fundamentals, NBER working paper 15008

Bacon, R., 1991, Modelling the price of oil. Oxford Review of Economic Policy 7, 17–34.

Baldwin, N., and Prosser, R., 1988, World oil market simulation, Energy Economics, 10, 185-198
102

Barsky, R. B., Kilian, L., 2002, Do We Really Know that Oil Caused the Great Stagflation? A Monetary
Alternative, NBER Macroeconomics Annual 2001, 137-183.

Barsky, R. B., Kilian, L., 2004, Oil and the macroeconomy since the 1970s, Journal of Economic
Perspectives 18(4), 115-134.

Baur, D. G. and B. M. Lucey, 2010, Is Gold a Hedge or a Safe Haven? An Analysis of Stocks, Bonds and
Gold, The Financial Review 45, 217–229

Baur, D.G., T.K. McDermott, 2010, Is gold a safe haven? International evidence, Journal of Banking &
Finance 34, 1886–1898

Benassy-Quere, A., V. Mignon and A. Penot, 2007, China and the Relationship Between the Oil Prices
and the Dollar, Energy Policy, 35, 5795–805.

Bergvall, A., 2004, What determines real exchange rates? The Nordic countries, Scandinavian Journal
of Economics 106(2), 315–337

Blanchard, O.J., and J. Galí, 2007, The Macroeconomic Effects of Oil Shocks: Why are the 2000s so
Different from the 1970s? NBER Working Paper No. 13368.

Breitenfellner, A., Cuaresma J. C., 2008, Crude Oil Prices and the USD/EUR Exchange Rate, Monetary
Policy & the Economy Q4

Brook, Anne-Marie, Robert Price, Douglas Sutherland, Niels Westerlund and Christophe André, 2004,
Oil Price Developments: Drivers, economic Consequences and Policy Responses, OECD Economics
Department Working Papers, No. 412

Bruno,S. and L. Chincarini, 2010, A Historical Examination of Optimal Real Return Portfolios for
Non-US Investors, Review of financial economics, Accepted manuscript

Bureau of Economic Analysis, 2009, News Release: U.S. International Transactions: First Quarter 2009,
http://www.bea.gov/newsreleases/international/transactions/2009/trans109.htm

Capie, F., Mills, T.C., Wood, G., 2005, Gold as a hedge against the dollar, Journal of International
Financial Markets, Institutions and Money 15, 343–352.

Cashin, P., L. F. Cespedes and R. Sahay, 2004, Commodity Currencies and the Real Exchange Rate.
Journal of Development Economics 75. 239–268.

Cavallo, M., Ghironi, F., 2002, Net foreign assets and the exchange rate: Redux revived, Journal of
Monetary Economics 49, 1057–1097
103

Chaudhuri, K., B.C. Daniel, 1998, Long-Run Equilibrium Real Exchange Rates and Oil Prices,
Economic Letters 58.231-238

Chen Y.C., K. Rogoff, 2003, Commodity currencies, Journal of International Economics 60.133–160

Chen, S.-S. and H.-C. Chen, 2007, Oil Prices and Real Exchange Rates. Energy Economics 29.
390–404.

Chen, Y.C., K. Rogoff and B. Rossi, 2008, Can Exchange Rates Forecast Commodity Prices? Working
Paper. Harvard University.

Cheng, K. C., 2008, Dollar Depreciation and Commodity Price. In: IMF (ed.). 2008. World Economic
Outlook. International Monetary Fund. Washington D.C. 48–50

Cheung, Y.W., M.D. Chinn, 2001, Currency traders and exchange rate dynamics: a survey of the US
market, Journal of International Money and Finance 20: 439–471

Cheung, Yin-Wong, Menzie D. Chinn and Antonio Garcia Pascual, 2005, Empirical Exchange Rate
Models of the Nineties: Are they Fit to Survive? Journal of International Money and Finance,
24:1150-1175.

Chevillon, G., and C. Rifflart, 2009, Physical market determinants of the price of crude oil and the
market premium, Energy Economics 31, 537–549

Chinn, Menzie D., and Richard A. Meese, 1995, Banking on Currency Forecasts: How Predictable Is
Change in Money? Journal of International Economics 38 (February): 161– 178.

Chua, J. and R.S. Woodward, 1982, Gold as an inflation hedge: a comparative study of six major
industrial countries, Journal of Business Finance & Accounting 9, 2, 191-197

Cox J., and Laurie Glapa, 2009, Credit Crisis Timeline, The University of Iowa Center for International
Finance and Development

Croushore, D., Tom Stark, 2001, A Real-Time Data Set for Macroeconomists, Journal of Econometrics
105, 111-130.

Dahl, C., and Yucel, M., 1990, Dynamic Modelling and Testing of OPEC Behaviour, Working Paper
EE11, Oxford Institute for Energy Studies, Oxford.

Darrat, A. F., Gilley, O.W., Meyer, D.J., 1996, US Oil Consumption, Oil Prices, and the Macroeconomy,
Empirical Economics 21:317-334

Dees, S., Gasteuil, A., Kaufmann, R. K., Mann, M., 2008, Assessing the Factors Behind Oil Price
Changes, European Central Bank Working Paper Series No. 855
104

Dees, S., Karadeloglou, P., Kaufmann, R. K., Sanchez, M., 2007, Modelling the world oil market:
assessment of a quarterly econometric model. Energy Policy 35, 178–191.

Dickey, D., and W. Fuller, 1979, Distribution of the Estimators for Autoregressive Time Series with a
Unit Root, Journal of the American Statistical Association, 74, 427–431.

Dickey, D., and W. Fuller, 1981, Likelihood Ratio Tests for Autoregressive Time Series with a Unit
Root, Econometrica, 49, 1057–1072.

Dornbusch, R., 1976, Expectations and Exchange Rate Dynamics. Journal of Political Economy 84,
1161–1176.

Dueker, Michael D., 1997, Strengthening the Case for the Yield Curve as a Predictor of U.S. Recessions,
Federal Reserve Bank of St. Louis Review, March/April, 79(2), 41-51.

Ehrmann, M., Fratzscher, M., 2005, Exchange rates and fundamentals: new evidence from real-time
data, Journal of International Money and Finance 24, 317–341

Elekdag, Selim, René Lalonde, Douglas Laxton, Dirk Muir, and Paolo Pesenti, 2007, Oil Price
Movements and the Global Economy: A Model-Based Assessment, Bank of Canada, Working Paper
2007–34.

Energy Modeling Forum, 1992, International Oil Supplies and Demands, EMF Report 11, vol. II.

Engel, Charles, Kenneth D. West, 2005, Exchange Rates and Fundamentals, Journal of Political
Economy 113, 485-517.

Engel, Charles, Nelson C. Mark, Kenneth D. West, 2007, Exchange Rate Models Are Not As Bad As
You Think, NBER Working Paper 13318

Engle, Robert F. and Clive W. J. Granger, 1987, Cointegration and Error Correction: Representation,
Estimation and Testing, Econometrica 55:251–276.

Engle, Robert F., David F. Hendry and Jean-Francois Richard, 1983, Exogeneity, Econometrica
51:277–304.

Evans, M., Lyons, R., 2002, Order flow and exchange rate dynamics, Journal of Political Economy 110:
170–180.

Evans, M., Lyons, R., 2008, How is macro news transmitted to exchange rates? Journal of Financial
Economics 88: 26–50

Faust, J., Rogers, J.H.,Wang, S.Y.B.,Wright, J.H., 2007, The high-frequency response of exchange rates
and interest rates to macroeconomic announcements. Journal of Monetary Economics 54, 1051–1068.
105

Federal Reserve Bank, 2008, Monetary Policy Press Release, October 29.
http://www.federalreserve.gov/newsevents/press/monetary/20081029b.htm

Fleming, M., 1962, Domestic Financial Policies under Fixed and Floating Exchange Rates, I.M.F. Staff
Papers 9 (November): 369-379.

Frankel, Jeffrey A., 2006, The Effect Of Monetary Policy On Real Commodity Prices, NBER Working
Paper 12713

Frankel, Jeffrey A., and Andrew K. Rose, 1995, Empirical Research on Nominal Exchange Rates,
Handbook of International Economics, vol. 3 (Elsevier).

Frenkel, J. A., 1976, A Monetary Approach to the Exchange Rate, Scandinavian Journal of Economics 2
(May): 200-221.

Ghosh, D., Levin, E. J., Macmillan, P., &Wright, R. E., 2004, Gold as an inflation hedge? Studies in
Economics and Finance (Charlotte, NC), 22, 1–25.

Golub, S. S., 1983, Oil Prices and Exchange Rates, The Economic Journal 93:576-93.

Gourinchas, P.O., Rey, H., 2007, International Financial Adjustment, Journal of Political Economy, vol.
115, no. 4, 665-703

Granger, C. W. J., 1969, Investigating Causal Relations by Econometric Models and Cross-Spectral
Methods, Econometrica, 37, 424-459.

Greene, William H., 2002, Econometric Analysis, Fifth Edition, Pearson Education, Inc.,

Gregory, A.W., Hansen, B.E., 1996, Residual-based tests for cointegration in models with regime shifts.
Journal of Econometrics. 70, 99-126.

Groen, Jan J. J., 2000, The Monetary Exchange Rate Model as a Long-Run Phenomenon, Journal of
International Economics 52 (December): 299–319

Guo, H. and Kevin L. Kliesen, 2005, Oil Price Volatility and U.S. Macroeconomic Activity, Federal
Reserve Bank of St. Louis Review, November/December, 87(6): 669-83.

Hamilton, J. D., 1983, Oil and the Macroeconomy Since World War II, Journal of Political Economy 91:
228-248.

Hamilton, J. D., 1996, Analysis of the Transmission of Oil Price Shocks through the Macroeconomy,
Working Paper, University of California, San Diego.

Hamilton, J. D., 2008, Understanding Crude Oil Prices, NBER Working Paper 14492
106

Hansen, B., 1992, Tests for parameter instability in regressions with I(1) processes. Journal of Business
and Economic Statistics 10, 321–335.

Hau, H., Rey, H., 2004, Can Portfolio Rebalancing Explain the Dynamics of Equity Returns, Equity
Flows and Exchange Rates? American Economic Review, Vol.94, No.2, 126-133

Hau, H., Rey, H., 2006, Exchange Rates, Equity Prices, and Capital Flows, The Review of Financial
Studies, 19:273-317

Johansen, S., and K. Juselius, 1990, Maximum Likelihood Estimation and Inference on Cointegration
-With Applications to the Demand for Money, Oxford Bulletin Of Economics And Statistics. 52,
169-210.

Jones, Donald W., Leiby, Paul N. and Paik, Inja K., 2004, Oil Price Shocks and the Macroeconomy:
What Has Been Learned Since 1996? The Energy Journal, 25(2), 1-32.

Kaul, A., Sapp, S., 2006, Y2K fears and safe haven trading of the U.S. dollar, Journal of International
Money and Finance 25, 760–779.

Kilian, L., 2008, The Economic Effects of Energy Price Shocks, Journal of Economic Literature,
46:871-909

Kilian, L., Rebucci, A., Spatafora, N., 2009, Oil shocks and external balances, Journal of International
Economics 77: 181–194

Kilian, L., Vega, C., 2008, Do Energy Prices Respond to U.S. Macroeconomic News? A Test of the
Hypothesis of Predetermined Energy Prices, Board of Governors of the Federal Reserve System,
International Finance Discussion Papers Number 957, November.

Krichene, N., 2002, World Crude Oil and natural Gas: A Demand and Supply Model, Energy
Economics, 24, 557-576

Krichene, N., 2005, A Simultaneous Equations Model for World Crude Oil and Natural Gas Markets,
IMF Working paper WP/05/32.

Krichene, N., 2007, An oil and gas model, IMF Working paper WP/07/135.

Krugman, P. R. 1980, Oil and the Dollar, NBER Working Paper.No.554

Levin, E.J. and Montagnoli, A. and Wright, R.E., 2006, Short-run and long-run determinants of the price
of gold, Research Study No. 32, World Gold Council.

Lynch, Michael C., 2002, Forecasting oil supply: theory and practice, The Quarterly Review of
Economics and Finance 42: 373–389
107

MacDonald, Ronald, and Mark P. Taylor, 1994, The Monetary Model of the Exchange Rate: Long-Run
Relationships, Short-Run Dynamics and How to Beat a Random Walk, Journal of International Money
and Finance 13 (June): 276–90.

Mark, Nelson C., 1995, Exchange Rates and Fundamentals: Evidence on Long-Horizon Predictability,
American Economic Review 85 (March): 201–18.

Mark, Nelson C., 2001, International Macroeconomics and Finance: Theory and Econometric Methods.
Oxford: Blackwell.

Mark, Nelson C., and Donggyu Sul, 2001, Nominal Exchange Rates and Monetary Fundamentals:
Evidence from a Small Post–Bretton Woods Sample, Journal of International Economics 53 (February):
29–52.

Mark, Nelson C., and Doo-Yull Choi, 1997, Real Exchange Rate Prediction Over Long Horizon, Journal
of International Economics 43: 29-60

McGuirk, Anne, 1983, Oil Price Changes and Real Exchange Rate Movements among Industrial
Countries, International Monetary Fund Staff Papers, December, 843-84.

Meese, R.A., Rogoff, K., 1983, Empirical exchange rate models of the seventies: do they fit out of
sample? Journal of International Economics 14: 3–24.

Möbert, J., 2007, Crude Oil Price Determinants, Darmstadt Discussion Papers in Economics, No. 186

Mundell, R. A., 1964, Exchange Rate Margins and Economic Policy, In Money in the International
Order, edited by C. Murphy. Dallas: Southern Methodist Univ. Press

Mundell, R. A., 1968, International Economics, New York: Macmillan

Mussa, M., 1976, The Exchange Rate, the Balance of Payments and Monetary and Fiscal Policy under a
Regime of Controlled Floating, Scandinavian Journal of Economics 2 (May): 229-248.

Newey, W.K., and K.D. West, 1987, A Simple Positive Semi-Definite, Heteroskedasticity and
Autocorrelation Consistent Covariance Matrix, Econometrica, 55, 703-708.

Obstfeld, M., Rogoff, K., 1995, Exchange rate dynamics redux. Journal of Political Economy 103,
624–660.

Park, Jungwook, Ronald A. Ratti, 2008, Oil price shocks and stock markets in the U.S. and 13 European
countries, Energy Economics 30: 2587–2608

Phillips, P., and P. Perron, 1988, Testing for a Unit Root in Time Series Regression, Biometrika, 75,
335–346.
108

Phillips, P.C.B., Hansen, B., 1990, Statistical inference in instrumental variable regression with I(1)
processes, Review of Economic Studies 57, 99–125.

Pindyck, R.S., 1994, Inventories and the short-run dynamics of commodity prices. RAND Journal of
Economics 25, 141–159.

Pindyck, R.S., 2001, The dynamics of commodity spot and futures markets: a primer. Energy Journal 22,
1–29.

Pindyck, R.S., 2004, Volatility and Commodity Price Dynamics, The Journal of Futures Markets, Vol.
24, No. 11, 1029–1047

Quandt, R., 1960, Tests of the Hypothesis That a Linear Regression System Obeys Two Separate
Regimes, Journal of the Amer- ican Statistical Association, 55, 324-330.

Ranaldo, A., Söderlind, P., 2010, Safe haven currencies, Review of Finance, 1–23

Roberts, S. W., 1959, Control Chart Tests Based on Geometrics Moving Averages, Technometrics, 1,
239-250.

Rossi, B., 2006, Are Exchange Rates Really Random Walks? Some Evidence Robust to Parameter
Instability, Macroeconomic Dynamics 10(1), 20-38.

Sadorsky, P., 2000, The empirical relationship between energy futures prices and exchange rates,
Energy Economics 22(2000)253-266

Schulmeister, S. 2000, Globalization Without Global Money: The Double Role of the Dollar as National
Currency and as World Currency and its Consequences. Journal of Post Keynesian Economics 22.
365-395

Sherman, E. J., 1982, Gold: A conservative, prudent diversifier, Journal of Portfolio Management, 8,
21–27.

Taylor, M.P., Peel, D.A., 2000, Nonlinear adjustment, long-run equilibrium and exchange rate
fundamentals, Journal of International Money and Finance 19, 33–53

US Geological Survey, 2000, World Petroleum Assessment, 2000, Digital Data Series 60.

Wikipedia, 2010, Subprime Crisis Impact Timeline,


http://en.wikipedia.org/wiki/Subprime_crisis_impact_timeline; Date of access: June 21, 2010.

Worthingtona, A. C. and M. Pahlavani, 2007, Gold investment as an inflationary hedge: cointegration


evidence with allowance for endogenous structural breaks, Applied Financial Economics Letters, 3,
259–262
109

Wright, J.H., 2008, Bayesian Model Averaging and exchange rate forecasts, Journal of Econometrics
146: 329–341

Ye, M., Zyren, J., Shore, J., 2002, Forecasting crude oil spot price using OECD petroleum inventory
levels. International Advances in Economic Research 8, 323–333.

Ye, M., Zyren, J., Shore, J., 2005, A monthly crude oil spot price forecasting model using relative
inventories. International Journal of Forecasting 21, 491–501.

Yousefi, A. and T. S. Wirjanto, 2004, The Empirical Role of the Exchange Rate on the Crude-oil Price
Formation. Energy Economics 26, 783–799.

Zhang Y.J. , Fan Y, Tsai H.T., Wei Y.M., 2008, Spillover effect of US dollar exchange rate on oil prices,
Journal of Policy Modeling 30, 973–991

Zhou, S., 1995, The response of real exchange rates to various economic shocks. Southern Economic
Journal, Vol. 61, No. 4. 936-954