i.
5.
. ‘
u. q
Q ..!.

 

L
1!..- .

‘ 2:1:-

xl. J“...- ulivi... .

.133: “taunt:

.3. 1A1!

. . 1
.1 \ y‘iilsiiav
21!: hit}.
511.2.

t.

 

 

"'"E'Ff‘ QARY
iv'iichigan State
University

-.~I~-m- m '

 

 

This is to certify that the
dissertation entitled

ECONOMIC ANALYSIS OF US. ETHANOL
EXPANSION ISSUES

presented by

Malika Chaudhuri

has been accepted towards fulﬁllment
of the requirements for the

PhD. degree in Agricultural Economics

 

 

M

Major Professor’s Signature

g/Z?/¢7

Date

’/

MSU is an Afﬁnnative Action/Equal Opportunity Employer

-.----—.---n-.---n-.-- - - -.- --

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 K:/ProilAcc&Pres/ClRC/DaIeDue.indd

ECONOMIC ANALYSIS OF US. ETHANOL EXPANSION ISSUES
By

Malika Chaudhuri

A DISSERTATION

Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Agricultural Economics

2009

ABSTRACT
ECONOMIC ANALYSIS OF US. ETHANOL EXPANSION ISSUES
By

Malika Chaudhuri

The dependency of the US. economy on crude oil imported from politically unstable
countries, escalating energy demand world wide, growing nationwide environmental
consciousness, and the Renewable Fuels Standards (RFS) government mandates are some
of the primary factors that have provided a favorable environment for the growth and
development of the US. ethanol industry.

The ﬁrst essay derives decision rules for a discrete-time dynamic hedging model
in a multiple commodity framework under expected utility maximization and basis risk.
It compares hedging performance of three types of hedging models, namely constant
hedging, time-varying static hedging model and the new dynamic hedging rule derived in
this study. Findings show that natural gas futures contracts were effective instruments for
hedging ethanol spot price risk before March, 2005, when ethanol futures trading was
initiated on the CBOT. However, post-March, 2005, corn and ethanol futures contracts
proved to be efﬁcient hedging instruments. Results also indicate that ethanol producers
may effectively decrease variance of cumulative cash ﬂows by hedging using ethanol,
natural gas and corn futures prices using the traditional techniques. The study concludes
that using the new dynamic hedge model in a three period and two commodity set up,
producers can effectively reduce variance of cumulative cash ﬂow by 13.2% as compared

to the ‘no hedge’ scenario.

In my second essay, I use choice based, conjoint analysis methods to estimate
consumers’ willingness to pay (WTP) for alternative transportation fuels in the US. In
this study, I consider unleaded gasoline and ethanol, which may be derived from corn or
three different sources of cellulosic biomass as alternative transportation fuels. Results
suggest that age and household income are some of the socioeconomic variables that
signiﬁcantly inﬂuence consumer’s choice behavior. Results indicate considerable
consumer preference heterogeneity. Welfare effects are analyzed when consumers are
faced with restricted choice sets. Results suggest that possible government mandates on
the consumption of E-lO and E-85 diminish welfare of individuals belonging to the
segment ‘Conventional Gasoline Acceptor’. Similarly, individuals belonging to ‘Ethanol
Acceptor’ segment experience welfare losses if corn grain ethanol is not available as an
alternative transportation fuel.

Ethanol is increasingly being used as a gasoline oxygenate and a volume extender
in the reﬁnery and blender industry in the US. This paper estimates reﬁnery and blender
factor demand and evaluates price responsiveness of inputs. The study also develops and
tests hypotheses regarding existence of structural change in the industry’s demand for
inputs. It determines whether there is a common shift point and adjustment rate for
structural change in all the reﬁnery and blender inputs by using gradual switching
multivariate regression techniques and maximum likelihood methods. Results suggest a
structural change in factor demand for inputs in the industry that occurs at different points
and rates. Results also suggest that the demand for inputs, except for capital and

unﬁnished oil, has become more inelastic over time.

T o my family, who oﬂered me unconditional love and support

throughout this journey.

iv

ACKNOWLEDGMENTS

I am deeply indebted to my dissertation committee- Dr. H. Christopher Peterson (Major
Professor), Dr. Glynn Tonsor (Dissertation Professor), Dr. Robert J. Myers, and Dr.
Roger Calantone for their insightful comments, helpful guidance, and unwavering
support. In particular, I owe a huge debt of gratitude to Dr. H. Christopher Peterson for
his encouragement, and unﬂinching support through all phases of the doctoral program. I
especially appreciate his guidance in helping me to keep things in perspective, whether in
regards to my dissertation, or my career aspirations. I am indebted, to Dr. Glynn Tonsor,
who, ﬁrst as a teacher and later as a coauthor, taught me much of what I know about my
ﬁeld of research, teaching me the art of conducting research and also that of presenting.
Without his constant guidance and support this dissertation would not have been possible.
I thank Dr. Robert J. Myers for his invaluable guidance, constructive criticisms and for
being instrumental in my development as an academic. I thank Dr. Roger Calantone for
his support and encouragement and also for being instrumental in opening up ‘Marketing’
as a new dimension in my research work. I thank the other professors of the Department
of Agricultural, Food, and Resource Economics for their encouragement and advice in
the process of completing my dissertation. Special thanks go to my colleague Joleen C.
Hadrich for being a close and dependable friend throughout this journey. I thank
Haimanti Banerji, my other friends, and all my fellow graduate students for all their help
and support.

On behalf of my husband, my daughter, and myself, I would like to express our
sincere gratitude for Dr. G. Geoffrey Booth and Elizabeth Booth for treating us like

family and for their help and support throughout our stay in East Lansing. I would like to

express my sincere appreciation to my brother, Udipta Mukherji, and my sister-in-laws
Dr. Sayanti Bhattacharya, Rini Banerji and brother-in-law Dr. Kunal Banerji for their
unwavering faith and conﬁdence in my abilities. Finally, none of this would have been
possible without the patience and encouragement of my family back in India, who have
been a constant source of love and concern, over all these years. Support, strength and
care from my parents, Dr. Raka Mukherji and Manas Mukherji and my mother-in-law,
Mrs. Renu Chaudhuri and father-in-law, Mr. R.K. Chaudhuri helped me overcome
setbacks and stay focused on my graduate studies. My deepest gratitude goes to my
husband Ranadeb and my daughter Esha for their understanding during all the time spent
away from them. Without their sacriﬁces, this dissertation would not have been possible.
Special thanks to Ranadeb for his constant guidance, helpful criticisms, unconditional

support and for being there when I needed him most.

vi

TABLE OF CONTENTS

 

List of Tables - - - ix
List of Figures ................................................................................................. x
Optimal Futures Market Hedging by Ethanol Producers ............................................ 1
1.1 Introduction ............................................................................................ 1
1.2 Rationale for Hedging ................................................................................ 4
1.3 Background Literature .............................................................................. 7
1.3.1 Background Literature on Hedging in the Ethanol Industry .......................................................... 9

1.4 Generalized Dynamic Hedging Model ........ . ...... .. .......................................... 11
1.4.1 Multi-Period Model Speciﬁcation ........................................................................ 13

1.5 Data and Preliminary Statistics ..... . ......................................................... ....22
1.6 Empirical Application and Results ............................................................... 26
1.6.1 Constant Hedge Model ..................................................................................... 26

1.6.2 Time Varying Static Hedge Model ....................................................................... 28

1.6.3 Generalized Dynamic Hedge Model ..................................................................... 30

1.6.4 Performance Test ........................................................................................... 33

1.7 Conclusion and Policy Implications 36

 

ChapterIAppendrx39
A1.1 Derivation ofCumulative Cash Flow40

Consumer Preference for Alternative Transportation Fuels: A Conjoint Analysis..............58

 

 

2.1 Introduction ....................................................................... . ........... . ...... 58
2.2 Previous Studies ...................................................................................... 60
2.3 Model of Fuel Choice ............................................................................... .63
2.4 Research Design ....... . .............................. . ............................................ ...66
2.4.1 Attributes of Alternative Fuels and Their Levels ........................................................ 67
2.4.2 Cheap Talk: 68
2.5 Empirical Speciﬁcation ............................................................................. 70
2.5.1 Factor Analysis ............................................................................................. 70
2.5.2 Estimation of the Number of Latent Segments ......................................................... 71
2.5.3 Willingness-to—Pay Estimates and Welfare Measures 72
2.6 Results and Discussion .............................................................................. 74
2.6.1 Characterizing the Segments and Comparing the Models ............................................. 74
2.6.2 Willingness to Pay .......................................................................................... 77
2.6.3 Consumer Welfare Evaluation ............................................................................ 80
2.7 Summary and Conclusions ........................................................................ 83

vii

Factor Demand Analysis for Ethanol in the US. Reﬁnery Industry ......................... .....96
3.1 Introduction .................................................. ............................. ..... 96
3.2 Motivatlon98

3.3 Literature Review .................................................................................. 100

3. 4 Model Speciﬁcation..... ............ ........ ................. ............. ......... 102
3.4.1 Speciﬁcation of the Translog Model .................................................................... 102

3. 4. 2 Structural change identiﬁcation allowing separate shift points and adjustment rates ............ 103

3.4.3 Structural change identiﬁcation assuming common join point and rate of adjustment .......... 104

3.4.4 Transition function speciﬁcations ....................................................................... 106
3.4.5 Imposing Concavity Constraints ......... . .................................................. ..108
3.4.6 Symmetry and Adding-up Constraints .................................................................. 109

3.5 DataOOOOOOIOOOOOOOI. .......... O. ..... O ..... O OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO 0.00.0111
3.5 Results ............................................. ........... . .................................... 113
3.6.1 Estimation of separate join points and rate of adjustments .......................................... 113

3.6.2 Estimation of common join points and rate of adjustments .......................................... 115

3.6.3 Own Price and Cross Price Elasticities ................................................................. 116

3.6.4 Morishima Elasticity of Substitution ................................................................... 119

3.7 Conclusion and Policy Implications ...................................................... ......121
Chapter 3 Appendix .................... . ............................................................. .....124
A3.1 Model details ................................................................................. 125
BIBLIOGRAPHY ........................................................................ 134

viii

List of Tables

1.1 Summary Statistics ............................................................................. 41

1.2 Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) Test Results for
Unit Roots ..................................................................................... 42

1.3 J ohansen Test Results for pair wise Cointegration between Spot and Futures
Prices, accounting for ‘Time to Maturity’ ................................................... 43

1.4 Johansen Cointegration Test Results among the Spot and Futures Price
Series, accounting for ‘Time to Maturity’ ................................................. 44

1.5 Temporal Causality Results Based on Vector Error-Correction Model (VECM). . . ..45
1.6 Constant Hedging Models ..................................................................... 46

1.7 Comparison of the Effects of Hedging on Cumulative Cash Flow across Models....47

A2.1 Distribution and Factor Analysis of Attitudinal Statements ............................ 87
2.1 Descriptive Statistics ........................................................................... 89
2.2 Attributes and Levels of Alternative Fuel Choice Experiment ........................... 90
2.3 Segment membership variables for the ﬁve segment model .............................. 91
2.4 Utility Function Parameters on Fuel Attributes ............................................. 92
2.5 Consumer Willingness to Pay [95% Conﬁdence Interval] for Transportation

Fuel Attributes .................................................................................... 93
2.6 Welfare Effects of Restricted Alternative Transportation Fuel Choice Sets

under Various Assumptions [95% Conﬁdence Interval] ................................... 94
A32 Parameter Coefﬁcients of the Translog model ......................................... 127
3.1 Summary Statistics of Select Variables ..................................................... 129
3.2 Estimated Individual Join Points and Adjustment rates ................................. 130
3.3 Own and Cross-Price Elasticities ............................................................. 131
3.4 Morishima Elasticities of Substitution ....................................................... 132

ix

List of Figures

1.1 Cross Hedging Ethanol Spot against Near-by Natural Gas Futures Contracts,
January-1991 to May, 2009 ................................................................ 48

1.2 Cross Hedging Ethanol Spot against Near-by Corn Futures Contracts, April-2005
to May, 2009 ................................................................................. 49

1.3 Hedging Ethanol Spot Price against Nearby Ethanol Futures, April, 2005 to May,
2009 ........................................................................................... 50

1.4 Hedging Natural Gas Spot Price against Nearby Natural Gas Futures contracts. . . .51

1.5 Hedging Corn Spot Price against Nearby Corn Futures contracts ..................... 52
1.6 Persistence in Ethanol Spot Price in Second Regime .................................... 53
1.7 Persistence in Natural Gas Spot Price in Second Regime ............................... 54
1.8 Ethanol Hedge Positions in Period I and II ................................................ 55
1.9 Natural Gas Hedge Positions in Period I and II ........................................... 56
1.10 Comparison of Proﬁt across Hedging Techniques ...................................... 57
2.1 An example of choice scenario from the choice experiment .......................... 95

3.1 Renewable Fuel Standard Mandates and Ethanol Production..........................133

Chapter 1

Optimal Futures Market Hedging by Ethanol Producers
1.1 Introduction

In the ethanol industry, proﬁt depends on a set of commodity prices (corn, natural gas,
ethanol and distiller’s dried grain with solubles (DDGS)). Of these, ethanol has an
illiquid futures market and DDGS does not have a futures market at all. With volatile
input and output prices, ethanol plant owners face signiﬁcant challenges in protecting
proﬁt margins given the unpredictability of commodity prices (CARD, 09). Additionally,
with the ﬁnancial risk stemming from initial capital outlays and increasing competition
for resources, existing and future ethanol producers may consider hedging as a risk
management tool. Hedging may allow market participants to increase the predictability of
future cash ﬂows, lock-in proﬁt margins, and reduce downside risk (V arangis, Thigpen
and Satyanarayan, 1994). Contractual agreements like futures contracts for ethanol were
introduced in March, 2005 (Funk et al., 2008). Given their relative newness, these
markets are comparatively illiquid and thinly traded (Franken and Parcell, 2002).
However, assuming that contracting in cash markets alone may leave ethanol producers
subject to price risk exposure, participation in the ﬁitures market may provide valuable
risk management strategies for ethanol producers.

This paper uses dynamic programming to develop a multiple-commodity,

multiple-period hedging model (henceforth referred to as Generalized Dynamic Hedge

(GDH) model) to manage ethanol proﬁt risk. We extend the current prevailing literature
(i.e. Myers and Hanson, 1996) by considering multiple commodities and multiple
maturities, and at the same time utilizing some of the standard features like dynamic and
cross hedging strategies. The intertemporal hedging model assumes that a typical ethanol
producer faced with unpredictable input and output prices hedges in the futures market to
mitigate price risk. He initiates a multiple period hedging strategy at an initial time period
for the next T periods of production. As he moves to the next period, he completes all
cash market transactions and liquidates the futures contract that has the closest maturity
date lying beyond the current month. With the advent of additional information, he
updates the futures position for the remaining futures contracts. We derive the sequence
of optimal hedging positions at each time period that maximizes utility of cumulative
cash ﬂow of the ethanol producer at the end of period T.

The hedging effectiveness of three types of hedging techniques is compared. The
ﬁrst technique is the constant hedge strategy where the hedge ratio is obtained by
regressing the returns from the spot market transaction on the returns to holding the
hedging instruments over the entire sample period. The second strategy is the time
varying static hedge technique where we use simple twelve month lagged historical
volatility to obtain the time varying hedging rule. We assume that at any given point, the
spot price of a commodity is a function of its futures prices and that of other related
commodities along with time to expiration of the nearest futures contract. In the third
case, we obtain the dynamic hedging rule based on the solution to the GDH model

described above. Next we compare results across the three hedging techniques.

The remainder of the paper is organized as follows. Section 1.2 explains the
rationale for hedging by the ethanol producers. Section 1.3 gives an overview of the
literature on the methodology utilized and relevant research on hedging in the ethanol
market. Section 1.4 describes the dynamic hedging model. Section 1.5 presents the
preliminary summary statistics of the data. Section 1.6 illustrates the empirical
application of the generalized dynamic hedging model and a comparison with constant
and time varying hedging model and also presents the empirical results. Section 1.7

concludes.

1.2 Rationale for Hedging

Ethanol producers are exposed to an array of risks, namely price risk, basis risk, yield
risk, storage risk, institutional risk, ﬁnancial risk and transportation cost risk (Coltrain,
2004). Price risk arises due to price volatility of the inputs and outputs related to ethanol
production. More speciﬁcally, price risk may arise due to the possibility of negative
proﬁtability due to unfavorable combinations of input prices (corn, natural gas) and
output prices (ethanol, distiller’s dried grain with solubles (DDGS)). Each of these
commodity prices present varying degrees of risk depending on their respective price
volatility, the relative amount of each commodity used or produced, and the risk
management tools available to the respective producers. Basis is deﬁned as the difference
between the spot price of a hedged asset and the price of an underlying futures contract.
Basis risk is the risk associated with imperfect hedging using futures contracts (Castelino,
1992). It arises when the settlement prices on the futures contracts at expiration are
correlated with, but not equal to, actual cash prices at which the producer makes
transactions (Lapan and Moschini, 1994). With high degrees of volatility in input and
output prices, producers may be exposed to signiﬁcant basis risk from multiple
commodities.

Institutional risk arises due to government intervention in the form of subsidies,
taxes and regulations. Adoption of unfavorable government policies may inject greater
degree of volatility in the proﬁt margin. Financial risk may also be signiﬁcant since it

limits the ﬁrm’s access to capital and the latest technology. Given the current technology,

yield risk is not a major concern for ethanol producers.l Ethanol producers usually haul
corn from elevators, or directly from corn farmers, to their plants and deliver ethanol and
DDGS to end users. Thus any uncertainties in freight rates expose the ethanol producers
to transportation price risk. In this paper, however, we assume that the producers are
primarily exposed to price and basis risk. Other categories of risk (i.e. yield and storage
risk, institutional and ﬁnancial risk and transportation risk) will be relatively less
important in the case of ethanol production and are left as topics for future research.

Ethanol producers may actively trade in the futures, swaps and options markets as
well as engage in forward contracts to manage risk. Ethanol currently has only two such
securities available for risk management, a futures contract traded on the Chicago Board
of Trade and another on the Brazilian Mercantile and Futures exchange (Paddrik and
Davis, 2008). The ethanol futures market lacks liquidity due to comparatively low
volume of contracts traded. However, with time, the ethanol market is increasingly
gaining efﬁciency with additional new entrants from producers as well as ﬁnancial ﬁrms
trading actively (CBOT, 2009). Currently hedgers in the ethanol markets are able to
hedge nearly two years of production. This enables them to create more long-term
protection and take on more leverage (EAR, 2008).

In a futures market transaction, delivery, or the possibility of delivery, promotes
convergence of the spot price and the futures price at the delivery point at expiration.
However, according to the CBOT, the predominant economic function of ethanol futures
contracts is risk transfer and price discovery, rather than merchandising or title transfer

for the underlying commodity. The CBOT’S ethanol futures contracts are designed to

 

' 1 bushel of corn and 165 thousand British thermal units of natural gas are needed to create 2.7 gallons of
ethanol and 17 pounds of DDGS (Hart, 2005).

provide high degrees of ﬂexibility for market participants throughout the delivery
process. For example, once the shipping certiﬁcate has been issued, there is no speciﬁed
timeframe in which the buyer and seller must exchange physical ethanol.2 Moreover,
small ethanol production cycles (i.e. 40 to 75 hours) and monthly ethanol futures
contracts might induce basis risk between the futures and cash markets.

Standard & Poor’s conducted a risk analysis of all the major factors of ethanol
production (government, yield margins, margin volatility, industry growth, and industry
dynamics). The report concluded that typical ethanol plants demonstrated characteristics
of a “B” rated entity, where “B” indicates that even though the ethanol industry primarily
has a strong foot hold it still has an appreciable margin of risk in its pricing (Paddrik and
Davis, 2008). Thus capital market imperfections coupled with high price volatility
provide incentives to ethanol producers to use hedging instruments to hedge against price

risk (Simmons, 2002; Panell et a1. 2008).

 

2 CBOT Ethanol Futures, The delivery Process, url: http://www.cbot.com/cbot/docs/7440 1 .pdf, viewed on
07/04/08

1.3 Background Literature

Historically, hedging has been one of the most effective risk management strategies used
by competitive ﬁrms to deal with price uncertainty (Peck, 1975; Johnson, 1960). An
optimal hedge ratio is usually deﬁned as the proportion of a cash position that should be
covered with an opposite position on a futures market (Myers, 1991). Static hedge ratios
are obtained by regressing returns in the spot market on the return in the futures market.
Given the restrictive assumptions underlying static hedging, several dynamic hedging
strategies have been developed where hedgers may be able to improve hedging
performance. Anderson and Danthine (1983) generalize the static hedging model to a
dynamic framework that allows the hedger to update the hedge as new information
arrives. The cash-holding period is divided into discrete time intervals such that at the
end of each interval, futures positions are ‘marked to market’ and the size of the hedge
may be adjusted. At the end of the ﬁnal period, the cash position is liquidated along with
any remaining futures positions. However, this study assumes no basis risk and negative
exponential utility with normally distributed random variables. Dufﬁe and Jackson
(1987) solve the optimal futures hedging problem in continuous time settings such that a
hedge is deﬁned as a vector stochastic process specifying a futures position in each
futures market. A hedge is considered optimal if it maximizes the expected utility of
terminal wealth, which is the sum of the market value of a committed portfolio of spot
market assets and the value of the margin account on termination. Mathews and
Holthausen (1991) developed a multiperiod hedging model that allows for periodic

adjustment of the hedge while minimizing the producer's proﬁt variance. The study

concluded that simple models may perform better for simple hedges, while the multi-
period model with adjustable hedges may perform better under complex hedging
situations such as those involving cross hedges. Karp (1987) developed a dynamic
hedging problem with stochastic production. Myers and Hanson (1996) derived an
optimal dynamic hedging rule where futures positions can be updated at regular intervals
over a cash-holding period. The paper assumes cash positions to be nonstochastic and
current futures price to be an unbiased predictor of upcoming futures prices.

To account for commodity price volatility, the possibility of autocorrelation in
spot and futures prices, and to investigate the effect on hedging strategy with the arrival
of new information, dynamic hedging methods may be more appropriate in many
applications. The commonly used approaches for estimating time-varying optimal hedge
ratios on futures markets are the moving sample variances and covariances approach and
generalized autoregressive conditional heteroscedasticity (GARCH) model of Bollerslev
(1986). Over the last decade, several studies on dynamic hedging have demonstrated that
hedging models, particularly GARCH, which factors in the dynamic relationship between
spot and futures contracts, outperforms traditional hedging models in hedging
effectiveness (Cecchetti, Cumby and Figlewski, 1988; Kroner and Sultan, 1991; Park and
Switzer, 1995). Baillie and Myers (1991) adopted the bivariate GARCH to estimate the
optimal hedge ratio. They found that the resulting optimal hedge ratio varies with time
and its effectiveness is better than those of the OLS constant hedge ratio models.

To avoid losing long-run information when dealing with the issues of
nonstationarity, Krehbiel and Adkins, (1993) utilized cointegration to illustrate the long-

run relationship between variables, as suggested by Engle and Granger (1987). If the spot

and ﬁrtures price series are nonstationary in levels, but stationary in ﬁrst differences and
additionally if a linear combination of the series is stationary, then the series are said to
be cointegrated. An Error correction model (ECM) combines the long run, cointegrating
relationship between the levels variables and the short run relationship between the ﬁrst

differences of the variables (Banerjee et al., 1993).

1.3.1 Background Literature on Hedging in the Ethanol Industry

Ethanol was introduced in the CBOT futures trading list in March, 2005. According to
Funk et al. (2008), CBOT ethanol contracts may add efﬁciency to the cash ethanol
market. CBOT futures contracts are a source of pricing information for producers,
marketers, and purchasers of ethanol. Dahlgram (2009) examines the effectiveness of
one-through eight-week ethanol hedges between 2005 and 2008. The study concludes
that ethanol inventory hedging effectiveness is signiﬁcant for two-week and longer
hedges and increases with the hedging horizon. The analysis also concludes that ethanol
futures are signiﬁcantly superior to gasoline futures for hedging ethanol price risk for
two-week and longer hedges. The study by Franken and Parcell (2002) estimated the
cross-hedge relationship between New York Mercantile Exchange (NYMEX) unleaded
gasoline futures price and Michigan spot ethanol price. The cross-hedge ratios varied
from 0.175 to 0.418 depending on the hedging period. The hedging effectiveness
increased for longer cross-hedging periods. The study also concludes that considerably
fewer gallons of unleaded gasoline futures are required to hedge a particular amount of

ethanol. However, with a typical 50 million gallon/year ethanol producing plant, the

quantities of gasoline futures required to implement an effective cross-hedge ratio should
not be a major deterrent for the producers to engage in cross hedging. A study by Brinker
et a1. (2007) used four locations for cash DDGS prices, namely Atlanta, Boston, Buffalo
and Chicago markets. The estimated hedge ratios for the four locations are similar in
value with very little variation in both the corn and soybean hedge ratios. The study
suggests that a combination of both corn and soybean meal futures contracts provides a

hedge that effectively manages DDGS price risk.

10

1.4 Generalized Dynamic Hedging Model

Consider an ethanol plant operating with ﬁxed production capacity. Corn and natural gas
are the main inputs used in the production process, with ethanol and Distiller's Dried
Grains with Solubles (DDGS) being the primary outputs. Other than DDGS, all the
commodities are traded in futures markets.

The producer faces a non-stochastic production process (i.e. no yield risk) with
ﬁxed scale of production. As discussed in the previous section, the producer is exposed to
price and basis risk. Suppose the producer has a ﬁxed cash position for corn, natural gas,
ethanol and DDGS. The cash position for the agent represents a commitment to buy
inputs or sell outputs, in known quantities, in the future. Future prices of the products are
stochastic at earlier periods so there is price risk over the cash-holding period (Myers and
Hanson, 1996). Producers may hedge price risk by establishing futures contracts. Even
though ethanol has been produced and traded in the cash market in the US. for the last
couple of decades, futures have been traded on the CBOT only Since March, 2005. Given
the high correlation between ethanol spot and gasoline and crude oil futures prices,
producers may cross-hedge ethanol price risk by establishing futures contracts in these
commodities. Similarly, DDGS is not traded in the futures market but the producer may
be able to reduce price risk through cross-hedging cash DDGS with any actively traded
futures market commodity (Anderson and Danthine, 1981).

Assuming the cash positions for corn, natural gas, ethanol and DDGS to be ﬁxed
and the futures positions of corn, natural gas, ethanol, crude oil and gasoline to be

stochastic, the decision process for an ethanol producer can be illustrated as follows:

11

t-i t

 

t +j
I I I
I I I
Establish Close out existing hedges All hedges are closed
- - for inputs (corn and natural out, ethanol and
antlcrpatory hedges gas), buy corn and natural DDGS are sold in the

with corn, ethanol,
natural gas, crude
oil and gasoline
futures contracts.

gas in cash market and cash market.
consider updating ethanol,

crude oil and gasoline

futures contract position.

where t —i is the initial time period, 1' is the time span before the production process

starts and j is the time required in the production of ethanol. The ﬁrst stage (t -—i )

corresponds to the initial decision making stage where the producer initially makes
decisions regarding futures positions. The second stage (t) corresponds to the beginning
of the production process when the inputs are acquired in the cash market. The third stage

(t + j) corresponds to the end of the production process where the ﬁnal outputs have

been produced and sold in the cash market.

With small production cycles (j = 40 to 75 hours) and monthly futures contracts,

multiple production cycles are completed between any two consecutive terminations of
ﬁitures contracts. We assume that the ﬁrm completes 10 production cycles in a month
and the entire spot market transactions are completed on the ﬁrst day of the month. The

decision process for the ethanol producer is therefore reduced to a two-period model:

1 . .
t +1+j

l I

I I

All hedges are closed out, corn

and natural gas are bought from

the cash market; ethanol and

ethznohlnatgral gals‘, DDGS are sold in the cash
cm 6 01 an 8330 me. market.

Establish anticipatory
hedges with com,

12

We assume no transaction costs and that the ﬁrm does not have a liquidity constraint.
However it is important to note that even with small production cycles, anticipatory
hedging by the ethanol producers is a rational risk management strategy given the

considerable sunk cost of building the plant (Schmit et a1, 2008).

1.4.1 Multi-Period Model Speciﬁcation

Consider an ethanol producer hedging in the ethanol, natural gas, corn, crude Oil and
gasoline futures markets. Since DDGS is not traded in the futures market, the producer
may mitigate DDGS spot price risk by cross-hedging using futures contracts of related
commodities. To keep the analysis simple, we model the optimal hedging strategy for
ethanol producers assuming ethanol to be the primary output. Let the producer initiate a
multiple period hedging strategy starting at I: 0 for input requirements and expected
outputs from time period t: lonwards, for the next T periods of production. The
producer establishes hedging positions for commodity bundles at each period against a
portfolio of futures contracts with maturities over each of the next T periods. We may
restrict our model by assuming that m inputs are being used to produce n =1 output.
Thus at t= O, the producer establishes positions in T x(m+1) futures contracts for
m + 1commodities for the next T periods. However, as he moves to the next period (i.e.
t =1), he completes all cash market transactions (i.e. buys inputs and sells output in the
spot market), liquidates the futures contract that has the closest maturity date lying

beyond the current month and updates the futures position for the

remaining (T — 1)x (m +1) futures contracts with additional available information. The

13

following timeline illustrates the futures positions established and updated at each period

where btﬂ is a vector of futures positions held at time t for maturity at r .3

 

2 T—3 T—2 T—l T
I l I I . I I
I I I | I | |
b
0.1 hm b2,3
b
b b b bT—3 T—2
”“2 l,T—2 2,T—-2 ’
1103-1 b1,T-—1 b2,T—l bT—3,T—1 bT-2,T—1
l’0,T b1“. b2,.r bT—3,T bT_2,T bT—1,T
Let ft; be the 1x (m+ 1) vector of futures prices corresponding to btﬂ at time t with
expiration at r .
Z Z Z
ft. =[f,3;, ,1 A}. ..... my] [,.(mHH (1.1)
Z Z Z
hm —_- [bag ,b,} 12,} , ..... ,b,,'," ] [lx(m+l)] (1.2)

where y and (21,..., 2”,) represent output and inputs respectively. Let Pt be the vector of

stochastic cash prices for the commodities at time t (positive for outputs and negative

for inputs), At be the vector of spot positions of the commodities at period t(assumed

given) and C t () be the non-stochastic cost function at t that depends on all inputs that

 

3 For example, b2,T represents the vector of futures positions held at time 2 maturing at T for all the

commodities that the producer is interested in hedging.

l4

cannot be hedged (i.e. labor, capital etc). Futures prices of the commodities at all maturity

dates (ft), hedge positions at all maturity dates (ht ), cash prices (Pt) and quantities

(At ) at period tare represented as follows:

ft = I‘LI’ftJ-I-I ’ft,t+2’°""ft,T LIX(m+I)X(T-I+I)] (13)

bt = Ibt,t+1’bt,t+2’"”’bt,T I[1x(m+l)x(T—t)] (1'4)

Pt -_— [13y ,—P,21 ,—P,22 , ..... ,—P,""] [H (m+1)I (1.5)

A =l z1 22 2m] (16)
t yr, t9 1’ """ 9 t [lx(m+l)] .

The futures position is negative for a short hedge and positive for a long hedge. We
assume that there is no initial cost of establishing futures positions because margin

requirements can be satisﬁed with interest-bearing government securities. Let us deﬁne

If” as:

ftL = Flt—13’ft—l,t+1’ft—l,t+2"""ft—1,T le(m+l)x(r_,+1)] (1'7)
The producer’s realized cumulative cash ﬂow (In) at period tis given by:

7:, = (1 + rer,_1 — C,_l )+ Artb',_l + PtA't (1.8)
where r is a constant per period interest rate representing the hedger’s Opportunity cost of

funds, ”H is the producer's cumulative cash ﬂow in period t-l and Aft = ft — ftL. The

producer’s objective is to choose a sequence of futures positions {bt_1 It; to maximize

the expected utility from cumulative cash ﬂow at the end of period t = T ,

15

Max E0u(7rT) (1.9)
{ht—1 IL]

where E0 denotes the expectation operator conditional on information available at t = 0 ,

and u is an increasing and strictly concave von Neumann-Morgenstem utility function.
The maximization is subject to the cash ﬂow constraint (1.8) and can be solved using
stochastic dynamic programming. Consistent with Myers and Hanson (1996), we make

the following four assumptions:

Assumption 1 (Al): The futures market is unbiased in that prices follow a martingale,
ft =1}L +°t where et is a vector of zero mean random shocks which cannot be
predicted and dim(et)= [1 x (m + 1)x (T —t + 1)]. This implies that on an average, hedgers

cannot predict upcoming ﬁitures prices more efﬁciently than the aggregate market
prediction. Given that most studies have found little or no systematic bias in futures

market, this assumption is reasonable for practical applications (Martin and Garcia,

1981).

Assumption 2 (A2): Expectations of cash prices at each period evolve according to

P1. = E,(P,)+ V”, r >t where v” is a zero mean random shocks which cannot be
predicted and dimIvm I: [1 x (m + 1)]. This is a very general assumption and implies that

expectations are consistent. Let us deﬁne v as

t

vt =[(1+ r)T—(‘+1)vt,t+1,....,(1 + r)vt,T_1,vt,TJ (1.10)

16

such that dim(vt ) = [1x (m + 1)x (T -t + 1)]. This might seem arbitrary, but will help us in

the following sections to obtain a more tractable form of the ﬁnal solution.

Assumption 3 (A3): Expectations of innovations in cash prices given by:

E,+h]vt,,]= vt,T + 11”.”, r > t + h where 'lt+h,1- is a zero mean random shocks which

cannot be predicted and dim(1|t+h,, )= [1 x (m + 1)].

Assumption 4 (A4): The innovations in futures prices and expectations of cash prices are

related according to vt = etdt +st , where at is a vector of zero mean random shocks
which cannot be predicted and is independent of etat all lags, and 5t is a matrix of
parameters that may vary over time. Note that dim(8t ) = 1 x (m +1)x (T -t +1) and ﬁt is

a [(m +1XT — t)x (m +1)(T — t + 1)] matrix. This assumption allows for a wide set of basis

relationship between the cash and futures price. Speciﬁcally, it does not require the

ﬁrtures price to be an unbiased predictor of the cash price at maturity.

We determine the optimal hedge ratio using backward recursion. At the last decision

period, T — 1, the hedging problem reduces to a one period problem:

Max ET-l [WI T )I
bT—I

st. ”7 .=(1+r)[7zT_1—CT_1]+AITIIIF_,+1)TA'T (1.11)

Second order conditions for maximization are satisﬁed by the concavity of the utility

function and the ﬁrst order condition is given by:

17

ET_1]u'(zzT)*eT]-.=o (1.12)
where AfT = eT according to assumption (A1). If 7rT and er are independent, then their

joint density factorizes and equation (1.12) may be written as:

ET—lIu'(”T)I* ET—l(eT):0 (1.13)
Thus we need to obtain an optimal decision rule for bT_1 which makes 7rT independent
of eT. Using assumptions (Al) through (A4), the cumulative cash ﬂow constraint may
be written as (for details please refer to Appendix Al .1):

”T = (1+rX7FT-1 “CT-1)+eTbT—l +IET—1(Pr)+eT5T +8TIAT (1-14)

In equation (1.14), er and ST are the stochastic components and by construction, 3T is
independent of err. Let bin] = —5TA:1~; this renders cumulative cash ﬂow independent

of movement of futures prices. Therefore b:r_1 = —6TAZI~ is the optimal hedging rule for
the two period problem. Note that it can be demonstrated that in a two-period scenario,
DTIS the matrix of the ratios of the covariance between one-period innovations in cash

and futures prices to the variance of one-period futures prices innovation, conditional on
information available at t = T —1. This is the standard one period cross hedging formula
that has been used previously in the literature (Kahl, 1983; Benninga, Eldor and Zilcha,
1984)

Following the dynamic programming algorithm, we move sequentially backwards
from the terminal decision date to derive decision rules for earlier periods. Assuming an
optimal strategy will be followed at T —1, the T— 2 period problem may be expressed

as:

18

Max ET_2 ]u((l + r)(7r7~_1 — CT—l )‘I’ [ET-1(PT)+ 8TIAT II
bT-z

SI. ”IT—I = (I + I‘Xﬂ'T_2 - CT-'2)+ AfT_le_2 + PT-IAT-I (1.15)
Assuming the second-order condition is satisﬁed, the necessary condition for the optimal

bT_2 is given by:

(1+r)ET_2]u'(7r7-)*e1~_2]=0 (1.16)
The Objective is to ﬁnd any bT_2 which makes n7 independent of eT_2 will satisfy

(1.16) and therefore be optimal. Let us consider the cumulative cash ﬂow in period

T assuming the optimal one-period rule is used in period T —1 , given by equation (1.14).

”T =(1+rX”T-1‘CT—l)+IET-1(PT)+3TIAT (1-17)
Substituting the expression for ”724 from equation (1.15),
= (1+ r)I(1+ r)(7r7._2 — CT~2 ) + AfT—le—Z + PT-IAT—I _ CT—1I+ [ET-1(PT)+ 8TIAT

2

=(1+ r)2 ”7-2 " 20 + rYCT-i +(1+ rkT—le—Z +(1+ rIPT-IAT—I

i=1

(1.18).
+IET—1IPT)+8TIAT

Following assumption (A2), we substitute PT_1 = ET_2(PT_1 )+v1~_2,1~_1 and

PT = ET-Z (PT ) + vT—2,T into (1.18) yielding.

2 ' V '
=(1+ ”)2 ”T—z — 2(1 + ”II CT—i + (I + rIET-IbT—z +(1+ r)ET—2(PT—1)AT—I (1 19)
i=1 .
+(1+ ’IVT-2,T—IAT—1 + ET—l [ET—2 (PT)+ VT—2,TIAT + STAT

Following the rule of iterated expectations we have:

19

2 ' I '
=(1+ rIZ ”T—z - 2(1 + ”)1 CT-r‘ +(1+ ”FT—IbT—z +(1+ r)ET—2(PT—1)AT—l (1 20)
1:1 . .

+ (1 + rIVT—2,T—1AT—1 + ET-Z (PT)A:1~ + ET—IVT—2,TAT + STAT

Following assumption (A3), we substitute ET_1 IVT—2,T ] = VT—2,T + ﬂT_1,T and obtain

2 ' V 0
=(1+ VIZ ”T—2 — 2(1 + r)I CT—r‘ +(1+ rIET-IbT—z +(1+ "IET-z (PT—1 )AT—l
i=1
+(1+ ’IvT—2,T—1AT-1 + ET—2IPTIAT + (VT—2,,T + 'IT-I,,TIAT + 8TAT
2 . . I
= (1+ r)2 ”T—z - Z (1 + r)! CT—i + (I + rICT-rbT—z +(1+ "IET-z (PT—1 )AT-l

i=1

. , (1.21).
I I AT_1 I
+ 8TAT + ET-z (PT)AT +l(1+ r)VT—2,T-l , vT—2,TI|:A. ]+ ﬂT—1,TAT
T
Let us deﬁne A? as:
T
At = [At , At+l ,..., AT J [lx(m+l)(T-t+l)] (I ..22)

1 A. -
Substituting v1~_l =](1+r)v1~_2,1~_1, VHS] and AL, =] T l]into(21)yields
AT

2 . . '
= (1 + r)2 ”F2 _ 2(1 + r)’ CT_,- + (1 + r)eT_1bT_z + (1 + 7)ET—2(PT—1)AT—l (1 23)
i=1 '

F O T ' V
+ 8TAT + ET-Z (PT)AT + VT—1A1~_1 + 'IT—1,TAT

Following assumption (A4), we substitute vT_1 = eT_16T_l + aT_1 yielding

2 . . '
= (1 + r)2 ”7—2 _ 2(1 + r)’ CT_,- +(1+ r)eT-1b1~-2 +(1+ r)ET—2(PT—1)AT—~l (1 24)
i=1 .

+3TAT + ET-Z (PTIAT +(eT—15T—1 +3T—1)A¥_1 + ﬂT—I,TAT

Thus to make 727* independent of e74 , we let

20

* '

bT—z = '(1+ r)_1 5T—1AT—1

More generally, considering a multi-period model, we have:

*

b',_l =—(1+r)"Ta,A,T (1.25).
These decision rules make terminal wealth independent of futures price movements at
every t, assuming optimal rules will be followed in the future when positions are

updated. According to equation (1.25), the optimal hedging rule is a function of ﬁt , the

matrix of the ratios of covariance to variance weighted by a discount factor, (1 + r)"T

and a matrix of cash positions. This reﬂects the fact that proﬁts or losses from futures
price movements are marked to market in every period. Moreover, the hedging rule
obtained generalizes the dynamic hedging rule obtained by Myers and Hanson (1996) by
considering multiple commodities, accommodating for multiple cash market transactions
and allowing for cross-hedging. It is to be noted that if we reduce the multivariate set up
to a single commodity frame work with single cash market transaction at the terminal
period, dynamic hedge rule obtained in equation (1.25) reduces to that obtained by

Myers and Hanson (1996) (page 16).

21

1.5 Data and Preliminary Statistics

Monthly price data from January 1993 to May 2009 has been used in the analysis.
Ethanol spot price data has been obtained from ofﬁcial Nebraska Government website
(Nebraska Government Website, 2009). The data source for corn and DDGS spot price is
the United States Department of Agriculture (USDA, 2009). Futures price data for corn,
natural gas, gasoline and crude oil has been obtained from Commodity Research Bureau
(CRB, 2009). Spot price data for natural gas and futures price data for ethanol was
obtained from Reuters (Reuters, 2009). At any given point of time, ethanol and natural
gas has twelve outstanding futures contracts and corn has ﬁve futures contracts. Since
multiple contracts usually trade simultaneously for a given commodity, futures prices
form multiple time series corresponding to different contracts. Researchers usually follow
‘nearbys’ or ‘constant-maturity price series’ to convert discontinual futures price data
into continual time series (Holton, 2003). Following Park and Switzer (1995) and Myers
and Thompson (1989), to estimate the constant hedge ratio and the time varying static
hedge ratios, we consider the closing price of the nearby futures contract on the ﬁrst
Wednesday of the month. Historic monthly risk free rate of return was Obtained from
Wharton Research Data Services (WRDS, 2009).

Analyses of the time series properties of the historic price data are important in
choosing the best model speciﬁcation (Koutmos and Pericli, 1999). Descriptive statistics
for the price series are reported in table (1.1). Initial data analysis revealed high
correlation between crude oil and gasoline futures price, even in ﬁrst differences. Thus, to

avoid the problem of multi-collinearity, our analysis does not include crude oil futures

22

contracts. The estimated skewness and kurtosis for the times series indicate departure
from normality. Corn spot and futures price series exhibit evidence of fat tails, since the
kurtosis exceeds 3, which is the value under normality. The time series data, in general,
exhibits positive skewness, which implies that the right tail is particularly extreme.

The Shapiro-Wilk test examines non-nonnality in small to medium sized samples.
Table (1.1) reports the test statistics (W). Normality is strongly rejected for the spot as
well as the futures price data of all commodities. The Anderson-Darling test for normality
is also designed to detect departures from normality. According to the test statistics
reported in table (1.1), with the exception of ethanol futures price series, the rest of the
time series data are likely to be from non-normal distributions. Non-normality may be
due to temporal dependencies in the return series. Next we also conduct the Ljung-Box
test that tests whether there is any autocorrelation in the series or whether the
observations are just a set of random, identically distributed variables (Baillie and
Bollerslev, 2002). Table (1.1) presents the Ljung- Box test statistic(Q). According to the
P- value, we can say that there is autocorrelation in the series.

Several diagnostic checks on the distributional properties of the data were
performed to test for stationarity and presence of unit root for each of the price series.
The Durbin-Watson Test for serial correlation assumes that the disturbances are
stationary and normally distributed with mean zero. Since the reported Durbin-Watson
test statistics are far from 2, there is strong evidence that the errors are strongly
autocorrelated.

Next, we conduct the augmented Dickey-Fuller test (ADF) and the Phillips-

Perron Test (PP) to test for presence of unit root in the univariate time series. Table (1.2)

23

reports the results for the spot and futures price series data for levels and the ﬁrst
differences. According to the ADF test results, some of the price series have unit roots
and become stationary on ﬁrst differencing. The Phillips-Perron test is a test of a unit root
hypothesis on a data series. It reports three types of test statistics corresponding to zero
mean, single mean and trend. According to the Phillips-Perron test statistics, all the price
series are non-stationary in their levels and that with ﬁrst differencing, price series are
rendered stationary. Thus the PP test conﬁrms the presence of unit root in the time series
data. The results, in general, suggest that all variables are integrated of order one
(i.e. 1(1)).

Prior to testing the existence of cointegration in the model, we conduct pair wise
cointegration test among spot and futures prices while accounting for time to maturity

(T—r) (Engle and Granger, 1987). Time to maturity is included in the equilibrium

relationship to reﬂect the fact that cash and futures prices are expected to converge at
maturity (Moschini and Myers, 2002). Table (1.3) presents the results of these tests.
According to the trace statistics and maximal eigen value statistics, with the exception of
ethanol, natural gas, and corn spot against DDGS spot; and corn spot against gasoline
futures, the rest of the spot and futures price series are cointegrated at the 5% signiﬁcance
level. To further explore possibility beyond pair wise cointegration, we test the
hypotheses whether spot and futures prices are cointegrated, sharing a common stochastic
trend (Engle and Granger, 1987).

Table (1.4) reports the Johansen Cointegration Test Results among the spot and
ﬁrtures price series. The ﬁrst row tests the null hypothesis r = 0 against r > Owhere

rrepresents the number of linearly independent cointegrating vectors; similarly the

24

second row tests r S lagainst r >1.The trace test statistics and the maximal eigen value
statistics are reported in columns 2 and 4 respectively. Results reported in Table (1.5)
indicate that there is one cointegrating relationship among the seven price series.
According to the results there is a feedback relationship between ethanol and gasoline
ﬁrtures and ethanol spot, corn and ethanol futures, Signiﬁcant at 1% level of signiﬁcance.
Causal relationships have also been revealed between corn spot and natural gas futures,
natural gas spot against DDGS spot, ethanol spot and natural gas futures at 5% level of

signiﬁcance. The rest of the feedback relationships may be interpreted accordingly.

25

1.6 Empirical Application and Results

1.6.1 Constant Hedge Model

Following Chiu et a1. (2005) and Myers and Thompson (1989), the traditional regression-

based hedging model may be represented as:

Mr" = a3 + aim + 6 Afi‘ + Arm:1 + ZrtAf.”(T)+ v... (1.26)
h

where AP,x and Aftx represents the rate of return for the spot and futures market for the

x'h commodity at period t, respectively, T represents the nearby expiration date, ,u,_l is

the error correction term, 1),“, is the prediction error, 6 is the hedge ratio,

and a; , alx , px , y}, are parameters to be estimated.

While estimating the constant hedge ratio, we partition the data set such that the
ﬁrst set corresponds to the time period when ethanol futures contracts were not available
to the producers to hedge risk (i.e. January, 1991 to March, 2005) and is referred to as
‘Regime 1’. The second data set refers to data from April, 2005 onwards, when ethanol
futures contracts are available and is referred to as ‘Regime 2’. We assume that the
ethanol producer may cross hedge ethanol spot price risk using natural gas, corn and
gasoline futures contracts in both the regimes.

Once the cointegrating relationships and the time to maturity effects have been
evaluated, we estimate equation (1.26) (Krehbiel and Adkins, 1993). We derive the hedge

ratio for ethanol, corn and natural gas assuming that in addition to the commodity’s own

26

futures contracts, the ethanol producer can also cross hedge ethanol spot price against
gasoline futures contracts.4 Results from constant hedge models are displayed in table
(1.6). In regime 1, we cross hedge returns from ethanol spot market against returns from
natural gas, corn and RBOB gasoline futures markets. After correcting for
autocorrelation, the constant cross-hedge ratio to hedge returns from ethanol spot market
using returns from natural gas futures market is 0.0842, signiﬁcant at 1% level of
signiﬁcance. This implies that for every gallon of ethanol produced and sold in the spot
market in regime 1, on an average, the producer may engage in 0.0842 thousand cubic
feet of natural gas futures contracts. Effect of error correction term with respect to natural
gas futures is signiﬁcant at 1% level of signiﬁcance. Results also indicate that lag of
ethanol spot price and natural gas futures price series are important in hedge ratio
determination, signiﬁcant at 1% level of signiﬁcance.

In the second regime ( i.e. April, 2005 to May, 2009), the ethanol producers may
use returns from ethanol futures market as the additional hedging instrument to hedge
against returns from the ethanol spot market. The constant hedge ratio using ethanol and
corn futures markets are 0.3018 and 0.0015 respectively, both signiﬁcant at 5% level of
signiﬁcance. This implies that for every gallon of ethanol produced and sold in the spot
market in regime 2, on an average, the producer may engage in 0.3018 gallons of ethanol
futures contracts and 0.0015 bushels of corn futures contracts. Since 1 bushel of corn
yields 2.7 gallons of ethanol, the cross-hedge ratio using corn futures contratcs is.

expected to be much less than unity (Hart, 2005). Results are consistent with those in

 

" No signiﬁcant difference in results was observed when the analysis was conducted using crude oil futures
contracts instead of gasoline futures contracts.

27

Dahlgram (2009) and Brinker et al. (2007). Effect of lag ethanol spot price is also
signiﬁcant at 1% level of signiﬁcance implying the persistence in spot prices (table 1.6).

It is to be noted that there is a sudden drop in constant cross-hedge ratio for hedging
ethanol spot prices against natural gas futures contracts from 0.0842 to 0.0167 as we
move across regimes (ﬁgure 1.2). This coincides with the time period when ethanol
futures contracts were introduced in the CBOT commodity trading list. Moreover, the
constant hedge ratio estimates for hedging returns from natural gas spot market against
natural gas futures contracts jumps up from 0.3798 in regime 1 to 0.6849 in regime 2,
both signiﬁcant at 1% level of signiﬁcance (ﬁgure 1.5). Similarly, results indicate that
there has been an increase in constant hedge ratio estimates for hedging returns from corn
spot prices against corn futures prices from 0.0025 in regime 1 to 0.009 in regime 2, both

signiﬁcant at 1% level of signiﬁcance (ﬁgure 1.6).

1.6.2 Time Varying Static Hedge Model

To estimate the time varying static hedge model, we estimate equation (1 .26) for different
sample periods of twelve month windows starting from January, 1991 to May, 2009
(Myers, 1991). For example, the ﬁrst estimate is obtained by estimating equation (1.26)
for the time period of January, 1991 to December, 1991. Similarly, the second estimate is
obtained by estimating equation (1.26) for the time period of February, 1991 to January,
1992.

Consistent with constant hedge ratio estimation, we derive the time varying static

hedge ratio for ethanol, corn and natural gas assuming that in addition to the

28

commodity’s own futures contracts, the ethanol producer can also cross hedge ethanol
spot price against gasoline futures contracts. Figure (1.2) illustrates estimated optimal
cross-hedge ratio paths for hedging ethanol spot prices against natural gas futures
contracts using moving sample variances and covariances over the ﬁrst and second
regime. Similarly, ﬁgure (1.3) illustrates the optimal hedge ratio paths for hedging
ethanol spot prices using nearby corn futures contracts. Given that com futures contracts
provide effective hedging instrument to hedge ethanol spot price risk in the second
regime (i.e. April, 2005 to May, 2009), for analysis of cross-hedge ratio using corn
futures contracts, we focus our attention exclusively on the second regime. Results
indicate that the time varying static hedge ratio towards the beginning of the second
regime is large and volatile. However, from end 2008 onwards, the hedge ratios seem to
converge to the corresponding constant hedge ratio estimate, (i.e. 0.0015). Figure (1.4)
displays the time varying hedge ratio path to mitigate ethanol spot price risk using nearby
ethanol futures contracts in regime 2. Even though the constant hedge ratio over the
sample period is 0.3018, the time varying static hedge ratios to hedge returns from
ethanol spot market using returns from ethanol futures market exhibits variability over
time. This may be due to the volatility in ethanol futures trading volume over the time
period (Reuters, 2009).

Figure (1.5) and (1.6) illustrates the optimal hedge ratio path for hedging natural
gas and corn spot price series against near-by natural gas and corn futures contracts
respectively. Results suggest a change in hedging strategy by ethanol producers to hedge

its input and output spot price risk.

29

1.6.3 Generalized Dynamic Hedge Model

For a simple illustration of the GDH rule derived in this paper (equation 1.25), we restrict
our empirical analysis to two commOdities (i.e. ethanol, the primary output and natural
gas, one of the primary inputs) and a three period time frame. Given that ethanol futures
contracts are available for trading only since March, 2005, we restrict our analysis in

regime 2. Consistent with equation (1.3), let us deﬁne ft =]ft,t+1,ft,t+2] where

ethn
ft 1+1 lft,t+lf tt,+lI' Thus

ft 41::th frng ,t+l frettﬁg’f ft ,t+2lI1X4I (1’27)'
In deriving the optimal hedging rule for the GDH model, key assumptions made are
given by (A1) through (A4). During empirical estimation, it may be desirable to impose
these assumptions. In a two commodity-three period framework, assumption 1 (A1) may

be written as:
ethn ethn
IAf t t+l ’ Aftng,t+1 ’ Aft, t+2’ Afrng t+2 (1.23).
The price vector given by equation (1.5) may be restated as:
th n
Pt = [pf *1 ,_p, 8]“le (1.29).

Following assumption 2 (A2), we can restate price in period t +1 and t + 2 as

(1+r)Pt+1=Er((1+r)Pt,+1)+(l+r)vtt+l and

PHI- _ EI(P t.‘+2)+vtt+2

Combination of the two price vectors may be stated as:

[(1 + VIP1+1sPt+2I= E1I(1+’)Pt+1aPt+2I+ I(1 +r)vt,t+lavt,t+2I

30

[Ft]=Et[l:t]+vt (1.30)

where Pt =I(1+ r)Pt+19Pt+2I and vt =I.(1+ r)"t,t+l th,t+2 I
Utilizing the relationship between vt and et under assumption 4 (A4), equation (1.30)

may be written as:
[Pt]=Et[Pt]+et8t +et (1.31).

A more appropriate way of expressing equation (1.30) would be to include an additional
set of control variables that include additional information available. More speciﬁcally,

equation (1.31) may be written as:

[Pt]=E,[Pr]+orXt +Aft8t +8, (1.32)

~

where X t is a set of variables in the information set predicting Pt ,0: is a vector of

unknown parameters and ﬁt is a 4 x 4 matrix. This is a very general representation of the

estimated model that has been implemented in this paper and can incorporate any

particular model explaining spot prices. For example, to model spot prices with AR(1)

speciﬁcation, we model E , [Pt] in the following way:

Pt =ﬂPt-1'I'Yt (1.33).

Then the estimated model would be

Pt =ﬂPt—I‘I'aXt +Aft5t+pt

31

The OLS estimate of ﬁt may then be substituted in equation (1.25) to obtain the optimal

*

dynamic hedge bt_1 in period t— 1.

Next we attempt to empirically examine the GDH model developed in this paper.
Figures (1.7) and (1.8) reports the persistence in ethanol and natural gas spot price. The
persistence in ethanol Spot price ranges from 0.1 to 0.89. However, the persistence in
natural gas spot price has a bigger range (i.e. from 0.07 to 1.2). Figure (1.9) illustrates the
dynamic path of hedge positions ethanol producers may establish to hedge returns from
ethanol spot market using returns from natural gas and ethanol futures contracts. Results
indicate that paths representing dynamic hedge positions set in period 0 with maturity in
period I (the graph represented by *) and hedge positions set in period 0 with maturity in
period 11 (the graph represented by +) follow each other closely. However, the optimal
hedge path representing updated hedge position set by the producer in period I with
maturity in period II (the graph represented by #) deviates from the former two graphs.
This may reﬂect the effect of new information arrival on the optimal hedge path,
assuming no transaction cost for updating hedge positions. Similarly, ﬁgure (1.10)
illustrates the optimal path of hedge positions to mitigate natural gas spot price risk using
natural gas and ethanol futures contracts. In this time period, the static hedge ratios to
mitigate ethanol and natural gas spot price risk are -1.1 and .03 respectively.

Figure (1.11) compares the cumulative cash ﬂow earned by the ethanol producers
using different hedging techniques. The graph (represented by ‘*’) illustrates the
cumulative cash ﬂow path under the three-period dynamic hedging model. Similarly, the
graphs (represented by ‘x’ and ‘#’) represent the cumulative cash ﬂow under no-hedge

and constant hedge strategies respectively.

32

1.6.4 Performance Test

Performance test results for constant hedge model, time varying static hedge model and
GDH model as compared to a ‘no hedge’ scenario are reported in table (1.7). Given the
ﬂuctuations in cumulative cash ﬂow caused by volatility in ethanol, natural gas and corn
spot prices, the ethanol producer establishes futures contracts to minimize spot price risk.
The producer’s wealth at the end of the period equals the sum of net gain on futures
transactions and net cash value from spot market transactions. The futures positions can
be adjusted on a monthly basis. Performance is evaluated in terms of effects on the
variance of the producer’s cumulative cash ﬂow position for the three types of hedging

techniques compared to ‘no hedge’ strategy.

1.6.4.1 Constant Hedge Model

Results indicate that in regime 1, producers may utilize constant hedging technique to
hedge returns from ethanol spot market using natural gas futures contracts. This
decreases variance of cash ﬂow returns by 0.2902%. Interestingly, in regime 2, utilizing
corn and ethanol futures contracts as hedging instruments, the producer may be able to
decrease variance of cash ﬂow returns by 3.2819%. However, magnitude of variance
reduction and direction of change varies across natural gas and com. If the producer
engages in constant hedging technique to hedge natural gas spot price against natural gas
futures contracts, variance of cumulative cash increases by 3.0354% in regime 1.
However, it decreases by 2.2805% in regime 2. Similarly, utilizing constant hedge

strategy, the producer may mitigate returns from corn spot market using returns from

33

corn futures contracts. Results indicate a decrease in variance of cumulative cash ﬂow
returns by 3.7722 and 8.7670 percentage points in regime l and 2 respectively.

Next we consider the effect of hedging on the portfolio comprising of ethanol as
the primary output. and natural gas and corn as the two primary inputs. Under constant
hedge strategy, variance of cumulative cash ﬂow decreases by 2.1216% and 0.9978% in

regime 1 and 2 respectively.

1.6.4.2 Time Varying Static Hedge Model

According to the analysis, in regime 1, hedging ethanol spot price using time varying
static hedge strategy decreases the variance of cash ﬂow returns by 0.2315%. Similarly,
utilizing this strategy to mitigate natural gas spot price risk using natural gas futures
contracts, the producer is able to decrease variance of cumulative cash ﬂow by 0.0944
and 1.0155 percentage points in regime l and 2 respectively. Results for hedging corn
spot prices may be similarly interpreted. Next we analyze the performance of time
varying static hedge strategy while considering the portfolio of commodities (i.e. ethanol,
natural gas and corn). Variance of cumulative cash ﬂow decreases by 1.9514 and 4.1662

percentage points in regime l and 2 respectively.

1.6.4.3 Generalized Dynamic Hedge Model

Next we analyze the performance of the GDH model developed in this paper. To
demonstrate a simple application of the model and test its effectiveness, we consider two
commodities, namely ethanol and natural gas and limit the analysis to three periods.

Results indicate that producers can effectively reduce variance of cash ﬂow return by

34

13.2105% by engaging in the three period hedging strategy as opposed to the ‘no
hedging’ technique. Similarly, the producer may reduce the variance by 11.4222% under

the constant hedge rule, assuming no transaction costs.

35

1.7 Conclusion and Policy Implications

This study develops a discrete-time dynamic hedging model for ethanol producers in a
multiple commodity-period framework under expected utility maximization and basis
risk. The basic assumptions made are futures markets to be unbiased and the size of the
cash positions to be nonstochastic. A typical ethanol producer, to mitigate price risk for
inputs and outputs, initiates a multiple period hedging strategy at initial time period for
the next T periods of production. As he moves to the next period, he completes all cash
market transactions and liquidates the futures contract that has the closest maturity date
lying beyond the current month. With new information available, he updates the futures
position for the remaining futures contracts. We derive the sequence of optimal hedging
positions at each time period that minimizes the variance of cumulative cash ﬂow of the
ethanol producer at the end of periodT . The model concludes that the Optimal dynamic
hedge depends on interest rates, the length of time remaining in the cash-holding period,
and a particular covariance-to-variance ratio. The hedging rule obtained generalizes the
dynamic hedging rule obtained by Myers and Hanson (1996) by considering multiple
commodities, accommodating for multiple cash market transactions and allowing for
cross-hedging.

Next the study compares dynamic hedge ratios and hedging effectiveness using
the optimal hedging rule derived in this paper and the dynamic error-correction hedging
model utilizing moving sample variances and covariances. Given that ethanol was
included in the CBOT futures trading list on March, 2005, the ethanol producers has the

option to hedge against ethanol futures contracts from March, 2005 onwards. Empirical

36

results Show that natural gas futures contracts may have been a good cross-hedging
instrument prior to this time period. However, post March, 2005, ethanol producers may
effectively hedge using ethanol futures contracts and cross hedge using corn futures
contracts. Natural gas spot and corn spot price risk may be hedged effectively using their
respective futures contracts. The performance test results indicate in regime 1, under
dynamic hedge strategy, variance of cumulative cash ﬂow of the portfolio decreases by
1.9514%, where as, it decreases by 2.1216% under constant hedging strategy. However,
in the second regime, with ethanol futures contracts being the additional hedging
instrument available, the variance decreases by 4.1662% and 0.9978% under dynamic
and static hedge strategies respectively.

Dynamic hedging model performs better than the ‘no hedge’ scenario in reducing
the producer’s variance of cumulative cash ﬂow. For a simple illustration of the hedging
model derived in this paper, we consider a three-period and two commodity framework.
Performance analysis indicates a 13.2105% decreases in variance of cumulative cash
ﬂow compared to the no-hedge scenario where the producer only engages in cash market
transactions. Similarly, the producer may effectively 11.4222% in variance of cumulative
cash ﬂow employing static hedging strategy. Given the data limitations, results are very
encouraging.

The solution to the dynamic hedging model derived under the generalized set up
is most effective provided there is little or no uncertainty surrounding the size of the cash
position being hedged and also if the assumption of an unbiased futures market is a good
approximation. We also assume no transaction cost for setting up and updating hedge

positions. An extension of the current work may be to derive the optimal dynamic

37

hedging rule relaxing the above assumptions. The empirical analysis is restricted to the
commodity set related to the ethanol industry. The effectiveness of the hedging strategy
may be tested using a wider set of commodities. Moreover, this paper considers futures
contracts as the risk management tool. Future research work may consider discussion and
analysis of options and forward contracts as alternative risk management tools by the
agricultural producers and analyze relative efﬁciency of each of these contracts with

respect to futures contracts.

38

Chapter 1 Appendix

39

A1 .1 Derivation of Cumulative Cash Flow

The culumlative cash ﬂow in terminal period T is given by equation (1.11):

7:, =(1+errT_1 —CT_1]+ArTIr1r_l +PTA'T (1.11)
Using assumptions (Al) and (A2), the cumulative cash ﬂow constraint may be expressed

as:
”T =(1+r)(”T—1 ‘CT—II'I‘eTbT—l + [ET—1(PT)+ VT—1,TIAT (1-11a)

”T =(1+ ’X”T—1 - CT—1)+ eTl’:1‘—1‘I'IET—1(PT)+ VTIAT (1-11b)

Using assumptions (A4), equation (1.11b) may be restated as:

”T =(1+rX”T—1 —CT—1)+eTbT—1+IET-1(PT)+9T5T +8TIAT (1-14)

40

b98238 £32 £1 use e\em £2 “a cacao—Lima $5852 I... use 3. .8

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

“mo“ 2883

ommd emod 2: .o mm_.o god 3: .o NNod «nod 1:5:5

note—oboo From 5% much

* * * _.. * ANCO “wow

:35: Immémﬁ 12.3.22 1.2.33 Immgmf 1.5—wig 13.32 13.2“: xomrwczﬁ

9.:qu

$de 1.03.”: ERONQS fLong aimwmwﬁz **omov.N “1.35.2 *Lﬂoei rcowsmvcax

93

Ineomod .Iiomhhd ,...,...vmwwd 3.1% y mud ***vmmw.o ***ow©o.o Simwvwd *iaaoahd xzatohamnm

38H b:mEoZ

33,— KEN anew; 2 56 move; 83d- mg : mnevé mﬁotsv—

ovmad wave; 2 mm; SwNN wmwv; Sand 38; food mmoEpB—m

255m vowﬂm omw _ .3 comma: 8me 88.03 83.2 83.6 825qu

89: 3:6 cow _ ._ 83.3: ooood 250.3 mum _ ._ oo E o.“ E3552

5:335

336 336 moEN hmcodw Q? We oMNNJVm vMow.~ wwewd vac—Sum

no: .N 236 was. mmmm. Km 8:; 855.2: 3 8% Rood :32
.0525 25030 mew 3.562 :80 655m mmvﬂ mew .8382 E00

325 3.32m mouth 8mm

 

 

 

moumuﬁm bmﬁﬁsm

H; 033.

41

£03838“ 20>“: .x; can fem .ﬁeo— an 858..“in 35858 1; was ..i ._._

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

*...*wo.h- ***wo.hr LWMr *va- €1.26- .I.*mN.mt *mm.mt *VON- 355m
***wo.NT 315.2- 3N- 3;. ...Ime. *uuovw- ...om.m- 3;. 2508mm
mm
3.3de ILQE- *mw.m- Mamm- ILNA T ***NN.:- ...vod- EN- 3532
star-.3- 12;va 3N- cm.” ...Ide- teamma- mN.N- 3.7 880
Saw worm 8:5an
12.0067 ***wm.o- ood- mN.N- ***wo.N- 1...; ~.NT **mN.vt ...Qo.m- Endgam—
***we.o7 .........$.oT SN- \rmN- ...Ivmd- 31.2w- wa. T mod- mOQQw
mm
1.693- 130.3- omd- no.7 traw- ***m¢.w- 59m- mm._- REEZ
IL.de 1.5de $4- EL- :LBB- 1.15.5- owd- ...NQN- :80
8% out; Sam
32% ESC- uw “SEE. use;
a. E0885 30225 50885 A 385:: as 3080:: 30885 do “380:: ﬁ 30885
3280me SEE 354 8055me SEE _o>o-H
3:62 “mom. mm 3.32 “mo-r LQ<

 

 

seem E5 5.2 Ease see EEC Seem-WEE Ea E95 LEE-+885 Sesame...
S 2.3

42

Table 1.3
J ohansen Test Results for pair wise Cointegration between Spot and Futures
Prices, accounting for ‘Time to Maturity’

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Pair of prices Number of Null Trace Maximal Eigen
lags in VAR hypothesis Statistics Value Statistics

Ethanol Spot/ Ethanol 2 r = 0 47.3 809** 38.0423M
futures r S1 9.3386 8.9990

r S 2 0.3395 8.9990
Ethanol spot/ DDGS spot 2 r = 0 10.0418 9.6158

r S 1 0.4260 0.4260
Ethanol spot/ Gasoline 2 r = 0 42.9541 ** 38.2552M
futures r S 1 4.6989 4.5759

r S 2 0.1230 0.1230
Ethanol Spot/ Natural gas 2 r = 0 46.8761 ** 35.0760“
futures rSl 11.8001 11.6121“

r S 2 0.1880 0.1880
Ethanol Spot/ Corn futures 2 r = 0 75.2361 ** 65.0783“

r S1 10.1578 10.0958

r S 2 0.0620 0.0620
Natural gas Spot/ Natural 2 = 0 67.9993” 51.6840"
gas futures r 31 16.3152“ 16.0424“

r S 2 0.2728 0.2728
Natural gas Spot/ Ethanol 2 = 0 49.2950“ 40.2994“
futures r s 1 8.9956 8.7722

r S 2 0.2234 0.2234
Natural gas Spot/ Com 2 r = 0 65 .4804" 60.83 80**
futures r 31 4.6424 4.6138

r S 2 0.0285 0.0285
Natural gas Spot/ DDGS 2 = 0 4.7151 4.7073
SPOt r S 1 0.0078 0.0078
Corn Spot/ Corn futures 2 r = 0 132.7806" 86.0461 **

r S1 46.7345“ 46.5533"

rS2 0.1812 0.1812
Corn Spot/ DDGS spot 2 = 0 22.3451 18.6640

r S1 3.6811 3.5583

r S 2 0.1229 0.1229
Corn Spot/ Gasoline futures 2 r = 0 15.9986 11.2633

r S 1 4.7354 4.6665

r S 2 0.0689 0.0689
Corn Spot/ Ethanol futures 2 = 0 42.5180“ 40.1154M

r S 1 2.4027 2.3425

r S 2 0.0601 0.0601
Com Spot/ Natural gas 2 r = O 35.8036M 28.7524M
futures r 31 7.0512 6.9845

r S 2 0.0667 0.0667

 

 

 

 

 

 

** implies signiﬁcance at 5% level

43

 

Table 1.4
J ohansen Cointegration Test Results among the Spot and Futures Price
Series, accounting for ‘Time to Maturity’

 

 

 

 

 

 

 

 

 

 

Null Trace 5% Critical Maximal Eigen 5% Critical
Hypothesis Statistics Value Value Statistics Value

r = 0 143.4892” 140.74 52.3455“ 47.99

r S1 91.1436 109.93 30.6705 41.51

r S 2 60.4731 82.61 22.5775 36.36

r S 3 37.8956 59.24 17.9234 30.04

r S 4 19.9723 39.71 10.8647 23.80

r S 5 9.1075 24.08 6.3021 17.89

r S 6 2.8054 12.21 2.3242 11.44

r S 7 0.4813 4.14 0.4813 3.84

 

 

 

 

 

 

** implies signiﬁcance at the 5% level

44

303838. 296. «x; can {an ..xoo on 8585:»? 35852 *1. can i. a.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

853m

- oom_.o- omNoo- woooo- Iimmo: Noooo momoo- wmomo 3535 <
85.55

195.? - ono- Eooo *Ioomﬁo :ooo- Noooo- moooo- oczomaO <
355

32.? Send. - voooo «do: 33.? mnovo good. 80 EBEZ<
855m

.1. ooﬁmom- Ivomﬁnm- EmoN - ...Iomnmow whoﬁo med. 2mg: :80 <
Sam

iknmﬁo- omooo ionooo- ﬂood - ooooo- anoo- who H .o- 656m <
mnmmd- wmoow- food- ohooo wwwooy - 2 Go- ommod 8mm mOQQ <
8am

omnNo- Somo- Iomvﬁo ovooo- imwwwo 135.? - «poo- 30 35324
mummo- mmo _ .o- 3N3 ﬁo- emooo- vwmvo *moooo oovoo- - Sam :80 <
85an BBSL 3:5an 95 moBSm 89m ..on 8mm 95 8mm Boat»;
BEBE < 0:28.90 < 3352 4 500 < 3.85m < mOQQ < Eh5mZ< :80 < E09530

 

 

©2080 6602 compooboU-Hobm Hog-00> so oommm 338m 5:830 3.895%
mg 2an

45

Table 1.6
Constant Hedging Models

 

Hedgi_ng ethanol spot price using Near-by futures contracts

 

 

(1991, January — 2005, March)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Intercept 0.8436***
Error correction term with respect to natural gas futures price 0.8716***
Difference in natural gas futures price 0.0842***
Difference in com futures Lice 0.0000
Difference in RBOB gasoline futures price -0.0337

_I_._agethanol spot price -0.8232***

3g natural gps futures price 0.0637***
(2005, April - 2009-May)

Intercept -0.0 129
gig ethanol spot price 0.3491 ***
Difference in ethanol futures price 0.3018"

Difference in natural gas futures price 0.0167
Difference in com futures price 0.0015“
Difference in RBOB gasoline futures price 0.0022
Lag natural gas futures mice 0.0167

 

Hedging corn spotJirice usingjear—by futures contracts

 

 

(1991, January — 2005, March)

 

 

 

 

 

 

 

 

 

 

 

 

 

Intercept 1.8598***
Error correction term with respect to natural gas futures price 0.7030***
Difference in corn futures price 0.0025***
Difference in natural gas futures price -0.0330***
_I:gg corn spot price -0.9051***
__Lagcom futures price 0.2088***
3g natural gas futures price -0.0440***
(2005, April - 2009-May)
Intercept 0. 1697
Difference in corn futures price 0.0090***
Difference in naturaljas futures price -0.0159
Lag corn spot price -1.0539***
3g corn futures price 0.9989***

 

 

Hedging naturaljgs spot price using Near-by futures contracts

 

 

(1991, January — 2005, March)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Intercept 3.4625***
Error correction term with respect to natural gas futures price 0.3489***
Error correction term with respect to corn futures price 0.2531 ***
Difference in natural gas futures price 0.3798 ***
Difference in corn futures price -0.0019***
rlpgg natural gas spot price -0.6602***
ﬁg natural gas futures price 0.3793***
ﬁgcom futures mice -0.1819***
(2005, April — 2009-May)
Intercept 0.0850
Difference in natural gas futures price 0.6849***
Difference in corn futures price 0.0012
3g natural gas spot price -0.64 l 8*"
Egnatural gas futures price 0.4135
Lag corn futures price 0.0255

 

*, ** and *** represents signiﬁcance at 10%, 5% and 1% levels respectively

46

 

Comparison of the Effects of Hedging on Cumulative Cash Flow across

Table 1.7

 

 

 

 

 

 

 

 

 

 

 

 

Models*

Conventional Hedge Models

[ Moving Sample Hedge l Constant Hedge
Regime 1 (January-1991 to March, 2005)
Ethanol Spot -0-2315 -0.2902
Natural Gas Spot -0.0944 3.0354
Corn Spot -1.1748 -3.7722
All commodities -1 .6814 -2. 1216
Regime 2 (April, 2005 to May, 2009)
Ethanol SJ)0t ~7.4565 ~3.2819
Natural Gas Spot -1.0155 -2.2805
Corn Spot -2.4579 8.7670
All commodities -4. 1662 -0.9978

 

Dynamic Two Period Hedge Model (Three Periodslgpril, 2005 to May, 2009)

 

Dynamic Model

Constant Hedge

 

 

Variance in Cash Flow

 

-13.2105

 

-1 1.4222

 

*Represents percentage change in variance of cumulative cash ﬂow using different hedging techniques

compared to ‘no hedge’ scenario

47

 

 

 

526m 698: 25m 26m 636... 2825 no;

 

to... on. auto...

 

 

 

wmmmmmmmmmmmmmmmm
wwwwwwwwwwwwwwwmw
5f 8
.. a j 5 xi... 36%....3», a... a» .. gm
. 4 «3% m
.36
-m.o
-v.o

88:62 - 33 66.136228
menus". new .9352 >a-emoz 9.33 3.5 .ocmcum 9.562.. macho

 

 

 

moom .432 o... goo-banana £68300 maﬁa-a $0 3.532 3-302 gamma 3mm 355m wamwoom $80
3 2:5

48

 

 

 

 

03mm manor oomuw 1:1 comm mono: 22.930 1014

 

ocean. 628:

ﬂGOVPVdGOVP-VdGOVPV
«mammnmmwmmnmmwmmn
mmmmmmmzmmmwmmwmww

 

 

ﬂea-5:00
moss... Eoo >ncmoz mEmD outn— uoaw .0535 9.502..

 

ones eBpeH

 

 

88 x22 2 38%?
.moombcoo 853-; E00 23-302 Hmﬁmwm 8mm 3.85m wﬁwoom $80
2 65w;

49

 

 

26m $81 3me IT 03mm @301 BEES IT
2:8; 232..

~ eo-qea
F 80-060
L80-100
— eo-ﬁnv
— so-unr
— 90-Jdv
~ eo-qed
_ Lo-OGQ
bzowo
~ zo-ﬁnv
H Lo-unr
» zo-Jdv
~ Ao-qed
F 90-060
L 90-100
— 90-5nv
— 9o-unr
90-Jdv

Nd-
w v.0:

 

T To
1 Nd
q Md
1 v.0
_ no
1 od
1 No
I wd

 

83%: 3 32423. .855".
.ocmﬁm 3-.me mEmD outn— uoaw .0525 9:990:

ones aﬁpeH

 

 

aoom Sn: .8 meow £23 £83an 355m 3302 “mammwm 85 8mm 355m wﬁwcoﬂ
2 2am

50

 

— LO'AON

_ 90-AON

 

 

ozmm wmumI £85 In: 03mm 381 259:5 IT

 

 

octan— 9:93:

— 90-A0N
r. vo-AON
» 90-A0N
_ zo-AON
F Lo-AON
h oo-AON
~ 66-A0N
_ 96-A0N
_ LG'AON
L 96-A0N
— 96-A0N
_ VG'AON

r QG'AON

_ ZG-AON
L6'A0N

 

 

a.

_ b... Pup»: ...P..Lr .
q a“. «4‘

3

33.200 3.3:".

_

.

u

,5

h a

mum 3552 32.32 5559‘ outn— aonm mum .2332 9.562..

I m.OI

I
o

1 md

 

5:

0923 aﬁpaH

 

 

385:8 mohﬁsm 30 3.532 3.802 “magma 8E 8mm mad 12:32 wﬁmuom

5 83mm

51

 

 

 

ozmm mmumI 033w IT 03mm mmumI 269:5 IT
Boron 9.3an

 

 

 

 

N N N N N N N N N N N N N N N N N
O O 0 0 O O O 0 O O O 0 O O O 0 O
m m m m m m m m m m ..n m m m m m ..n
O 0 0 0 0 0 0 0 6 6 6 6 6 6 6 6 6
I. 9 9 IV 8 Z L 0 6 8 I. 9 9 IV 8 Z L .
_ o r _ _ L _ _ _ _ _ r o _ _ _ mFo OI
, I 36.

.

r — I mooo-
.. .. I o
r. 4.: J] I ...1 1‘ «a I.“ 2AM mooo
A 1.3 .. . . ..Foo
I moo
I moo
I omoo

 

£02200
3.3:... :30 3-532 mEm: morn. :35 :30 9:93...

ones aBpaH

 

 

385:8 $35,.» :80 3802 omﬁowo 03E 8mm :80 wcmwwom
3 2:3

52

Figure 1.6
Persistence in Ethanol Spot Price in Second Regime

 

 

Persistence in Ethanol Spot Price in Second Regime

o

 

I
‘—

ﬂi

i l
°°. ".
o o

0.6 1

O?
O
iuegomeoo

"7!
o
9

 

VI “'7! N. ". o
O O O O
ouaisgswd

6002/ L/ L

8002/L/L L

8002/ L/6

8002/ LIL

8002/ US

8002/ Ll€

rlod

8002/L/L '

Fe

1002/ L/ L L

mg

[002/ US

Hedg

1002/ L/L

1002/ HS

1002/ US

1002/ L/ L

9002/L/LL

9002/ L/6

 

53

 

Figure 1.7
Persistence in Natural Gas Spot Price in Second Regime

 

 

Persistence in Natural Gas Spot Price in Regime 2

V.
‘_

°°. ‘0. V. N.
O O O O
iuagomaoo oouaisgsmd

N‘-
‘—

~ 6002/L/L
» 900mm
~ soozn/s
I SOOZ/L/L
I soozn/s
— soozn/e
— 8002/L/L
L LOOZ/L/LL
— LOOZ/L/S
— LOOZ/L/L
— LOOZIL/S
— LOOZ/L/S
— LOOZ/L/L

* 9002/L/LL

 

‘r—— 9002/ US
0

Hedging Period

 

54

 

 

 

 

o Eton. E “mm mm _ Eton. E coEmon. mono:
o cocoa 5 “mm mm __ voted 5 coEmon. mono: I¢I
_ coron. c_ “mm mm = “Eton. 5 Samoa manor 0233: IT

 

 

 

eaten. omuo:

— so-uer
~ so-AON
~ eo-des
— eo-Inr

— eo-Aew
— 80-Jew
~ eo-uer
— AO'AON
~ LO'dQS
~ LO'lnl‘

~ Lo-Aew
~ zo-Jew
I LO'UBI‘
~ 90-A0N

 

__ Ea _ carom :_ 22:3“. ooooz 355m

 

I md

suomsod aﬁpaH

 

 

: o5 o 85o a 283$ owoom 3%on
3 05mm

55

 

 

 

o oozed s o8 8 _ Eton. E Eamon. woof
o oozed E o8 no __ Boom c_ 8.28.”. move... I-I
_ Eton 5 own on __ 85m c_ Eamon. mono: 883: IT

 

 

I eouer

” 80'A0N

— sodas

r 801“?

veron— wave:

~ so-Kew
— eo-Jew
— 90—uer
— AO'AON
r LO‘dSS
— AO'INI‘

— Loﬁew
I LO-JBW
— zo-uer

* 90'A0N

90-des

 

 

= Ucm _ Boron. E mcoEmOn— om—uoz no.0 .9552

 

mod-

I moo

rd
mod
Nd
mNd
md
mmd

suoggsod eBpaH

 

 

= o5 o BE a 20:38 88: 30 3532
3 03E

56

 

 

 

 

88% coma .225 “.00
o 8: oz 595 Loo IT
.805. 9.58: 2585 Soon. 8.5 9o 525 “.00 IT

 

 

 

toromomuoz
NS WWAINS WWFNS
Mowaomwowawmwo
mmwmmmmmmmmmmm
Hor___ir__ulde o
md
IF
Ime
IN
/ ImN

mosgczooh mEm—oor
E2950 mEmD £09 30E :wmo 035.2553 coming—coo

 

 

:lOO

 

 

moswmgooI—I wcmwdom mmohoo 32m ammo 9:33:50 mo coﬂﬁmﬁoo
o: Room

57

Chapter 2

Consumer Preference for Alternative Transportation Fuels: A
Conjoint Analysis

2.1 Introduction

Recent decades have witnessed persistent effort by the US. government to enhance
ethanol production and promote ethanol blended gasoline as an alternative transportation
fuel (Rask, 1998; Yacobucci 2006, 2008). Successful implementation of such mandates
and market acceptance of fuel alternatives may aid the government in solving ongoing
energy crises by slowing U.S. gasoline demand growth rates. Use of ethanol, considered
thecleanest liquid fuel alternative to fossil fuels, may also reduce greenhouse gas (GHG)
emissions, help prevent radical climate change and bolster local economies (Lin and
Tanaka, 2005; Niven, 2004). However, less energy efﬁciency and limited availability as
compared to conventional gasoline are some of the major drawbacks of alternative ﬁiels.
The pros and cons of ethanol blended gasoline as an alternative transportation fuels may
result in consumers holding heterogeneous preferences. Thus for policy prescriptions and
for strategic product positioning in the US. market, analyses of consumer preference for
alternative transportation fuels and welfare analyses are of vital importance.

We evaluate the characteristics of consumer preferences for alternative
transportation fuels with differing fuel attributes including fuel cost and availability,

emission level, bio-feedstock source, and purchase and maintenance cost of the

58

IF—‘T-“Zl

appropriate vehicle. Alternative transportation fuels considered in this study are E-lO and
E-85 where E-x implies motor fuel blends of x% ethanol with (100 - x) % gasoline.
Ethanol may be derived from corn or cellulosic biomass including com stalks, wood
stalks and switch grass. Thus the choice set consists of a set of hypothetical fuels and a
set of real transportation fuel alternatives available in the market. We employ choice
based conjoint analysis to estimate consumer willingness to pay (WTP) to switch from
conventional gasoline to alternative transportation fuels in a US. based survey (Louviere,
1988; Lusk et al., 2003; Alfnes and Rickertsen, 2003). A latent segmentation approach is
employed to allow for and evaluate consumer preference heterogeneity (Boxall and
Adamowicz, 2002). Our analyses of consumers’ WTP for alternative transportation fuels
uses observed variables (i.e. the demographic and socio-economic characteristics) as well
as unobserved or latent variables. We also evaluate whether willingness to pay for
alternative fuels varies across consumer segments and how availability of alternative
fuels impacts social welfare. Three particular hypothetical situations are considered
where due to varied reasons, either conventional gasoline, corn grain based ethanol or
cellulosic ethanol is not available for consumption. We identify welfare effects across
consumer segments in each of these three cases.

The remainder of the paper is organized as follows. Section 2.2 gives an overview
on the existing research on alternative fuel vehicles and the methodology. Section 2.3
describes the latent class model. Section 2.4 presents a description on the research design.
Section 2.5 gives the empirical speciﬁcations. Section 2.6 presents the empirical results

and Section 2.7 concludes.

59

2.2 Previous Studies

Studies on consumer demand for alternative transportation fuels are few and limited in
scope. Bhattacharjee et al. (2008) measured willingness to pay for E-lO fuel by US.
consumers employing a contingent valuation (CV) technique in a simultaneous latent
variable equation framework. The authors found self-described liberals have signiﬁcantly
higher WTP, WTP is higher for males, and WTP increases as familiarity with ethanol
increases. However, supporters of alternative fuels, but who are not sympathetic to
ethanol, have signiﬁcantly lower WTP for E-lO. Results show that the mean WTP for
E10 is almost .049 cents higher than the prevailing national average price of regular
gasoline during the time of survey ($2.55/gallon,).

Several studies on the potential demand for alternative fuel vehicles have been
conducted over the last decade using discrete choice analyses (Ahn et al., 2008; Home et
al., 2005;). Potoglou and Kanaroglou (2007) examined the potential demand for
alternative fuelled vehicles in Canada. The study concluded that purchase tax reliefs and
low emission levels may attract potential buyers whereas limited fuel availability may
prove to be a major deterrent factor. Ewing and Sarigollu (2000) assessed preferences for
clean-fuel vehicles in Canada using a discrete choice experiment. The study concluded
that even though consumers value the environmental impact, vehicle performance
characteristics are critical decision variables. It also concluded that government
regulations are not sufﬁcient to create a market for clean-fuel vehicles.

Following Boxall and Adamowicz (2002) and Greene and Hensher (2008), we use

a latent segmentation model to analyze consumer preference heterogeneity for alternative

60

transportation fuel in the US. using an internet based survey and employing choice based
conjoint analysis. This is the ﬁrst known study to assess US. consumer preference for
alternative transportation fuels besides conventional gasoline.

Choice experiments present respondents with the task of choosing one alternative
among multiple product proﬁles, each described in terms of a generally common attribute
set. Traditionally, resulting choice data are modeled through models such as the
Multinomial Logit (MNL), Nested MNL, and Probit models (Swait and Adamowicz,
2001). However, each of these models suffers from several shortcomings. The MNL has
an error structure assuming Independence from Irrelevant Alternatives (IIA). The IIA
property states that ‘the relative odds of one alternative being chosen over a second
should be independent of the presence or absence of unchosen third alternatives’
(McFadden, 1974). Moreover, conventional MNL assumes homogeneity in consumer
preference. That is, all respondents have the same preferences across alternative fuel
attributes (Lusk et al., 2003). To account for consumer preference heterogeneity, Boxall
and Adamowicz (2002) use a latent class model and introduces sociodemographic
variables to understand systematic heterogeneity. The approach assumes that consumers
can be grouped into discrete segments where each segment is characterized by
homogenous preferences while preferences are heterogeneous across segments (Swait,
1994). The model simultaneously assesses the inﬂuence of latent variables such as
attitudes, perceptions, socioeconomic effects and choice-based attributes in the estimation
of the latent segments and predicting product choice (McFadden, 1986; Swait 1994).

Internet surveys offer the possibility of large sample sizes allowing the researcher

to investigate a range of methodological questions (Berrens, 2003). Internet-based data

61

Fm"!

collection method has the potential to yield higher-quality data with lower nonresponse
rates and is less costly than traditional methods (Fleming and Bowden, 2009). According
to Lindhjem and Navrud (2008), data quality and welfare estimates are not signiﬁcantly
different or substantially biased using Internet as the data collection mode of choice
compared to in-person interviews. A study by Kypri et a1. (2004) deals with nonresponse
bias in internet surveys of alcohol use. The ﬁndings from the study indicate that the bias
resulting from nonresponse is arguably too small to be of concern with respect to
estimating other important attributes such as consumption levels, the incidence of
alcohol-related problems etc.

In response to the need for more rapid and iterative feedback on customer
preferences, researchers are increasingly employing web-based adaptive conjoint analysis
(ACA) methods that adapt the design of conjoint questions based on a respondent’s
answers to previous questions as opposed to the traditional non-adaptive design (Fixed).
A study by Dahan et a1. (2002) reveal that results from both methods exhibit a remarkable

level of consistency and this ﬁnding holds across estimation methods.

62

2.3 Model of Fuel Choice

Random utility models are based upon the assumption that individual i derives utility

U ,j (X ,1) from selecting option j where X i]- is a vector of attributes of j as perceived by

individuali. The utility function, U i]- (X ) is composed of two components, the

ij

deterministic component that depends on the attributes of the alternatives Vij- (X ,1) and

the random component represented by the error term, 8,] :

U1.j = Vi].(X,.j)+ 3,]. (2.1)

In this study, V,~ is the systematic portion of the utility function determined by the

alternative fuel attributes and their levels:

VI,- = ﬂI-XIj v j = A,B,C (2.2)
V,-j=6 Vj=D (2.3)
where X g. is the vector of choice attributes, ‘A’, ‘B’ and ‘C’ are alternative choices

available to the consumer, ‘D’ refers to the option of not purchasing either of the three

alternatives and 5 is a constant. The fuel attributes (X ij) considered in the analysis are

fuel cost, fuel availability, level of emission, purchase and maintenance cost, and bio-
feedstock or not.

The latent class approach assumes that latent variables such as general attitudes
and perceptions and observable variables such as socioeconomic characteristics inﬂuence
segment membership and classiﬁes the respondent into one of S segments (Swait, 1994).

The probability that a randomly chosen individual i chooses j and lies in segment

s where s = 1,....,S is given by (McFadden, 1974; Swait, 1994):

63

 

”iU)=i exp(a23Z,-) * exp(ysﬂng)

_ S CXP/J [3 X-f
3’1 Zexp(a/LsZi) _jeA§C,D( S s y)

LS=1

(2.4)

 

 

 

 

 

 

..I

where the ﬁrst expression gives the probability of membership for respondent i in

segment 3 (i.e. ﬁgs) and the second expression gives the choice probabilities conditional

on segment membership (i.e. 7r,-IS(j)). us and A, are segment speciﬁc scale and utility

parameters respectively, ﬁts are segment membership parameters and Z ,- is a vector of
both the socioeconomic variables (1;) as well as latent characteristics. Socioeconomic

characteristics used in this analysis are gender, age, level of education, income,
employment, race, family size, state of residency, vehicle characteristics and willingness
to buy a Flex Fuel Vehicle (FFV) in the next ﬁve years. This model permits choice
attribute data and individual consumer characteristics to simultaneously explain choice
behavior. Moreover, latent class approach relaxes the requirement of the analyst to make
speciﬁc assumptions about the distributions of parameters across individuals (i.e.

normality not assumed). The parameters, i.e. ls , [35, ,us ,a are estimated via maximum

likelihood methods. The log likelihood function may be written as:

lnL(auBlS)=ﬁZ Z aimj ln[i7ri|s(j)*”is] (2-5)

i=1 m jeA,B,C,D
where N is the total respondents who participated in the survey, m represents the total

number of choice sets per person for which the choice data were provided, j represents

the alternatives from the choice experiment and 6,-1- equals 1 if individual i chooses j and

64

0 otherwise. The scale parameters a and y s are set to unity (Boxall and Adamowicz,

2002)

65

2.4 Research Design

Our unique consideration of cellulosic biomass based fuel, which is not currently
available to consumers, necessitates our use of hypothetical choice experiments.
Following Gao and Schroeder (2009 AJAE), e-Rewards, Inc. a marketing research
company conducted an intemet based survey and collected complete response from 1,400
US. residents in January, 2009. Table (2.1) gives the descriptive proﬁle of the stated
choices sample versus the 2000 US. Census. As is evident from the table, the sample is a
good representation of the population.

A questionnaire was developed that gathered information on the socio-
demographic characteristics of the respondents, their degrees of awareness and their
attitudes and perceptions on related issues such as the blender’s credit offered by the
government for every gallon of ethanol blended with gasoline, possibility of industry
level production of cellulosic ethanol etc. The demographic characteristics include age,
gender, education, annual income, number of adults and children in the family, state of
residency, vehicle characteristics, average number of miles driven annually, average fUCl
cost etc.

Six additional pieces of information were collected that were used in the latent
segment model. These involved a series of 22 statements (Appendix 2A.1) that
represented reasons why individuals may favor the usage of ethanol as an alternative fuel
based on environmental issues with reference to air pollution, level of GHG emission,
and probable effects on climate change, deforestation and loss of grasslands. The other

issues considered were the effect of production of corn based ethanol on the local

66

economy in terms of job creation, income generation and opportunity for corn farmers.
Some of the geo-political issues considered were the nation’s dependency on foreign oil
and the possibility of food price volatility injected in the system due to increased corn
based ethanol production. Respondents were asked to rate the level of importance of each
statement on a 5 point Likert scale ranging from “Strongly Agree” to “Strongly Disagree”
(Crandall, 1980; Beard and Ragheb, 1983). The scores of the respondents were analyzed
with factor analysis to measure consumer’s attitude towards using ethanol as the
alternative transportation fuel.

The second part of the survey was an application of a choice experiment where
the respondents provide their choices among sets of alternative transportation fuels or
opting out from choosing any of these products (Alfnes, 2004; Lusk et al., 2003). We
personalized the survey to include choice experiments revealing resident preferences for
alternative transportation fuels based on information provided by respondents (i.e., car
type, miles driven per year, purchase/lease status) earlier in the survey. Table (2.1) lists

the attribute-level speciﬁcation for the alternative fuel choice experiment:

2.4.1 Attributes of Alternative Fuels and Their Levels:

1. Fuel cost refers to the total annual cost of buying transportation fuel. It is a function
of fuel per gallon prices, fuel efﬁciency, and the average number of miles driven by
the respondent annually. Price of fuel per gallon refers to the price of the
transportation fuel per gallon. Fuel efﬁciency refers to the energy content/ mileage

per gallon of alternative transportation fuels.

67

PI

Level of pollution and GHG emissions refers to the amount of health-damaging and
smog-forming airborne pollutants the vehicle emits and the level of green house gas
emitted in terms of the annual tons of C02 due to combustion of given volume of
ethanol blends as compared to gasoline.

Fuel Availability: Fuel availability is deﬁned as the proportion of existing stations
offering fuels other than gasoline in the area. It may be measured by the percentage of
gas stations offering the alternatives within a 5 mile radius from the respondent’s
residence.

Cost ofbuying and maintaining F F V: Includes purchase price of a FFV, annual repair
and maintenance cost.

Feedstock: Feedstock refers to the basic ingredient the altemative transportation fuel
is derived from. At present, ethanol used in the commercial sector is predominantly
corn based. Cellulosic ethanol may be produced from a wide variety of cellulosic
biomass feedstocks including agricultural plant wastes (corn stover, cereal straws,
sugarcane bagasse), plant wastes from industrial processes (sawdust, paper pulp) and
energy crops grown speciﬁcally for fuel production, such as switchgrass (Moller,

2005)

2.4.2 Cheap Talk:

Current literature suggests willingness-to-pay is overstated in hypothetical valuation

questions as compared to when actual payment is required (Lusk, 2003). We introduced

cheap talk in our survey to eliminate the potential bias in hypothetical valuation

questions. Cheap talk refers to process of explaining hypothetical bias to individuals prior

to asking a valuation question.

68

The experience from previous similar surveys is that people often state a
higher willingness to pay than what one actually is willing to pay for the
good. It is important that you make your selections like you would if you
were actually facing these fuel-vehicle choice decisions, noting that

allocation of funds to these alternative fuels means you will have less money

available for other purchases. Please place an “X” in the “I choose” box,

below the option that you would choose from each of the following

scenarios:

69

2.5 Empirical Speciﬁcation

In this choice experiment, a typical choice scenario (Figure 2.1) consisted of four
alternatives. The ﬁrst three choices present the respondent with multiple different sets of
hypothetical alternative transportation fuels and vehicle choice. The underlying
assumption being that standard vehicles run on conventional gasoline or E-lO where
FFVs have the choice of running on gasoline, E-lO or E-85. The fourth option gives the
respondents an option to opt out. The conventional gasoline constituted the base
alternative with the levels of the attributes being held constant, excepting that of
respondent’s fuel cost. Attributes of the rest of the alternatives were varied in the design.
A design matrix was developed using SAS for designing discrete choice experiments and
considering 33"‘24*24"‘24"‘24 orthogonal main-effects design, yielding 31 scenarios
(Louviere et al. 2000; Kuhfeld, 2005). The scenarios were randomly assigned into 4

blocks each consisting of 8, 8, 8 and 7 choice exercises respectively.

2.5.1 Factor Analysis:

The ﬁrst step in developing the latent segmentation model involved a factor analysis of
the 22 statements (motivational indicators) to provide estimates of the latent motivational
constructs which enter the membership likelihood function. The scores from the 22
statements were factor analyzed using principal component analysis with varimax
rotation (Sprotles And Kendall, 1986). We used conﬁrmatory factor analysis to verify the
factor structure of the set of observed variables (Suhr, 2006). The analysis identiﬁed four

components of motivations for choice of alternative transportation fuels and the factor

70

structure thus obtained explains 91.68% of the variance in the data. These motivational
components were labeled based on magnitudes of the loadings of individual statements
(Appendix 2A.1).

The ﬁrst component was called “Economic and Environmental Effect” because
statements related to the positive biofuel economic impact of production and usage of
ethanol on the US. economy as well as the positive environmental effects loaded highly
in this factor. The second factor was labeled “Positive intentions toward bioﬁiel” because
statements related to how positive individual effort may initiate changes towards
adoption of eco-friendly transportation fuel and also make other individuals
environmentally consciousness loaded highly in this factor. The third factor involved
statements related to the impact of production of corn based ethanol on food supply and
on the effect on food price volatility. This factor was labeled “Negative biofuel impact on
food”. The fourth factor was called “Crisis Overrated”. Scores of the four latent factors
were then calculated for each individual in the sample yielding four variables to be

included in the Z ,- vector in equation (2.4), to explain latent segment membership.

2.5.2 Estimation of the Number of Latent Segments

In estimating the latent segment models, statistical criteria are used to select the ‘optimal’
number of segments in a set of estimations and the number of segments imposed varies
across estimations. The underlying idea being that as the number of segments increase,
the log likelihood value improves. However, the model ﬁt gets compromised as the
number of parameters increases due to additional segments. Following Kamakura and

Russell (1989), this study uses two criteria to assist in determining the optimal number of

71

 

segments, the minimum Akaike Information Criterion (AIC) and the minimum Bayesian

Information Criterion (BIC) (Boxall and Adamowicz, 2002; Allenby, 1990).

2.5.3 Willingness-to-Pay Estimates and Welfare Measures

Consumer willingness to pay (WTP) estimates are of particular interest. Following Lusk
et al. (2003) and Tonsor et al. (2009, CJAE), we quantify the value that consumers place
on each of the fuel-vehicle attributes by taking the ratio of the attribute coefﬁcients to the
purchase and maintenance cost coefﬁcient (multiplied by two if effects coding is
applicable) (Boxall and Macnab, 2000). In order to consider statistical variability in
parameter estimates, we utilize simulation techniques and generate 20,000 values of each
WTP estimate using a bootstrapping procedure proposed by Krinsky and Robb (1986).
More speciﬁcally, 20,000 observations were drawn from a multivariate normal
distribution parameterized by using the coefﬁcients and estimated variance-covariance
matrix. The simulated WTP statistics from each model were utilized to empirically test
for differences in WTP preferences across segments. First, mean WTP estimates and 95%
conﬁdence intervals were identiﬁed incorporating both statistical and preference
variability in LCM. We repeat similar simulation procedure to obtain the mean WTP
estimates and 95% conﬁdence intervals for the MNL model.

We use the utility estimates to examine how restricted alternative fuel choice sets
affect consumer welfare. Other than unleaded conventional gasoline, our choice
experiment contains four different attribute levels for bio-feedstock: corn grain ethanol
and also ethanol derived from cellulosic biomass (i.e. wood, corn stalks and switch

grass). We estimate welfare impacts of government policies mandating the use of ethanol

72

blended gasoline (i.e. unleaded conventional gasoline is no longer an option) and of
banning usage of corn grain ethanol considering the controversy linking food price
volatility with its increased production. Lastly, considering the fact that production of
cellulosic ethanol is still in the experimental stage, we examine changes in consumer
welfare if it never becomes available to consumers. Following Lusk et al. (2006), Small
and Rosen (1978) and Tonsor, Olynk and Wolf (JAAE, forthcoming), expected

maximum utility (EU) from consumeri ’3 choice set selection is given by:
EU,- = ln[z exp(Vl-j )] + K (2.6)
1'

whereK is Euler’s constant and Vij is deﬁned by equations (2.2) and (2.3). Under the

assumption of no income effects, general welfare changes may be given by:

CV,- = l 1n[}:exp(V,-} )] — ln[2exp(V,-JQ)] (2.7)
7 j j

where C Viis the compensating variation for individual i,7 is the marginal utility of

income, the 0 subscript refers to the initial state andl superscript refers to the new state

following some change in Vij (Hanemann, 1982; Boxall and Adamowicz, 2002).

73

2.6 Results and Discussion

In estimating the latent segment models, 1 to 6-segment solutions were attempted. The
log likelihood values at convergence reveal improvement in the model ﬁt as segments are
added to the procedure, particularly with the 2, 3 and 4 segment models. This is evident
in the pseudo R-squared values which increase from the base of 0.074 to 0.341 with the 4
segment model. The minimum BIC and AIC statistics are clearly associated with the four

segment model (Boxall and Adamowicz, 2002; McLachlan and Basford, 1988).

2.6.1 Characterizing the Segments and Comparing the Models

The segment membership parameters (is) for the 4-segment solution are displayed in

Table (2.3). According to the segment membership probabilities, segment 1 and 2 each
had 20.9% members of the total 1400 respondents, 13.8% were members of segment 3,
and the remaining 44.4% belong to segment 4. Only the second and fourth segments have
signiﬁcant, negative coefﬁcients on the ‘opt out’ parameter. This indicates that the
members who belong to the ﬁrst and third segments (approximately one-third of
consumers) are statistically indifferent to all options available.

The data revealed high correlation between the number of adults and number of
children in the household and also number of vehicles owned by the household. It also
revealed high correlation between age and employment status of the respondent. Thus we
included age of the respondent, plans of buying a FFV. in the next ﬁve years, number of

adults in the household and annual pre-tax household income in the set of demographic

74

factors used in the analysis. Note that the parameters for the fourth segment are
normalized during estimation.

The utility function parameters for the multinomial logit (MNL) model and the 4-
segment model are displayed in table (2.4). We labeled the segments based on their
preferences (e.g., utility coefﬁcients) as well as their demographic and socioeconomic
characteristics (e.g., segment membership variables) displayed in tables (2.4) and (2.3)
respectively. Segment one was labeled ‘Cellulosic Ethanol Acceptor’ since the members
who belong to this segments exhibit signiﬁcant preference for ethanol derived from
cellulosic biomass (i.e. wood, corn stalk and switch grass). The second segment was
labeled ‘Corn Grain Ethanol Rejecter’ because members lying in this segment exhibit
disutility from the use of corn grain based ethanol. These segment members are however
indifferent to the use of cellulosic ethanol. Finally the third segment was labeled
‘Conventional Gasoline Acceptor’ because members in this segment exhibit disutility
from the consumption of wood and corn stalks based cellulosic ethanol. Members of this
segment however are indifferent towards corn grain ethanol and switch grass based
cellulosic ethanol. The fourth segment was labeled ‘Ethanol Acceptor’ since members
who belong to this segment exhibit preference for ethanol as the alternative fuel,
independent of the source for ethanol. They also exhibit sensitivity towards all the fuel
attributes, signiﬁcant at 1% level of signiﬁcance. More speciﬁcally, the parameters on
feedstock for production of ethanol (i.e. corn, wood, stalk and switch grass) are all
unifome positive and statistically signiﬁcant unlike the other three segments.

According to the parameter estimates displayed in table (2.3), older individuals

and those belonging to lower household income with higher number of adults in the

75

household are more likely to reject corn grain ethanol. Based on the latent variables,
individuals who have ‘Positive intentions toward biofuel’ (factor 2), ‘Negative biofuel
impact on food’ (factor 3) and ‘Economic and Environmental Effect’ (factor 1) are also
more likely to reject corn grain ethanol. However, individuals who believe that the
biofuel related crisis is overrated (factor 4) are less likely to reject corn grain based
ethanol. Similarly, individuals who have plans of buying or leasing a FF V in the next ﬁve
years are less likely to lie either in segment ‘Corn Grain Ethanol Rejecter’ (segment 2) or
in ‘Conventional Gasoline Acceptor’ (segment 3).

It is to be noted that there are some anomalies in results that may be attributed to
data error. For example, everything else remaining the same, individuals characterized by
‘Positive intentions toward biofuel’ (factor 2) are more likely to lie in ‘Ethanol Acceptor’
(segment 4) as opposed to ‘Conventional Gasoline Acceptor’ (segment 3) or ‘Corn Grain
Ethanol Rejecter’ (segment 2) as reﬂected in table (2.3). This could be due to poor factor
analysis or poor random utility model setup or both and it may be difﬁcult to separate out
the effects. Moreover, our choice experiment requires the respondent to make tradoffs
amongst multiple desirable alternatives. On the other hand, Likert scale requires them to
agree or disagree with the statement. Given the inherent methodological differences,
choice experiment results may not be completely at sink with Likert-scale judgments,
though both are derived from a common data set.5

According to the preference information displayed in table (2.4), parameters for
level of emission and purchase and maintenance cost are negative and statistically

signiﬁcant for both MNL and the 4-segment models. Similarly, parameters for fuel

 

5 It is to be noted that we had similar conclusions from the latent class model run exclusively using the fuel-
vehicle choice variables and ignoring the latent factors.

76

availability are positive and statistically signiﬁcant across models. These results are
consistent with economic theory. However, parameters for fuel cost are variable across
the models and segments. Results from the 4-segment model imply that preference for
alternative transportation fuel increases for ‘Cellulosic Ethanol Acceptor’ (segment 1),
‘Corn Grain Ethanol Rejecter’ (segment 2) and ‘Ethanol Acceptor’ (segment 4) with a
decrease in fuel cost. However, the ‘Conventional Gasoline Acceptor’ (segment 3) set of
respondents are indifferent towards fuel cost. Moreover, the ‘Cellulosic Ethanol
Acceptor’ (segment 1) and ‘Ethanol Acceptor’ (segment 4) respondents have preference
for cellulosic ethanol (i.e. ethanol derived from wood, corn stalks and switch grass). On
the contrary, ‘Conventional Gasoline Acceptor’ (segment 3) exhibit disutility for wood
and corn stalks based ethanol and is indifferent towards the rest. Similarly, ‘Corn Grain
Ethanol Rejecter’ (segment 2) and ‘Ethanol Acceptor’ (segment 4) have reversed
preference for corn grain ethanol with the later segment exhibiting a preference for the
fuel. Differing magnitudes of estimates and degrees of signiﬁcance across fuel attributes
and segments suggest preference heterogeneity among respondents. These effects are
absent in the single segment model which suggests that individuals would prefer
alternative fuel with lower fuel cost universally. According to the MNL model estimates,
consumers exhibit signiﬁcant preference for cellulosic ethanol and derive disutility from

consumption of corn grain ethanol.

2.6.2 Willingness to Pay

In the latent class model, WTP estimates for each attribute speciﬁc to each segment is

simply the ratio of the parameter on the attribute to the corresponding parameter on the

77

purchase and maintenance cost coefﬁcient (multiplied by two because of effects coding,
when ever applicable).6 Table (2.5) presents the WTP estimates and corresponding 95%
conﬁdence intervals for fuel attributes. According to the MNL model estimates, the
representative consumer is willing to pay 79 cents less in purchase and maintenance fees
for every annual dollar spent in fuel cost. They are also willing to pay an additional
$9,573.77 in purchase and maintenance cost for 100% increase in alternative fuels
availability in the existing gas stations within 5 miles radius from their respective
residence. Similarly, consumers reveal signiﬁcant discount for level of emissions as they
are willing to pay $205.1 less in purchase and maintenance fees for every additional
pound of green house gas (GHG) emission caused by the combustion of unit gallon of
fuel. Additionally, the MNL estimates reveal that consumers exhibit signiﬁcant
preference for cellulosic ethanol and derive disutility from consumption of corn grain
ethanol. The representative consumer is willing to pay $1,698.02 and $2,299.81 more in
purchase and maintenance cost for vehicles run on corn stalks and switch grass based
ethanol relative to unleaded gasoline. On the other hand, they are willing to pay
$1,243.17 less in purchase and maintenance fees for vehicles run on corn grain based
ethanol relative to unleaded gasoline.

Examining preference heterogeneity using a latent class model lays out more
intricate details. Even though mean WTP is consistently high across all segments, the
WTP for all other fuel attributes vary signiﬁcantly across segments. In particular, the
latent class model indicates that the ‘Ethanol Acceptor’ (segment 4) exhibit preference
towards consumption of corn based ethanol over gasoline. Individuals who belong to this

segment are willing to pay an additional $1,074.49 in purchase and maintenance cost for

 

6 Willingness-to-pay values are in $ per purchase and maintenance cost units.

78

vehicles run on corn grain ethanol over unleaded gasoline. However, ‘Cellulosic Ethanol
Acceptor’ (segment 1), ‘Conventional Gasoline Acceptor’ (segment 3) and ‘Corn Grain
Ethanol Rejecter’ (segment 2) exhibit indifference for corn grain ethanol over unleaded
gasoline. Similarly, ‘Cellulosic Ethanol Acceptor’ (segment 1), ‘Corn Grain Ethanol
Rejecter’ (segment 2) and ‘Conventional Gasoline Acceptor’ (segment 3) exhibit
disutility from higher fuel cost (mean WTP of -$10.59, -$2.72 and -$3.89 respectively) as
opposed to ‘Conventional Gasoline Acceptor’ (segment 3) being indifferent towards the
fuel attribute.

The four classes are also very diverse in their valuations for ethanol derived from
cellulosic biomass over unleaded gasoline. Segments ‘Cellulosic Ethanol Acceptor’
(segment 1) and ‘Ethanol Acceptor’ (segment 4) exhibits high WTP for cellulosic ethanol
(mean WTP of $1,434.95 and $2,186.39 for wood based ethanol, mean WTP of
$2,917.57 and $3,404.62 respectively for corn stalks based ethanol and mean WTP of
$1,447.72 and $3,891.97 respectively for switch grass based ethanol). ‘Conventional
Gasoline Acceptor’ (segment 3) exhibit disutility from consumption of wood based and
corn stalks based ethanol (mean WTP of -$13,137.82 and -$6,731.9l respectively) and
indifferent towards the consumption of switch grass based ethanol. ‘Corn Grain Ethanol
Rejecter’ (segment 2) is indifferent towards consumption of ethanol as alternative
transportation fuel in general. The willingness-to-pay values also convey some important
characteristics about consumer preferences for different fuel attributes. Consumers, in
general, are willing to pay large premiums for higher fuel availability. Similarly,
respondents who belong to ‘Conventional Gasoline Acceptor’ (segment 3) are willing to

pay large negative premiums so as to prevent wood based and corn stalks based cellulosic

79

ethanol from being viable options as alternative transportation fuel. Even though the
estimated premiums appear large and reﬂect some overstatement of willingness to pay in
a hypothetical setting, we can be more conﬁdent about their relative magnitudes
assuming that hypothetical-bias is similar across attributes (Cummings et al., 1995; Fox

et al., 1998; Fox et al., 2001).

2.6.3 Consumer Welfare Evaluation

Table (2.6) presents estimates of welfare impacts across segments of restricting
alternative transportation choice sets in three distinct cases. The ﬁrst scenario assumes a
hypothetical situation where government policy mandates the use of E-lO or E-85 (i.e.
ban use of unleaded conventional gasoline). Welfare estimates capture the change in
welfare that would result from moving from a world where consumers could choose
between conventional gasoline, E-lO or E-85 with ethanol derived either from corn grain
or cellulosic biomass and none to a world where the choice was only between E-10, E-85
and none. The welfare estimates can be interpreted as the maximum amount consumers
would pay to keep unleaded conventional gasoline in their choice set. The welfare
estimates from the MNL model implies that restricted choice set results in welfare gain of
approximately $346.09 for the representative consumer. This implies that on an average,
the consumers are willing to pay $346.09 to prevent the consumption of unleaded
conventional gasoline. Estimates for the four segments identiﬁed by the LCM reveal
mixed welfare effect across segments. Segments ‘Conventional Gasoline Acceptor’
(segment 3), comprising approximately 14% of the population have signiﬁcantly large

welfare losses with restricted choice set (mean welfare loss being $2,000.00). On the

80

   

contrary, ‘Cellulosic Ethanol Acceptor’ (segment 1) and ‘Ethanol Acceptor’ (segment 4),
comprising approximately 65% of the population, experience welfare gain as a result of
the ban (mean welfare gain being $546.38 and $1,089.53 respectively). However, the
welfare impact for individuals belonging to ‘Corn Grain Ethanol Rejecter’ (segment 3) is
not different from zero. It is to be noted that the above analysis assumes no income effect
as the set of alternative transportation fuel available to the consumer changes.7

In the second scenario, we consider the case where due to controversial debate
relating world food price volatility with the production of corn grain ethanol, the later
ceases to be one of the options in the alternative transportation fuel choice set (Trostle,
2008; Leibtag, 2008). According to the estimates of the MNL model, the new restricted
choice set leads to welfare enhancement for the representative consumer by
approximately $89.38. However, the 4 segments LCM portray a different story. Only the
‘Ethanol Acceptor’ segment, comprising of 44.44% of the population, experiences
welfare loss (mean welfare loss being $77.25). The welfare impacts for rest of the three
segments when faced with the restricted choice set are not different from zero.

At present, in the U.S., ethanol is predominantly produced by fermentation of
corn glucose. Given the nation’s signiﬁcant agronomic base, technology development for
ethanol production from nonfood-plant sources, also known as cellulosic biomass, which
can be converted into fuel ethanol, holds promise (Lin and Tanaka, 2006). However,
current technology is still in the experimental stage. In the third scenario, we analyze
consumer welfare if cellulosic ethanol never becomes available to consumers. The 1-

segment model indicates signiﬁcant welfare loss by the consumers if faced with the

 

7 Another reason explaining consumer’s gain in welfare when faced with restricted choice set may be
reduction in search cost. Moreover, in reality, the assumption of no income effect might not hold well.

81

restricted fuel choice set (mean welfare loss being $330.78). As expected, the variation in
consumer welfare levels across the four segments reveals preference heterogeneity.
‘Cellulosic Ethanol Acceptor’ (segment 1) and ‘Ethanol Acceptor’ (segment 4),
constituting 65% of the evaluated consumer population have preferences for cellulosic
ethanol and hence experience welfare loss with the restricted choice set (mean welfare
loss being $362.64 and $674.87 respectively). On the other hand, ‘Conventional Gasoline
Acceptor’ (segment 3) experiences a welfare gain when exposed to the restricted choice
set (mean welfare gain being $1,420.00). The welfare impact for individuals belonging to

‘Corn Grain Ethanol Rejecter’ (segment 2) is not different from zero.

82

2.7 Summary and Conclusions

In recent years, production of fuel ethanol by com grain fermentation or by conversion of
cellulosic biomass has gained momentum. Use of ethanol blended gasoline is also a
rapidly increasing alternative transportation fuel in the US. market. This study examines
consumer preference heterogeneity and estimates consumers’ willingness to pay (WTP)
to switch from conventional gasoline to alternative transportation fuels by employing
choice based conjoint analysis and applying latent segmentation approach.

F uel-vehicle attributes used in the analysis are fuel cost and availability, emission
level, bio feedstock and purchase and maintenance cost of the appropriate vehicle.
Evaluated consumer base was segmented into four broad classes based on the
socioeconomic characteristics, latent constructs of the consumers as well as their fuel
attribute choice. Results from the latent class model indicate strong consumer
heterogeneity for fuel attributes. Analysis of consumer preference heterogeneity and
WTP may be of considerable value to policy makers.

Age, annual pre-tax household income and plans of buying or leasing a ﬂex fuel
vehicle (FF V) in the next 5 years are some of the socioeconomic variables that inﬂuenced
the respondents’ choice of alternative transportation fuels. Our analysis indicates that
state of residency, speciﬁcally whether the respondent is from the Mid-West or from
California did not have a signiﬁcant impact on decision choice made by the respondent.
Similarly, degree of awareness as indicated by the respondents did not play a signiﬁcant
role in their decision making process. According to the parameter estimates of the latent

variables, individuals who are more economically and environmentally conscious are

83

more likely to lie in the ‘Corn Grain Ethanol Rejecter’ segment. Similarly, individuals
who believe that production of corn based ethanol adversely affects production of corn
for human consumption and also affects overall food prices are more likely to belong
either to the ‘Cellulosic Ethanol Acceptor’ or the ‘Corn Grain Ethanol Rejecter’ segment.
Additionally, individuals who believe that the crisis involving production and usage of
transportation fuels is overrated are less likely to lie either in ‘Corn Grain Ethanol
Rejecter’ or in ‘Conventional Gasoline Acceptor’ segments.

The WTP estimates validate suspected consumer preference heterogeneity.
Results from the MNL model and the 4-segment model indicate that on an average,
consumers are willing to pay signiﬁcantly higher amounts for higher fuel availability and
pay less for fuel cost and Green House Gas (GHG) emissions. Moreover, the WTP for
bio-feedstocks vary across the segments in the latent class model. Three particular
hypothetical situations are considered where due to varied reasons, either conventional
gasoline, corn grain based ethanol or cellulosic ethanol is not available for consumption.
The consumer welfare estimates indicate notable diversity among consumers in each of
the three cases. ‘Conventional Gasoline Acceptor’, comprising approximately 14% of the
evaluated consumer population experiences welfare loss when faced with a restricted
choice set with unleaded conventional gasoline being not an option, with mean welfare
loss being $2,000.00. Individuals belonging to ‘Ethanol Acceptor’ segment,
approximately 44.44% of the evaluated consumer population experiences welfare loss if
corn grain ethanol is not made available in the US. market as alternative transportation
fuel choice, with mean welfare loss being $77.25. On the other hand, if cellulosic ethanol

is not available as the alternative transportation fuel, individuals lying in segments

84

‘Cellulosic Ethanol Acceptor’ and ‘Ethanol Acceptor’, constituting approximately 65%
of the sample set experience welfare loss under the restricted choice set, with mean
welfare loss being $362.64 and $674.87 respectively.

Thus analysis of consumer preference heterogeneity for alternative transportation
fuel and their WTP estimates may be important in determining marketing opportunities
for farmers and rural small businesses, bioﬁrel producers, petroleum suppliers, and fuel
marketers. Such analysis may also aid the policy makers. Future research work may
consider some of the alternative transportation fuels available in the market, namely
hydrogen, liqueﬁed natural gas, Liqueﬁed Petroleum Gas (LPG), Compress Natural Gas
(CNG), Bio-diesel etc to estimate consumer preference structure and their WTP for

various fuel-vehicle attributes.

85

Chapter 2 Appendix

86

 

 

.co_mw_Eo 010
do .65. 33905 new omcmco 295.0 3 £82 88:35 m: 8:;

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

wow _ .o- mag. T _mS.N- .3365 6550 05 9 tomazm EBEEE m.EoEEo>ow 05 comma 8: ow _
oven—83w mo 82 EE cougar—8% 8 muse. :ozozcoom m: 85m
mﬁvod- mm; T 3. SN. Emsdﬁ 3550 on“ 3 20.5.5 Egan—m m.EoEEo>ow 2: tomazm 8: ow _
5:529 coca doom 303:: 5:03.05 m: 02%
ooomd- 58. T 26m. T .05st 3550 2: 8 tong—om E655» o.o:oEEo>om 05 tommzm Ho: ow _
.moo_>._om v.8 mwoow .8
deﬁned new 058:. waxes—cow .noCo 2.2880 no 2E3 E .oanooo :30.
Sand Nmmod good 05 :6 «onto :38 2on a on: 35.3 3550 2: .3 wooooocow oEooE
.ooobo m:
333 m 30o- 0326. Bone voEoocoo 8: Eu _ 28 36:06 252 S... m_ max—=29 nova doom.
.o>§mom m_ : wmmEsmcoo 28 05.0%
UoEmEOo ma 3 mﬁﬁzmcoo 2:... $83880“. 3550 wEozcoE do sous—Ba
momod womwd good 533 93 to we 3:3 5 8on 35:28:25 Ho: 2: :2: 96:3 _
demogm 0:0 mo 35.
was coca—Em be 8.62 bozoooho EB 33 cocotoamcab ozooEozm
m _ mud N_ mod ended lg mo sochﬁnmcs 05 08205 8 take 02..» core: a “my: o>o=on _
€255.35 05 :0 Some: 025% 2:8 08: EB 32% 5%
Goad 302. 836 035:8? mo coco—83:8 2: 3.855 8 tobo 1323?: >2
..oa :ochtommcob
83d ooood good bucoEéoo mo :osnﬁnmcoo 8m $6958 \omngo 8 95:3 Em _
goth .Azeﬂwo—ooo
Namod 885 good 20:. E $5 33 533.822» b3 8 0.5.: and 8 m===3 ES
doom 3305
:o 83:: @838 Hootm
voooto>0 $803 22885 .ScoEco:>:m
ﬂow—U ozoowoz 328m Ea o_Eo=oom

 

 

 

 

 

3588on 3:55:68. mo mambo: swoon was 2095185

~.N< 293.

87

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

08$ _ SN; _ 0E: _ $03 _ 02553.5

33.: 0 02: 0 9.9% _ 5.3 _ 0350.5

3000.0 .000m .3 00:003xm 00:00:0>

.=0 :w_0:00 :0 00:00:0000 0.0050: 05 0:0 :0? E0305

300.0 88.0 336 3:050 05 00 0:009; 003:0:0 m0:0EE0>0w .m.D 05 005 0>0=00 0

000000 $0 5000

'3 waned- N306- 280.0 00E00:00 00: E0 0 0:0 0000::0>0 0:02 :0 00 03 :w00:00 :0 00:00:00.3—u

.00:0::0>0

[E 88.0 ooood 88.0 m: 0000 5500:0020: 050E003 00 00: 05 0: 0000—0: 0030 :05230

3000500 :03 00:00:: :0 000:0

00 _0>00 .06 :0 $380 :0 00:00:0000 0: 0002000 00 :050: 05 03:0 :05

303.0 - , - 030.0 0mm; 0300 5500300005 0200:0000 00 0050:5200 0005:0000: 005 05:00 0

000:0 0000 3:03 w:500000

300.0 33.0 «and- .n_0m:0>00 m: 3:050 00000-E00 00 :0500005 :00 E00 5 0:0800 :wE

.E00 03::

30% 0:0 3500005 .3003 00 0023005 0:05:00 w:_0=00:m E00-:0:

mmamd .. 003.0 33.0 w:00_>0:0 00 3:050 00000-E00 00 5500090 :00 E00 :_ 0:0800 :wE

.:0_EE:m:00 0000 :55; :00 E0000 5500080 9500000

mwwmd ommmd mm 20.0- .m_0m:0>00 m: 3:050 00000::00 00 :0500005 :00 E00 5 000800 :wE

0:08:00 E00 05:00 000» m0 :0:0:00:0M:

gum: g .0. 25.0 00:7. b00000: 3:050 05 00 0:009; 33:05.0 m,::0::E0>0w 05 E0003 0

..AEO:000 300— 05:00 000m m0 :0500005 0:

03m.— 020; mama: 005m b00005 3:050 05 00 200000 33:05.0 0.0:0EEO>0m 05 E0003 0

.300 03:05.80 :0 m: :0

m 0 EV.— _N_v.0 $00.0 00:00 0:00:05 3:050 05 8 E0003 33:05.0 00:0EE0>0w 05 E0003 0

030500000

0: 3:050 505 0::0m0w _0:05:0>:00 00 00535300 30:00 0 005 30:00

wmwmd 03.0.0 q i. 4 050.0 0 .300 550030205 :00 0:0E00 0:00:30 :0 0:05 30:50 05 :030
0:05:00 E00 .m.D 05 :00 000

vcmnd 008.0 M: mud m_ 3:050 00m00-::00 00 :0500005 :00 020:0:0w E00 :00 05:50 :wE

 

 

080080 :0... 2.5

88

Table 2.1
Descriptive Statistics

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Total Number of Household Respondents (1400)
Survey 2000 US
Data Census
Statistics
Gender Females 46.49 49.1
Males 53.51 50.9
Age 20 years and less 1.21 28.60
21 - 30 years 15.83 13.63
31 — 401ears 16.96 15.36
41 —50 years 14.55 15.11
51-60ygars 12.99 11.03
61 — 70 years 27.04 7.23
71 — 80 Lears 10.93 5.78
81 years and over 0.50 3.26
Income Less than $20,000 12.63 22.10
$20,000 to $39,999 25.05 25.29
$40,000 to $59,999 22.07 19.66
$60,000 to $79,999 14.83 13.34
$80,000 to $99,999 9.30 8.37
$100,000 to $119,999 4.90
$120,000 - $139,999 2.91
$140,000 - $159,999 2.63 1 1.24
$160,000 - $179,999 1.49
$180,000 or more 4.19
Education Did not graduate from high school 0.78 19.60
Graduated from 13h school, Did not attend college 14.48 28.60
Attended college, no degree earned 23.85 21.10
Attended college, associates or trade degree earned 12.70 6.30
Attended college, Bachelor’s (BS. or BA.) degree 28.46
earned 15.5
Graduate or advanced degree (M.S., M.A., Ph.D., 17.89
Law School) 8.90
Other 1.85 -
Race White, Caucasian 80.27 75.1
Black, African American 9.01 12.3
Asian, Paciﬁc Islander 1.49 3.7
American Indian 0.21 .9
Other 2.20 -
Mexican, Latino 6.81 12.5
Employmen Full-time employed 44.22
t Part-time employed 10.50 63.9
Unemployed 4.26
House maker 4.12
Retired 30.02 36.1
Student 4.47
Other 2.41

 

Source: US. Census, 2000 ; url : hip://factﬁnder.censusgoWhome/saff/main.html?_lang=en.

89

 

Table 2.2

Attributes and Levels of Alternative Fuel Choice Experiment

 

 

 

 

 

 

 

 

 

Experimental Gasoline E-10 E-85
Design
Attributes
Fuel Cost ($) First level/ base calculated in the program based on available
information on primary vehicle class, approximate miles driven
per year and the average cost of fuel per gallon. The other 2
levels are 15% lower and higher than the base.
Fuel 100% (1) 90% (1) 90%
Availability(% (2) 75% (2) 75%
of existing (3) 50% (3) 50%
stations within (4) 25% (4) 25%
5 miles radius
from the house)
Emission 19.40 (1) -10% than the base (1) -10% than the
(pounds/ gallon) pounds/ gallon (2) -20% than the base base
(base) (3) -50% than the base (2) -20% than the
(4) -75% than the base base
(3) -50% than the
base
(4) -75% than the
base
Feedstock Not (1) Corn (1) Corn
Applicable (2) Cellulose: Wood (2) Cellulose: Wood
(3) Cellulose: Corn (3) Cellulose: Corn
Stalks Stalks
(4) Cellulose: Switch (4) Cellulose: Switch
Grass Grass
Cost of buying $21,500 (base) (1) -20% than the base (1) -20% than the
and maintaining (2) -10% than the base base
the vehicle ($) (3) base (2) -10% than the
(4) +10% than the base base

 

 

 

(3) base
(4) +10% than the
base

 

90

 

Table 2.3
Segment membership variables for the four segment model'

 

 

 

 

 

 

 

 

 

 

 

 

 

Variables Segment 1‘” Segment 2 Segment 3 Segment 4
(Cellulosic (Corn Grain (Convention (Ethanol
Ethanol Ethanol a1 Gasoline Acceptor)

Acceptor) Rej ecter) Acceptor)

Intercept -3.7264*** -6.0427*** -5.6414*** 0

Age -0.0112 0.0400*** 0.0040 0

Plans of buying/ 0

leasing a FFV in the

next 5 years -0.2528 -1.1224*** -0.8360***

Number of adults in the 0

household 0.0824 0.2623” -0.0407

Annual pre-tax 0

household income 0.0600 -0. 1467* * 0.0040

Factor 1 0

(Economic and

Environmental Effect) 0.0571 ** 0.0658“ 0.1487**"‘

Factor 2 0

(Positive intentions

toward biofuel) 0.0417 0.2008" * 0.1869***

Factor 3 0

(Negative biofuel

impact on food) 0.0546 0.0776** -0.0621

Factor 4 0

(Crisis Overrated) —0.0080 -0.1755*** -0.2825***

 

 

 

*, **, *** indicates 10%, 5% and 1% level of signiﬁcance respectively
1 Log likelihood at convergence (LL): -5,690.359
‘1’ Average probabilities of lying in class 1, 2, 3 and 4 are .209, .209, .138 and .444 respectively.

Pseudo R-squared: 0.341

AIC: 11,517.72

91

 

 

 

303838. 858

saw? mo .05. .x; 28 $m .o\..o_ 88065 1:. .2. ..._

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

...Iawocs- ommmg *3mwwﬁm- 286m- 1*m85m- 20:90 SO “no
Iawwvd moved- Swod ioﬁmd IEvvmd $.80 nozam
*Iwnmvd Iwmmmd- Sued- 2&3 :3 :Eomﬁo mﬁﬁm :80
1.9.26 ,...,...wﬁmod- mde- :cwomd “.1386 woo?
.13de wowed iwvvmd- Svod *INNEd- Eco
... .. a. Seed. ... ... 2.2159? ... * .._ moood- _.. ... *voood- * ... *Noood- “moo oo§q8§2
do 838an
***ombo.on **©~mc.o- tiwhvod- ***oon.ou ***©mvo.ou dommmmam—
,...,...Vmwm; ,...vaoé Iawgd “.13on ...Imomod b25332“ 35m
***OHO0.0- mo-m5m.m ***wooo.o- ***Ovoo.ou ***Nooo.o- 500 ESL
©8303,. A8383,
CoEo8< cam—ammo €88“wa 355m 855m
3555 3:233:08 E80 :88 ammo—5:08
v Eoﬁwom m Eoﬁwom N EoEmow M Eoﬁmom
E65
3608 EoEwom v “aoﬁwom _ 03atw>

 

 

 

 

moSnEE :35 no 330883 85055 355
«N 2an

92

 

 

 

 

 

 

 

 

 

 

 

 

 

 

48.28.8- 8.238 :8. am.”- €o+mm$ 8.28.2
2 8.84. _ 2 2 8.88.3 2 8.8.2: 2 8253 - 2 8.3.2-:
8.28.8- 2.5.2 8.22 _- 8.22.8: 8.22.:- 880 so 25
88.2.3. .82? H8828 8m. 82
2 8. 5...: 2 833.2 2 8.82-: .8882 2 2.8: 2 8.2m: E2228:
8. .23 8.88- R28 ~32: $.83 220 225m
83m 3. 84.28.. 88.3 3 28.2.2.2
2 8.28.: - 2 828.2: 2222.1 832% 2 8. 8.: 2 5.8: 3.2822
8.22% a. _ MS. 8.8.... 3.83 8.83 22m 28
28.83 8.8% 82.23 8382
2 8. an: - 2 8.22. :1 2 218.3 832.3 2 8. EN: 2 m _ .2 Z 3282:
8.22.2” 2.5.2- 2.8. 383 8.2% 803
838; 838.2 8.8.33 838
2 8.2: 2 8.3.2 5.2 2 8.58.2 2 8.88. 1 - 2 8.23.1 22235
21.82 3.33 8.85- aim 2.3-m. _- 28
83mm 2%: Gm. §
- 2 8.2.2-: 2 8.2.3-: :22 2 $28-: :32- 2 82mm: - 2 8.83 A882:
2.8. 3.8.... 2.2.- $2.:- SSN- 2.2222
828.0 3:82 So. _ 82 83.8. a 8.82: 3:52;:
2 8.82.: 2 8.3%: 2 82 5.: 2 8.25.: 2 8.83: .02 8252026
88.8.8 R. _ 8.2. 8.83 8.22.: :83 $5222 22
EA- 2 8.1 .32 2 8.1 .8. _- 2 8.2 E;- 2 8.2-: .88- 2 2.1
8 m- 38 NE. 8.2- 8.8- 28 2.2
Coumooo<
€8.5on 3:036 Couooﬁm Coumooo<
3:253 3:225:08 Eamﬁm E80 E08 3535 0:23:08
v EoEwow m Ens—mom N EoEMom _ EoEwom
.oon EoEMom v :62: Eon—wow _ oEatm>

 

 

 

 

8233?. 63m catatommcﬁh Sm 53285 oocowmcou $3: ham 8 mmocwﬁzm? Hoﬁsmcoo
Wu 2an

93

 

 

 

 

 

 

 

 

 

838 :83 388 5.88- 285.5 28:?“
- 2 8.82 2 8.8: $32 2 8.82 - 2 828-: 2 8.81 28 68220 283:8
8+8- 8+m82 $2 $88- 888- 8 22288
:38 $38 :8. a A8225 03255
- 2 8.8 1 2 $88.: $8.08 2 :82 2 8.83 god; 2 8.8: 28 6520 £28 88
3E- 8+m8. _- 8.5 8.2- 8.8 “N €2.88
€85va
838; 82.8 2 _ FEE .882. 22252 8: 2:88
2 8.8: - 2 8. 5.3 $98 2 8.81 2 8.8: 2 8.8: 382528 823:5
880.2 8.2288- 8.87 8.2% 8.28 2 22.58
Coﬁoooxx Coﬁooo<
Coaooo< 230me €88 mom 3:25 355m
855me 12535288 580 E08 3.0.2::qu
v EoEwom m :58me N EoEwom _ Eoﬁwom
6:08 :58me v BEE Eoﬁmom H 032:5»

 

 

 

 

8222 85380 2.8:

82385842» 8335; $95 Sum 8630 ~25 nous-8&8“; 033532 uoaownmom mo mace-mm Emu—u?

ed 2an

94

Figure 2.1
An example of choice scenario from the choice experiment

 

 

 

 

 

 

 

 

Gasoline E-10 E-85 Not Purchase
Fuel Cost $731.25 $989.34 $731.25 I choose not to
Fuel Availability(% 100% 50% 25% purchase either
of existing stations) of these
Emission 19.40 9.70 14.55 products.
(pounds/ gallon)
Bio-Feedstock Not Cellulose: Cellulose:
Applicable Switch Corn Stalks
Grass
Cost of buying and $21,500 $19,350 $19,350
maintaining the
vehicle
I Choose [3 D D D

 

 

 

 

 

95

 

 

 

Chapter 3

Factor Demand Analysis for Ethanol in the US. Reﬁnery Industry

3.1 Introduction
The demand for ethanol as a complement to and substitute for gasoline has increased
signiﬁcantly in the last decade. This is primarily in response to government policies such
as the Clean Air Acts, the Federal Reformulated Gasoline (RFG) programs, mandatory
Renewable Fuel Standard (RFS), and a series of tax exemptions and subsidies targeted at
the ethanol and gasoline industries. Other contributing factors inﬂuencing the demand for
ethanol are high demand for fuel, volatile gasoline prices, technological advancement and
environmental awareness. Methyl tertiary butyl ether (MTBE) was initially used as the
primary petroleum-based oxygenate. However, given that MTBE is bio-accumulative and
an active polluting agent, the major environmental hazard involved with its expanded use
was ground water contamination (RFA, 2008). A gradual phase out of MTBE as an
oxygenate and replacement by ethanol was initiated in 1999. This provided an additional
boost to the factor demand for ethanol at the reﬁnery industry in the US. (Miranowski,
2007). These policies may lead to short and long term effects on the US. economy and
may even result in a permanent structural change in factor demand in the country’s
reﬁnery and blender industry.

The objectives of this study are three fold. First, factor demand relationships for
material inputs from the US. reﬁnery industry are estimated using a static translog

model. Second, we examine whether there has been a structural change in the factor

96

demand of reﬁnery inputs. Finally we examine whether the gradual switching process has
been simultaneous for all factor inputs with a common adjustment rate or individual
switching paths with unique adjustment rates.

In our analysis, we employ two alternative switching regression models. The ﬁrst
model is a gradual switching regression model that allows the independent factors
affecting the share equations to shift at different change-points and at individual
adjustment rates. The second model is a restricted version of the ﬁrst model and assumes
the shift point (or join point) and rate of adjustment for all independent factors to be
identical for the set of share equations of the translog system. Results obtained across the
two models are compared. We then proceed to obtain bootstrap distributions and
conﬁdence intervals of the common shift point and the rate of adjustment.

The remainder of the paper is organized as follows. Section 3.2 provides a
discussion of the factors inﬂuencing the demand for ethanol in the reﬁnery and blender
industry. Section 3.3 gives an overview on the existing research on structural change
analysis, speciﬁcally those using switching regression models. Section 3.4 describes the
translog cost function, two alternative gradual switching regression methods and the
associated assumptions. Section 3.5 describes the data. Section 3.6 presents the empirical

results and Section 3.7 concludes.

97

3.2 Motivation

Federal and some state level governments have offered an array of tax credits,
exemptions and investment incentives to boost the demand for and future development of
the ethanol industry as well as the domestic production of ethanol (Dufﬁeld and Collins,
2006). Volumetric ethanol excise tax credit (VEETC) is a production-linked subsidy
offered to the US. ethanol industry by the Federal government. It ranges from
approximately $4.5 to $6.4 billion per year with no limit or dependence on the price of
fuels for which ethanol may be considered to be a substitute (Koplow, 2006).

Blender’s credit is an incentive policy offered by the government to the petroleum
reﬁning industry for every gallon of ethanol blended with gasoline. This provides an
additional boost to the factor demand for ethanol. Initiated in 1980, reﬁners and blenders
receive an income tax credit for every gallon of ethanol blended with gasoline (Energy
Information Administration, 2008). Petroleum blenders currently receive 51 cents per
gallon income tax credit, which is scheduled to expire in 2010.

Renewable Fuel Standard (RFS) is a mandatory requirement aimed at reducing
the nation’s dependence on foreign oil by increasing the supply of alternative fuel
sources. Figure (3.1) demonstrates the effect of mandates on ethanol production. The
government mandates and initiatives have successfully increased the production of
ethanol by more than 400% over the last ten years (Anderson and Coble, 2008).

Crude oil is the major raw material used in the US. petroleum reﬁning industry,

with other inputs being capital, labor, liqueﬁed petroleum gases (lpg), oxygenates,

98

unﬁnished oilss, motor gasoline blending components, etc. The factor demand for ethanol
at the reﬁnery level has increased from 1,857 thousand barrels in January, 2002 to 14,768
thousand barrels in May, 2008. At the same time, the factor demand for MTBE has
decreased by 99.77% where as liqueﬁed natural gases exhibited a cyclical pattern.
However, the factor demand for crude oil has been comparatively stable experiencing
only a 2.15% increase in this time period (Energy Information Administration, 2008).
The capital intensity of the US. reﬁning industry, measured by the ratio of total
capital expenditure to reﬁnery capacity, registered a sharp rise between January, 2000
and January, 2002, reﬂecting the surge in investments in upgrading reﬁneries. The
industry registered a second phase of steady increase in investments from January, 2006
to January, 2007. The capital intensity in the interim period (i.e. 2002 to 2006) has
ﬂuctuated. Labor intensity in the petroleum reﬁning industry, measured by the ratio of
the total number of production workers to operating crude oil distillation capacity,
registered a steady decrease from 1993 onwards. However, it has experienced an upward
trend from 2006 onwards. Thus there is evidence to suggest structural change in the
factor demand in the US. reﬁnery and blender industry due to several contributing
factors such as high demand for oil, volatile gasoline prices, governmental policies and

environmental awareness (Rask, 1998).

 

8 In the petroleum reﬁning and processing industry, unﬁnished oil refers to all oils that require further
processing, except those oil that require only mechanical blending. They are produced as a result of partial
reﬁning of crude oil. Naphthas and lighter oils, kerosene and light gas oils, heavy gas oils, and residuum
fall under this category of unﬁnished oils (Energy Information Administration, 2008).

99

3.3 Literature Review

Extensive research has been conducted on identifying, measuring and testing the
hypotheses of structural change in demand systems. Goodwin and Brester (1995) utilized
gradual switching regression techniques to evaluate structural change in factor demand
relationships in the US. food industry. Rickertsen (1995) investigated structural change
in the Norwegian demand for meat and ﬁsh utilizing a dynamic switching speciﬁcation of
the almost ideal demand system. Stern et al (1979) analyzed evidence of structural
change in the demand for aggregate U.S. imports and exports. Schafer (2005) illustrates
structural change in the energy sector for eleven world regions using time series data and
discusses their impact on the decline in energy intensity. This current paper evaluates the
structural change in factor demand in the US. petroleum reﬁnery industry, a study being
conducted for the ﬁrst time known.

Since Quandt (1958) proposed the switching regression model, it has often been
utilized to test for structural change in demand systems. Tsurumi (1980) and Katayama et
al ( 1987) considered the switching regression model assuming that structural change was
gradual. Ohtani and Kakimoto (1989) and Ohtani and Katayama (1990) extended the
switching regression model by setting a switching regression model with a transition path
speciﬁed by a polynomial of time and considering a gradual switching regression model
with autocorrelated errors respectively. Although the majority of the literature on
switching regression models assumes that coefﬁcients shift at a common shift point,
Toyoda and Ohtani (1989) developed a less restrictive switching regression model such

that the individual coefﬁcients may have differing change-points. In this paper, we extend

100

the literature by setting up a switching regression model where the inputs follow
individual transitional paths characterized by unique shift points and adjustment rates for

the US petroleum reﬁnery industry.

101

3.4 Model Speciﬁcation

3.4.1 Specification of the Translog Model

In this study, a static translog model is used to estimate factor demand of inputs for the
US. petroleum reﬁnery industry. We use time as a proxy for technological progress.
Since the translog model provides a test for structural change while identifying the exact
nature of any shifts, this empirical approach proves to be advantageous over other
approaches given its ﬂexible functional form and its representation of the cost structure
of an industry (Goodwin and Brester, 1995; Fischer et al, 2001). The translog cost model
may be expressed as:

N N N
Inc, (a.Q,.t)= ao +Zai 1111’” my 1119: +azt+%22aa lnPn 111sz
i=1 i=1j=l

(3.1)

N N
+Zaiq lnPit an, +ZaittlnPi, +%ayy an, an, +ayttan, +%a,,12
i=1 i=1

where C: ( ) is total cost; Pj, is price of input j in period t; a ’s are parameters to be

estimated and Q, is output in period I, measured by the total amount of gasoline

produced in that period. The cost-share equations are estimated rather than the cost
function to ease potential estimation problems and conserve degrees of freedom (Huang,
1991). The cost-minimizing factor demands are obtained by applying Shephard’s lemma
(Shephard, 1953),

xf : aC*(Pi9Qt’t)
r aPi

 

(3.2).

102

um

The input share equation is:

N

w,-, = ai + 2a,] In Pj, + aiq ln Q, + ant; i =1,....,N (3.3),
j=1

where w" is the cost share of input iat timet and subscript i stands for i = 1 for crude oil

(co ),i = 2 for ethanol (e ),i = 3 for MTBE (mtbe ), i = 4 for liqueﬁed petroleum gases

(ng ), i = 5 for labor (l),i = 6 for capital (k) and i = 7 for unﬁnished oil( uo ).

3.4.2 Structural change identiﬁcation allowing separate shift points & adjustment

rates

Consider multiple regression models for each of the N linear share equations as given by
equation (3.3). A structural change can be interpreted as a shift in parameters from one
regime to another. We allow the system of share equations to follow separate transition
paths with respect to coefﬁcients on each of the individual independent variables such
that they may have different join points and adjustment rates. Building from equation

(3.3) we assume that the parameters (i.e.a ’5) shift gradually from one regime to the
other along the transition path trni(s,-, / 77,-) at an unknown join point (1;) and rate of

adjustment (77,-) such that the gradual switching translog demand system may be

expressed as:

103

Wit =a,‘ +2019- lnPj, +aiq anI +ai,t+a,-trn,-(s,-, /77i)
/

+ Zay-trni(s,-, /77,-)ln Pj, +aiqtrn,(s,-, /77,-)ln Q, +ai,trn,-(s,-, /77,~)r+ Sit; (3.4)
f
i=1,....,N —1 t =1,..,T

where

Sit = {0 for t 5 ti } (3.5).

t — t; otherwise
The transition function, (rm-(Si, / 17,.) satisﬁes (Tsurumi, 1982)
“ms—m "‘"ilsit / 771'): 1;
1im,7_,0 Inn (3;, /77,-)=1; and (3.6).
limtrni (0) = 0
We assume that the error term is normally and independently distributed with zero mean

and constant covariance 03.2 and remains constant over the whole period

(i.e. 8,, ~ NID(0,0,?' j i = 1,..., N ) (For more details on model speciﬁcations please refer to

Appendix A3.1).

3.4.3 Structural change identiﬁcation assuming common join point and rate of
adjustment

In this section, we assume that that the parameters in equation (3.3) shift gradually from
one regime to the other at an unknown join point (*and an unknown gradual rate of

adjustment 77 common to all shares (i.e.t: =t;;77,- =77j Viij ). Here we use the

104

transition function trn(s, / 77) (i.e.trni = trnj‘v’ i at j ). Equation (3.3) being re—expressed
as:

w: Xﬂ+trn(s, /77)X6+e, t= 1,...,T (3.7),
where w is an T (N —l)xl vector of observations on the dependent variable; X is an
T (N — 1)x k matrix of observations on the explanatory variables, ,6 and6 are

k xlvectors of unknown demand parameters and unknown transitional parameters
respectively used to transform the parameters to their post-shift values. Since only
(N -l) of the share equations are independent, one equation is deleted. We impose the
adding up constraints by eliminating the cost share equation for uo (unﬁnished oil) and
recovering these parameters post-estimation. Please refer to Tsurumi et al (1986),
Tsurumi and Wago (1986) for details.

Following Goodwin and Brester (1995), we conduct an F-test on the joint

signiﬁcance of the transitional parameters to test for structural change. Assuming

structural change on factor demand to be initiated at the optimal join point parameter t*

at a slow adjustment rate 77 , we also test for structural change using the following

multiple regression model:
w=z*w+z*d,y +5.1, +5 (3.8)

where Z * is the matrix of price and output data, y/ is the vector of unknown parameters,
y and 6 are the coefﬁcient vectors and a is the vector of random disturbances. d, is a

dummy variable such that

1f0rtZt*
dt = * (3.9).
0f0rt<t

105

Deﬁning d, by equation (3.9) implies that the coefﬁcient vector shifts fromw to y/ + y

at an unknown change interval given by t* (Toyoda and Ohtani, 1989). To test for
structural change, we test the following hypothesis:

H0:y=0 vs. ley¢0 (3.10).

3.4.4 Transition function specifications

Estimation of the gradual switching translog system requires the selection of a speciﬁc
functional form for the transition function, which satisﬁes the conditions given by
equation (3.6). Since the choice of transition functional form is not believed to have a
signiﬁcant impact on the outcome, we represent the transition between alternative

regimes with the hyperbolic tangent function given by equation (3.11) (Tsurumi and

Wago, 1986):

expls it / 771' ) — exp(‘ Sit / 771')
CXP(Sit /77i)+ CXP(’ Sit /77i)

 

trn(s,~,/77,~)= (3..11)

Assuming that a is distributed as a multivariate normal distribution with mean 0

and variance-covariance matrix 0217, the log-likelihood function is (Ohtani and

Katayama, 1986):

L(W,a|,*,n*)_ _[T_(__A;— 110>J g2ﬂ_ [(__T2—1))10 gza __(w—Zw)(w-Zw) (3.12)

202

 

* . . . .
where 1* and 77 are the vectors of jom pomt and rate of adjustment parameters

4' 1!

respectively (i.e. tat =(t;,...1;) and 77* =[n1 ,....,777 j) and l/l(t*,77*) is the set of

106

*
regression coefﬁcients and is a function of t* and 77 . Differentiating (3.12) with respect

I: t 2 . . . .
to VII ,7] and a , and equating resultant equations to zero, we obtain the maxrmum

III *
likelihood estimates (MLE) of t/l(t*,7] )and 0'2 conditional on 1* and 77 :

 

(IA/=(Z'Zilz'w (3.13)
A [w-ZVCJ [w—ZQ]
2 _

Substituting (3.13) and (3.14) into (3.12), we obtain the concentrated log-likelihood

function:

Lm,x(z*,.7*)= _(_T(_Ni;1_))(1 +log27r)—[T(A;_l)]log0'2 (3.15).

t t
/\ /\

at: * t
The MLE of t and 77 , say t and 77 are obtained by a grid search over I and 77* that

 

maximizes (3.15) (Toyoda and Ohtani, 1989; Moschini and Meilke, 1989). More

speciﬁcally, let the inputs be characterized by distinctive join point (i.e. t: ) and rate of
adjustment (i.e. 77; ) for i: 1,...,7. We use SAS program to conduct an extensive grid

. . . . . * a-
search over the sample data set to determine the unique comblnatlon of t and 77 for all

the inputs that maximizes the value of the concentrated log-likelihood function given by

equation (3.15).9

 

9 The SAS program is available on request.

107

 

3.4.5 Imposing Concavity Constraints

The translog model has been noted to suffer from certain limitations. It may exhibit ‘well
behaved’ results for only a restricted range of relative prices. However, outside this
limited range, the model may fail to satisfy the theoretical regularity conditions of
negative own-price elasticity (quasi-concavity) and positive shares (non-negativity)
(Caves and Christensen, 1980, Lutton and LeBlanc, 1984). Moreover, given that the
translog function’s curvature property is data dependent, under typical estimation
procedures, concavity is not guaranteed. Thus, we have imposed local concavity in the
translog function following Ryan and Wales (1998, 2000). While this approach does not
provide a global solution it improves the curvature property of the estimated function,
while still preserving the ﬂexibility of the translog functional form (Chua, Kew and

Yong, 2005).

Let X * = (P*,Q*,t*) denote the chosen normalization point. We scale the prices,

state of nature variables, and output (but not the time and dummy variables) such that

they are equal to one atX*. The Hessian (H) of the translog cost function will be
negative semi-deﬁnite, providing C *(P, Q,t)> Oif and only if the matrix G as deﬁned
by (16) below is negative semi-deﬁnite (Diewert and Wales, 1987). The ij’h element
of G , evaluated at t* , is deﬁned as

gij=a,-j—a,-*K,-j+a,-*aj i,j=1, ..... ,n (3.16),
where gij denotes the ij’h element of G matrix, ay- , aiand aj are the estimated

parameters shown in equation (3.3) and Ki]- =1 if i = jand 00therwise. Imposing

108

curvature at the reference point is attained by setting G = -—DD'with D being a triangular
matrix. Let (DD'L- denote the ij’h element of DD', and substituting (DD'L into (3.16)
and rearranging the terms, we obtain:

aij=—(DD'),-j+a,-*K —a,-*a- i,j=1, ..... ,n (3.17),

ij 1
where a ,1- denotes the ij’h element of A . The a)!- ’s in (3.17) are replaced in the system of

cost equation (3.1) and the cost share equations (3.3), thus guaranteeing that G , and

hence H, is negative semi-deﬁnite. Negative semi-deﬁniteness of G guarantees

. . . ..
concavrty of the cost functron at the reference pomt, t .

3.4.6 Symmetry and Adding-up Constraints

In order for the demand system to be consistent with neoclassical production theory, the

following restrictions are imposed:

at] =aji Viij,i,j=1,...,N; (3.18),
N
2a,- :1 (3.19).
i=1
N
Zag- = o v i =1,....,N (3.20),
i=1
N
2a,, = o v i = 1,....,N (3.21),
i=1
N
201,-, = 0 v i = 1,....,N (3.22).

109

Equation (3.18) gives the symmetry condition, while equations (3.19)-(3.22) gives the

necessary and sufﬁcient adding up conditions ensuring that C (.) is linearly homogeneous

in input prices. In addition, for a well behaved technology, the cost function must be
concave in input prices and the underlying technology must be monotonic. Concavity
implies that the matrix of second derivatives of the cost function with respect to prices is
negative semi-deﬁnite. Monotonicity requires that the predicted cost shares be positive

(Goodwin and Brester, 1995).

110

3.5 Data

Monthly time series data, October, 1993 through May, 2008 was collected for crude oil,
ethanol, liqueﬁed petroleum gases (lpg), MTBE and unﬁnished oil used as reﬁnery and
blender net inputs. Historic price data of crude oil, lpg and unﬁnished oil were obtained
from Energy Information Administration publications (Energy Information
Administration, 2008). Monthly average ethanol rack prices and MTBE prices were
collected from Nebraska website and Platts respectively (Nebraska Government Website,
2008; Platts, 2008). Labor costs were derived from the number of total production
workers in the US. petroleum reﬁnery industry and the average weekly earnings of
production workers from the reports published by the US. Department of Labor Bureau
of Labor Statistics (2008).

Capital expenditure was measured as the product of total capital stock and rental
price of capital. In this study total productive capital stock in the petroleum reﬁnery
industry has been used as a proxy for measure of capital (Rasmussen, 1997). Productive
capital stocks attempt to measure the total productive capacity of capital assets in
existence at a point in time. However, the capital stock and cost of capital measure should
be formed taking into account the loss in productive capacity of capital assets and the
decline in asset prices as the age of the asset increases.

Total nominal investment in the petroleum reﬁnery industry (V,) was assumed to

be the sum of the additional property, plant, equipment and activity during the year and
additional investment and advances made in the reﬁnery industry in that year. The

volume of investment in the asset (1,) was obtained by dividing the nominal investment

111

series by the. corresponding price series. Assuming that information regarding
investments in a particular asset type was available for period t— s to period t such that
s = 0,1 ....,S , the productive capacity of the asset in period t that is now 5 periods old is
given by:

R1,. = 2.1.... (3.23)
where (153 is the age-efﬁciency schedule representing the relative productive capacity of a

s period vintage asset to the productive capacity of -a new asset. We also assumed that the

life of the asset is ten years and the loss of productive capacity is linear. Thus the

productive capital stock K, for a particular asset type in period t was calculated as

follows (McLellan , 2004; Diewert, 1990; Diewert and Lawrence, 2000):
S
Kt = Z¢sIt-s (324)-
3:0

The depreciation rate used in the study was consistent with the petroleum reﬁning
industry depreciation rate as published by the Bureau of Economic Analysis (2008).
Annual data on the cost of capital, the weighted average cost of equity and after-tax cost
of debt, weighted by the market values of equity and debt, was used as the proxy for
rental rate of capital (Value Line database, 2008). Data on additional property, plant,
equipment and activity during the year and additional investment and advances made in
the reﬁnery industry in that year were available from Energy Finance (Energy
Information Administration, 2002). Table 3.1 presents summary statistics with respective
cost shares. Reviewing the statistics we observe that the expenditure on capital
contributes to 75.3% of the total cost with expenditure on crude oil being the second

largest component.

112

3.6 Results

Speciﬁcation tests, such as the Wu-Hausman test was conducted to test for the exogeneity
of prices and aggregate output in the translog system. Iterated SUR estimates and iterated
3SLS estimates were compared using the Hausman (1978) speciﬁcation test. Instruments
employed were lagged input prices, ending input and output stocks, lagged and
contemporaneous reﬁnery capacity, and crude oil accrual costs. Natural gas and corn are
the major inputs for the production of ethanol while butane is the main input for MTBE
production. Prices for natural gas, corn and butane were included in the list of
instruments. The Hausman test suggests prices are endogenous, and hence iterated 3SLS

are used.

3.6.1 Estimation of separate join points and rate of adjustments

Following the approach of Moschini and Meilke (1989), the separate join points and
speed of adjustment parameters were estimated via a grid search over all possible values
that maximized the concentrated log-likelihood function given by equation (3.15). Table
3.2 reports the separate individual shift points and rate of adjustments of the reﬁnery
inputs. The estimated change point for factor demand of capital was September, 1998,
which led to a more capital intensive reﬁnery and blending operation. This increased
expenditure on capital may reﬂect the surge in investments made in the reﬁnery industry
to upgrade reﬁneries to be able to respond to government mandates, implement

technological changes and increase product yields (Rasmussen, 1997). The corresponding

113

steady decline in the employment of labor in the industry over the last decade resulted in
a structural shift in the demand for labor and the estimated shiﬁ point is January, 2001.
With time, labor has been increasingly substituted by capital leading to a higher capital-
labor ratio. The reﬁnery industry also experienced a structural change in the demand for
oxygenate with a fall in the demand for MTBE and a substantial increase in the demand
for ethanol. The government policy of a gradual phase out of the use of MTBE as
gasoline oxygenates was initiated in 1999 with a targeted deadline of December 2002.
Given that the requirement was not effective immediately and the deadline was extended,
the change in factor demand were observed with a lag. According to the estimates,
structural change in the factor demand of ethanol occurred in November, 2003 followed
by that of crude oil and MTBE in January, 2004 and March, 2004 respectively.
Moreover, once the process of structural change has been initiated, the rate of adjustment
has been consistently very slow for the reﬁnery inputs. Results are in accordance with the
ﬁndings of the American Petroleum Institute (API) which states that the phase down of
MTBE use and increase in the use of ethanol as an oxygenate added to gasoline occurred
on a time schedule that allowed reﬁners and marketers to make an ‘orderly transition’
(API, 2008). Additionally, the estimates indicate that the adjustment rate for capital is
higher than the rest of the inputs.

So far, we have identiﬁed unique join points and rates of adjustment speciﬁc to
each of the inputs using the gradual switching model. Even though this is the preferred
model of estimation, to keep presentation practical, we have used a restrictive model in

the following sections.

114

3.6.2 Estimation of common join points and rate of adjustments

We have used a restrictive version of the original model to estimate the common shift
point and adjustment rates. It is assumed that reﬁnery and blender inputs experienced a

structural change in factor demand simultaneously. The estimated common shift point

It It

A A
(t ) is March, 2004. The speed of adjustment (77 ) has been estimated to be 0.0025. We

would refer to regime 1 as the time period from October, 1993 to March, 2004 and to

regime 2 as the time period from April, 2004 to May, 2008. Next we used bootstrap

techniques to estimate the statistical properties of t* and 77*. After obtaining the MLE

2

estimates of 177,0 ,t*,77*the original data set is randomly sampled with replacement

10,000 times with each sample containing exactly 176 observations (our original sample

size). The common joint point (t...) and adjustment rate (77*) have been estimated for

each resulting bootstrap sample. Statistical properties of 7* and 77* have been estimated

from the set of 10,000 bootstrap estimates generated (Krinsky and Robb, 1986). At 99%
conﬁdence interval, the joint point ranges from December, 2003 to June, 2004. The
narrow width of the range of conﬁdence interval suggests small standard error.

As a test for structural change, the hypothesis that the transitional parameters are
jointly equal to zero is strongly rejected based on the results from F-test.lo The hypothesis

that the coefﬁcient vector for the structural change model, 7! is equal to zero is strongly

rejected based on the results from t-test. The signiﬁcance of structural break was

 

‘0 The F-test statistic for structural change is 2,180.7 and is signiﬁcant at 1% level of signiﬁcance.

115

conﬁrmed by comparing the parameter estimates of iterated 3SLS in regime l and 2
(Appendix A.3.2). Conﬁrmation of structural change indicates two regimes separated by

a signiﬁcant, gradually occurring structural shift.ll

3.6.3 Own Price and Cross Price Elasticities

Own price and cross price elasticities of the two regimes were estimated using the
restricted gradual switching regression model which assumes a common shift point and
rate of adjustment.12 The elasticity estimates provide valuable insights into the nature of
the structural change. Own price elasticities were estimated at the normalized point by
using Allen’s partial elasticities of substitution for a translog function (Uzawa, 1962;

Pindyck, 1979):

E _ (777 +W,'Wj —5,]°W,')
if ' 2
W1

 

(3.25)

where 6,,- is the Kronecker delta, 7'". represents the estimated second order derivatives on

the diagonal of Hessian Matrix and w, represents the ﬁtted cost share for the i th input.
The bootstrap technique was used to estimate the statistical properties of the own

price and cross price elasticities (Douglas, 1996; Douglas and Guilkey, 1996). With the

common join point t* being identiﬁed as March, 2004, the original data set was split into

two regimes as mentioned above. The original sample section pertaining to regime 1 was

 

” Structural change may also be detected by testing for the constancy of the parameters. Chow test,
CUSUM test or the Cusum-of-squares test may be used to test for break points or structural changes in a
model. Assuming a common shift point and adjustment rate, Chow test result indicates occurrence of
structural change in July, 2002.

'2 Elasticity estimates for the seven inputs at different switch points and rates is difﬁcult to illustrate and
space consuming.

116

1‘33

'5‘

randomly sampled using the same bootstrap method as before with replacement 126
times for exactly 126 observations. Similarly, the original sample section pertaining to
regime 2 was randomly sampled with replacement 50 times each containing exactly 50
observations using the bootstrap approach. Each of the bootstrap samples generated is

normalized according to the following rule: the 1",? sample was normalized by

choosing X * = (P*,Q*,t* = i) as the normalization point and scale the prices, state of

nature variables, and output (but not the time and dummy variables) such that they are

equal to unity atX*. For each of the resulting, normalized data sets, we estimated the
system of equations using 3SLS and consequently estimate the elasticities, including
cross-price and own-price elasticities. Thus regime 1 has 126 estimates of elasticities and
regime 2 has 50 estimates of estimates, from where we obtain the statistical properties of
the elasticities.

Table 3.3 reports the own-price elasticities and the Hicks-Allen cross price
elasticities of the seven commodities for regime 1 and 2. We have reported the elasticities
correct to two decimal places. In both the regimes, all the estimates are statistically
signiﬁcant at 10% level of signiﬁcance or lower except Hicks-Allen cross price elasticity
between labor and MTBE in regime 2. The absolute magnitude of own-price elasticities
of all the inputs except capital and unﬁnished oil has decreased across regimes, implying
that demand for these goods have become less responsive to own price changes. In
particular, the demand for ethanol and MTBE has become less elastic with the absolute
value of own price elasticity decreasing from 2.25 and 2.03 in the ﬁrst regime to 2.03 and
1.91 in the second regime respectively. This has occurred at a time when MTBE’s share

of total production cost has fallen dramatically and that of ethanol increased substantially.

117

ﬁm~ffl

The decrease in ethanol and MTBE own price elasticity across regimes may be due to the
compounded effect of the Federal Reformulated Gasoline programs, mandatory RFS set
by the government and the ban on the use of MTBE has a gasoline oxygenate. Moreover,
across regimes, the factor demand for capital and unﬁnished oil has become more
responsive to own price changes with their respective own price elasticities increasing
from -0.46 and -1.76 in the ﬁrst regime to -0.59 and -1.82 in the second regime.

The cross-price effects suggest complementarity between labor and the set of
inputs including crude oil, ethanol, MTBE and lpg in the ﬁrst regime. However, in the
second regime, estimates suggest complementarity between crude oil and labor, ethanol
and labor, MTBE and labor only. The cross price elasticity estimates between ethanol and
MTBE, lpg, labor and capital have decreased from the ﬁrst regime to the second regime,
implying that ethanol demand tends to be less elastic to the price changes of these inputs.
On the other hand, the cross price elasticities between capital and crude oil, ethanol and
unﬁnished oil have increased across regimes whereas that between capital and MTBE,
lpg and labor have decreased over time. Additionally, the cross price elasticities between
MTBE and the rest of the inputs, excepting labor and capital have increased across the
regimes. Similarly, with the exception of capital, the cross price elasticities of all the
inputs with respect to lpg have increased from regime l to regime 2. Thus according to
the Hicks-Allen cross price elasticity estimates, with exceptions, the degree of
substitutability across inputs in the reﬁnery industry has increased as we move across

regimes.

118

3.6.4 Morishima Elasticity of Substitution

The deﬁnition of factors as substitutes or complements based on the signs of the Hicks-
Allen elasticities of substitution has been criticized (Goodwin and Brester, 1995). In a
more-than-two factor scenario, a scalar measure of curvature and hence relative shares is
inherently asymmetric. According to Blackorby and Russell (1989), Morishima elasticity
of substitution (MES) yields a measure of the curvature of the isoquants, rendering itself
to be a natural extension of the Hicksian two-variable elasticity. It measures the
percentage change in a ratio of two factors in response to a 1% change in the
corresponding relative price ratio. It is the logarithmic derivative of a factor quantity ratio
with respect to the corresponding factor price ratio. The Morishima elasticity of

substitution for the translog function is:

0'17 =[(}’ij +W1Wj)]_[(7ii +w,2 —w,)] (3.26).

 

 

W1 W1

Substitutability among factors in the reﬁnery and blender industry as revealed by the
Morishima elasticity of substitution for regime (l) and (2) has been reported in table 3.4.
We have reported elasticity estimates correct to three decimal places and across regimes,
the estimates are signiﬁcant at 1% level of signiﬁcance or lower. The elasticity estimates
between labor, capital and unﬁnished oil against the set of the rest of the inputs indicate
increasing substitutability between these inputs across regimes. However, the elasticity
estimates between crude oil, ethanol, MTBE and lpg against the set of the rest of the
inputs indicate decreasing substitutability with time. With the array of government

policies and regulations imposing a ban on the use of MTBE as a gasoline oxygenate and

119

simultaneously boosting the demand for ethanol as a gasoline oxygenate as well as a
volume extender, the demand for ethanol has become less responsive to own and cross
price changes. The Morishima elasticity of substitution estimate between capital and all
other inputs in both the regimes are small compared to the rest of the estimates, implying
signiﬁcantly less degrees of substitutability between capital and the other inputs.
Additionally, the Morishima elasticity of substitution estimates indicates labor and crude

oil to be complementary inputs with the degree of complementarity decreasing over time.

120

3.7 Conclusion and Policy Implications

In this paper we have uniquely utilized the multivariate gradual switching regression
techniques and translog factor demand model to estimate factor demand relationships for
inputs used in the US. reﬁnery and blender industry. Next we have identiﬁed structural
changes in factor demand relationships in the industry and have evaluated separate shift
points and rates of adjustment for the set of inputs. The underlying assumption of the
model being that given external inﬂuencing factors such as government mandates,
technological progress, changes in tastes and preferences, prices etc, the inputs tend to
follow separate transitional path, each initiated at different join points with different rates
of adjustment. The ﬂexibility of the empirical method used allows the identiﬁcation of
the change for each of the inputs while allowing speed of adjustment between alternative
regimes to be gradual and vary across inputs. We found structural change in factor
demand of inputs to exist and occur for the seven inputs at different points and
adjustment rates.

The structural change in the demand for capital and labor was initiated in late
1998 and early 2001 at slow rates of adjustment. There has been a gradual increase in the
cost share of capital followed by a gradual decrease in the cost share of labor. This was
reﬂective of the fact that the industry made additional capital expenditures to adhere to
the heightened pollution abatement requirements for the reﬁning operations. Structural
change for the rest of the inputs was initiated only in late 2003 and early 2004 at typically
slow rates of adjustment. The Renewable Fuels Standard (RFS) mandates (’02, ’05 and

‘07) along with the ‘Clean Air Act’ and the phase out of the use of MTBE initiated in

121

1999 and be substituted by ethanol have played important roles in the demand for ethanol
in the reﬁnery industry. Even though government policy requiring the gradual phase out
of the use of MTBE as the gasoline oxygenate was initiated in 1999, the requirement was
not mandatory. Thus the change in factor demand was observed with a lag. More over,
the time line of the change in factor demand in MTBE and ethanol occurred on a
schedule that allowed reﬁners and marketers to make an ‘orderly transition’.

We have used a restrictive version of the original model to estimate the common
shift point and adjustment rates assuming that all the seven reﬁnery and blender inputs
experienced a structural change in factor demand simultaneously. The estimated common
shift point is March, 2004 and the rate of adjustment is of gradual nature. Given the
statistical signiﬁcance of the common join point and adjustment rate, such an exercise is
practical and relevant. The demand for capital is less elastic than for the other factor
inputs. According to the own-price elasticity estimates, with the exception of capital and
unﬁnished oil, demand responsiveness of inputs to price changes decreased across
regimes. As we move across regimes, the own price elasticity of ethanol and MTBE have
decreased. This may be a consequence of the compounded effect of the Federal
Reformulated Gasoline programs, the mandatory Renewable Fuel Standard set by the
government and the ban on the use of MTBE has a gasoline oxygenate. The Hicks-Allen
cross price elasticity estimates suggest complementarity between labor and rest of the
inputs including crude oil, ethanol, MTBE and lpg in the ﬁrst regime and crude oil,
ethanol and MTBE in the second regime. Morishima elasticity of substitution estimates

suggests a decrease in the degree of substitutability among factors in the reﬁnery industry

122

across regimes in general. However, capital, labor and unﬁnished oil have been
increasingly substituted by other inputs as we move from regime 1 to regime 2.

Results indicate that the array of government policies and mandates imposing a
gradual phase out of the use of MTBE as a gasoline oxygenate and simultaneously
encouraging the use of ethanol as the alternative oxygenate as well as the gasoline
volume extender has successfully altered the structure of factor demand in the reﬁnery
and blender industry in US. Considerable decrease in the factor demand of MTBE was
conﬁrmed and can be expected to curb associated future environmental hazards.
Simultaneously, the signiﬁcant increase in the factor demand of ethanol is consistent with
the goal of US. policies aimed at decreasing U.S. dependency on foreign oil.

This research work is the ﬁrst known study that evaluates structural change in
factor demand in the petroleum reﬁning industry in the US. We use industry speciﬁc
aggregate data in our analysis and impose the symmetry and homogeneity constraints in
the cost function analysis. However, given the nature of the data, some constraints that
are satisﬁed at the micro units may no longer be satisﬁed at the aggregate level. Hence
following Illmakunnas (1986), an extension of the current research work may be to
impose the constraints on the parameters stochastically in the system of cost share
equations. Moreover, our methods can be applied in future work to identify separate
switch points and rates of adjustment for individual factors used in other industries while

evaluating structural changes in demand.

123

Chapter 3 Appendix

124

A3.1 Model details

Equation (3.4) may be represented as:
w* = Z*1,u + g, (3.27)

where w* is a T(N — 1)x 1 vector of shares,t// is the vector of unknown parameters and a

is the vector of random disturbances (Tsurumi et a1, 1986). Since only N -— l of the share

equations are independent, one equation is deleted,
w" = (w; , w'2 ,...,w',(,_1 ), (3.28)

where w,- =(w1,-,w2,-,...,wT,-)', i=l,...,N—1. Z*is the matrix of price and output data

given by the following equation:

 

 

P(X1,trn10X1) O ...... O —
O (X2,trn2 0X2)
2" =
_O 0 ...... (XN_1,trnN_1 0 XN"1)4T(N-1)X2k
(3.29)
— le trnj 0x111
x '2 trn - ox '2 .
(Xj,trnj 0X1): 1 J J _] =1,2,...,7 (3.30)
_ ij trnjoxJ-T‘TXZ

 

 

125

 

Let w = (3.31)

,BN—i

-6N -1 _ 2kxl

 

 

where ,6,- and 19,- are vectors of regression coefﬁcients such that

31 = (ﬁlis-H-aﬂki)and6i = (91;. ----- .91“). V i = 1,.--.N-1and

8=[8] £2 ........ gN-l].T(N-l)><l'

126

Table A3.2
Parameter Coefﬁcients of the Translog model

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Coefﬁcient Estimates Regime 1 Regime 2

(Pooled model (1993,0ctober to (2004,April to

results) 2004,March) 2008, May)
or, 0.29* 0.28* 0.25*
a2 3.25E-3* 1.37E-3* -l.24E-2*
a3 l.25E-2* 1.09E-2* -2.93E-3
a4 1.93E-3 ** 4.27E-3 * l.64E-2*
a5 7.00E-6* 5.92E-6* 1.71E-6***
0‘6 0.67* 0.69* 0.75*
“crude oil, crude oil 012* 011* 0-18’“
acrude oil, ethanol '1'92E'4 '9-6E'4 '3 -09E-2* *
aonndo oil, lpg 29213-3 4.08E-3*** -1.97E-2
“crude oil, mtbe -4.98E-3 * -4.37E-3 ** 1.28E-3
acrude oil, unﬁnished oil 8’51E'4 ”2'855’3 '6-965'3
acrude oil, labor 4535'“ -3-01E-6* 6.76E-7
acrude oil, capital '0-12* -0.1 1* -0. 17*
“crude oil, gasoline 5‘81E'4 6-72E'4 -2.00E-5
“crude oil, year 5-55E-4* -4.4lE-4* -5.4lE-4***
aonndo oil, month 3-005-5 -2-32E-4 6.58134
“crude oil, month 3-005'5 432134 -5.58E-4
aethanol, ethanol 6-495'3" 14713-3“ 7.82E-3*
aethanol, lpg -1-6lE-4 -9.1 113.4 1.62E-4
aethanol, mtbe '3 725-3 "‘ -l .64E-3 * 9.9E-4**
“ethanol, unﬁnished oil “95‘3" 3235-3” 4.03E-4
“ethanol, labor '2'47E'7 -8-OlE-7** 8.759E-8
aethanol, capital '6'92E'3* “2.1933" -4.3 113-3*
aethanol, gasoline l "005-5 -9. 1 213-6 -1 .3 E-4* *

 

 

 

 

 

 

Table A3 .2 Continued

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

acapital, month

 

 

 

Coefﬁcient Estimates Regime 1 Regime 2
(Pooled model (1993,0ctober to (2004,April to
results) 2004,March) 2008, May)
aethanol, year 3-5'13'6 1515-5” 1.28E-4*
aethanol, month ‘Z‘OOE'S "62436 -4.2 1 B7
alpg, lpg 15213-2“ 1.33E-2* 1.29E-2*
alpg, mtbe -2.81E-3** -1.9113—3** -l.08E-3
“lpg. unﬁnished oil '4'025'3*** '6-3913-3“ -3.58E-3**
alpg, labor 2.05E-6 3.93E-7 -2.04E-6**
a lpg, capital -1.21E-2* -8.21E-3* -6.52E-3*
a ,pg, gasohne -6.03E-6 -7.38E-6 4,415-4"
a lpg, year '3-00E-5* -7.00E-5* 4.22E-5**
a lpg, month -3-4E-5** 7.5113-5 9.81E—5***
a ”we, mbe 6.78E—3* 6.54E-3* 3.63E—3*
amtbe,unﬁnished oil '9'61B'4 6-015'4 -1.23E-3
amtbe, labor '1‘6E'7 '8-365'8 4325-7
a mtbe, capital 5'68E'3* 7-795'4 -1.64E-3**
a mtbe, gasoline '2’05 E-S '5-24E'6 15325-4“
a mtbe, year 1-71E-5* 8.94E-6 -1.111=..4*
a mtbe, month 1-64E-6 -3.7E-6 21313-5
alabor, unﬁnished oil 7-335'7 39313-7 -2.46E-7
“labor, labor l-81E-6* 2.68E-6* 5.15E-6*
atahor, capital '1-715'6" -7-71E-8 -3.2E-6*
alabor, gasoline 36913-8“ 3.86E-8** -2.86E-9
atahor, year -2-97E-8* -2.26E-8 6.08E-9
alabor, month '7-275'9 -1-61E-8 -3 2213-9
“capital, unﬁnished oil '8'99E'3 * '2-61E'4 'l-OlE-Z"
acapital, capital 0'14,“ 0.12* 0.19*
acapital, gasoline '3'01E'4 '3-415-4 6.13E-4
acapital, year 6-345-4" 4.82E—4* 4.14E-4
-3.4E-4 14113-4 -1.03 5-3

 

*, ** and *** represents 1%, 5% and 10% level of signiﬁcance respectively

128

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

94:885. 8
Ehnmvwhm $m.~mﬁoo: 355?.va N3; maﬁmc £2820
Ron? Ed mod Sod mod 81m £2.30
€8.83 2335.5
wdm 0.8 NOW 3.3 3:850
3% 5m 8
$38? 362 bwdwm No.3 ﬂ ..mmm 02.5 cons
935m @5325
wow; 09.; Sad—cw 533.: £525
£055 5m 3
{cad vm; _ 3N3 modm Snow BE :0 cocmEccb
Eaton cannon:
mmmﬁ owo.E Sonia 2.2% _ .w bingo
cobwm con @
Xomd ocﬁm 3.3 3.: 3.3. 8cm on:
@otmm 9832C
2 new.» Deoﬁm omdmmﬁ 3:550
9095 Sn 9
$33 3 .3 3.9: mwdm 3.3 8E mmHE
28.3 @5325
omm $5.3 365$ 5.85% Q2530
22.8m— nom 8
$93 wsm £52 5. _ N ovdo BE .ocnﬁm
32am U532:
omoﬁom 235m 8.25.3 oodnmgmv bucnso
autum— hom mv
$m.- mm: www— nmdm whim SE :0 02:0
9:25
:60 :5:— 55.3%;—
uwano>< £2 :2 c.5255 582 5:58:80

 

 

 

 

 

 

 

gig—Sr 820m mo moumuﬁm 355m
ﬁm 033.

129

 

Table 3.2
Estimated Individual Join Points and Adjustment rates

 

 

 

 

 

 

 

 

Refinery and Individual Shift Point Adjustment
Blender Input Rate
Crude oil January, 2004 0.0025
Ethanol November, 2003 0.0025
LPG September, 2003 0.0025
MTBE March, 2004 0.0025
Unﬁnished Oil January, 2004 0.0025
Labor January, 2001 0.0025
Capital September, 1998 0.0125

 

 

 

130

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3,0380%“: 852MB»? mo _o>o_ OR: was o\..w ..x; 35832 1;. can I .*

*3. T EN; goanﬁvé *m-mwm.~ *m-mmv.a *N-mmv._ ..de :0 85?ch
...N-mmm.m .30? 3.8%...“ “8&de ...N-m:o._ 3-5%; *mmd 3:me
*Nlmmm N *had *mN.N- m-mmm.h- ..cIKAXO ...NHnKbﬁ .13.. _©.o- gonad
*N-mmv.m 3H; :ngé. ... a. T ...Imﬁnﬁﬁ .w 3.9%; ...omd mmtz
$-52 :2 33mm; .38.” *5. _- at"; 2 :2 3.5 so saga 85%:
*Nﬂwm N ..va *m-m:N.T ...mﬁnmd *N-m:o._ *mo.m- Lmd 355m
...Nkmmm N * S: ...mﬂoﬁc- ...m-mSV.N 3-9:: 3&3: ...vWT :0 23:0
3.2 .88 s .Eﬂvgs N 253m
EFT ...omg ...o-m:w.w ...mﬂNOd 3-9.56 *m-mo_.m *wmd :0 3.3:ch
*N-m om; *cvd- “his; ...N-m:o.~ 3-5:: *m-m: Wm *Nvd 3:30
*N-mwm._ .Lmd LON- ..:.T ...mmd- *wwd- ...oodT “one:
Rana»: *omg *m-m3 . T ...mON- ...mﬂmmd *Ymmaé ...mmd mmhz
3.9%; *9: *Ymaod- *m-m=w.m .5». T ...mﬁmn; ...mmd REV .30 Easy—05$ BEBE;
...N-m=...m._ *Nng «Mum—cod- ...imﬁaoé adv—mm; ...mNN- ...de 655m
...NHn—Ql *3; ...mRnGQV. 3&8“: *m-m Nod *m-m~ ﬂ .N ...E..T :0 09:0
EnhazJécﬂ 3 ..onSXﬁmaaG u oﬁﬁom

Ea: $0

:0 .553me
uoamEucD 3:me gonad mmtz 35:75 .0523. :0 09:0 moszcmzo

 

 

 

 

 

 

 

 

 

856335 8.5-380 ES :30
m.m 2an

131

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

bo>toommoh ooﬂwocmawﬁ ..‘o ~o>o_ go— Us o\om ..XL mucumohmou 3.2., Us as." ._._

.2 3 La: RN“: ES; :5: *2 3 E 852%:
5:3 *82 *32 *8: $85 *2: 358
32 38a .33 *2: 38.” :amS- :53
:8; *8? “.83 33 RN? ii may:
*5: *8: ".22 *2: *3: a2 8% 26 53.23 8:55
$25 $35 :8.“ 38a 33 $3 655
302 33a .32 *3? if am: :0 025
an: .23 3 .Eﬂzau N 259m
*8; *8: at; E: aw: LEN :0 BEES
3:3 .336 35d $53 333 *8: .336
* and a“? $8; a”: *2 2 $8.2- :53
:2: Reg *5: *3: an: Emu m5:
:8.” *8? $3: $9: Ea; 3de am: 30 E=2e§ 85%:
$02 $36 Lia 1&3 163 .32 62:3
*3: * $2 *8: as: .8: *8: =0 255
£932.33 3 585.83 ﬂ 253m

an:

:0 mao ESQ—0.50m
Bﬁécs 3:5 .83 m3: 85%: .253 =0 2:5 8:530

 

 

 

 

 

 

 

 

 

compacmnsm mo moumoumﬁm «EEmtoE

Tm

23

132

Figure 3.1
Renewable Fuel Standard Mandates and Ethanol Production

 

RFS Mandates and Ethanol Production

20.00 —
18.00 —
16.00 -
14.00 -
12.00 -
10.00 -

Billions of Gallons

    

PPS”?

0000

0000
P.

 

0.00

L4

— l

@§:¢§:&weﬁ§§§§§§§§§§$§$§¢

Yea rs
—-— Total Bhanol Production (Billions of Gallons) —I— Dermnd for Bhanol (Billion gallons)
—t—Energy Policy Act, 2005 —0— Energy Policy Act, 2007
—0— Energy Policy Act, 2002

 

 

 

 

 

 

Source: Energy Policy Act of 2002, Senate Amendment 2917,

MW ”MPH“! senate any," ‘ ’ uvbill/2917 analvsk ndf‘ Energy Policy Act Of 2005, url:
ht_tp://www.doi.gov/iepa/EnergyPolicyActon005 pdf,

Energy Independence and Security Act of 2007, H.R.6, SEC. 202.

Renewable Fuel Standard (2008).

133

BIBLIOGRAPHY

134

Ahn J, Jeong G, Kim Y. A forecast of household ownership and use of alternative fuel
vehicles: A multiple discrete-continuous choice approach. Energy Economics 2008;30;
2091—2104.

Alfnes F, Rickertsen K. European Consumers' Willingness to Pay for US. Beef in
Experimental Auction Markets. American Journal of Agricultural Economics 2003;85;
396-405.

Alfnes F. Stated preferences for imported and hormone-treated beef: application of a
mixed logit model. European Review of Agricultural Economics 2004;3 1; 19-37.

Allenby GM. Hypothesis Testing with Scanner Data: The Advantage of Bayesian
Methods. Journal of Marketing Research 1990;27; 379-3 89.

Anderson, Ronald W. and Danthine, Jean-Pierre, “Cross Hedging”, The Journal of
Political Economy, Vol. 89, No. 6. (Dec., 1981), pp. 1182-1196.

Anderson, Ronald W. and Danthine, J can-Pierre, “The Time Pattern of Hedging and the
Volatility of Futures Prices”, The Review of Economic Studies, Vol. 50, No. 2 (Apr.,
1983), pp. 249-266

Anderson, J. D, Coble, K. H, 2008. Impact of renewable fuels standard ethanol mandates
on the corn and livestock markets. Working paper, Department of Agricultural
Economics, Mississippi State University.

American Petroleum Institute, 2008.
http://www.api.org/_?gclid=CK6bxdizOchFREeDOodLOhjm. Cited 22 September
2008.

Baillie, Richard T. and Myers, Robert J ., “Bivariate Garch Estimation of the Optimal
Commodity Futures Hedge”, Journal of Applied Econometrics, Vol. 6, No. 2. (Apr. -
Jun., 1991), pp. 109-124.

Baillie, Richard T. and Bollerslev, Tim, “The Message in Daily Exchange Rates: A
Conditional-Variance Tale”, Journal of Business & Economic Statistics, Vol. 20, No. 1,
Twentieth Anniversary Commemorative Issue (Jan., 2002), pp. 60-68

Banerjee, A., J .J . Dolado, J .W. Oalbraith and DP. Hendry, 1993, Co-integration, error-
correction, and the econometric analysis of non-stationary data (Oxford University Press,
Oxford).

Beard JG, Ragheb MG. Measuring Leisure Motivation. Journal of Leisure Research
1983;15; 219-228.

Berrens R. “The Advent of Internet Surveys for Political Research: A Comparison of
Telephone and Internet Samples. Political Analysis 2003,11; 1—22.

135

Benninga, Simon, Eldor, Rafael and Zilcha, Itzhak, “The Optimal Hedge Ratio in
Unbiased Futures Markets”, Journal of Futures Markets; Summer84, Vol. 4 Issue 2,
p155-159, 5p

Bhattacharjee S, Petrolia D, Hemdon CW. Estimating Willingness to Pay for E10 fuel: a
contingent valuation method. Selected Paper prepared for presentation at the Southem
Agricultural Economics Association Annual Meeting, Dallas, Texas, February 2-5, 2008

Blackorby, C., Russell, R. R., 1989. Will the real elasticity of substitution please stand
up? (A comparison of the Allen/Uzawa and Morishima elasticities). The American
Economic Review 79, 882—888.

Bollerslev, Tim (1986), "Generalized Autoregressive Conditional Heteroskedasticity,"
Journal of Econometrics, 31, 307-327.

Boxall PC, Adamowicz WL. Understanding Heterogeneous Preferences In Random
Utility Models: A Latent Class Approach. Environmental And Resource Economics
2002;23; 421—446.

Boxall PC, MacNab B. Exploring The Preferences of Wildlife Recreationists for Features

of Boreal Forest Management: A Choice Experiment Approach. Canadian Journal of
Forest Research 2000;30; 1931-1941.

Brinker,A., J. Parcell, and K. Dhuyvetter. 2007. “Cross-Hedging Distillers Dried Grains:
Exploring Corn and Soybean Meal Futures Contracts.”Proceedings of the NCCC-134
Conference on Applied Commodity Price Analysis, Forecasting, and Market Risk
Management. Chicago, IL. [http://www.farmdoc.uiuc.edu/nccc134].

Bureau of Economic Analysis, 2008. lmJ/wwwbeagovﬁ Cited 6 June 2008.
Castelino, Mark G., “Hedge Effectiveness: Basis Risk and Minimum-Variance Hedging”,
The Journal of Futures Markets, Vol. 12, No. 2, 187-201 (1992)

Caves, D. W., Christensen, L. R., 1980. Global properties of ﬂexible functional forms.
The American Economic Review 70, 422—432.

Cecchetti, Stephen G., Cumby, Robert E. and F iglewski, Stephen, “Estimation of the
Optimal Futures Hedge”, The Review of Economics and Statistics, Vol. 70, No. 4 (Nov.,
1988), pp. 623-630.

Center for Agricultural and Rural Development (CARD), url:
http://www.card.iastate.edu/research/bio/tools/hist_eth gm.aspx viewed on 07/10/09

Chicago Broad of Trade (CBOT), 2009
Chiu, Chien-Liang, Wu, Pei-Shan, Chen, Chun-Da and Cheng, Wan-Hsiu, “Hedging with

F [cor-Traded and E-Mini Stock Index Futures”, Journal of Business and Economics,
2005, Vol. 44, pg. 49

136

Chua, C. L., Kew, H., Yong, J ., 2005. Airline code-share alliances and costs: imposing
concavity on translog cost function estimation. Review of Industrial Organization 26,
461--487. ‘

Coltrain, D., D. Barton, and E. Dean. 2004. Risk Factors in Ethanol Production.
Available online at
http://www.agmanager.info/agribus/energy/Risk%20Factors%20in%2OEthano1%20Produ
ction%204-O4.pdf.

Commodity Research Bureau (CRB), 2009, url: http://www.crbtrader.com/. Cited
05/27/2009.

Crandall R. Motivations for Leisure. Journal of Leisure Research l980;2; 45-53.

Cummings RG, Harrison GW, Rutstrom EE. Homegrown Values and Hypothetical
Surveys: Is the Dichotomous Choice Approach Incentive-Compatible?. The American
Economic Review 1995; 85; 260-266.

Dahan E, Hauser JR, Simester D, Toubia 0. Application and Test of Web-based
Adaptive Polyhedral Conjoint Analysis. Working Paper No. 146. Center for
eBusiness@MIT.

Dahlgram, R.A., 2009, “Inventory and Transformation Hedging Effectiveness in Corn
Crushing”, Journal of Agricultural and Resource Economics, 34(1): 154-171.

Diewert, W. E., 1980. Aggregation problems in the measurement of capital, in: Usher, D.,
(Ed.), The measurement of capital, The University of Chicago Press, Chicago.

Diewert, W. E., Wales, T. J ., 1987. Flexible functional forms and global curvature
conditions. Econometrica 55, 43--68.

Diewert, W. E, Lawrence, D. A., 2000. Progress in measuring the price and quantity of
capital, in: Lau, L. J. (Ed.), Econometrics and the cost of capital: Essays in Honor of Dale
W. Jorgenson, The MIT Press, Cambridge MA.

Douglas, S., 1996. Bootstrap conﬁdence intervals in a switching regressions model.
Economics Letters 53, 7--15.

Douglas, S., Guilkey, D. K., 1996. Bootstrap standard error estimates in a switching
regression model with unknown switch point. Communications in Statistics - Simulation
and Computation 25, 1 -- 23.

Dufﬁe, Darrell and Jackson, Matthew 0, “Optimal Hedging And Equilibrium In

A Dynamic Futures Market”, Journal of Economic Dynamics and Control 14 (1990) 21-
33. North-Holland

137

Dufﬁeld, J. A., Collins, K. 2006. Evolution of renewable energy policy. Choices 21, 9--
14.

Economic Analysis and Research (EAR), LLC, 2008

Energy Information Administration, 2008. US. Department of Energy.
http://www.eia.doe.gov/. Cited 15 April 2008.

Engle, Robert F. and Granger, C. W. J ., “Co-Integration and Error Correction:
Representation, Estimation, and Testing”, Econometrica, Vol. 55, No. 2. (Man, 1987),
pp. 251-276.

Ewing G, Sarigollu E. Assessing consumer preferences for clean-fuel vehicles: A discrete
choice experiment. Journal of Public Policy & Marketing 2000;19;106-118.

Fischer, P. A., Bitzan, J ., Tolliver, D., 2001. Analysis of economies of size and density
for short line railroads. Dissertation, North Dakota State University.

F ranken, Jason R. V. and Parcell, Joe L., “Cash Ethanol Cross-Hedging Opportunities”,
Journal of Agricultural and Applied Economics, Dec 2003

Fleming CM, Bowden M. Web-based surveys as an alternative to traditional mail
methods. Journal of Environmental Management 2009;90; 284-292.

Fox JA, Shogren J F, Hayes DJ, Kliebenstein JB. CVM-X: Calibrating Contingent Values
with Experimental Auction Markets. American Journal of Agricultural Economics 1998;
80; 455-465.

Funk SM, Zook J E, Featherstone AM. Chicago Board of Trade Ethanol Contract
Efﬁciency. Selected Paper prepared for presentation at the Southern Agricultural
Economics Association Annual Meeting, Dallas, TX, February 2-6, 2008.

Gao Z, Schroeder T. Effects of Additional Quality Attributes on Consumer Willingness-
To-Pay for Food Labels. American Journal of Agricultural Economics 2009;91; 795—809.

Goodwin, B. K., Brester, G. W., 1995. Structural change in factor demand relationships
in the US. food and kindred products industry. American Journal of Agricultural
Economics 77, 69--79.

Greene WH, Hensher DA. A Latent Class Model For Discrete Choice Analysis: Contrasts
With Mixed Logit. Working Paper, lTS-WP-02-08, Institute Of Transport Studies 2002;

Hanemann WM. Applied Welfare Analysis with Qualitative Response Models. Working
Paper No. 241, University of California, Berkley, 1982;

138

Hart, Chad E., “Agricultural Situation Spotlight: Ethanol Revisited”, Center for
Agricultural and Rural Development, Iowa Ag Review, Summer 2005, Vol. 11, No. 3

Hausman, J. A., 1978. Speciﬁcation tests in econometrics. Econometrica 46, 1251--1271.

Holton, Glyn A., “Value-at-Risk Theory and Practice”, Sample Chapter 6, Copyright ©
Academic Press, 2003, url: http://www.value-at-risk.net/content/Holton_6.pdf

Horne M, J accard M, Tiedemann Ken. Improving behavioral realism in hybrid energy-
economy models using discrete choice studies of personal transportation decisions.
Energy Economics 2005;27; 59— 77.

Huang, K. S., 1991. Factor demands in the US. food-manufacturing industry. American
Journal of Agricultural Economics 73, 615--620.

Ilmakunnas P., 1986. Stochastic constraints on cost function parameters: mixed and
hierarchical approaches. Empirical Economics 11, 69--80.

Johnson, Leland L., “The Theory of Hedging and Speculation in Commodity Futures”,
The Review of Economic Studies, Vol. 27, No. 3 (Jun., 1960), pp. 139-151.

Kahl, K.H., “Determination of the recommended hedging ratio”, American Journal of
Agricultural Economics, Vol. 65, No. 3, (Aug, 1983), pp. 603-605

Kamakura WA, Russell GJ. A Probabilistic Choice Model for Market Segmentation and
Elasticity Structure. Journal of Marketing Research 1989;26; 379-390.

Karp, Larry 8., “Methods for Selecting the Optimal Dynamic Hedge When Production Is
Stochastic”, American Journal of Agricultural Economics, Vol. 69, No. 3 (Aug, 1987),
pp. 647-657.

Katayama, S., Ohtani, K., Toyoda, T., 1987. Estimation of structural change in the import
and export equations: an international comparison. The Economic studies quarterly 38,
148--158.

Koplow, D., 2006. Biofuels — at what cost? government support for ethanol and biodiesel
in the United States. The Global Studies Initiative (GSI) of the International Institute for
Sustainable development (IISD). http://www.globalsubsidies.org/ﬁles/assets/Brochure_—
_US_Update.pdf. Cited 07 July 2008.

Koutrnos, Gregory and Pericli, Andreas, “Hedging GNMA Mortgage-Backed Securities

with T-Notes Futures: Dynamic versus Static Hedging”, Real Estate Economics, Vol. 27,
No. 2, pp. 335-363

139

 

Krehbiel, Tim and Adkins, Lee C., “Cointegration tests of the unbiased expectations
hypothesis in metals markets”, Journal of Futures Markets, Volume 13 Issue 7, 753 —
763.

Krinsky, I., Robb, A. L., 1986. On approximating the statistical properties of elasticities.
The Review of Economics and Statistics 68, 715--7l9.

Kroner, Kenneth F. and Sultan, Jahangir, “Time-Varying Distributions and Dynamic
Hedging with Foreign Currency Futures”, The Journal of Financial and Quantitative
Analysis, Vol. 28, No. 4. (Dec., 1993), pp. 535-551.

Kuhfeld W. Marketing Research Methods in SAS: Experimental Design, Choice,
Conjoint and Graphical Techniques. SAS Institute Inc. Cary, NC, USA: SAS Institute
Inc. 2003.

Kypri K, Stephenson S, Langley J. Assessment Of Nonresponse Bias In An Internet
Survey Of Alcohol Use. Alcoholism: Clinical And Experimental Research 2004;28; 630-
634.

Lapan, Harvey E. and Moschini, Giancarlo, “Futures Hedging Under Price, Basis and
Production Risk”, American Journal of Agricultural Economics, Vol. 76, No. 3 (Aug,
1994), pp. 465-477

Leibtag E. Corn Prices Near Record High, But What About Food Costs?. Amber Waves,
2008, February. url:
http://www.ers.usdagov/AmberWaves/FebruaryO8/Feagures/CornPriceshtm

Lin Y, Tanaka S. Ethanol fermentation from biomass resources: current state and
prospects. Applied Microbiology and Biotechnology 2006;69; 627-642.

Lindhjem H, Navrud S. Internet CV surveys - a cheap, fast way to get large samples of
biased values?. 2008; Available at SSRN: http://ssm.com/abstract=1295521

Louviere JJ, Hensher DA, Swait JD. Stated Choice Methods: Analysis and Application.
Cambridge. UK: Cambridge University Press; 2000.

Louviere J J . Conjoint Analysis Modeling of Stated Preferences: A Review of Theory,
Methods, Recent Developments and External Validity. Journal of Transport Economics
and Policy 1988;22; 93-119.

Lusk J L, Fox JA, Schroeder TC, Mintert J, Koohmaraie M. In-Store Valuation of Steak
Tenderness. American Journal of Agricultural Economics 2001; 83; 539-550.

Lusk J L. Effects of Cheap Talk on Consumer Willingness-to-Pay for Golden Rice.
American Journal of Agricultural Economics 2003;85; 840-856.

140

Lusk J L, Roosen J, Fox JA. Demand For Beef From Cattle Administered Grth
Hormones or Fed Genetically Modiﬁed Corn: A Comparison of Consumers in France,
Germany, the United Kingdom and the United States. American Journal of Agriculture
Economics 2003;85;16-29.

Lutton, T. J ., LeBlanc, M. R., 1984.,Comparison of multivariate logit and translog
models for energy and nonenergy input cost share analysis. The Energy Journal 5, 35--44.

Martin, Larry and Garcia, Philip, “The Price-Forecasting Performance of Futures Markets
for Live Cattle and Hogs: A Disaggregated Analysis ”, American Journal of Agricultural
Economics, Vol. 63, No. 2, (May, 1981), pp. 209-215

Mathews, Kenneth H. and Holthausen, Duncan M. Jr., “A Simple Multiperiod Minimum
Risk Hedge Model”, American Journal of Agricultural Economics, Vol. 73, No. 4, (Nov.,
1991), pp. 1020-1026.

McFadden D. The choice theory approach to market research. Marketing Science 1986;5;
275-297.

McFadden D 1974. Conditional logit analysis of qualitative choice behavior. In:

Zarembka P (Eds), Frontiers in Econometrics, New York: Academic press; l974.p.105-
142.

McLachlan GJ, Basford KE. Mixture Models: Inference and Applications to Clustering.
Marcel Dekker, New York; 1988.

McLellan, N., 2004. Measuring productivity using the index number approach: an
introduction. New Zealand Treasury Working Paper 04/05.

Miranowski, J. A., 2007, Biofuel incentives and the energy title of the 2007 farm bill.
Staff General Research Papers, Iowa State University, Department of Economics.

Moller RM. Brief on Biomass and Cellulosic Ethanol. California Research Bureau 2005;
Prepared on the request of Senator Richard Alarcon, url:
http://dpw.lacountj.gov/epd/tf/Attachments/CRB_Biomasstt 12-2005.pdf, cited on
04/04/09.

Moschini, G., Meilke, K. D.,1989. Modeling the pattern of structural change in US. meat
demand. American Journal of Agricultural Economics 71,251--61.

Moschini, GianCarlo and Myers, Robert J ., “Testing for constant hedge ratios in

commodity markets: a multivariate GARCH approach”, Journal of Empirical Finance 9
(2002) 589— 603

141

 

Myers, Robert J. and Thompson, Stanley R. “Generalized Optimal Hedge Ratio
Estimation”, American Journal of Agricultural Economics, Vol. 71, No. 4 (Nov., 1989),
pp. 858-868

Myers, Robert J, “Estimating Time-Varying Optimal Hedge Ratios on Futures Markets”,
The Journal of Futures Markets (1986-1998); Feb 1991; 11, 1; ABI/TNFORM Global

Myers, Robert J. and Hanson, Steven D., “Optimal Dynamic Hedging in Unbiased
Futures Markets”, American Journal of Agricultural Economics, Vol. 78, No. 1, (Feb.,
1996), pp. 13-20

Nebraska Government Website, 2008. http://www.neo.ne.gov/statshtml/66.htrnl. Cited 15
July 2008.

New York Mercantile Exchange (N YMEX), url: http://www.nvmex.com/index.a_sp_x_,
cited on 06/04/2009

Niven RK. Ethanol in gasoline: environmental impacts and sustainability review article.
Renewable and Sustainable Energy Reviews 2005;9; 535—555.

Ohtani, K., Kakimoto, S., 1990. A gradual switching regression model with a ﬂexible
transition path. Economics Letters 32, 43--48.

Ohtani, K., Katayama, S., 1986. A gradual switching regression model with
autocorrelated errors. Economics Letters 21, 169--172.

Paddrik, Mark E. and Ginger, M. Davis, “Strategies for Hedging and Trading in the
Emerging Ethanol Commodities Market”, Proceedings of the 2008 IEEE Systems and
Information Engineering Design Symposium, University of Virginia,Charlottesville, VA,
USA, April 25, 2008

Pannell, David J., Hailu, Getu, Weersink, Alfons, and Burt, Amanda, “More reasons why
farmers have so little interest in futures markets”, Agricultural Economics 39 (2008) 41—
50

Park, Tae H. And Switzer, Lorne N., “Bivariate Garch Estimation Of The Optimal
Hedge Ratios For Stock Index”, The Journal Of Futures Markets (1986-1998); Feb 1995;
15, 1; Abi/Inform Global, Pg. 61

Peck, Anne E., “Hedging and Income Stability: Concepts, Implications, and an
Example”, American Journal of Agricultural Economics, Vol. 57, No. 3. (Aug, 1975),
pp. 410-419.

Pindyck, R. S., 1979. Interfuel substitution and the industrial demand for energy: an
international comparison. The Review of Economics and Statistics 61, 169--179.

142

 

Platts, 2008. http://www.platts.com/. Cited 23 June 2008.

Potoglou D, Kanaroglou PS. Household demand and willingness to pay for clean
Vehicles. Transportation Research 2007;12; 264-274.

Quandt, R. E., 1958. The estimation of the parameters of a linear regression system
obeying two separate regimes. Journal of the American Statistical Association 53, 873--
880.

Rask, K. N., 1998. Clean air and renewable fuels: the market for fuel ethanol in the US.
from 1984 to 1993. Energy Economics 20, 325--345.

Rasmussen, J. A., 1997. The impact of environmental compliance costs on US. reﬁning
proﬁtability, key ﬁndings. Energy Information Administration.
http://www.eia.doeggov/emeu/permro/refgi/keyﬁnd.html. Cited 03 June 2008.

Renewable Fuels Association, 2008. http://www.ethanolrfa.org[. Cited 12 December
2007.

Reuters 3000 Xtra Hosted Terminal Platform, url:
https://portal.hpd.global.reuters.corn/auth/login.aspx, Cited 06/10/2009

Rickertsen, K., 1996. Structural change and the demand for meat and ﬁsh in Norway.
European Review of Agricultural Economics 23, 316--330.

Ryan, D. L., Wales, T. J ., 1998. A simple method for imposing local curvature in some
ﬂexible consumer demand systems. Journal of Business and Economic Statistics 16, 331-
-338.

Ryan, D. L., Wales, T. J ., 2000. Imposing local concavity in the translog and generalized
Leontief cost functions. Economics Letters 67, 253--260.

Schafer, A., 2005. Structural change in energy use. Energy Policy 33, 429--437.

Schmit, Todd M., Luo, J ianchuan and Tauer, Loren W., “Ethanol Plant Investment using
Net Present Value and Real Options Analyses ”, Working Paper Department of Applied
Economics and Management Cornell University, Ithaca, New York 14853-7801 USA.

Shephard, R. W., 1953. Cost and production ﬁmctions. Princeton University Press,
Princeton.

Simmons, Phil, “Why do farmers have so little interest in futures markets?”, Agricultural
Economics 27 (2002) 1-6

Small KA, Rosen HS. Applied Welfare Economics with Discrete Choice Models.
Econometrica 1978;49; 43-46.

143

Sproles GB, Kendall EL. A Methodology for Proﬁling Consumers’ Decision-Making
Styles. Journal of Consumer Affairs 1986; 20; 267- 279.

Suhr, Diana D., “Exploratory or Conﬁrrnatory Factor Analysis?”, Statistics and Data
Analysis, SUGI 31 Proceedings, March 26-29, 2006.

Standard & Poor’ 3, url:
1m: //www2. standardandpoors. conﬂportal/site/sp/en/us/pa aeg ehome/home/O.0.O.0,OL0.O.0.
0.0.0.0.0.0.0.0.html Cited 04/13/2009

Stern, R. M., Baum, C. F ., Greene, M. N., 1979. Evidence on structural change in the
demand for aggregate U.S. imports and exports. The Journal of Political Economy 87,
179--192.

Swait J. A structural equation model of latent segmentation and product choice for cross-
sectional revealed preference choice data. Journal of Retailing and Consumer Services
1994;l; 77-89.

Swait J, Adamowicz W. The Inﬂuence of Task Complexity on Consumer Choice: A
Latent Class Model of Decision Strategy Switching. Journal of Consumer Research
2001;28; 135—148.

Tonsor, G.T., T.C. Schroeder, J .M.E. Pennings, and J. Mintert. (2009). “Consumer
Valuation of Beef Steak Food Safety and Quality Assurances in Canada, Japan, Mexico
and the United States.” Canadian Journal of Agricultural Economics. 57:395-416.

Tonsor, G.T., N. Olynk, and C. Wolf. “Consumer Preferences for Animal Welfare
Attributes: The Case of Gestation Crates.” Journal of Agricultural and Applied
Economics. Forthcoming.

Toyoda, T., Ohtani, K., 1989. A switching regression model with different change-points
for individual coefﬁcients and its application to the energy demand equations for Japan.
Emperical Economics 14, 93--103.

Trostle R. Global Agricultural Supply and Demand: Factors Contributing to the Recent
Increase in Food Commodity Prices.” Outlook Report No. WRS-080l, Economic
Research Service, US. Department of Agriculture, May 2008. available at:
(http://www.ers.usda.gov/Publications/WRSO801/WRSO801 .pdf. cited on 04/7/09.

Tsurumi, H., 1980. A Bayesian estimation of structural shifts by gradual switching
regressions with an application to the US. gasoline market, in: Zellner A (Ed.), Bayesian

analysis in econometrics and statistics, North-Holland, Amsterdam.

Tsurumi, H., 1982. A Bayesian and maximum likelihood analysis of a gradual switching
regression in a simultaneous equation framework. Journal of Econometrics 19, 165--182.

144

 

Tsurumi, H., Wago, H., 1986. Gradual switching multivariate regression models with
stochastic cross-equational constraints and an application to the KLEM translog
production model. Journal of Econometrics 31, 23 5--253.

US. Census Bureau. 2000. available at
http://factﬁnder.censusipv/home/saff/main.html? lang=en. cited on 12/17/2008

United States Department of Agriculture (USDA) (2009), url:
http://www.usda.gov/wps/portal/usdahome. Cited 05/27/2009.

US. Department of Labor Bureau of Labor Statistics, 2008. http://www.bls.gov/. Cited
21 May 2008.
Value Line, 2008. htmz//www.valueline.com/. Cited 2 May 2008.

Varangis, Panos, Thigpen, Elton And Satyanarayan, Sudhalza, “The Use Of New York
Cotton Futures Contract To Hedge Cotton Price Risk In Developing Countries”, Policy
Research Working Paper 1328, The World Bank, International Economics Department,
International Trade Division, July 1994.

Wharton Research Data Services, url: http://wrds.wharton.upenn.edu/, viewed on
06/24/2009.

 

Yacobucci BD. Fuel Ethanol: Background and Public Policy Issues. Congressional
Research Service Issue Brief, RL33290; 2006; available at
http://ncseonline.org/NLE/CRSreports/O6Apr/RL3329049df. cited on 02/16/09.

Yacobucci BD. Biofuels Incentives: A Summary of Federal Programs. CRS Report for
Congress RL33572; 2008. available at
http://vienna.usembasg.ng/en/download/pdf/biofuels.pdf; cited on 02/16/09.

145

,jlltlllulllll

llllll

S"
u
H
"